file
stringlengths 27
28
| text
stringlengths 1
5.17M
|
|---|---|
PMC009xxxxxx/PMC9390333.txt
|
==== Front
Bioconjug Chem
Bioconjug Chem
bc
bcches
Bioconjugate Chemistry
1043-1802
1520-4812
American Chemical Society
35833631
10.1021/acs.bioconjchem.2c00174
Article
Radiation Cleaved Drug-Conjugate Linkers Enable Local Payload Release
Quintana Jeremy M. †
https://orcid.org/0000-0002-1740-551X
Arboleda David †
Hu Huiyu †‡-
Scott Ella †
Luthria Gaurav †
Pai Sara †‡
Parangi Sareh ‡
https://orcid.org/0000-0003-0828-4143
Weissleder Ralph †§∥
https://orcid.org/0000-0001-7638-8898
Miller Miles A. *†§
† Center for Systems Biology, Massachusetts General Hospital Research Institute, Boston, Massachusetts 02114, United States
‡ Department of Surgery, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts 02114, United States
§ Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, Massachusetts 02114, United States
∥ Department of Systems Biology, Harvard Medical School, Boston, Massachusetts 02115, United States
* [email protected].
14 07 2022
17 08 2022
14 07 2023
33 8 14741484
© 2022 The Authors. Published by American Chemical Society
2022
The Authors
https://pubs.acs.org/page/policy/termsofuse.html Made available for a limited time for personal research and study only License.
Conjugation of therapeutic payloads to biologics including antibodies and albumin can enhance the selectively of drug delivery to solid tumors. However, achieving activity in tumors while avoiding healthy tissues remains a challenge, and payload activity in off-target tissues can cause toxicity for many such drug-conjugates. Here, we address this issue by presenting a drug–conjugate linker strategy that releases an active therapeutic payload upon exposure to ionizing radiation. Localized X-ray irradiation at clinically relevant doses (8 Gy) yields 50% drug (doxorubicin or monomethyl auristatin E, MMAE) release under hypoxic conditions that are traditionally associated with radiotherapy resistance. As proof-of-principle, we apply the approach to antibody– and albumin–drug conjugates and achieve >2000-fold enhanced MMAE cytotoxicity upon irradiation. Overall, this work establishes ionizing radiation as a strategy for spatially localized cancer drug delivery.
National Institutes of Health 10.13039/100000002 DP2CA259675 China Scholarship Council 10.13039/501100004543 201906370164 National Natural Science Foundation of China 10.13039/501100001809 81974423 National Cancer Institute 10.13039/100000054 U01CA206997 National Cancer Institute 10.13039/100000054 T32CA079443 document-id-old-9bc2c00174
document-id-new-14bc2c00174
ccc-price
==== Body
pmcIntroduction
Numerous approaches have been developed to improve the therapeutic index of potent small molecule compounds by linking them as inactive prodrugs to biologics or nanoparticles.1,2 For the treatment of metastatic cancers, cytotoxic agents have been bound to serum albumin3 and nanoparticles4 to improve systemic pharmacokinetics and, in principle, to promote tumor accumulation via molecularly targeting and/or “enhanced permeability and retention” (EPR) mechanisms of uptake. These mechanisms include oncogene-driven macropinocytosis,5,6 permeable tumor vasculature,7,8 dysfunctional tumor lymphatics, and other features that contribute to what is collectively referred to as the EPR effect.9−11 Alternatively, molecularly targeted strategies based on antibodies12,13 or functionalized nanoparticles2,14,15 have been designed to bind surface receptors selectively overexpressed by cancer cells. Multiple antibody–drug conjugates (ADCs) and therapeutic nanoparticles have received FDA-approval for clinical use. Yet despite their successes, these agents still accumulate in off-target tissues and elicit systemic toxicities.16−18 Drugs receive black box warnings when they exhibit potentially serious and deadly adverse effects, and toxicities affecting the bone marrow, liver, and other organs have led to boxed warnings on the FDA package inserts of nearly all drug-conjugates in oncology.
How can off-target payload activity be minimized for drug-conjugates? Optimization of the delivery vehicle—the biologic or nanoparticle to which the payload is conjugated—represents one set of strategies. Vehicle accumulation in clearance organs and the mononuclear phagocyte system can be minimized through PEGylation or FcRn engineering, for instance,19,20 but is nonetheless difficult to completely eliminate.21 Choosing appropriate tumor-specific molecular targets can improve selective accumulation, but one would need to first identify a tumor-specific target, and often only a subset of tumors or tumor cells may preferentially express such targets.22,23 Optimizing the chemistry by which a drug payload attaches to its delivery vehicle represents another set of strategies.24−26 Premature or off-target payload-release contributes to nonspecific and/or systemic exposure,27 while noncleavable linkers may insufficiently yield fully active payloads in tumors.28 Unfortunately, most current strategies for controlling payload release rely on pH, lysosomal degradation, protease activity, and other processes that are not reliably unique to tumor cells.29
Bio-orthogonal approaches aim to overcome limitations of biological specificity by modulating drug activity in a manner that is independent of naturally occurring chemical processes. Previously studied strategies for triggered activation of cancer prodrugs have used electrochemical,30 ultrasonic,31 optical,32,33 copper-free click chemical,34,35 and transition-metal catalytic36,37 activation schemes. It has generally been a challenge to achieve highly localized control of drug activation deep through tissue in a noninvasive manner. Methods such as heat, ultrasound, and laser irradiation may not easily penetrate deep through tissue, and bio-orthogonal chemical triggers with increased selectivity and compatibility are still in development.
In contrast to the above, ionizing radiation offers an attractive solution: radiation is routinely delivered deep through tissue via focused beams of gamma, proton, and X-ray radiation. Recent reports have described caged prodrugs that become activated with ionizing radiation, but initial studies relied on small molecule modifications with either suboptimal caging of drug activity or do not support strategies to more efficiently deliver compound to tumors, for instance, as achievable by drug conjugation to antibodies, albumin, or other delivery vehicles.38,39 To overcome these limitations, here we present a modular prodrug design containing a self-immolative linker with a cytotoxic payload, anchor for protein conjugation, and a radiation-activated trigger. A 3,5-dimethylbenzyl alcohol (DMBA) caging moiety undergoes 1–4/1–6 elimination upon radical hydroxylation that is initiated via clinically relevant doses of X-ray irradiation and, under hypoxic conditions, results in >50% localized release of the active drug and a >2000-fold increase in drug-induced cytotoxicity. The generic linker design offers flexibility in both targeting vehicles and their payload and promises to be broadly applicable to therapeutic applications that may synergize with standard-of-care radiation in the treatment of hypoxic tumors.
Results and Discussion
We designed the radiation-cleavable linker based on a tripartite structure containing three components—a drug payload, a radiation-responsive DMBA caging moiety, and a reactive anchor for conjugation to the delivery vehicle. Serum albumin (Alb) was used as an initial model vehicle, since cancer cells can accumulate high levels via oncogene-driven macropinocytosis,1,5,6 multiple albumin-based drug formulations are clinically used in the treatment of solid cancers,40,41 but off-target accumulation contributes to dose-limiting toxicity.3 A self-immolative linker (SIL) bridges the three components37 and was designed to release the drug payload upon radiation-induced reaction with DMBA, which prior work has shown to be mediated primarily by hydroxyl radical.38 We used monomethyl aurisatin E (MMAE) and doxorubicin (DOX) as model drug-conjugate payloads, since they are used in clinical-stage drug-conjugates but face known systemic toxicities. Combining these components together yielded Alb-DMBA-SIL-MMAE and Alb-DMBA-SIL-DOX as radiation-activated albumin–drug conjugates.
The overall synthesis of the Alb-DMBA-SIL-MMAE or Alb-DMBA-SIL-DOX was performed in 5 steps (Scheme 1). Self-immolative linker 3 was synthesized as previously described,42 and installation of the X-ray activated trigger was achieved by first reacting benzyl alcohol 1 with triphosgene to produce the chloroformate 2, which was then reacted directly with aniline 3 to yield DMBA-SIL 4. The benzylic alcohol on this molecule was activated by reaction with bis(4-nitrophenyl) carbonate under basic conditions with diisopropyl ethyl amine to generate carbonate 5. This carbonate could then be reacted with either MMAE or DOX to yield compounds 6 and 8, respectively.
Scheme 1 Synthesis of 3,5-Dimethyloxybenzyl Alcohol (DMBA) Prodrugs with Self-Immolative Linker (SIL) and Maleimide Anchor
Prodrugs using monomethyl auristatin E (MMAE) and doxorubicin (DOX) payloads used shared DMBA-SIL structures and precursors in the synthesis.
Maleimide was chosen as the reactive group to conjugate the drug payload to its delivery vehicle, since it is a widely used robust strategy shown to relatively selectively react with free thiol at Cysteine-34 on Alb, thereby forming a homogeneous product modified at a single site.43 It is also used for antibody conjugations, including in clinical products.44 Maleimide was installed via hydrolysis followed by amide coupling to obtain products 7 and 9. The maleimide-containing prodrugs were then conjugated to Alb by incubation at room temperature in PBS (pH 7.4, Gibco) for two hours, followed by spin filtration (30 kDa MWCO) to remove unreacted prodrug and further purification by size exclusion chromatography.
Initial experiments were performed to assess linker cleavage and payload release upon exposure to ionizing radiation. We hypothesized that the amount of dissolved oxygen in the solution would impact reaction rates, in part due to production of reactive oxygen species (ROS) other than the desired hydroxyl radicals. Although prior literature has hinted at oxygen dependent reaction,38,39 there is little data on actual effects. This is relevant for in vivo drug action, since advanced tumors are often hypoxic, and hypoxia is associated with resistance to radiation therapy in patients.45 Therefore, we tested the impact of dissolved oxygen on linker cleavage by deoxygenating solutions of DMBA-SIL-MMAE under vacuum or by bubbling the solution with an inert gas (argon) immediately prior to irradiation. Using previously reported procedures,46 we estimated that dissolved oxygen partial pressure decreased from 0.21 atm under ambient air (21% oxygen) to 0.025 atm with vacuum and to 0.005 atm with inert gas (Figure 1), as determined by the Winkler titration method. As a reference, dissolved oxygen is proportionate to its partial pressure (pO2) according to Henry’s law; tumor pO2 values can be >100-fold lower than ambient air pO2, and some tumor cells and xenografts can survive <0.1% oxygen.47 Thus, 0.005 atm oxygen partial pressure achieved by inert gas bubbling (a 40-fold decrease over ambient conditions) is likely relevant to hypoxic tumor tissue.
Figure 1 Drug release initiation via radical hydroxylation under hypoxia. (A) Radiation-induced radical hydroxylation, followed by 1,6-elimination and subsequent loss of the SIL caging group, releases the caged drug payload. (B) Measured MMAE release from DMBA-SIL-MMAE following 8 Gy irradiation with either an X-ray or gamma ray source, as a function of oxygen partial pressure. Data are means ± s.e. (n = 3). Pearson’s correlation coefficient and two-tailed t test are reported.
Results showed a correlation between oxygen levels and the efficiency of linker cleavage (R2 > 0.97, p = 0.02) at a radiation dose of 8 Gy, which is relevant to clinical radiation protocols, especially those using hypofractionated treatment schedules.48,49 No difference in cleavage efficiency was found between gamma irradiation from a sealed 137Cs source (primarily 662 keV photons at a dose rate of roughly 0.5 Gy/min) and X-ray irradiation (from a 4000 W tube generating 320 keV photons at 3 Gy/min) (Figure 1b), and little background release was observed under non-irradiated conditions (5 days, 37 °C; Figure S1). Neither MMAE nor doxorubicin themselves were substantially affected by 8 Gy irradiation, according to LC/MS chromatograph analysis (Figure S2). These data thus show that radiation can trigger drug-conjugate cleavage in a hypoxia-dependent manner without adversely degrading the drug payload.
Prior work has implicated hydroxyl radicals as the primary ROS formed from ionizing radiation that activates the DMBA trigger.38 The mechanisms for the release of doxorubicin from these prodrugs (Figure 2a) consist of a radical hydroxylation (either position 2 or 4) of the dimethoxybenzyl moiety, followed by a 1–4/1–6 elimination and loss of carbon dioxide. The self-immolative linker then releases via a further 1–6 elimination and subsequent loss of carbon dioxide to provide the free drug.37 This mechanism is supported by the observed masses of the expected intermediates in our X-ray irradiated prodrug, shortly after irradiation (Figure S3a,b). Aminophenyl fluorescein (APF) is often used as an indicator of highly reactive oxygen species including hydroxyl radicals,50 and both hydrogen peroxide and X-ray irradiation decage APF to enhance its fluorescence (Figure S4). Standard Fenton reaction conditions (100 μM H2O2 and 100 μM Fe2+) are also known to generate hydroxyl radicals and therefore can serve as a benchmark comparison to ionizing radiation effects. X-ray irradiation under the most hypoxic conditions was more efficient than Fenton conditions at cleaving the DMBA-based linker (Figure S3c), and these irradiation conditions were used in all subsequent in vitro experiments. The observed dependence on hypoxia may be considered unexpected due to the known role of oxygen in promoting ROS generation. One possible explanation may be that the absence of oxygen enhances the desired hydroxylation mechanism by reducing the quantity of alternative reactive oxygen species being formed, although further exploration of this phenomenon is needed to test this hypothesis and dissect underlying mechanisms.
Figure 2 Dose–response of payload release from drug–albumin conjugates. (A) Percentage of MMAE and DOX released from albumin-conjugates after treatment with varying doses of X-ray irradiation compared to free drug (10 μM). Data are means ± s.e. (n = 3). (B,C) Representative LC-MS chromatographs of MMAE (b; +ESI, isolated mass of 718.8 Da) and DOX (c; +ESI, isolated mass of 544.4) released from albumin.
We next examined the sensitivity of linker cleavage to varying doses of radiation. Solutions of both conjugates at a concentration of 100 μM were irradiated at an exposure of 1–16 Gy, and LC-MS analysis was performed to determine the released drug concentrations (Figure 2). Alb-DMBA-SIL-MMAE and Alb-DMBA-SIL-DOX exhibited similar dose–response behaviors, suggesting the approach generalizes to diverse drug payloads. With a standard protocol used in clinical treatments, patients typically receive ∼2 Gy radiation per day, 5 days a week, for ≥5 weeks, therefore yielding cumulative radiation doses of ≥50 Gy. Our data show that even low doses of 1–2 Gy elicit detectable drug release, and higher doses of 8–16 Gy, which are less common but still used clinically in hypofractionated schedules, release 66 ± 6% and 69 ± 8% of MMAE and DOX, respectively. These results thus show release under a range of clinically relevant doses.
We next performed experiments with cancer cells to assess the degree to which our linker strategy could shield biological activity of the drug payload under non-irradiated conditions and restore drug activity once irradiated. We focused on cancer cell lines with constitutively active oncogenic signaling in the mitogen activated protein kinase (MAPK) RAS/RAF/MEK/ERK pathway, which has been previously shown as important for oncogene-driven uptake of serum albumin;5 we furthermore focused on cancer types that are treated with cytotoxic agents and radiation therapy, including anaplastic thyroid cancer (ATC), a rare but aggressive malignancy that is associated with high degrees of resistance to traditional chemotherapy and radiation therapy. Drugs were irradiated prior to cancer cell treatment to allow assessment of drug action (caging and uncaging) independent of the biological effects of radiation to the cancer cells, which even at <10 Gy doses in cells considered radioresistant, such as human 8505c ATC cells, can substantially impact proliferation in vitro.51 Compared to the parent MMAE drug, the drug-conjugate Alb-DMBA-SIL-MMAE exhibited 5700-fold lower cytotoxicity in 8505c cells (Figure 3). Anchoring the DMBA-SIL-MMAE to its protein vehicle, Alb, enhanced the caging of drug activity: Alb-DMBA-SIL-MMAE exhibited ∼40-fold lower cytotoxicity than the unanchored compounds DMBA-SIL-MMAE and MMAE-DMBA (Figure 3). In contrast, all three prodrugs exhibited similar cytotoxicity following 8 Gy irradiation, consistent with the behavior of the parent drug and the known efficiency of drug release at this dose (52 ± 9%). Compared to prior DMBA caging approaches,38 these data show that our linker strategy exhibits a superior ability to limit drug activity under non-irradiated conditions, which is likely important for minimizing systemic toxicity. Furthermore, cytotoxicity of the parent drug is restored to a level commensurate with the fraction of released drug as measured by LC/MS, indicating that irradiation appropriately releases the payload in a fully active and intact form. Similar observations were found in cells derived from a genetically engineered mouse model of ATC (the TBP-3743 cell line) and other cancer cell lines (Figure 3c, Figure S5).
Figure 3 Caged and conjugated MMAE is selectively cytotoxic and activated by X-ray irradiation. (A) Chemical structures of Alb-DMBA-SIL-MMAE and other radiation-activated derivatives. (B) Cytotoxicity of non-irradiated and irradiated (8 Gy) MMAE and prodrug derivatives in anaplastic thyroid cancer cells (8505c), measured 72 h post-treatment by a resazurin-based assay (data are means ± s.e., n = 4). (C) Half-maximal inhibitory concentration (IC50) of non-irradiated or irradiated Alb-DMBA-SIL-MMAE in cancer cell lines of anaplastic thyroid cancer (TBP), oral squamous cell carcinoma (MOC-2), colon adenocarcinoma (MC38), and pancreatic adenocarcinoma (iKRAS). See Figure S2 for full dose responses.
Given that cytotoxicity measurements suggested parent MMAE activity was being restored following irradiation, we subsequently assessed whether the molecular effects of MMAE were similarly affected. MMAE blocks the polymerization of tubulin into microtubules, which are critical cytoskeletal components that mediate mitotic cell division and metastatic invasion of cancer cells.52,53 To directly visualize microtubule dynamics in live cancer cells, we used the HT1080 EB3-mApple cell line, which transgenically expresses the fluorescent protein mApple fused to the protein EB3 (microtubule-associated protein RP/EB family member 3, MAPRE3). EB3 binds plus-end tips of growing microtubules, and time-lapse microscopy allows growing microtubules to be quantified for their abundance, growth velocities, and other features.37 Using this approach, non-irradiated Alb-DMBA-SIL-MMAE elicited no significant impacts on cancer cell microtubule dynamics; in contrast, dynamics were totally eliminated with irradiated drug (Figure 4). These results confirm that the linker strategy efficiently cages drug activity and show that the activity of MMAE to disrupt microtubule dynamics is restored upon radiation-mediated drug release.
Figure 4 Radiation restores the microtubule-disrupting activity of caged MMAE. (A) Reporter cell line for tracking +TIP (microtubule plus-end tracking protein) was imaged over time via confocal microscopy to visualize microtubule dynamics. After drug treatment, microtubule “comets” were automatically detected, computationally tracked, and visualized with pseudocoloring according to comet speed. (B) Corresponding to representative data in part A, features of microtubule dynamics were averaged across individual cells (n > 10 per condition). Prodrug = Alb-DMBA-SIL-MMAE, n.d. = none (no comets) detected.
Similar experiments were performed to test the caging and activation efficiencies of Alb-DMBA-SIL-DOX, compared to the parent compound doxorubicin (Figure 5). The concentration at which drug inhibited 50% of ATC cell growth was >10 μM for non-irradiated Alb-DMBA-SIL-DOX and DMBA-SIL-DOX. This was roughly 100-fold higher than that observed for the parent drug (IC50 = 94 nM). Irradiation restored prodrug cytotoxicity to a level expected based on the fraction of drug released (0.5 ± 0.1). Similar results were observed in other cancer cell lines (Figure 5c, Figure S6). Compared to MMAE, the DMBA-SIL-DOX did not benefit further from conjugation to serum albumin, in terms of limiting the activity of non-irradiated compound. This is potentially due to the distinct DOX mechanism of action compared to MMAE (Figure 5b).
Figure 5 Caged and conjugated DOX is selectively cytotoxic and activated by X-ray irradiation. (A) Structures of DOX prodrug derivatives. (B) Cytotoxicity of each prodrug/conjugate ±8 Gy X-ray irradiation in comparison to free DOX in anaplastic thyroid cancer cells (8505c). Data are means ± s.e., n = 4. (C) Half-maximal inhibitory concentration (IC50) of non-irradiated or irradiated Alb-DMBA-SIL-DOX in cancer cell lines of anaplastic thyroid cancer (TBP), oral squamous cell carcinoma (MOC-2), and colon adenocarcinoma (MC38). See Figure S3 for full dose responses.
Doxorubicin and prodrug derivatives exhibit intrinsic fluorescence that is visible by confocal microscopy, and imaging can therefore be used to assess drug accumulation and colocalization with its target in the nuclei of live cancer cells.54 Doxorubicin is an anthracycline that intercalates DNA, inhibits topoisomerase II, and therefore generates DNA damage leading to cell death. The 3′-amino of the daunosamine moiety on doxorubicin forms a covalent bond with the exocyclic amino of guanine, and this site is frequently modified to cage drug activity,37 as done here as well. ATC cells show nuclear accumulation of doxorubicin and irradiated Alb-DMBA-SIL-DOX (Figure 6, Figure S7). In contrast, non-irradiated prodrug is confined to the cytoplasm, therefore suggesting the non-irradiated drug-conjugate remains intact, such that doxorubicin is unable to freely enter the nucleus and interact with DNA. Overall, these results indicate that the radiation-cleavable linker blocks the ability of doxorubicin to intercalate DNA and elicit cytotoxic effects, and that drug irradiation can appropriately release it to facilitate nuclear localization and cytotoxicity.
Figure 6 Intrinsic doxorubicin fluorescence quantifies nuclear uptake following X-ray mediated payload release. (A) Representative images of intrinsic DOX fluorescence in 8505c cells treated with DOX, non-irradiated Alb-DMBA-SIL-DOX, or X-ray-irradiated (8 Gy) conjugate for 24 h; Hoechst 33342 counterstains cell nuclei. (B) Ratio of nuclear to cytoplasmic DOX fluorescence was quantified to evaluate the subcellular drug accumulation. Data are means ± s.e., n > 15 single cells per condition (one-way ANOVA with Dunnett’s T3 multiple comparisons test).
To examine the generalizability of the linker strategy beyond albumin-conjugates, we next tested whether the approach could also yield an antibody–drug conjugate (ADC) with a radiation-cleavable cytotoxic payload. We prepared conjugates of the DMBA-SIL-DOX/MMAE prodrugs to a model tumor-targeted monoclonal antibody that binds epidermal growth factor receptor (αEGFR mAb), since multiple αEGFR-mAb are used clinically including cetuximab and panitumumab for a variety of solid tumors, and αEGFR-ADC are under development. We first reduced antibody disulfiide bonds with tris(2-carboxyethyl)phosphine hydrogen chloride (TCEP-HCl), followed by conjugation with the respective prodrugs via thiol–maleimide Michael addition. The resulting mAb-DMBA-SIL-DOX and mAb-DMBA-SIL-MMAE were then subjected to the same in vitro analyses as the albumin conjugates to demonstrate X-ray activation and determine drug release efficiencies. These experiments showed that antibody-conjugates performed similarly to the previous albumin-conjugates, releasing with radiation 64 ± 7% and 56 ± 4% of the estimated MMAE and DOX, respectively (Figure 7a). In a cytotoxicity assay using the 8505c ATC cell line, the mAb-DMBA-SIL-MMAE conjugate demonstrated a 70-fold increase in cytotoxicity after X-ray irradiation (8 Gy) compared to the nonirradiated conjugate (Figure 7b).
Figure 7 Radiation-mediated release of antibody–drug conjugate (ADC) payloads. (A) Chemical structures of ADCs, here using anti-EGFR mAb as a model tumor-targeted antibody (shown as purple). (B) MMAE and DOX payload release from their respective antibody conjugates after X-ray irradiation (8 Gy), measured by LC-MS. Data are means ± s.e., n = 3. (C) Cytotoxicity of non-irradiated or X-ray-irradiated (8 Gy) mAb-DMBA-SIL-MMAE conjugate in comparison to free MMAE in anaplastic thyroid cancer (8505c cell line). (D) Ratio of nuclear to cytoplasmic DOX fluorescence was quantified as in Figure 6 in anaplastic thyroid cancer cells (8505c) treated for 24 h with DOX, non-irradiated conjugate, or irradiated (8 Gy) conjugate. Prodrug = mAb-DMBA-SIL-DOX. Data are means ± s.e., n > 16 single-cells per condition (one-way ANOVA with Dunnett’s T3 multiple comparisons test).
Similarly, irradiated mAb-DMBA-SIL-Dox demonstrated an increased nuclear-to-cytoplasm ratio in the subcellular distribution of the doxorubicin payload relative to its non-irradiated counterpart (Figure 7c). Taken together, these data show that mAb conjugation is effective in blocking payload activity and that radiation exposure releases payload from the mAb-conjugate and unleashes payload activity in the cancer cells.
Conclusions
This work extends radiation-activated chemistry to support triggered release of biologic drug-conjugates. As with traditional ADC linkers, the approach can be applied to diverse delivery vehicles such as serum albumin or antibodies as we demonstrate here; it furthermore can be applied to diverse drug payloads, including the widely used cytotoxic chemotherapies doxorubicin and MMAE. We show that the linker cleavage can be triggered by multiple forms of ionizing radiation and is accelerated under hypoxic conditions thought to stimulate macropinocytosis of albumin-bound agents.55 Radiation itself has been shown to improve drug penetration into tumors8,37,56 and therefore offers the possibility of self-amplifying the local payload activity in this case. This work sets the stage for future work to investigate mechanisms of in vivo drug activation and potential off-target payload release, which will require (i) careful analysis of chemical and biological drug behaviors in tumors and off-target tissue, with and without radiation; (ii) comparison with free (nonconjugated) drugs and prodrugs, as well as noncleavable drug-conjugates; and (iii) assessment of drug behavior in orthotopic, patient-derived, and/or autochthonous tumor models that recapitulate the hypoxic tumor microenvironments often found in patients.57 This work also establishes a foundation from which to assess drug-conjugate payloads, beyond MMAE and doxorubicin, that may synergistically combine with the biological effects of radiation. Although the therapeutic radiation doses used here are known to elicit strong impacts on cancer cell behavior in vitro—for instance, by reducing proliferation of the cancer cells used in this work (8505c, MC38, HT1080) by >95% in clonogenic assays—such doses nonetheless fail to eradicate tumors and block disease progression.8,51,58 To this end, the modular design strategy lends itself to further optimization of the X-ray activated trigger, the conjugation anchor, and the drug payload.
Our presented method of prodrug engineering may be applied to a variety of therapeutics including immunomodulatory agents and targeted inhibitors that may be chosen to synergistically combine with radiation administered as part of the standard of care. Extensive research has already gone into understanding how DNA- and microtubule-targeted drugs may best combine with radiation therapy, with special focus on optimizing cancer cells to be in the radiosensitive G2/M phase of their cell cycle and on tumor microenvironment effects that maximize oxygenation. In contrast, our linker approach offers a distinct way of considering synergistic effects under conditions of hypoxia, and local drug delivery may allow higher local drug concentrations or more potent drug payloads to be considered.
Experimental Procedures
Synthetic Procedures
Detailed synthetic protocols and characterization data can be found in the Supporting Information.
Conjugate Preparation
To a solution of bovine serum albumin (BSA, 47.1 mg in 1 mL PBS) was added 100 μL of a 10 mg/mL solution of Mal-containing prodrug (9 or 11) in DMF. These mixtures were then incubated at room temperature with gentle mixing for 3 h, followed by removal of the unreacted prodrugs by spin filtration (20,000g for 5 min, × 3) using 30,000 kDa MWCO spin filters (Amicon, 0.5 mL). The conjugates were further purified by size-exclusion chromatography in PBS using a SEC column on an Agilent 1260 HPLC. Collected fractions eluting between 8.5 and 10.5 min were concentrated by further spin filtration to provide a final conjugate concentration of 10 mg/mL in PBS. Alb is generally well tolerated in patients, given at high doses, but can undergo accelerated clearance if extensively modified; therefore, we performed conjugation with a low degree of labeling, achieving on average 0.12 drug molecules per Alb molecule (Table S1).
Conjugates of the DMBA-SIL-DOX/MMAE prodrugs to an anti-epidermal growth factor receptor (EGFR) antibody (BioXcell, mab225) were prepared by first reducing the disulfiide bonds in the antibody by incubating in 100 μM tris(2-carboxyethyl)phosphine hydrogen chloride (TCEP-HCl) for one hour at room temperature, followed by removal of the reducing agent by spin filtration (10,000g, 50,000 kDa MWCO, Amicon/Sigma). While leaving a small fraction of the TCEP-HCl (∼50 μL) to maintain the reduced state of the antibody, the prodrugs were added to separate aliquots and incubated at room temperature for 2 h, after which unreacted prodrug was removed by further spin filtration (10,000g × 3, 50,000 kDa MWCO). The degree of labeling was estimated to be on average 5.6 drug molecules per antibody calculated following manufacturer guidelines and using a NanoDrop spectrophotometer (Table S1).
Drug Release
Determination of drug release efficiency was performed by preparing solutions of each prodrug in deionized water (0.1% DMF) to a final concentration of 10 μM. Vacuum degassing was performed by gently stirring the samples at ∼200 mbar for 30 min at room temperature, while degassing by argon was performed by bubbling ultrahigh-purity-grade gas (Airgas, Lynn, MA, 01902, USA) into the solution at a rate of approximately 5 mL s–1. After these preparations, X-ray irradiation was performed using a Precision (Madison, CT, USA) X-Rad320 at a rate of 385 ± 10 cGy/min until the desired dosage was achieved. Gamma irradiation was similarly performed on a dual source 137Cs Gammacell 40 Exactor (Best Theratronics) with a dose rate of roughly 50 cGy/min. After irradiation or other treatment, the concentrations of the released drug and the intact prodrug were determined by LC-MS on a Waters instrument equipped with a Waters 2424 ELS Detector, Waters 2998 UV–vis Diode Array Detector, and a Waters 3100 Mass Detector. Drug concentrations were determined by comparing the AUC from the ELSD or isolated mass chromatographs (+ESI, 718.8 Da for MMAE, 544.4 Da for DOX) of the analyzed sample to a standard calibration curve.
For the measurement of released products relative to the X-ray dosage, DMBA-SIL-MMAE protein conjugate was prepared to a concentration of 50 μM in PBS (0.1% DMF, 5 mL), then purged with ultrapure-grade argon (5 mL s–1) for 15 min. The solution was then X-ray irradiated using the X-Rad320 to deliver 8 Gy irradiation and analyzed by LC-MS as described above.
Determination of the oxygen concentrations for each degassing condition was achieved through the use of the Winkler titration method46 and is reported as parts per million (ppm, mg O2 per kg dI H2O). Briefly, water samples were degassed by either vacuum or argon purging as described above, with untreated deionized water used as a control. To the samples were then added solutions of manganese sulfate monohydrate (100 μL, 2 mM) and alkaline potassium iodide (100 μL, 12.5 M NaOH, 0.8 M KI, 0.15 M NaN3). The solutions were mixed and allowed to stand for 15 min before concentrated sulfuric acid (100 μL) was added and again mixed. Addition of 100 μL of a starch solution (50 mg mL–1) turned the solution blue, and the mixture was then titrated with a solution of sodium thiosulfate (2.5 mM) until it reached a colorless end point. The volume of the thiosulfate solution added (in mL) was equivalent to the initial concentration of dissolved oxygen in ppm.
Cytotoxicity Assays
TBP-3743 (derived from a genetically engineered mouse model of anaplastic thyroid cancer) and 8505c (human anaplastic thyroid cancer) are described previously,59 as are HT1080-EB3-mApple cells.52 Raw264.7 (mouse macrophage model, Raw-MΦ) were from ATCC. MC38 cells provided by M. Smyth (Peter MacCallum Cancer Centre, Victoria, Australia). iKras cells were derived from a genetically engineered mouse model of KrasG12D pancreatic adenocarcinoma and were routinely cultured in Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 media (DMEM/F12, Invitrogen) supplemented with 2 μg mL–1 doxycycline (Sigma) to maintain mutant Kras expression (provided as a gift from H. Ying, MD Anderson Cancer Center, by way of N. Bardeesy, Massachusetts General Hospital (MGH)).60 MOC2 (mouse oral cavity squamous cell cancer – aggressive growth phenotype) (RRID:CVCL_ZD33) cell lines from Kerafast and were cultured according to provider guidelines using IMDM/F12, supplemented with 5 ng/mL EGF (EMD Millipore), 400 ng/mL hydrocortisone (Sigma-Aldrich), and 5 mg/mL insulin (Sigma-Aldrich). All cells were routinely evaluated for mycoplasma contamination and cultured following provider guidelines using 10% FBS (Bio-Techne Sales), 100 IU mL–1 penicillin, 100 μg mL–1 streptomycin (Invitrogen), with incubation at 37 °C and 5% CO2. Cytotoxicity experiments were performed by seeding 5000 cells per well overnight in a 96-well plate (Corning) before the addition of each drug/conjugate. Empty wells with only media or vehicle-treated cells were used as controls. After addition of the corresponding drugs and a 72 h incubation, the number of live cells was determined by PrestoBlue (ThermoFisher, USA) staining according to the provider’s protocols.
Cell Imaging
Subcellular localization of DOX in the 8505c cell line was measured using a modified BX63 (Olympus) inverted microscopy system equipped with an environmental chamber and robotic stage. DOX images were collected using a 40× air objective (PLAPO 40×/0.95 numerical aperture), with excitation and emission wavelengths of 489 and 508 nm, respectively. All images were processed using cellSens Dimension 3.1.1 (Olympus, USA) and ImageJ 1.53k (NIH, USA) software. Nuclear to cytoplasm fluorescence ratios were calculated and plotted using Excel (Microsoft) and Prism (GraphPad). 20 μM concentration of free (or released) DOX was used, and non-irradiated prodrug used the same total (caged and uncaged) DOX concentration as in the irradiated sample. Following treatment for 24 h, live cells were immediately imaged.
EB3 imaging was performed on an FV1000 confocal laser scanning microscope equipped with a 37 °C heated stage, XLUMPLFLN 20× (NA 1.0) water-immersion objective, 559 nm diode laser, and BA575-620 emission filter (all Olympus America). Cells were treated with 1 μM MMAE, Alb-DMBA-SIL-MMAE, or Alb-DMBA-SIL-MMAE after argon purging and X-ray irradiation (8 Gy) 1 h prior to imaging. MT tracks were obtained by detecting and linking EB3 comets using the U-track software.61 Cell masks were constructed using ImageJ. For MT tracks to be included in the downstream analyses, they must pass a strict set of filters: (1) located within a cell boundary, (2) be present in a minimum of 3 consecutive frames, (3) have a path length less than 10 μm, (4) track persistence greater than 0.5 (measured on a scale of 0 to 1, with 1 indicating a line). Additionally, tracks with outlier speeds (greater than or less than 1.5 *IQR) within their respective cells were not included in the analysis.
Supporting Information Available
The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.bioconjchem.2c00174.Figure S1, MMAE release from DMBA-SIL-Mal after degassing and irradiation; Figure S2, Drug stability after irradiation; Figure S3, Release intermediates after irradiation; Figure S4, Release via reactive oxygen species; Figure S5, MMAE conjugate cytotoxicity; Figure S6, DOX conjugate cytotoxicity; Figure S7, Subcellular DOX conjugate distribution; Figure S8, LC chromatographs of DMBA prodrugs; Table S1, Conjugate degree of labeling; Organic Synthesis and Characterization (PDF)
Supplementary Material
bc2c00174_si_001.pdf
The authors declare the following competing financial interest(s): RW is a consultant to ModeRNA, Tarveda Pharmaceuticals, Lumicell, Seer, Earli, Aikili Biosystems and Accure Health. MAM is a scientific advisor for January Therapeutics and has received past support from Ionis Pharmaceuticals, Pfizer, and via material transfer agreement from Genentech/Roche. None of these activities are related to the manuscript. The other authors declare that they have no competing interests.
Acknowledgments
This work was supported in part by NIH grants T32CA079443 (J.Q.), DP2CA259675 (M.A.M.), and U01CA206997 (R.W.). H.H. was supported by the China Scholarship Council 201906370164, and the National Natural Science Foundation of China 81974423 and is affiliated with the Department of General Surgery, Xiangya Hospital, Central South University, China. We acknowledge Rainer H. Kohler, Ph.D. (Massachusetts General Hospital) for expertise and assistance in microscopy experiments. We also acknowledge the Harvard Center for Mass Spectrometry (Harvard University) for conducting high-resolution mass spectrometry measurements of selected prodrugs.
==== Refs
References
Liu H. ; Qian F. Exploiting macropinocytosis for drug delivery into KRAS mutant cancer. Theranostics 2022, 12 , 1321–1332. 10.7150/thno.67889.35154489
Liu L. ; Kshirsagar P. G. ; Gautam S. K. ; Gulati M. ; Wafa E. I. ; Christiansen J. C. ; White B. M. ; Mallapragada S. K. ; Wannemuehler M. J. ; Kumar S. ; et al. Nanocarriers for pancreatic cancer imaging, treatments, and immunotherapies. Theranostics 2022, 12 , 1030–1060. 10.7150/thno.64805.35154473
Hoogenboezem E. N. ; Duvall C. L. Harnessing albumin as a carrier for cancer therapies. Adv. Drug Delivery Rev. 2018, 130 , 73–89. 10.1016/j.addr.2018.07.011.
Mitchell M. J. ; Billingsley M. M. ; Haley R. M. ; Wechsler M. E. ; Peppas N. A. ; Langer R. Engineering precision nanoparticles for drug delivery. Nat. Rev. Drug Discovery 2021, 20 , 101–124. 10.1038/s41573-020-0090-8.33277608
Li R. ; Ng T. S. C. ; Wang S. J. ; Prytyskach M. ; Rodell C. B. ; Mikula H. ; Kohler R. H. ; Garlin M. A. ; Lauffenburger D. A. ; Parangi S. ; et al. Therapeutically reprogrammed nutrient signalling enhances nanoparticulate albumin bound drug uptake and efficacy in KRAS-mutant cancer. Nat. Nanotechnol. 2021, 16 , 830–839. 10.1038/s41565-021-00897-1.33958764
Commisso C. ; Davidson S. M. ; Soydaner-Azeloglu R. G. ; Parker S. J. ; Kamphorst J. J. ; Hackett S. ; Grabocka E. ; Nofal M. ; Drebin J. A. ; Thompson C. B. ; et al. Macropinocytosis of protein is an amino acid supply route in Ras-transformed cells. Nature 2013, 497 , 633–637. 10.1038/nature12138.23665962
Inoue Y. ; Matsumoto Y. ; Toh K. ; Miyano K. ; Cabral H. ; Igarashi K. ; Iwasaki S. ; Kataoka K. ; Yamasoba T. Manipulating dynamic tumor vessel permeability to enhance polymeric micelle accumulation. J. Controlled Release 2021, 329 , 63–75. 10.1016/j.jconrel.2020.11.063.
Miller M. A. ; Chandra R. ; Cuccarese M. F. ; Pfirschke C. ; Engblom C. ; Stapleton S. ; Adhikary U. ; Kohler R. H. ; Mohan J. F. ; Pittet M. J. ; et al. Radiation therapy primes tumors for nanotherapeutic delivery via macrophage-mediated vascular bursts. Sci. Transl. Med. 2017, 9 , eaal0225 10.1126/scitranslmed.aal0225.28566423
Fang J. ; Islam W. ; Maeda H. Exploiting the dynamics of the EPR effect and strategies to improve the therapeutic effects of nanomedicines by using EPR effect enhancers. Adv. Drug Delivery Rev. 2020, 157 , 142–160. 10.1016/j.addr.2020.06.005.
Shi Y. ; van der Meel R. ; Chen X. ; Lammers T. The EPR effect and beyond: Strategies to improve tumor targeting and cancer nanomedicine treatment efficacy. Theranostics 2020, 10 , 7921–7924. 10.7150/thno.49577.32685029
Nia H. T. ; Munn L. L. ; Jain R. K. Physical traits of cancer. Science 2020, 370 , aaz0868 10.1126/science.aaz0868.
Chau C. H. ; Steeg P. S. ; Figg W. D. Antibody–drug conjugates for cancer. Lancet 2019, 394 , 793–804. 10.1016/S0140-6736(19)31774-X.31478503
Beck A. ; Goetsch L. ; Dumontet C. ; Corvaïa N. Strategies and challenges for the next generation of antibody-drug conjugates. Nat. Rev. Drug Discovery 2017, 16 , 315–337. 10.1038/nrd.2016.268.28303026
Bertrand N. ; Wu J. ; Xu X. ; Kamaly N. ; Farokhzad O. C. Cancer nanotechnology: the impact of passive and active targeting in the era of modern cancer biology. Adv. Drug Delivery Rev. 2014, 66 , 2–25. 10.1016/j.addr.2013.11.009.
Fu Z. ; Xiang J. Aptamer-Functionalized Nanoparticles in Targeted Delivery and Cancer Therapy. Int. J. Mol. Sci. 2020, 21 , 9123 10.3390/ijms21239123.
Birrer M. J. ; Moore K. N. ; Betella I. ; Bates R. C. Antibody-Drug Conjugate-Based Therapeutics: State of the Science. J. Natl. Cancer Inst. 2019, 111 , 538–549. 10.1093/jnci/djz035.30859213
Drago J. Z. ; Modi S. ; Chandarlapaty S. Unlocking the potential of antibody-drug conjugates for cancer therapy. Nat. Rev. Clin. Oncol. 2021, 18 , 327–344. 10.1038/s41571-021-00470-8.33558752
Joubert N. ; Beck A. ; Dumontet C. ; Denevault-Sabourin C. Antibody-Drug Conjugates: The Last Decade. Pharmaceuticals 2020, 13 , 245 10.3390/ph13090245.
Tedeschini T. ; Campara B. ; Grigoletto A. ; Bellini M. ; Salvalaio M. ; Matsuno Y. ; Suzuki A. ; Yoshioka H. ; Pasut G. Polyethylene glycol-based linkers as hydrophilicity reservoir for antibody-drug conjugates. J. Controlled Release 2021, 337 , 431–447. 10.1016/j.jconrel.2021.07.041.
Lu S. ; Xia R. ; Wang J. ; Pei Q. ; Xie Z. ; Jing X. Engineering Paclitaxel Prodrug Nanoparticles via Redox-Activatable Linkage and Effective Carriers for Enhanced Chemotherapy. ACS Appl. Mater. Interfaces 2021, 13 , 46291–46302. 10.1021/acsami.1c12353.34558902
Blanco E. ; Shen H. ; Ferrari M. Principles of nanoparticle design for overcoming biological barriers to drug delivery. Nat. Biotechnol. 2015, 33 , 941–951. 10.1038/nbt.3330.26348965
Wang Y. ; Giaccone G. Challenges in Cancer Molecular Targets and Therapeutics. Frontiers in Oncology 2011, 1 , 4 10.3389/fonc.2011.00004.22655226
Kinneer K. ; Meekin J. ; Tiberghien A. C. ; Tai Y.-T. ; Phipps S. ; Kiefer C. M. ; Rebelatto M. C. ; Dimasi N. ; Moriarty A. ; Papadopoulos K. P. ; et al. SLC46A3 as a Potential Predictive Biomarker for Antibody–Drug Conjugates Bearing Noncleavable Linked Maytansinoid and Pyrrolobenzodiazepine Warheads. Clin. Cancer Res. 2018, 24 , 6570–6582. 10.1158/1078-0432.CCR-18-1300.30131388
Su D. ; Kozak K. R. ; Sadowsky J. ; Yu S.-F. ; Fourie-O’Donohue A. ; Nelson C. ; Vandlen R. ; Ohri R. ; Liu L. ; Ng C. ; et al. Modulating Antibody–Drug Conjugate Payload Metabolism by Conjugation Site and Linker Modification. Bioconjugate Chem. 2018, 29 , 1155–1167. 10.1021/acs.bioconjchem.7b00785.
Pillow T. H. ; Sadowsky J. D. ; Zhang D. ; Yu S.-F. ; Del Rosario G. ; Xu K. ; He J. ; Bhakta S. ; Ohri R. ; Kozak K. R. ; et al. Decoupling stability and release in disulfide bonds with antibody-small molecule conjugates. Chem. Sci. 2017, 8 , 366–370. 10.1039/C6SC01831A.28451181
Cilliers C. ; Menezes B. ; Nessler I. ; Linderman J. ; Thurber G. M. Improved Tumor Penetration and Single-Cell Targeting of Antibody–Drug Conjugates Increases Anticancer Efficacy and Host Survival. Cancer Res. 2018, 78 , 758–768. 10.1158/0008-5472.CAN-17-1638.29217763
Shen B.-Q. ; Xu K. ; Liu L. ; Raab H. ; Bhakta S. ; Kenrick M. ; Parsons-Reponte K. L. ; Tien J. ; Yu S.-F. ; Mai E. ; et al. Conjugation site modulates the in vivo stability and therapeutic activity of antibody-drug conjugates. Nat. Biotechnol. 2012, 30 , 184–189. 10.1038/nbt.2108.22267010
Lambert J. M. ; Berkenblit A. Antibody-Drug Conjugates for Cancer Treatment. Annu. Rev. Med. 2018, 69 , 191–207. 10.1146/annurev-med-061516-121357.29414262
Mckertish C. M. ; Kayser V. Advances and Limitations of Antibody Drug Conjugates for Cancer. Biomedicines 2021, 9 , 872 10.3390/biomedicines9080872.34440076
Norman D. J. ; González-Fernández E. ; Clavadetscher J. ; Tucker L. ; Staderini M. ; Mount A. R. ; Murray A. F. ; Bradley M. Electrodrugs: an electrochemical prodrug activation strategy. J. Chem. Soc., Chem. Commun. 2018, 54 , 9242–9245. 10.1039/C8CC04151B.
Bezagu M. ; Clarhaut J. ; Renoux B. ; Monti F. ; Tanter M. ; Tabeling P. ; Cossy J. ; Couture O. ; Papot S. ; Arseniyadis S. In situ targeted activation of an anticancer agent using ultrasound-triggered release of composite droplets. Eur. J. Med. Chem. 2017, 142 , 2–7. 10.1016/j.ejmech.2017.03.057.28416362
Lerch M. M. ; Hansen M. J. ; van Dam G. M. ; Szymanski W. ; Feringa B. L. Emerging Targets in Photopharmacology. Angew. Chem., Int. Ed. 2016, 55 , 10978–10999. 10.1002/anie.201601931.
Zang C. ; Wang H. ; Li T. ; Zhang Y. ; Li J. ; Shang M. ; Du J. ; Xi Z. ; Zhou C. A light-responsive, self-immolative linker for controlled drug delivery via peptide- and protein-drug conjugates. Chem. Sci. 2019, 10 , 8973–8980. 10.1039/C9SC03016F.31762977
Wang Y. ; Zhang C. ; Wu H. ; Feng P. Activation and Delivery of Tetrazine-Responsive Bioorthogonal Prodrugs. Molecules 2020, 25 , 5640 10.3390/molecules25235640.
Ji X. ; Pan Z. ; Yu B. ; De La Cruz L. K. ; Zheng Y. ; Ke B. ; Wang B. Click and release: bioorthogonal approaches to “on-demand” activation of prodrugs. Chem. Soc. Rev. 2019, 48 , 1077–1094. 10.1039/C8CS00395E.30724944
Weiss J. T. ; Dawson J. C. ; Macleod K. G. ; Rybski W. ; Fraser C. ; Torres-Sánchez C. ; Patton E. E. ; Bradley M. ; Carragher N. O. ; Unciti-Broceta A. Extracellular palladium-catalysed dealkylation of 5-fluoro-1-propargyl-uracil as a bioorthogonally activated prodrug approach. Nat. Commun. 2014, 5 , 3277 10.1038/ncomms4277.24522696
Miller M. A. ; Mikula H. ; Luthria G. ; Li R. ; Kronister S. ; Prytyskach M. ; Kohler R. H. ; Mitchison T. ; Weissleder R. Modular Nanoparticulate Prodrug Design Enables Efficient Treatment of Solid Tumors Using Bioorthogonal Activation. ACS Nano 2018, 12 , 12814–12826. 10.1021/acsnano.8b07954.30550257
Fu Q. ; Li H. ; Duan D. ; Wang C. ; Shen S. ; Ma H. ; Liu Z. External-Radiation-Induced Local Hydroxylation Enables Remote Release of Functional Molecules in Tumors. Angew. Chem., Int. Ed. 2020, 59 , 21546–21552. 10.1002/anie.202005612.
Geng J. ; Zhang Y. ; Gao Q. ; Neumann K. ; Dong H. ; Porter H. ; Potter M. ; Ren H. ; Argyle D. ; Bradley M. Switching on prodrugs using radiotherapy. Nat. Chem. 2021, 13 , 805–810. 10.1038/s41557-021-00711-4.34112990
Yardley D. A. nab-Paclitaxel mechanisms of action and delivery. J. Controlled Release 2013, 170 , 365–372. 10.1016/j.jconrel.2013.05.041.
Wagner A. J. ; Ravi V. ; Riedel R. F. ; Ganjoo K. ; Van Tine B. A. ; Chugh R. ; Cranmer L. ; Gordon E. M. ; Hornick J. L. ; Du H. ; et al. nab-Sirolimus for Patients With Malignant Perivascular Epithelioid Cell Tumors. J. Clin. Oncol. 2021, 39 , 3660–3670. 10.1200/JCO.21.01728.34637337
Venkatesan A. M. ; Smith R. A. ; Kumar S. S. ; Bakshi R. K. ; Damle N. K. ; Odera A. V. Self-Immolative Linkers Containing Mandelic Acid Derivatives, Drug-Ligand Conjugates for Targeted Therapies and Uses Thereof. Internation Patent Applications WO 2015/038426, PCT/US2014/054236, March 19, 2015.
Kratz F. ; Warnecke A. ; Scheuermann K. ; Stockmar C. ; Schwab J. ; Lazar P. ; Drückes P. ; Esser N. ; Drevs J. ; Rognan D. ; et al. Probing the Cysteine-34 Position of Endogenous Serum Albumin with Thiol-Binding Doxorubicin Derivatives. Improved Efficacy of an Acid-Sensitive Doxorubicin Derivative with Specific Albumin-Binding Properties Compared to That of the Parent Compound. J. Med. Chem. 2002, 45 , 5523–5533. 10.1021/jm020276c.12459020
Gong J. ; Yan J. ; Forscher C. ; Hendifar A. Aldoxorubicin: a tumor-targeted doxorubicin conjugate for relapsed or refractory soft tissue sarcomas. Drug Des., Dev. Ther. 2018, 12 , 777–786. 10.2147/DDDT.S140638.
Sørensen B. S. ; Horsman M. R. Tumor Hypoxia: Impact on Radiation Therapy and Molecular Pathways. Frontiers in Oncology 2020, 10 , 562 10.3389/fonc.2020.00562.32373534
Butler I. B. ; Schoonen M. A. A. ; Rickard D. T. Removal of dissolved oxygen from water: A comparison of four common techniques. Talanta 1994, 41 , 211–215. 10.1016/0039-9140(94)80110-X.18965910
McKeown S. R. Defining normoxia, physoxia and hypoxia in tumours-implications for treatment response. Br. J. Radiol. 2014, 87 , 20130676 10.1259/bjr.20130676.24588669
Rodin D. ; Tawk B. ; Mohamad O. ; Grover S. ; Moraes F. Y. ; Yap M. L. ; Zubizarreta E. ; Lievens Y. Hypofractionated radiotherapy in the real-world setting: An international ESTRO-GIRO survey. Radiother. Oncol. 2021, 157 , 32–39. 10.1016/j.radonc.2021.01.003.33453312
Murray Brunt A. ; Haviland J. S. ; Wheatley D. A. ; Sydenham M. A. ; Alhasso A. ; Bloomfield D. J. ; Chan C. ; Churn M. ; Cleator S. ; Coles C. E. ; et al. Hypofractionated breast radiotherapy for 1 week versus 3 weeks (FAST-Forward): 5-year efficacy and late normal tissue effects results from a multicentre, non-inferiority, randomised, phase 3 trial. Lancet 2020, 395 , 1613–1626. 10.1016/S0140-6736(20)30932-6.32580883
Setsukinai K.-i. ; Urano Y. ; Kakinuma K. ; Majima H. J. ; Nagano T. Development of Novel Fluorescence Probes That Can Reliably Detect Reactive Oxygen Species and Distinguish Specific Species * 210. J. Biol. Chem. 2003, 278 , 3170–3175. 10.1074/jbc.M209264200.12419811
Oweida A. ; Phan A. ; Vancourt B. ; Robin T. ; Hararah M. K. ; Bhatia S. ; Milner D. ; Lennon S. ; Pike L. ; Raben D. ; et al. Hypofractionated Radiotherapy Is Superior to Conventional Fractionation in an Orthotopic Model of Anaplastic Thyroid Cancer. Thyroid 2018, 28 , 739–747. 10.1089/thy.2017.0706.29774792
Luthria G. ; Li R. ; Wang S. ; Prytyskach M. ; Kohler R. H. ; Lauffenburger D. A. ; Mitchison T. J. ; Weissleder R. ; Miller M. A. In vivo microscopy reveals macrophage polarization locally promotes coherent microtubule dynamics in migrating cancer cells. Nat. Commun. 2020, 11 , 3521 10.1038/s41467-020-17147-y.32665556
Dumontet C. ; Jordan M. A. Microtubule-binding agents: a dynamic field of cancer therapeutics. Nat. Rev. Drug Discovery 2010, 9 , 790–803. 10.1038/nrd3253.20885410
Miller M. A. ; Askevold B. ; Mikula H. ; Kohler R. H. ; Pirovich D. ; Weissleder R. Nano-palladium is a cellular catalyst for in vivo chemistry. Nat. Commun. 2017, 8 , 15906 10.1038/ncomms15906.28699627
Zhang M. S. ; Cui J. D. ; Lee D. ; Yuen V. W. ; Chiu D. K. ; Goh C. C. ; Cheu J. W. ; Tse A. P. ; Bao M. H. ; Wong B. P. Y. ; et al. Hypoxia-induced macropinocytosis represents a metabolic route for liver cancer. Nat. Commun. 2022, 13 , 954 10.1038/s41467-022-28618-9.35177645
Kim H.-Y. ; Li R. ; Ng T. S. C. ; Courties G. ; Rodell C. B. ; Prytyskach M. ; Kohler R. H. ; Pittet M. J. ; Nahrendorf M. ; Weissleder R. ; et al. Quantitative Imaging of Tumor-Associated Macrophages and Their Response to Therapy Using 64Cu-Labeled Macrin. ACS Nano 2018, 12 , 12015–12029. 10.1021/acsnano.8b04338.30508377
Ng T. S. C. ; Hu H. ; Kronister S. ; Lee C. ; Li R. ; Gerosa L. ; Stopka S. A. ; Burgenske D. M. ; Khurana I. ; Regan M. S. ; et al. Overcoming differential tumor penetration of BRAF inhibitors using computationally guided combination therapy. Sci. Adv. 2022, 8 , eabl6339 10.1126/sciadv.abl6339.35486732
Jones K. I. ; Tiersma J. ; Yuzhalin A. E. ; Gordon-Weeks A. N. ; Buzzelli J. ; Im J. H. ; Muschel R. J. Radiation combined with macrophage depletion promotes adaptive immunity and potentiates checkpoint blockade. EMBO. Mol. Med. 2018, 10 , e9342 10.15252/emmm.201809342.30442705
Ng T. S. C. ; Gunda V. ; Li R. ; Prytyskach M. ; Iwamoto Y. ; Kohler R. H. ; Parangi S. ; Weissleder R. ; Miller M. A. Detecting Immune Response to Therapies Targeting PDL1 and BRAF by Using Ferumoxytol MRI and Macrin in Anaplastic Thyroid Cancer. Radiology 2021, 298 , 123–132. 10.1148/radiol.2020201791.33107799
Ying H. ; Kimmelman A. ; Lyssiotis C. ; Hua S. ; Chu G. ; Fletcher-Sananikone E. ; Locasale J. ; Son J. ; Zhang H. ; Coloff J. ; et al. Oncogenic Kras Maintains Pancreatic Tumors through Regulation of Anabolic Glucose Metabolism. Cell 2012, 149 , 656–670. 10.1016/j.cell.2012.01.058.22541435
Applegate K. T. ; Besson S. ; Matov A. ; Bagonis M. H. ; Jaqaman K. ; Danuser G. plusTipTracker: Quantitative image analysis software for the measurement of microtubule dynamics. J. Struct. Biol. 2011, 176 , 168–184. 10.1016/j.jsb.2011.07.009.21821130
|
PMC010xxxxxx/PMC10338269.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
CD220090
10.2337/cd22-0090
Feature Articles
Effectiveness of RADAR: An Innovative Model to Organize Diabetes Care in First Nations Communities
https://orcid.org/0000-0003-2197-0463
Eurich Dean T. 1 2
Wozniak Lisa A. 1 2
Soprovich Allison 1 2
Minhas-Sandhu Jasjeet K. 1 2
Crowshoe Lynden 3
Johnson Jeffrey A. 1 2
Samanani Salim 4
1 School of Public Health, University of Alberta, Edmonton, Alberta, Canada
2 Alliance for Canadian Health Outcomes Research in Diabetes, University of Alberta, Edmonton, Alberta, Canada
3 Cumming School of Medicine and Indigenous, Local, and Global Health Office, University of Calgary, Calgary, Alberta, Canada
4 OKAKI Health Intelligence, Inc., Calgary, Alberta, Canada
Corresponding author: Dean T. Eurich, [email protected]
Summer 2023
27 1 2023
27 1 2023
41 3 351358
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
Challenges exist for the management of diabetes care in First Nations populations. RADAR (Reorganizing the Approach to Diabetes through the Application of Registries) is a culturally appropriate, innovative care model that incorporates a disease registry and electronic health record for local care provision with remote coordination, tailored for First Nations people. This study assessed the effectiveness of RADAR on patient outcomes and diabetes care organization in participating communities in Alberta, Canada. It revealed significant improvements in outcomes after 2 years, with 91% of patients achieving a primary combined end point of a 10% improvement in or persistence at target for A1C, systolic blood pressure, and/or LDL cholesterol. Qualitative assessment showed that diabetes care organization also improved. These multimethod findings support tailored diabetes care practices in First Nations populations.
The Lawson Foundation Canadian Institutes of Health Research 10.13039/501100000024 143562 Alberta Innovates - Health Solutions 10.13039/501100000145
==== Body
pmcIn Canada, chronic diseases have reached critical imbalances for First Nations (FN) people (1). Diabetes is three to five times more common in FN than in non-FN populations (2,3). Higher rates of diabetes-related complications are observed in FN populations (4,5). The high burden of disease, complications, and comorbidity that FN people with diabetes experience is compounded by the significant and historical social, physical, and economic challenges they face (5,6).
Primary care for people with diabetes is typically delivered using a chronic care model (7), although its implementation within FN communities is challenging. FN communities are often remote, with limited access to comprehensive diabetes care. Current federal and provincial approaches to managing diabetes care among FN people has led to suboptimal and fragmented care for people living both on and off reserves. Indeed, we previously found that such diabetes care was reactive, relying on FN people to navigate a complicated health care system (8).
Within a chronic care model, one framework for organizing and delivering optimal diabetes care is called the 5Rs, which stands for Recognize (performing screening/risk factor assessment), Register (systematically tracking patients), Relay (facilitating information-sharing), Recall (providing timely review and reassessment), and Resource (supporting self-management) (7). Furthermore, FN-specific clinical practice guidelines (CPGs) recommend that diabetes prevention, care, and education be grounded in communities’ social, cultural, and health service contexts (6). The organization of diabetes care (7) and the principles of self-determination and governance of FN communities (9) guided the development of RADAR (Reorganizing the Approach to Diabetes through the Application of Registries) (10), in partnership with FN communities and OKAKI Health Intelligence, Inc., a private, social enterprise in Alberta, Canada, with >12 years working with FN communities.
The RADAR model (10) uses remote care coordinators (CCs), who are registered nurses or dietitians, to support local health care providers (HCPs) (e.g., registered nurses, licensed practical nurses, and dietitians) in FN communities through telehealth to 1) populate an electronic medical record (EMR) diabetes registry called the CARE platform, which was developed specifically for FN communities; 2) coordinate population-level care to identify priorities and gaps in care; and 3) apply current CPGs (11) through regular client review and case conferencing to consider therapeutic changes and coordinate referrals (Figure 1).
FIGURE 1 RADAR infographic.
RADAR was implemented collaboratively with FN communities from Treaty 6, 7, and 8 territories in Alberta, Canada, where local HCPs found the model appropriate, acceptable, and valuable for both HCPs and patients (12). Here, we describe the effectiveness of RADAR with regard to patient outcomes and the organization of diabetes care in these communities.
Design and Methods
Design
We used a modified stepped-wedge design supplemented with a descriptive qualitative assessment to evaluate RADAR’s effectiveness (10). Eligible patients were ≥18 years of age, diagnosed with type 2 diabetes, and actively engaged in and received care from the FN health facility on a reserve and had provided verbal consent for HCPs to manage their diabetes (10). For the qualitative assessment, we purposefully sampled remote care coordinators and local HCPs who were implementing RADAR.
The decision to participate in RADAR was made by FN leaders (e.g., chiefs and council, chief executive officers, and/or community health directors) in each participating community, who reviewed and endorsed the project. OCAP (Ownership, Control, Access, and Possession) principles were followed (13,14). Findings were shared with FN health leadership to incorporate their feedback. RADAR was approved by the University of Alberta’s Health Research Ethics Board (Study ID Pro00048714).
Data Collection and Analysis
The primary outcome was a 10% improvement in A1C, systolic blood pressure (SBP), or LDL cholesterol during the 2-year follow-up relative to baseline control periods, representing a clinically important difference (15). For patients who were already at target at baseline, we considered persistence at target (i.e., maintaining values within 10% of baseline) as achieving the primary outcome (10). Outcomes were measured continuously throughout the follow-up period and at the time of each community transition (i.e., step) as per traditional stepped-wedge designs.
The primary outcome was first analyzed using an intention-to-treat (ITT) framework, which included all patients identified by the communities in their baseline assessments, irrespective of the follow-up care they received. Additionally, we used a per-protocol (PP) framework, removing patients who died, moved, or subsequently refused care at FN community health centers. We fit a generalized linear mixed model with random effect for cluster (i.e., community) and fixed effect for each step (i.e., transitions from baseline to follow-up intervention) (16). We also included calendar year in our models as a fixed effect to control for any temporal confounding effects, although calendar year was not significant in the models (P >0.10). Patients with insufficient primary outcome data were assumed to be nonresponders (i.e., individuals who had failed to achieve the primary end point). In addition, the mean change in A1C, SBP, and LDL was assessed for patients with A1C ≥7.5% (58 mmol/mol), SBP ≥140 mmHg, and LDL cholesterol ≥2.5 mmol/L (i.e., at-risk patients), after accounting for the baseline value, year, and community clustering in alignment with Diabetes Canada CPGs (10,11). Outcome data are not reported by community to maintain confidentiality as per our data-sharing agreements; however, the consistency of effects across the communities was assessed and is expressed as ranges. Quantitative analyses were performed using Stata, v. 14.2, statistical software (StataCorp, College Station, TX).
We used a descriptive qualitative approach (17) to assess the effectiveness of RADAR on the organization of diabetes care within participating communities, which were diverse by treaty, geography, population, and proximity to urban centers. Data sources included individual interviews with remote CCs and local HCPs. Qualitative data were managed using ATLAS.ti, v. 8, software (Scientific Software Development, Berlin, Germany) and analyzed using summative content analysis within the 5Rs framework, including negative case analysis for meaning saturation to understand the diversity of experiences (18,19). We followed the Consolidated Criteria for Reporting Qualitative Research framework (Supplementary Table S1) (20).
Results
Primary Outcomes
The quantitative assessment included seven communities at the end of 2 years. At baseline, 516 patients with type 2 diabetes were registered in RADAR, ranging from 28 to 129 per community. The mean age was 60 years (SD 13.6 years, mean community range [MCR] 55–65 years), 57% were female (MCR 50–66%), the mean A1C was 8.3% (67 mmol/mol) (SD 2.0%, MCR 8.0–8.7%), the mean SBP was 131 mmHg (SD 19.1 mmHg, MCR 122–137 mmHg), and the mean LDL cholesterol was 2.0 mmol/L (36 mg/dL) (SD 0.9, MCR 1.7–2.3 mmol/L [30.6–41.4 mg/dL]) (Table 1).
TABLE 1 Clustered Baseline Demographics of RADAR Participants (n = 516)
Characteristic Value
Age, years 60 ± 13.6
Female sex 292 (57)
Smoker 119 (23.1)
A1C, % 8.3 ± 2.0
Cholesterol, mmol/L
LDL cholesterol
HDL cholesterol
Triglycerides
Total cholesterol–to–HDL cholesterol ratio
2.0 ± 0.9
1.1 ± 0.4
2.5 ± 2.7
3.9 ± 1.8
Estimated glomerular filtration rate, mL/min/1.73 m2*
15–30
30–45
45–60
60–90
>90
15 ± 3.1
31 ± 6.5
44 ± 9.2
143 ± 29.9
246 ± 51.4
Data are mean ± SD or n (%).
* n = 479.
Overall, after 2 years of RADAR across all communities, we found significant improvements for the primary combined end point and in all three outcomes individually (Figure 2). Achievement of the combined end point was 91% (95% CI 89–94%) in the ITT population (n = 516) and 93% (95% CI 91–95%) in the PP population (n = 419) (intracluster correlation ρ = 0.008). Within the combined primary end point (ITT population), the majority (377 of 465, or 80%) had a 10% improvement in any one parameter (A1C, SBP, or LDL cholesterol) over baseline, while 78% of the remaining patients (40 of 51) were persistently at target (Supplementary Figure S1). When examining the primary combined outcome by community, achievement ranged from 87 to 100% (Supplementary Figure S2).
FIGURE 2 Primary combined end point of achieving a 10% reduction or maintaining values within 10% of target for A1C, SBP, or LDL cholesterol during 2 years of follow-up.
When examining the individual components of the primary end point, the biggest driver of the improvement was A1C. Two-thirds of patients achieved a 10% reduction in or persistence at target for A1C (ITT 66% [95% CI 55–77%] and PP 65% [95% CI 54–76]). The proportion of patients achieving a 10% reduction (221 of 336, or 64%) was similar to the number of patients with persistence at target (130 of 180, or 71%) in the ITT population. The mean change in A1C during follow-up was −0.93% (95% CI −0.59 to −1.28%, P = 0.001; MCR −1.26 to −0.16%). Among those with an A1C ≥7.5% (58 mmol/mol) at baseline (mean 9.70%, SD 1.66%; n = 303), mean change in A1C during follow-up was −1.62% (95% CI −2.18 to −1.06%, P <0.001; MCR −2.43 to −0.54%).
With regard to the other individual outcomes, SBP improved for 54% (95% CI 39–68%) of the ITT population and 58% (95% CI 46–69%) of the PP population, and LDL cholesterol improved for 40% (95% CI 32–48%) of the ITT population and 42% (95% CI 34–50%) of the PP population. Unlike A1C, achieving the SBP or LDL cholesterol target was driven by achieving a 10% reduction: 134 of 187 (70%) for SBP and 159 of 236 (66%) for LDL cholesterol in ITT populations, whereas a smaller proportion persisted at target for SBP (150 of 329, or 45%) or LDL cholesterol (55 of 280, or 18%). The mean change in SBP during follow-up was −6.03 mmHg (95% CI −2.68 to −9.38 mmHg, P = 0.005; MCR −19.64 to −0.82 mmHg). Among those with an SBP ≥140 mmHg at baseline (mean 156 mmHg, SD 14.9 mmHg; n = 193), mean change in SBP during follow-up was −14.19 mmHg (95% CI −22.55 to −5.82, P = 0.006; MCR −36.17 to −3.45 mmHg). The mean change in LDL cholesterol during follow-up was −0.17 mmol/L (95% CI −0.10 to −0.24 mmol/L, P = 0.001; MCR −0.38 to −0.02 mmol/L). Among those with LDL cholesterol ≥2.5 mmol/L at baseline (mean 3.17 mmol/L, SD 0.57 mmol/L; n = 215), mean change in LDL during follow-up was −0.33 mmol/L (95% CI −0.46 to −0.21 mmol/L, P = 0.01; MCR −0.71 to −0.03 mmol/L).
Organization of Diabetes Care
We conducted 21 semistructured interviews with 11 individual participants: three remote CCs and eight local HCPs (e.g., registered nurses, licensed practical nurses, and dietitians) from May 2015 to February 2019 (Table 2). Overall, we found improvements in the organization of diabetes care by the 5Rs, as presented below, with illustrative quotes including position (i.e., CC or HCP) and anonymous study code number for each participant.
TABLE 2 Interview Timing and Participant Roles
Participant Role Interviewed at 6 Months, n Interviewed at 24 Months, n Total, n (%)*
Remote CC 2 3 3 (27)†
Local dietitian — 3 3 (27)
Local licensed practical nurse — 2 2 (18)
Local registered nurse — 3 3 (27)
* Percentages do not sum to 100 due to rounding.
† Two CCs were interviewed at both 6 and 24 months.
Recognize and Register
Overall, RADAR facilitated the recognition and registration of patients with diabetes. RADAR “identified our [diabetes] population” (HCP 5), including “younger people that we didn’t know had diabetes” (HCP 4), thus capturing “a lot more of the population” (HCP 7). Nevertheless, RADAR was limited in identifying new patients because only physicians can diagnose. “I don’t know if RADAR had any impact on recognizing clients . . . [because] we never can diagnose” (HCP 6). As such, “the registry was only those [patients] already diagnosed with type 2 diabetes” (HCP 3).
Regardless, RADAR resulted in a “client registry that has been created” (HCP 1). The act of centrally registering patients in CARE with CC support confirmed the type of diabetes for some patients because “many patients had been indicated as ‘other’ diabetes because the HCPs weren’t sure [of the diagnosis]” (CC 2). CCs helped identify type of diabetes by “looking on [the provincial EMR] through discharge summaries to find data to support whether [patients had] type 2 diabetes” (CC 2) or asking “the nurse to confirm with [the patient’s] physician” (CC 1). However, centrally registering patients also resulted in “increasing our overall client load” (HCP 1) or feeling overwhelmed by “seeing my whole population and saying, ‘I should be doing all these things for everybody’” (HCP 6) in light of limited resources.
Relay
Overall, RADAR improved HCP access to and sharing of information. Within health centers, local HCPs described efficient and comprehensive access to information through CARE versus relying on patients’ recollections or paper charts. “You don’t have to go into the [paper] chart every time. Even when you’re unfamiliar with a client, all that information is [in CARE] . . . . Because, sometimes, your clients are not great historians” (HCP 7). CCs also enabled access of local HCPs to clinical information contained in other EMRs. “It can take a lot of time to get access to [the provincial EMR], if we are given it at all. With RADAR, [the CCs] populated CARE with medical information that is important to the level of care we can provide” (HCP 1). Moreover, CARE resulted in “improved communication amongst the team” (HCP 7), allowing local HCPs “to communicate everything you’ve done with a client” (HCP 5). This feature was especially important given staffing challenges. “We really have a good flow [of information] regardless of the staff turnover” (CC 1).
Finally, RADAR improved communication with providers outside the health centers. “Prior to RADAR, there was no communication between on- and off-reserve medical services,” but the CCs helped “connect us to the off-reserve services, such as the physicians . . . [and] the communication has become more open” (HCP 1). Furthermore, CARE facilitated communication and coordinated diabetes specialty care such as ophthalmology (HCP 7) and endocrinology (HCP 6) through report and letter templates. “I use the [patient] report summary, and I can easily fax that to the [physician] clinic” (HCP 2). However, improved relay of information was not reported in all communities, indicating that more work is needed.
Recall
RADAR improved timely review and reassessment of patients with diabetes by local HCPs through CARE’s features (e.g., tasks, reminders, and patient summaries) and CC support. “You have very specific tasks that you need to accomplish to improve people’s health, and they’re right in the forefront [in CARE]” (HCP 7). CCs facilitated recall by “looking at the total population, including the ones not accessing care” (CC 1) and identifying for HCPs which patients “should be targeted at this time” (HCP 4). As a result, “we have been able to reach out and connect with a lot more community members who may not have been utilizing our services for diabetes care” (HCP 1). In addition, CCs reviewed “the diabetes guidelines and asked us to follow up on as much as we can with that client” (HCP 6). As a result, processes of care were completed. “Blood pressures were taken; foot exams and lab work were done” (CC 1). Recall support was crucial in the context of busy clinics. “It gets busy, and you get sidetracked, so it helps to have that person saying, ‘This is who you’ve got to follow up with next; this person needs [this]’” (HCP 4).
However, not all HCPs believed CARE improved recall beyond what existed “because we already set up regular reminders, like [for A1Cs], foot exams, eye exams . . . into the [physician] EMR” (HCP 3). Furthermore, in some communities, “we only case-reviewed half the patients” (CC 1), thus limiting recall. Some HCPs were uncomfortable recalling patients, saying, “That’s not our job’” (CC 1). HCPs resisted recalling patients they believed did not want their care. “Some of these clients that [the CCs] asked me to review were ones that don’t access care here regularly. We know their [glucose] is high [and] that they have other things going on in their life, so they don’t come in for appointments . . . . Chasing them down, [is] not the best use of my time” (HCP 3). Another HCP explained, “We tell [the CC] this person is not engaged . . . and [to] just leave them alone” (HCP 6).
Resource
RADAR helped improve local HCPs’ ability to support patients’ diabetes self-management by informing patients of health center services and supports, developing care plans, and increasing local HCPs’ diabetes knowledge. For example, one respondent said, “Based on the registry, we would call clients, introduce ourselves, [and] introduce the program” (HCP 1) or inform patients that “we have a diabetes management support system . . . [and ask] ‘Would you like some help?’” (CC 2). This resulted in patients accessing services. “There’s probably 50% more people coming for their monthly foot care” (HCP 5).
Furthermore, CCs helped local HCPs develop care plans informed by CPGs. “[The CC] reviewed the guidelines with us and helped us incorporate them into our day-to-day care with clients” (HCP 1). This, in part, helped to increase local HCPs’ diabetes knowledge and “build local capacity” (CC 1). For example, one respondent said, “[The CC showed us] the gold standards for diabetes treatment . . . . I am not an expert on diabetes. I have learned so much” (HCP 7). Diabetes education was especially valuable to HCPs who did not have time or resources to stay current in their knowledge. “Even though we’re close to the nearest town and physicians, you’re still isolated from information . . . . There may be new information, and [the CC is] helpful that way, with new stuff or if we have a question” (HCP 4). As a result, local HCPs reported increased confidence in their diabetes knowledge. “[The CC] has made a huge difference in my confidence with diabetes knowledge” (HCP 1). However, not all HCPs believed RADAR improved their diabetes knowledge. “I have a lot of knowledge in diabetes, so it is not anything new” (HCP 2). One respondent noted that the community already had a certified diabetes care and education specialist (HCP 3). Finally, some HCPs requested further education, such as “10-minute education sessions on new [CPGs]” (HCP6), indicating an ongoing need for up-to-date diabetes knowledge.
Discussion
We found that RADAR improved patient-level outcomes through improved organization of diabetes care in FN communities. In addition, we found significant achievement of our primary outcomes for RADAR participants over 2 years, with the largest effect observed for change in A1C. This finding is similar to an intervention study of rural (non-FN) patients with type 2 diabetes, in which improvements in SBP and/or LDL cholesterol were smaller than improvements in A1C (21). Although glycemic control is essential, further improvements in SBP and/or LDL cholesterol would provide substantial macro- and microvascular benefits (22).
Our results also demonstrated improved diabetes care organization (8), which is foundational to improving patient-level outcomes (7). With regard to the 5Rs, RADAR resulted in central registries, helping communities verify, visualize, and recall their diabetes population. The use of patient registries for diabetes management is associated with better processes and outcomes of care (23,24). Furthermore, FN health directors can use their own registry data as a resource to inform decision-making for local program planning, meet federal reporting requirements, and support funding requests based on the needs of and trends in the local patient population. RADAR allows for a community-led response based in data sovereignty (25) and aligned with the principles of self-governance and self-determination (9).
Second, RADAR facilitated relay of clinical information essential for continuity and coordination of team-based diabetes care (11). This effort included overcoming the restricted access of local HCPs to patients’ provincial EMR data in the current fragmented system, with multiple isolated technologies hindering clinical information-sharing necessary for diabetes care (26,27). Within health centers, centralized compilation of patients’ clinical information and care plans in CARE was crucial in the context of the high staff turnover that FN communities face (27). In addition, team-based care was enhanced through relationship-building among HCPs through CCs’ professional networks, including diabetes specialists. This was especially important for remote communities with irregular access to services necessary for comprehensive diabetes care (11,27).
Finally, RADAR increased local HCP capacity (i.e., knowledge and confidence) to support diabetes self-management (resource) through education (e.g., of current CPGs) and peer support, promoting the long-term sustainability of primary outcomes. This is important because lack of adequate diabetes knowledge among local HCPs is a barrier to diabetes care and is largely affected by time/competing priorities, staff turnover, and changing guidelines (4,28). Furthermore, local HCPs, who are the experts in their communities, can apply RADAR, including enhanced knowledge of diabetes care, with local patients in appropriate ways that recognize contextual and cultural factors (12).
Although RADAR improved diabetes care organization overall, challenges remain. Centrally registering patients does not necessarily recognize new patients through screening, thereby limiting the population health impact. Future RADAR activities may include recommended earlier and more frequent screening of type 2 diabetes for FN people (6). Furthermore, centrally registering patients with recall functionality increased local HCPs’ workload, contributing to feelings of stress. In addition, although RADAR encouraged local HCPs to actively engage or re-engage patients, some HCPs resisted recalling patients, thus limiting the proactive care essential to improving patient outcomes (11). This challenge is not unique to FN communities, as HCPs must balance proactive care with patients’ right to decline care. However, it can become a greater challenge in FN communities, as patients may have competing priorities that take precedence over diabetes self-management (resource), such as housing or food insecurity related to historic and social injustices (5,9). Although local HCPs are experts in their communities, including knowing which patients want to engage or re-engage with diabetes care, some HCPs may believe it is patients’ responsibility to seek care and outside of their role as care providers to facilitate patient engagement proactively. This opinion may be related to local HCPs’ feelings of stress, which is understandable in the context of their limited resources. Nonetheless, this belief may limit community-level comprehensive diabetes care and requires further exploration. Indeed, we previously found that the 5Rs of diabetes care organization are interrelated and influenced by financial and human resources (the sixth R) (8). Regardless, RADAR created the conditions for local HCPs and patients to collaborate in diabetes self-management in this resource-challenged context.
Strengths and Limitations
Our incorporation of quantitative and qualitative components to comprehensively measure effectiveness for both clinical outcomes and diabetes care organization is a strength. However, this study is not without limitations. First, given our follow-up of 2 years, we used intermediate outcomes as opposed to hard clinical end points (e.g., occurrence of heart attack or stroke). Nevertheless, A1C, SBP, and LDL cholesterol are well established diabetes care targets within guidelines (11) and have been shown in many large-scale randomized controlled trials (RCTs) to confer substantial benefits with regard to macro- or microvascular outcomes (11,21). Second, we designed a pragmatic, controlled, community-based intervention aimed at local HCPs as opposed to patients per se. Individual patients or local HCPs within the community could not be randomized to care with and without RADAR because of the threat of “contamination” via exposure to some aspects of the intervention within the same community team or health facility. Moreover, our community partners were not interested in a traditional RCT; a stepped-wedge design was used, as few communities would accept being the control community for an extended duration. Although the participating FN communities were diverse, they may not be representative of all FN communities. Additionally, the participating local HCPs who were interviewed for the qualitative component of the study are a sample of the health care workers involved in the implementation of RADAR. Although almost all FN communities are dealing with similar issues in providing diabetes care, it is possible that some of our findings may not be applicable to all communities, depending on local resources.
Conclusion
RADAR is an effective, tailored approach to support the organization of and capacity building in diabetes care for FN communities. From our findings, it is reasonable to assume that this intervention could be applied to other indigenous populations in jurisdictions around the globe.
Article Information
Acknowledgments
The authors thank their FN partners for their ongoing and generous support and recognize that this work was implemented in the traditional territories of their Treaty 6, 7, and 8 FN partners. The authors thank OKAKI for their support and participation in the evaluation of RADAR. They also acknowledge the significant contribution of Dr. Sumit (Me2) R. Majumdar, who died in 2018, to the study design.
Funding
This work was funded by the Canadian Institutes of Health Research (MOP #143562), Alberta Innovates Health Solutions, and the Lawson Foundation.
Duality of Interest
No potential conflicts of interest relevant to this article were reported.
Author Contributions
D.T.E. and S.S. conceived the study, designed its protocols, and received ethics approval and funding. D.T.E. led the quantitative data collection and analysis, with support from J.K.M.-S. L.A.W. led the evaluation design and conducted qualitative data collection and analysis. D.T.E., L.A.W., and A.S. drafted the manuscript. L.C. and J.A.J. actively contributed to the study design. All authors read and approved the final manuscript. D.T.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
This article contains supplementary material online at https://doi.org/10.2337/figshare.21872040.
==== Refs
References
1. King M. Chronic diseases and mortality in Canadian Aboriginal peoples: learning from the knowledge. Prev Chronic Dis 2011;8 :A07 21159219
2. Oster R, Toth EL, King M, Crowshoe L. Diabetes and the Status Aboriginal Population in Alberta. Edmonton, Alberta, Institute of Health Economics, 2011
3. Turin TC, Saad N, Jun M, . Lifetime risk of diabetes among First Nations and non-First Nations people. CMAJ 2016;188 :1147–1153 27647609
4. Harris SB, Naqshbandi M, Bhattacharyya O, Hanley AJ, Esler JG; CIRCLE Study Group. Major gaps in diabetes clinical care among Canada’s First Nations: results of the CIRCLE study. Diabetes Res Clin Pract 2011;92 :272–279 21376415
5. Oster RT, Virani S, Strong D, Shade S, Toth EL. Diabetes care and health status of First Nations individuals with type 2 diabetes in Alberta. Can Fam Physician 2009;55 :386–393 19366950
6. Diabetes Canada Clinical Practice Guidelines Expert Committee; Crowshoe L, Dannenbaum D, Green M, Henderson R, Hayward MN, Toth E. Type 2 diabetes and indigenous peoples. Can J Diabetes 2018;42 (Suppl. 1 ):S296–S306 29650108
7. Clement M, Filteau P, Harvey B, .; Diabetes Canada Clinical Practice Guidelines Expert Committee. Organization of diabetes care. Can J Diabetes 2018;42 (Suppl. 1 ):S27–S35 29650106
8. Wozniak LA, Johnson JA, Eurich DT. Working towards a better understanding of type 2 diabetes care organization with First Nations communities: a qualitative assessment. Arch Public Health 2020;78 :7 32025300
9. Katz A, Enns J, Kinew KA. Canada needs a holistic First Nations health strategy. CMAJ 2017;189 :E1006–E1007 28790055
10. Eurich DT, Majumdar SR, Wozniak LA, . Addressing the gaps in diabetes care in First Nations communities with the reorganizing the approach to diabetes through the application of registries (RADAR): the project protocol. BMC Health Serv Res 2017;17 :117 28166804
11. Diabetes Canada Clinical Practice Guidelines Expert Committee. Diabetes Canada 2018 Clinical Practice Guidelines for the Prevention and Management of Diabetes in Canada. Can J Diabetes 2018;42 (Suppl. 1 ):S1–S325 29650079
12. Wozniak LA, Soprovich AL, Johnson JA, Eurich DT. Adopting and implementing an innovative model to organize diabetes care within First Nations communities: a qualitative assessment. BMC Health Serv Res 2021;21 :415 33941176
13. Schnarch B. Ownership, Control, Access, and Possession (OCAP) or self-determination applied to research: a critical analysis of contemporary First Nations research and some options for First Nations communities. J Aborig Health 2004;1 :80–95
14. First National Information Governance Centre. The First Nations principles of OCAP. Available from https://fnigc.ca/ocap-training. Accessed 24 May 2022
15. Majumdar SR, Guirguis LM, Toth EL, Lewanczuk RZ, Lee TK, Johnson JA. Controlled trial of a multifaceted intervention for improving quality of care for rural patients with type 2 diabetes. Diabetes Care 2003;26 :3061–3066 14578240
16. Hussey MA, Hughes JP. Design and analysis of stepped wedge cluster randomized trials. Contemp Clin Trials 2007;28 :182–191 16829207
17. Sandelowski M. What’s in a name? Qualitative description revisited. Res Nurs Health 2010;33 :77–84 20014004
18. Hsieh HF, Shannon SE. Three approaches to qualitative content analysis. Qual Health Res 2005;15 :1277–1288 16204405
19. Hennink MM, Kaiser BN, Marconi VC. Code saturation versus meaning saturation: how many interviews are enough? Qual Health Res 2017;27 :591–608 27670770
20. Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups. Int J Qual Health Care 2007;19 :349–357 17872937
21. Johnson JA, Eurich DT, Toth EL, Lewanczuk RZ, Lee TK, Majumdar SR. Generalizability and persistence of a multifaceted intervention for improving quality of care for rural patients with type 2 diabetes. Diabetes Care 2005;28 :783–788 15793173
22. Diabetes Canada Clinical Practice Guidelines Expert Committee; Stone JA, Houlden RL, Lin P, Udell JA, Verma S. Cardiovascular protection in people with diabetes. Can J Diabetes 2018;42 (Suppl. 1 ):S162–S169 29650091
23. Renders CM, Valk GD, Griffin SJ, Wagner EH, Eijk Van JT, Assendelft WJJ. Interventions to improve the management of diabetes in primary care, outpatient, and community settings: a systematic review. Diabetes Care 2001;24 :1821–1833 11574449
24. Stroebel RJ, Scheitel SM, Fitz JS, . A randomized trial of three diabetes registry implementation strategies in a community internal medicine practice. Jt Comm J Qual Improv 2002;28 :441–450 12174408
25. Canadian Alliance for Healthy Hearts and Minds First Nations Cohort Research Team. “All about us”: indigenous data analysis workshop—capacity building in the Canadian Alliance for Healthy Hearts and Minds First Nations cohort. CJC Open 2019;1 :282–288 32159122
26. Misra V, Sedig K, Dixon DR, Sibbald SL. Prioritizing coordination of primary health care. Can Fam Physician 2020;66 :399–403 32532718
27. Crowshoe LL, Henderson RI, Green ME, Jacklin KM, Walker LM, Calam B. Exploring Canadian physicians’ experiences with type 2 diabetes care for indigenous patients. Can J Diabetes 2018;42 :281–288 28821421
28. Minore B, Boone M, Katt M, Kinch P, Birch S, Mushquash C. The effects of nursing turnover on continuity of care in isolated First Nation communities. Can J Nurs Res 2005;37 :86–100
|
PMC010xxxxxx/PMC10338270.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
CD220085
10.2337/cd22-0085
Case Study
Weight Loss With Rising Blood Glucose: Challenges in Distinguishing Conventional Type 2 Diabetes From Pancreatic Cancer–Associated Hyperglycemia
Johannes Laura 1
https://orcid.org/0000-0002-4321-3992
Westcott Gregory P. 2
1 Independent medical journalist, Boston, MA
2 Division of Endocrinology, Diabetes & Metabolism, Beth Israel Deaconess Medical Center, Joslin Diabetes Center, and Harvard Medical School, Boston, MA
Corresponding author: Gregory Westcott, [email protected]
Summer 2023
20 1 2023
20 1 2023
41 3 477480
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
==== Body
pmcDiabetes is both a risk factor for pancreatic cancer and a potential early warning sign of a growing tumor. Approximately 1% of people with diabetes who are ≥50 years of age will be diagnosed with pancreatic cancer within 3 years of first meeting the criteria for diabetes, a rate that is eight times higher than would have been expected for people of similar age and sex in the general population (1). Distinguishing typical diabetes from hyperglycemia associated with pancreatic cancer can be difficult. There is no validated, widely available, noninvasive test for pancreatic cancer, and, unlike colon and breast cancer, there is no standard screening process. Additionally, given its low prevalence compared with type 2 diabetes, pancreatic cancer is often low on the list of differential diagnoses when patients present with hyperglycemia. While the lifetime risk of diabetes in U.S. adults is estimated at ∼40% (2), the lifetime risk of pancreatic cancer is 1.7% (3). However, the prevalence of hyperglycemia in patients with pancreatic adenocarcinoma may be as high as 75% (4), indicating a relationship between pancreatic cancer and diabetes.
In many cases, hyperglycemia may precede detection of a tumor and be the first clinical indication of malignancy, so hyperglycemia can provide a critical clue to diagnosis if it prompts early evaluation. Although surgical resection of pancreatic cancer offers the best chance at a cure, some 80% of pancreatic tumors are not detected until they are unresectable. The transition of a growing pancreatic tumor from resectable to unresectable occurs over a period of ∼6 months before diagnosis, suggesting that detection even as little as 6 months earlier would lead to an increase in the resectability rate (5).
Case Presentation
A 56-year-old female recreational cyclist presented to her primary care provider (PCP) with postprandial hyperglycemia, new-onset upper abdominal pain worse with fatty meals, and weight loss.
She had a history of type 2 diabetes, with an A1C of 7.4% 3 years before presentation. At that time, she had a BMI of 36.5 kg/m2 and subsequently lost 51 lb via lifestyle interventions, with a corresponding reduction in A1C to 5.0% and durable remission of diabetes.
At presentation, her A1C had risen to 6.1%, with associated postprandial glucose excursions into the 200-mg/dL range. Her fasting C-peptide concentration was 0.84 ng/mL and tests for pancreatic autoantibodies (GAD65, IA-2, and anti-insulin) were negative. Her PCP was initially concerned about gastritis and prescribed omeprazole. The patient had lost 23 lb in 11 weeks since the onset of abdominal pain, although this was thought to be explained by dietary modifications to address her gastritis. A DEXA body composition scan demonstrated that her weight loss had been entirely accounted for by loss of lean mass when compared with prior studies.
Four months after presentation, the patient reported that, despite a weight loss of an additional 5 lb, or a total of 28 lb since the abdominal pain began, a home A1C measurement was 6.4%, and she was recording fingerstick blood glucose values as high as 321 mg/dL. She was prescribed metformin.
Because of severe hyperglycemia from glycemic excursions, worsening abdominal pain, and pruritus, the patient presented to a local emergency department, where urinalysis revealed moderate bilirubin. In follow-up with the PCP, her A1C had risen to 7%, her serum C-peptide concentration was 1.39 ng/mL, and her blood glucose was 285 mg/dL. Insulin glargine 20 units daily was initiated. A Helicobacter pylori urea breath test was negative, and upper endoscopy was recommended. Given her high postprandial blood glucose levels, insulin aspart with meals was initiated, and metformin was discontinued because of abdominal pain and loose stools. Her fasting glucose levels and weight trends over time are shown in Figure 1.
FIGURE 1 Weight and fasting blood glucose trends from day of initial presentation (day 0). Blood glucose trendline represents 7-day running average.
The patient was subsequently noted by friends to appear jaundiced, and she presented to the emergency department, where an abdominal CT scan revealed a hypodense mass in the pancreatic head, highly concerning for malignancy, as well as several ill-defined lesions in the liver, enlarged lymph nodes, and small nodules in the lungs. She was admitted and received a biliary stent. The tumor was biopsied and found to be adenocarcinoma, which was deemed nonresectable, and she sought specialist care for her malignancy and diabetes.
Questions
When should pancreatic cancer be suspected in a patient with hyperglycemia?
How should hyperglycemia be evaluated and managed in patients with suspected pancreatic cancer?
Commentary
Distinguishing typical type 2 diabetes from tumor- related hyperglycemia can be challenging in clinical practice. This case highlights how difficult it is to make the connection between new-onset hyperglycemia and pancreatic cancer, as classic symptoms of cancer such as abdominal pain and weight loss can easily be attributed to other causes.
In certain clinical circumstances, new-onset diabetes or sudden and unexplained worsening of glucose control may be a red flag for possible underlying malignancy. As previously noted, ∼75% of pancreatic cancer patients present with either diabetes or impaired glucose tolerance. In this case, significant postprandial glucose excursions appear not to have been reflected by the A1C, which remained mostly in the prediabetes range despite intermittent blood glucose readings >300 mg/dL. The mechanisms by which pancreatic cancer contribute to hyperglycemia are not fully understood but appear to include both an increase in insulin resistance and an impaired β-cell response (6).
Type 2 diabetes is a gradually progressive disease and does not typically advance as rapidly or dramatically as seen in this case, and the worsening in blood glucose despite significant weight loss and a healthy low-glycemic diet should suggest a possible secondary cause. Given that patients are often instructed to lose weight, this sign can be overlooked by both patients and clinicians, but successful weight loss—16% of body weight in this case—should have resulted in a significant improvement in glycemia. The onset of abdominal pain, although nonspecific, is also not typical of type 2 diabetes and, in the setting of unexplained worsening glycemia, should be investigated more aggressively.
Despite the established link between pancreatic cancer and new-onset diabetes, the American Diabetes Association recommends against routine evaluation for pancreatic cancer in patients without clinical signs or symptoms such as weight loss or abdominal pain (7), and the low prevalence of the disease paired with the current lack of a low-cost screening modality limits the practicality of widespread screening. Clinical tools such as the Enriching New-Onset Diabetes for Pancreatic Cancer (END-PAC) score have been developed to aid clinicians in identifying patients who may warrant further workup (8), but this score is not in widespread clinical use and requires knowledge of blood glucose values before the onset of diabetes, which may limit its practicality. In this case, the patient had a complete remission of diabetes for 3 years before her diagnosis of cancer, and, at the time of diabetes relapse, an END-PAC score would have been 9, suggesting a high risk of malignancy (although this score has not been validated in the setting of diabetes relapse). The glucagon-to-insulin ratio has been proposed as a method for discriminating pancreatic cancer–associated hyperglycemia from type 2 diabetes (9), but the specificity of this approach is not adequate to be used in clinical practice, and this ratio was not determined in this patient. There is interest in developing better biomarkers to identify patients with new-onset hyperglycemia who are at high risk of pancreatic adenocarcinoma; a large clinical study is currently enrolling patients to determine the role of measuring carbohydrate antigen 19–9 and potentially other markers in this population (10), and Bluestar Genomics is developing a blood test to screen for pancreatic cancer based on DNA hydroxy-methylation signatures (11). For now, however, it is up to health care providers to clinically assess patients with a new diagnosis of diabetes or suddenly worsening diabetes for signs or symptoms of pancreatic cancer that warrant further evaluation.
In patients with suspected or confirmed pancreatic adenocarcinoma, it is important to recognize that the secretion of multiple islet hormones may be affected, with therapeutic implications. Extensive pancreatic destruction results in pancreatogenic, or type 3c, diabetes, which is characterized not only by decreased insulin production, but also by lower glucagon and pancreatic polypeptide levels (12). In terms of therapy, studies of particular antihyperglycemic agents, including incretin-based therapy (i.e., dipeptidyl peptidase 4 inhibitors and glucagon-like peptide 1 receptor agonists), generally have been unable to establish a strong causal relationship between any agent and pancreatic cancer (13). It may be helpful to estimate insulin secretory function at the time of cancer diagnosis by measuring fasting or stimulated C-peptide level to assess the need for insulin therapy, although this can also be determined based on a patient’s response to noninsulin therapy and the clinical scenario, as patients who are perioperative or have impaired renal or hepatic function may have limited therapeutic options. Note that patients with substantial metastatic liver involvement may have diminished gluconeogenic capacity and therefore may be prone to fasting hypoglycemia, and, rarely, paraneoplastic IGF-II production can result in noninsulin-mediated hypoglycemia (14), presenting management challenges.
In conclusion, risks of pancreatic cancer and diabetes are bidirectional; pancreatic cancer is a risk factor for new-onset diabetes, and longstanding type 2 diabetes is a risk factor for pancreatic cancer. Careful clinical assessment is critical at the time of diabetes diagnosis or a change in diabetes status that is characterized by atypical features such as unintentional weight loss, abdominal pain, or dramatic postprandial glucose excursions out of proportion to overall glycemic control, since early detection of pancreatic adenocarcinoma is likely to be lifesaving.
Clinical Pearls
At the time of diabetes diagnosis, worsening blood glucose despite significant weight loss with negative pancreatic autoantibodies, particularly in patients >65 years of age, should prompt assessment for secondary causes, including pancreatic cancer.
Abdominal pain in the context of new-onset or worsening diabetes should trigger expedient assessment.
Patients with pancreatic cancer and hyperglycemia can be treated with standard diabetes therapy, although there may be high risk for hypoglycemia, particularly in patients with substantial pancreatic destruction or resection.
Article Information
Duality of Interest
No potential conflicts of interest relevant to this article were reported.
Author Contributions
L.J. drafted the initial manuscript. G.P.W. critically reviewed and revised the manuscript. G.P.W. is the guarantor of the work and, as such, had full access to all data in the case study and takes responsibility for the contents of this article.
==== Refs
References
1. Chari ST, Leibson CL, Rabe KG, Ransom J, de Andrade M, Petersen GM. Probability of pancreatic cancer following diabetes: a population-based study. Gastroenterology 2005;129 :504–511 16083707
2. Gregg EW, Zhuo X, Cheng YJ, Albright AL, Narayan KMV, Thompson TJ. Trends in lifetime risk and years of life lost due to diabetes in the USA, 1985–2011: a modelling study. Lancet Diabetes Endocrinol 2014;2 :867–874 25128274
3. National Cancer Institute Surveilance, Epidemiology, and End Results Program. Cancer stat facts: pancreatic cancer. Available from https://seer.cancer.gov/statfacts/html/pancreas.html. Accessed 2 August 2022
4. Permert J, Ihse I, Jorfeldt L, von Schenck H, Arnqvist HJ, Larsson J. Pancreatic cancer is associated with impaired glucose metabolism. Eur J Surg 1993;159 :101–107 8098623
5. Chari ST. Detecting early pancreatic cancer: problems and prospects. Semin Oncol 2007;34 :284–294 17674956
6. Hart PA, Bellin MD, Andersen DK, .; Consortium for the Study of Chronic Pancreatitis, Diabetes, and Pancreatic Cancer (CPDPC). Type 3c (pancreatogenic) diabetes mellitus secondary to chronic pancreatitis and pancreatic cancer. Lancet Gastroenterol Hepatol 2016;1 :226–237 28404095
7. American Diabetes Association Professional Practice Committee. 4. Comprehensive medical evaluation and assessment of comorbidities: Standards of Medical Care in Diabetes—2022. Diabetes Care 2022;45 (Suppl. 1 ):S46–S59 34964869
8. Sharma A, Kandlakunta H, Nagpal SJS, . Model to determine risk of pancreatic cancer in patients with new-onset diabetes. Gastroenterology 2018;155 :730–739.e3 29775599
9. Kolb A, Rieder S, Born D, . Glucagon/insulin ratio as a potential biomarker for pancreatic cancer in patients with new-onset diabetes mellitus. Cancer Biol Ther 2009;8 :1527–1533 19571666
10. ClinicalTrials.gov. A study to establish a new onset hyperglycemia and diabetes cohort (NOD). Available from https://clinicaltrials.gov/ct2/show/NCT03731637. Accessed 2 August 2022
11. ClinicalTrials.gov. EpiDetect study: clinical validation of a pancreatic cancer detection test in new-onset diabetes patients. Available from https://clinicaltrials.gov/ct2/show/NCT05188573. Accessed 2 August 2022
12. Cui Y, Andersen DK. Pancreatogenic diabetes: special considerations for management. Pancreatology 2011;11 :279–294 21757968
13. Egan AG, Blind E, Dunder K, . Pancreatic safety of incretin-based drugs: FDA and EMA assessment. N Engl J Med 2014;370 :794–797 24571751
14. Dynkevich Y, Rother KI, Whitford I, . Tumors, IGF-2, and hypoglycemia: insights from the clinic, the laboratory, and the historical archive. Endocr Rev 2013;34 :798–826 23671155
|
PMC010xxxxxx/PMC10338274.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
36920750
CD220128
10.2337/cd22-0128
Practical Pointers
Personalized Virtual Care Using Continuous Glucose Monitoring in Adults With Type 2 Diabetes Treated With Less Intensive Therapies
Reddy Sushma
Wu Calvin C.
José Aimée
Hsieh Jennifer L.
Rautela Shetal Desai
Carbon Health Virtual Diabetes Care, Oakland CA
Corresponding author: Sushma Reddy, [email protected]
Summer 2023
15 3 2023
15 3 2023
41 3 452457
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
Abbott Diabetes Care provided funding for the development of this manuscript
==== Body
pmcDuring the past 5 years, an increasing number of individuals with diabetes have transitioned from traditional blood glucose monitoring (BGM) to continuous glucose monitoring (CGM). Numerous studies have demonstrated the safety and efficacy of CGM use in improving overall glycemic control in type 1 diabetes and intensively treated type 2 diabetes (1–12).
When used in conjunction with virtual telehealth visits and telemonitoring interventions, CGM not only improves glycemic management (13–18), but has also been shown to reduce diabetes-related distress (19) and increase medication adherence (20). Importantly, use of CGM in conjunction with telehealth visits and remote monitoring can serve to connect patients remotely with their health care providers and facilitate timely clinical care (21–23).
Despite innovations in glucose monitoring and other diabetes technologies, suboptimal glycemic control persists (24–26) and continues to worsen (27) among a substantial percentage of individuals with type 2 diabetes, ∼95% of whom are treated with basal insulin only, noninsulin medications, and/or lifestyle interventions (28–30). As reported in numerous studies, therapeutic inertia is a significant driver of poor diabetes management (31–36). Several factors contribute to therapeutic inertia, including lack of sufficient data to make clinical decisions, restrictions on clinicians’ time and resources, lack of training/education, and patient-level obstacles to treatment adherence (37).
Carbon Health (San Francisco, CA) recently launched a virtual, endocrinologist-led diabetes program that addresses these factors. In this article, we discuss how this approach was used to personalize therapy and improve outcomes in two patients with newly diagnosed type 2 diabetes treated with nonintensive therapies.
Evidence for CGM Use in Nonintensively Treated Type 2 Diabetes
Large database analyses have demonstrated strong associations between CGM use and significant reductions in A1C, acute diabetes events, and diabetes- related hospitalizations in people with diabetes treated with basal insulin only or noninsulin medications (38,39). Although these types of studies have limitations, their findings are supported by one randomized controlled trial (40) and smaller observational cohort studies (18,19,23,41).
In a large retrospective database study of 1,034 adults with nonintensively treated type 2 diabetes, Wright et al. (38) reported a significant A1C reduction (from 10.1% to 8.6%, P <0.001) with the use of CGM. Importantly, individuals treated with noninsulin medications showed a greater A1C reduction compared with those treated with basal insulin only (−1.6 vs. −1.1%, respectively, P <0.001).
A similar database analysis by Miller et al. (39) reported significant reductions in acute diabetes events (ADEs) and all-cause hospitalizations (ACHs) among 10,282 adults with nonintensively treated type 2 diabetes using CGM. As reported, the ADE rate decreased from 0.076 to 0.052 events per patient-year (P <0.001). The rate of ACHs decreased from 0.177 to 0.151 events per patient-year (P = 0.002) during the 6 months after acquisition of a CGM system.
In the randomized, controlled MOBILE study, Martens et al. (40) followed 175 adults with type 2 diabetes treated with basal insulin with or without noninsulin medications who were randomized to CGM or BGM. Mean baseline A1C was 9.1%. At 8 months, A1C decreased 1.1% in the CGM group compared with a 0.5% reduction in the BGM group (P = 0.02). The mean percentage of time in the established target glucose range (time in range [TIR]) of 70–180 mg/dL (42) was 59% with CGM versus 43% with BGM. Although neither group achieved the recommended goal of >70% TIR, the difference was significant (P <0.001).
More recently, Grace and Salyer (41) reported findings from a 6-month, prospective, interventional, single-arm study of 38 adults with type 2 diabetes and a mean baseline A1C of 10.1%. Twenty-two (58%) were treated with noninsulin medications, and 16 were treated with basal insulin with or without noninsulin medications. At 6 months, mean A1C decreased to 7.3% (P <0.001), with significant reductions in body weight (−3.1 kg, P = 0.002) and significant increases in TIR, from 57.0% at baseline to 72.2% at study end.
Persistent CGM use (>6 days/week) is generally recommended for individuals treated with intensive insulin regimens (42). However, this level of CGM use may not be necessary for those treated with less intensive regimens.
A recent observational study of 594 adults with type 2 diabetes who were enrolled in the Onduo Virtual Diabetes Care (VDC) clinic found that this cohort achieved significant A1C reductions (mean 0.6%, P <0.001) with intermittent use of CGM supported by coaching and clinical advice delivery via telehealth visits (23). Survey results revealed that 94.7% of respondents agreed or strongly agreed that they were comfortable inserting the sensor via remote training, and the majority agreed or strongly agreed that intermittent use of CGM improved their understanding of the impact of eating (97.0%), increased their diabetes knowledge (95.7%), and helped them improve their diabetes control when not wearing the CGM sensor (79.4%). Additional surveys of the Onduo cohort have found similar glycemic improvements (18,43), and one study by Polonsky et al. (19) showed associations between participation in the VDC and reductions in diabetes-related distress.
Clinical Setting
Carbon Health designed its endocrinologist-led program to integrate with primary care physicians (PCPs) throughout its primary care and urgent care (UC) network. The program uses virtual telehealth visits, telemonitoring of CGM data, nutrition and physical activity counseling, and support in making other lifestyle changes through a completely virtual platform. A specialized care team of endocrinologists and diabetes care and education specialists (DCESs) interprets the data and creates individualized treatment plans optimized for each patient’s health status, lifestyle, and preferences. The digital delivery of this care provides patients with greater accessibility, with no waiting or travel time involved. The following case examples demonstrate how we used this approach to address the needs of two patients with newly diagnosed type 2 diabetes.
Case Example 1
At presentation, patient 1 was a 56-year-old man with a BMI of 31.2 kg/m2, weight of 97.8 kg, fasting glucose of 431 mg/dL, and blood pressure of 136/78 mmHg. On 1 November 2021, the patient presented to his PCP with fatigue and polyuria. Laboratory results showed that his A1C was 12.9%. The PCP provided basic education on diabetes management, started the patient on metformin 1,000 mg once daily, and prescribed CGM, which was initiated the next day with an intermittently scanned CGM system.
On 16 November, he returned to his PCP for follow-up. His blood pressure was significantly lower (110/66 mmHg) and his albumin-to-creatinine ratio was normal. The patient explained that he was no longer consuming sugar, had switched to a low-carbohydrate diet (≤35 g at each meal), and was exercising almost daily. He indicated that he did not want to do fingerstick BGM testing and liked being able to see his glucose levels at any time with CGM. His PCP increased his metformin dose to 1,000 mg twice daily; however, the patient declined treatment with an angiotensin receptor blocker for high blood pressure or a statin for elevated lipids.
His CGM glucose profiles revealed notable improvements in glycemic control during the first 4 days of sensor wear (Figure 1A), which corresponded with daily progressive reductions in average glucose of 275, 227, 180, and 167 mg/dL on 2–5 November, respectively. These improvements continued throughout the next week (Figure 1B). During the same period, his TIR increased from 63.2 to 99.4%.
FIGURE 1 Glucose profiles during patient 1’s first 4 days of CGM use (A) and during the subsequent week (B).
Additional history was obtained when the patient met with the endocrinologist on 30 November 2021 in a telehealth visit. The endocrinologist learned that the patient had been “addicted to soda pop” (drinking six bottles per day) and had his glucose checked annually because, he said, he “knew he would get diabetes some day.” The patient explained that he had given away all of his candy and kept only one can of sugar-sweetened soft drink for possible episodes of low glucose. Additionally, the patient said he had stayed at a health and wellness retreat and met with a nutritionist and physician to learn how to eat more healthfully. The patient also expressed interest in using health-related technologies and wears a sleep and fitness tracking ring, which provides personalized health metrics. Laboratory results obtained in February 2022 revealed that the patient’s A1C had decreased from 12.9% to 6.7%.
Case Example 2
At presentation, patient 2 was a 54-year-old man with a BMI of 24.2 kg/m2 and weight of 63.5 kg. On 6 April 2021, he presented at a Carbon Health UC clinic complaining of pain in his left hand with cramping and numbness, which had started 3 weeks before the visit. The patient also described symptoms of polyuria and polydipsia. Random glucose was 467 mg/dL. The UC provider scheduled a virtual telehealth visit with the endocrinologist for the next day.
On 7 April, the endocrinologist provided education on the pathophysiology of diabetes and appropriate lifestyle changes and prescribed the patient CGM. Laboratory results showed that the patient’s A1C was 14.5%. The patient was started on metformin 750 mg extended release (ER) once daily and insulin glargine 15 units (0.24 units/kg) once daily to treat the glucotoxicity. A telehealth visit with the DCES was scheduled for the next day.
At that visit, the DCES reviewed proper insulin injection technique, instructed the patient on CGM use and sensor placement, and provided additional nutrition guidance. The patient started an intermittently scanned CGM system on 11 April.
At the follow-up visit 4 days later, the patient reported that he had discontinued the insulin on 11 April. As shown in Figure 2, the patient’s glucose profile showed significant improvement, with 99% of CGM values within the target range.
FIGURE 2 Patient 2’s glucose profile during the first 2 weeks of CGM use.
Throughout April, the DCES kept in touch with the patient through frequent messaging and provided patient-appropriate articles about various aspects of diabetes self-management. At a follow-up visit with the endocrinologist on 1 August, the patient’s A1C was 5.8%, a decrease of 8.7% from his initial visit.
Discussion
Through use of telehealth visits in conjunction with CGM data, our diabetes program has been effective in improving glycemic outcomes in our patients with type 2 diabetes. Our patients with a baseline A1C >9.0% have achieved an average 4.3% reduction through participation in our program. Our approach allows us to provide comprehensive, personalized diabetes care that is timely and efficient for both patients and our health care team. Whereas the average wait time for an initial nonurgent endocrinologist consultation is 37 days (44), patient 2 was able to meet virtually with an endocrinologist 1 day after being seen by the UC provider. Importantly, our approach facilitates personalized care in a way that allows patients to engage with their self-management.
With patient 1, engagement was immediate. Although initiating metformin effectively lowered his overall glucose levels throughout the day, the notable reductions in glycemic spikes and rapid improvement in TIR suggests that he quickly recognized the causes and effects of his eating pattern and physical activity level and responded appropriately with significant lifestyle changes. Moreover, given his significant improvement in overall diabetes management and his affinity for using health technologies, it is likely that he will persist in using CGM.
With patient 2, we recognized his need for ongoing follow-up, education, and reassurance. We were able to provide this with two telehealth visits with the DCES and frequent messaging during his first month of diabetes care, which reinforced his successes and kept him engaged with his diabetes self-management.
Use of CGM data is an essential component of our approach to care. As we have seen with our patients, CGM facilitates their understanding of diabetes and promotes needed changes in their lifestyle behaviors in addition to providing our team with meaningful information to guide therapy decisions. The patients described here traditionally would have required multiple medications and/or insulin, but with the use of this care model, they achieved optimal glycemic control within 2 weeks with metformin alone by incorporating CGM into their regimens.
Based on our experiences, we believe that CGM should be an integral part of the treatment algorithm for individuals with type 2 diabetes regardless of their treatment regimen. Unfortunately, many individuals with diabetes who are treated with nonintensive therapies are denied access to CGM because of overly restrictive insurance coverage policies. Given the significant and increasing worldwide prevalence of type 2 diabetes, it only makes sense that current and future diabetes technologies be made available to all people with diabetes regardless of type and therapy.
Article Information
Acknowledgments
The authors thank Christopher G. Parkin, MS, of CGParkin Communications, Inc., for his assistance in developing this article.
Funding
Abbott Diabetes Care provided funding for the development of this article.
Duality of Interest
S.R. receives speaker fees from Dexcom. J.L.H. receives speaker fees from Dexcom and Eli Lilly. No other potential conflicts of interest relevant to this article were reported.
Author Contributions
S.R. and C.C.W. conceptualized the article and contributed to writing, reviewing, and editing the manuscript. All authors curated the data. S.R. is the guarantor of this work and, as such, takes responsibility for the integrity of the data and the accuracy of the information included in the article.
==== Refs
References
1. Lind M, Polonsky W, Hirsch IB, . Continuous glucose monitoring vs conventional therapy for glycemic control in adults with type 1 diabetes treated with multiple daily insulin injections: the GOLD randomized clinical trial. JAMA 2017;317 :379–387 28118454
2. Beck RW, Riddlesworth T, Ruedy K, .; DIAMOND Study Group. Effect of continuous glucose monitoring on glycemic control in adults with type 1 diabetes using insulin injections: the DIAMOND randomized clinical trial. JAMA 2017;317 :371–378 28118453
3. Ruedy KJ, Parkin CG, Riddlesworth TD; DIAMOND Study Group. Continuous glucose monitoring in older adults with type 1 and type 2 diabetes using multiple daily injections of insulin: results from the DIAMOND trial. J Diabetes Sci Technol 2017;11 :1138–1146 28449590
4. Šoupal J, Petruželková L, Flekač M, . Comparison of different treatment modalities for type 1 diabetes, including sensor-augmented insulin regimens, in 52 weeks of follow-up: a COMISAIR study. Diabetes Technol Ther 2016;18 :532–538 27482825
5. Šoupal J, Petruželková L, Grunberger G, . Glycemic outcomes in adults with T1D are impacted more by continuous glucose monitoring than by insulin delivery method: 3 years of follow-up from the COMISAIR study. Diabetes Care 2020;43 :37–43 31530663
6. Kröger J, Fasching P, Hanaire H. Three European retrospective real-world chart review studies to determine the effectiveness of flash glucose monitoring on HbA1c in adults with type 2 diabetes. Diabetes Ther 2020;11 :279–291 31833041
7. Charleer S, De Block C, Van Huffel L, . Quality of life and glucose control after 1 year of nationwide reimbursement of intermittently scanned continuous glucose monitoring in adults living with type 1 diabetes (FUTURE): a prospective observational real-world cohort study. Diabetes Care 2020;43 :389–397 31843948
8. Fokkert M, van Dijk P, Edens M, . Improved well-being and decreased disease burden after 1-year use of flash glucose monitoring (FLARE-NL4). BMJ Open Diabetes Res Care 2019;7 :e000809
9. Tyndall V, Stimson RH, Zammitt NN, . Marked improvement in HbA1c following commencement of flash glucose monitoring in people with type 1 diabetes. Diabetologia 2019;62 :1349–1356 31177314
10. Paris I, Henry C, Pirard F, Gérard A-C, Colin IM. The new FreeStyle Libre flash glucose monitoring system improves the glycaemic control in a cohort of people with type 1 diabetes followed in real-life conditions over a period of one year. Endocrinol Diabetes Metab 2018;1 :e00023 30815557
11. Pintus D, Ng SM. FreeStyle Libre flash glucose monitoring improves patient quality of life measures in children with type 1 diabetes mellitus (T1DM) with appropriate provision of education and support by healthcare professionals. Diabetes Metab Syndr 2019;13 :2923–2926 31425957
12. Charleer S, Mathieu C, Nobels F, .; RESCUE Trial Investigators. Effect of continuous glucose monitoring on glycemic control, acute admissions, and quality of life: a real-world study. J Clin Endocrinol Metab 2018;103 :1224–1232 29342264
13. Faruque LI, Wiebe N, Ehteshami-Afshar A, .; Alberta Kidney Disease Network. Effect of telemedicine on glycated hemoglobin in diabetes: a systematic review and meta-analysis of randomized trials. CMAJ 2017;189 :E341–E364 27799615
14. Salehi S, Olyaeemanesh A, Mobinizadeh M, Nasli-Esfahani E, Riazi H. Assessment of remote patient monitoring (RPM) systems for patients with type 2 diabetes: a systematic review and meta-analysis. J Diabetes Metab Disord 2020;19 :115–127 32550161
15. Tchero H, Kangambega P, Briatte C, Brunet-Houdard S, Retali GR, Rusch E. Clinical effectiveness of telemedicine in diabetes mellitus: a meta-analysis of 42 randomized controlled trials. Telemed J E Health 2019;25 :569–583 30124394
16. Wang X, Shu W, Du J, . Mobile health in the management of type 1 diabetes: a systematic review and meta-analysis. BMC Endocr Disord 2019;19 :21 30760280
17. Charpentier G, Benhamou PY, Dardari D, .; TeleDiab Study Group. The Diabeo software enabling individualized insulin dose adjustments combined with telemedicine support improves HbA1c in poorly controlled type 1 diabetic patients: a 6-month, randomized, open-label, parallel-group, multicenter trial (TeleDiab 1 Study). Diabetes Care 2011;34 :533–539 21266648
18. Dixon RF, Zisser H, Layne JE, . A virtual type 2 diabetes clinic using continuous glucose monitoring and endocrinology visits. J Diabetes Sci Technol 2020;14 :908–911 31762302
19. Polonsky WH, Layne JE, Parkin CG, . Impact of participation in a virtual diabetes clinic on diabetes-related distress in individuals with type 2 diabetes. Clin Diabetes 2020;38 :357–362 33132505
20. Lee WC, Balu S, Cobden D, Joshi AV, Pashos CL. Medication adherence and the associated health-economic impact among patients with type 2 diabetes mellitus converting to insulin pen therapy: an analysis of third-party managed care claims data. Clin Ther 2006;28 :1712–1725; discussion 1710–1711 17157128
21. Gal RL, Cohen NJ, Kruger D, . Diabetes telehealth solutions: improving self-management through remote initiation of continuous glucose monitoring. J Endocr Soc 2020;4 :bvaa076 32864542
22. Vigersky RA, Velado K, Zhong A, Agrawal P, Cordero TL. The effectiveness of virtual training on the MiniMed 670G system in people with type 1 diabetes during the COVID-19 pandemic. Diabetes Technol Ther 2021;23 :104–109 32678672
23. Bergenstal RM, Layne JE, Zisser H, . Remote application and use of real-time continuous glucose monitoring by adults with type 2 diabetes in a virtual diabetes clinic. Diabetes Technol Ther 2021;23 :128–132 33026839
24. Carls G, Huynh J, Tuttle E, Yee J, Edelman SV. Achievement of glycated hemoglobin goals in the US remains unchanged through 2014. Diabetes Ther 2017;8 :863–873 28646411
25. Bin Rakhis SA Sr, AlDuwayhis NM, Aleid N, AlBarrak AN, Aloraini AA. Glycemic control for type 2 diabetes mellitus patients: a systematic review. Cureus 2022;14 :e26180 35891859
26. Ceriello A, deValk HW, Guerci B, . The burden of type 2 diabetes in Europe: current and future aspects of insulin treatment from patient and healthcare spending perspectives. Diabetes Res Clin Pract 2020;161 :108053 32035117
27. Fang M, Wang D, Coresh J, Selvin E. Trends in diabetes treatment and control in U.S. adults, 1999–2018. N Engl J Med 2021;384 :2219–2228 34107181
28. International Diabetes Federation. IDF Diabetes Atlas 2021. 10th ed. Available from https://diabetesatlas.org/atlas/tenth-edition. Accessed 31 January 2022
29. World Health Organization. Diabetes. Available from https://www.who.int/news-room/fact-sheets/detail/diabetes. Accessed 25 November 2022
30. Basu S, Yudkin JS, Kehlenbrink S, . Estimation of global insulin use for type 2 diabetes, 2018-30: a microsimulation analysis. Lancet Diabetes Endocrinol 2019;7 :25–33 30470520
31. Ampudia-Blasco FJ, Palanca A, Trillo JL, Navarro J, Real JT. Therapeutic inertia in patients with type 2 diabetes treated with non-insulin agents. J Diabetes Complications 2021;35 :107828 33436143
32. Andreozzi F, Candido R, Corrao S, . Clinical inertia is the enemy of therapeutic success in the management of diabetes and its complications: a narrative literature review. Diabetol Metab Syndr 2020;12 :52 32565924
33. Khunti K, Gomes MB, Pocock S, . Therapeutic inertia in the treatment of hyperglycaemia in patients with type 2 diabetes: A systematic review. Diabetes Obes Metab 2018;20 :427–437 28834075
34. Khunti K, Millar-Jones D. Clinical inertia to insulin initiation and intensification in the UK: A focused literature review. Prim Care Diabetes 2017;11 :3–12 27727005
35. Khunti K, Nikolajsen A, Thorsted BL, Andersen M, Davies MJ, Paul SK. Clinical inertia with regard to intensifying therapy in people with type 2 diabetes treated with basal insulin. Diabetes Obes Metab 2016;18 :401–409 26743666
36. Pantalone KM, Misra-Hebert AD, Hobbs TM, . Clinical inertia in type 2 diabetes management: evidence from a large, real-world data set. Diabetes Care 2018;41 :e113–e114 29678811
37. American Diabetes Association. Summary of proceedings of the American Diabetes Association summit “Overcoming Therapeutic Inertia: Accelerating Diabetes Care FOR_LIFE.” Available from https://professional.diabetes.org/sites/default/files/media/ada_therapeutic_inertia_interior_final.pdf. Accessed 12 April 2020
38. Wright EE Jr, Kerr MSD, Reyes IJ, Nabutovsky Y, Miller E. Use of flash continuous glucose monitoring is associated with A1C reduction in people with type 2 diabetes treated with basal insulin or noninsulin therapy. Diabetes Spectr 2021;34 :184–189 34149259
39. Miller E, Kerr MSD, Roberts GJ, Nabutovsky Y, Wright E. Flash CGM associated with event reduction in nonintensive diabetes therapy. Am J Manag Care 2021;27 :e372–e377 34784145
40. Martens T, Beck RW, Bailey R, .; MOBILE Study Group. Effect of continuous glucose monitoring on glycemic control in patients with type 2 diabetes treated with basal insulin: a randomized clinical trial. JAMA 2021;325 :2262–2272 34077499
41. Grace T, Salyer J. Use of real-time continuous glucose monitoring improves glycemic control and other clinical outcomes in type 2 diabetes patients treated with less intensive therapy. Diabetes Technol Ther 2022;24 :26–31 34524013
42. Battelino T, Danne T, Bergenstal RM, . Clinical targets for continuous glucose monitoring data interpretation: recommendations from the International Consensus on Time in Range. Diabetes Care 2019;42 :1593–1603 31177185
43. Majithia AR, Kusiak CM, Armento Lee A, . Glycemic outcomes in adults with type 2 diabetes participating in a continuous glucose monitor–driven virtual diabetes clinic: prospective trial. J Med Internet Res 2020;22 :e21778 32856597
44. Vigersky RA, Fish L, Hogan P, . The clinical endocrinology workforce: current status and future projections of supply and demand. J Clin Endocrinol Metab 2014;99 :3112–3121 24940655
|
PMC010xxxxxx/PMC10338275.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
CD220099
10.2337/cd22-0099
Feature Articles
Severe Hypoglycemia and the Use of Glucagon Rescue Agents: An Observational Survey in Adults With Type 1 Diabetes
https://orcid.org/0000-0003-1054-7096
Hughes Allyson S. 1
Chapman Katherine S. 2
Nguyen Huyen 2
Liu Jingwen 2
Bispham Jeoffrey 2
Winget Melissa 3
Weinzimer Stuart A. 4
Wolf Wendy A. 2
1 Department of Primary Care, Ohio University Heritage College of Osteopathic Medicine, Athens, OH
2 T1D Exchange, Boston, MA
3 Zealand Pharma A/S, Boston, MA
4 Department of Pediatrics, Yale School of Medicine, New Haven, CT
Corresponding author: Allyson S. Hughes, [email protected]
Summer 2023
06 3 2023
06 3 2023
41 3 399410
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
Severe hypoglycemia (SH) is the most frequent and potentially serious complication affecting individuals with type 1 diabetes and can have major clinical and psychosocial consequences. Glucagon is the only approved treatment for SH that can be administered by non–health care professionals (HCPs); however, reports on the experiences and emotions of people with type 1 diabetes associated with SH and glucagon rescue use are limited. This survey study demonstrated that an increasing number of individuals with type 1 diabetes have current and filled prescriptions for glucagon and have been educated about glucagon rescue use by an HCP. Despite this positive trend, challenges with SH remain, including a high level of health care resource utilization, considerable out-of-pocket expenses for glucagon kits, a high prevalence of hypoglycemia unawareness, and a negative emotional impact on individuals with diabetes. Nocturnal and exercise-related hypoglycemia were concerns for most survey participants.
Zealand Pharma A/S
==== Body
pmcInsulin therapy carries an increased risk of hypoglycemia, a significant and potentially fatal complication of diabetes management (1–4). The American Diabetes Association (ADA) defines level 2 hypoglycemia as a blood glucose level <54 mg/dL (3.0 mmol/L). It is associated with impaired counterregulatory mechanisms that may lead to reduced awareness of hypoglycemic events, potentially leading to a cycle of recurrent hypoglycemia (3,4). Level 3 hypoglycemia (also called severe hypoglycemia [SH]) is defined by an altered mental and/or physical status that requires assistance from another person for recovery (3). In the United States, SH results in almost 300,000 emergency department visits per year (5). Untreated SH may lead to adverse clinical outcomes that can include cognitive dysfunction, loss of consciousness, seizures, coma, and death (2,3,6).
Long-term clinical studies have shown that SH remains an ongoing challenge for people with type 1 diabetes, occurring at an average rate of 0.36–0.41 episodes per person per year (7). A history of SH is a strong indicator for the risk of subsequent SH events (7), and people with type 1 diabetes are likely to experience several SH events in their lifetime. A clinical history of SH is associated with an increased risk of all-cause mortality and cardiovascular disease (CVD) in people with type 1 diabetes (8,9), which may lead to substantial direct and indirect costs (10–12) and constitutes a significant part of the economic burden associated with diabetes (13–16). Hypoglycemia can also significantly compromise a person’s emotional state and quality of life (2,3).
The potential negative impacts of SH are significant sources of distress for many individuals with type 1 diabetes and may create a strong aversion to situations in which SH might occur. This fear of hypoglycemia is prevalent among individuals with type 1 diabetes and may have substantial implications for diabetes management and subsequent health outcomes (3,17). Many people with diabetes who are worried about hypoglycemia maintain their blood glucose at higher than recommended levels to avoid the adverse effects of hypoglycemic episodes (18), thus contributing to the metabolic complications of hyperglycemia.
Glucagon has been the first-line emergency treatment for SH in insulin-treated people with diabetes since the 1960s (19). The ADA recommends prescribing glucagon for all individuals at an increased risk of level 2 and level 3 hypoglycemia so that it is available when needed (3). The ability to administer glucagon for the treatment of SH is not limited to health care professionals (HCPs); it can be administered by family members, caregivers, or bystanders (3). The significant burden of SH could be partially alleviated with the use of glucagon as recommended by the ADA (3). However, despite its favorable safety and efficacy profile, glucagon is underutilized as a rescue treatment for SH owing to a lack of routine prescribing of glucagon among HCPs, people with diabetes not filling their glucagon prescriptions, and caregivers not administering or not correctly administering glucagon in the event of SH (5,20–23). First-generation glucagon kits (24,25) require reconstitution of lyophilized native human glucagon before use. The complexity of this multistep process is a known barrier to the rapid and efficient administration of glucagon (21,26,27). Other barriers include lack of appropriate education and training (21,22,26,27) and reluctance to administer rescue glucagon in an emergency for fear of harming the individual with SH (21,22,27).
Over many decades, there has been an unmet need for easy-to-use glucagon products to treat SH to aid in alleviating fear related to hypoglycemia, reduce training needs, and increase overall utilization of glucagon (27). Second-generation glucagon products do not require reconstitution and have been available on the U.S. market since 2019. These products include Baqsimi (Eli Lilly and Company) (28), a native human glucagon powder for nasal administration; Gvoke (Xeris Pharmaceuticals) (29), a ready-to-use premixed solution of native human glucagon for subcutaneous injection; and Zegalogue (Zealand Pharma) (30), a ready-to-use pre-mixed solution of the glucagon analog dasiglucagon for subcutaneous injection.
To date, there has been limited research on the experiences and emotions of individuals with diabetes regarding hypoglycemia and glucagon rescue use. This dearth of research is concerning considering the potentially severe consequences that a lack of glucagon use can have for SH recovery (19). To address these questions, we surveyed adults with type 1 diabetes to gain a real-world understanding of their experiences with hypoglycemia and the use of glucagon rescue products, identify barriers to glucagon use, and determine their preferences for glucagon rescue product characteristics. The objective of this study was to provide an updated real-world perspective on how the landscape of emergency glucagon use has changed since the introduction of second-generation, ready-to-use glucagon products in 2019.
Research Design and Methods
Focus Groups Study to Inform Survey Development
The survey questions for this research were informed by focus groups conducted virtually from October to December 2020. In total, 38 adults with type 1 diabetes consented and participated. Seven focus groups with five to seven participants each were asked about their experiences with hypoglycemia and glucagon use. Participants were encouraged to interact with each other to gain robust insights. Each 90-minute session was recorded and transcribed for analysis. Each focus group participant was remunerated with a $100 Amazon gift card. The questions included in the survey for this study were developed from the key concepts that emerged during these focus groups.
Recruitment
Participants for the survey were recruited through the T1D Exchange registry (31–33), a longitudinal study collecting information on type 1 diabetes management and outcomes. To be enrolled in the registry, individuals with type 1 diabetes must either receive insulin or have had a pancreas or islet cell transplant. In May 2021, recruitment emails were sent to the adult members of the T1D Exchange registry community, and posts were set up on social media (Facebook and Twitter) containing a link directing potential participants to a brief description of the research study.
Informed Consent
An institutional review board (IRB) exemption as well as a partial waiver of the Health Insurance Portability and Accountability Act authorization signature requirement for use and disclosure of protected health information were obtained from the WCG IRB on 22 April 2021. Potential study participants from the T1D Exchange registry community were provided with a link to an electronic consent form for their review and electronic signature. Participants who provided consent were then emailed a copy of their informed consent form for their records.
Inclusion Criteria
Participants enrolled in the study were included in the data analysis if they satisfied the following criteria: ≥18 years of age, diagnosis of type 1 diabetes for at least 1 year, at least one episode of SH defined as a low blood glucose event during which assistance was required, familiarity with current glucagon options on the market, currently residing in the United States, fluent in written and spoken English, and agreed to provide digital informed consent. Participants were excluded from the analysis if they were currently pregnant or refused to declare whether they were pregnant.
Survey
Consenting participants were directed to the online survey conducted via Alchemer (34). After completing the survey, each participant was remunerated with a $25 Amazon gift card. For the purposes of the survey, participants’ experiences with hypoglycemia were classified as “mild” (blood glucose <70 mg/dL [3.9 mmol/L] to ≥54 mg/dL [3.0 mmol/L]), “moderate” (blood glucose <54 mg/dL [3.0 mmol/L]), and “severe” (an altered mental and/or physical status requiring assistance for treatment of hypoglycemia). Definitions of these classifications were provided to participants as they completed the survey. Some survey questions allowed for multiple answers to be given, resulting in an overall percentage in excess of 100%. These questions are noted in the results tables.
Statistical Analysis
All statistical analyses were performed using R software, v. 4.0.5 or later (R Core Team, Vienna, Austria). Summary statistics, including mean, SD, frequency, and percentage, were calculated for general demographic and diabetes-related health information. Bivariate statistical analyses were performed using Welch two-sample t tests, Pearson χ2 tests, and Fisher exact tests to assess differences in demographic and clinical characteristics between participants who, at the time of the survey, had a current and filled glucagon prescription and those who did not and between participants who had impaired awareness of hypoglycemia and those who did not. A P value ≤0.05 was considered statistically significant.
Results
Participants’ Demographics and Clinical Characteristics
In total, 428 individuals consented and were enrolled in the study. Of these, 316 individuals met the inclusion criteria and completed the survey. The baseline participant characteristics are shown in Table 1.
TABLE 1 Participants’ Demographics (N = 316)
Characteristic Value
Age, years 35.6 ± 8.9
Duration of type 1 diabetes, years 17.2 ± 12.7
Gender
Female
Male
Other
187 (59.2)
128 (40.5)
1 (0.3)
Race (multiple answers possible)
White
Black or African American
Asian
American Indian or Alaskan Native
Native Hawaiian or other Pacific Islander
Other
290 (91.8)
12 (3.8)
9 (2.8)
8 (2.5)
1 (0.3)
4 (1.3)
Ethnicity
Not Hispanic or Latino
Hispanic or Latino
302 (95.6)
14 (4.4)
Employment status (multiple answers possible)
Working full time
Working part time
Student
Temporarily unemployed or on leave from work
Unemployed, looking for work
Volunteer
Disabled
Unemployed, not looking for work
Retired
195 (61.7)
65 (20.6)
36 (11.4)
13 (4.1)
12 (3.8)
9 (2.8)
8 (2.5)
6 (1.9)
2 (0.6)
Health insurance (multiple answers possible)
Private health insurance (e.g., commercial, fee-for-service, HMO, PPO, or POS)
Medicare
Medicaid
Other government-sponsored health coverage plan
Affordable Care Act plan
Medigap
Single service plan (e.g., dental, vision, or prescriptions)
Military health care (e.g., TRICARE, CHAMPUS, CHAMPVA, or VA)
Other state-sponsored health coverage plan
No health insurance or health care coverage of any type
Not known
187 (59.2)
71 (22.5)
34 (10.8)
12 (3.8)
8 (2.5)
7 (2.2)
6 (1.9)
6 (1.9)
5 (1.6)
4 (1.3)
1 (0.3)
Education
Bachelor’s degree
Some college
Master’s degree
Associate’s degree
Doctoral degree
High school graduate/GED
Some high school
111 (35.1)
77 (24.4)
57 (18.0)
33 (10.4)
14 (4.4)
12 (3.8)
12 (3.8)
Annual household income
<$30,000
$30,000 to <$50,000
$50,000 to <$75,000
$75,000 to <$100,000
$100,000 to <$200,000
≥$200,000
Do not wish to provide
Not known
44 (13.9)
41 (13.0)
85 (26.9)
53 (16.8)
54 (17.1)
18 (5.7)
12 (3.8)
9 (2.8)
Data are mean ± SD or n (%). CHAMPUS, Civilian Health and Medical Program of the Uniformed Services; CHAMPVA, Civilian Health and Medical Program of the Department of Veterans Affairs; GED, general education diploma; HMO, health maintenance organization; POS, point of service; PPO, preferred provider organization; VA, Veterans Affairs.
Participants had a mean ± SD age of 35.6 ± 8.9 years and a mean duration of type 1 diabetes of 17.2 ± 12.7 years since their diagnosis. Of the participants, 59.2% were female, most were White and identified as not Hispanic or Latino, and the majority were in full-time or part-time employment. Overall, 59.2% had private health insurance; 22.5% and 10.8% were enrolled in Medicare or Medicaid, respectively; 1.3% did not have health insurance of any type; and the remainder received health insurance from other sources. In terms of education, 57.5% reported having a bachelor’s degree or higher, and 66.5% had an annual household income ≥$50,000.
The clinical characteristics of participants are summarized in Table 2. The most frequently used method of glucose monitoring was real-time continuous glucose monitoring (CGM; 80.4%), followed by traditional fingerstick blood glucose monitoring (BGM; 47.2%) and intermittently scanning CGM, formerly called flash CGM (FGM; 0.9%). Most participants (76.6%) reported using an insulin pump, 30.4% used injectable insulin, and 4.1% used inhalable insulin. Participants reported comorbid conditions, the most frequent of which were joint issues (27.5%), hypothyroidism (25.3%), CVD (19.3%), and retinopathy (19.3%). Overall, 69.3% were seen by an adult or pediatric endocrinologist, and 22.8% were seen by other specialized diabetes HCPs, with the remainder seen by primary care HCPs.
TABLE 2 Participants’ Clinical Characteristics (N = 316)
Characteristic Value
Glucose monitoring method (multiple answers possible)
CGM
BGM
FGM
254 (80.4)
149 (47.2)
3 (0.9)
Duration of CGM use (n = 254)
<6 months
6 months to <1 year
1 year to <3 years
3 years to <5 years
≥5 years
10 (3.9)
20 (7.9)
105 (41.3)
49 (19.3)
70 (27.6)
Blood glucose tests
A1C, %
Number of daily fingerstick blood glucose checks
Number of daily glucose checks by CGM/FGM
7.4 ± 2.5
2.4 ± 2.6
20.7 ± 25
Insulin delivery method (multiple answers possible)
Insulin pump
Multiple daily injections using an insulin pen
Multiple daily injections using vial/syringe
Inhalable insulin
Other
242 (76.6)
59 (18.7)
37 (11.7)
13 (4.1)
1 (0.3)
Reported diabetes-related comorbidities (multiple answers possible)
Joint issues
Hypothyroidism
Retinopathy
CVD
Gastroparesis
Neuropathy
Sexual dysfunction
Nephropathy
Hyperthyroidism
87 (27.5)
80 (25.3)
61 (19.3)
61 (19.3)
51 (16.1)
46 (14.6)
44 (13.9)
22 (7.0)
18 (5.7)
Type of diabetes HCP
Adult endocrinologist
Diabetes nurse practitioner
Diabetes physician assistant
Pediatric endocrinologist
Primary care physician
Primary care, nurse practitioner
Primary care, physician assistant
203 (64.2)
48 (15.2)
24 (7.6)
16 (5.1)
14 (4.4)
7 (2.2)
4 (1.3)
Frequency of visits with diabetes HCP
Every month
Every 2–3 months
Every 6 months
Once per year
Once every 1–2 years
Other
9 (2.8)
201 (63.6)
92 (29.1)
6 (1.9)
2 (0.6)
6 (1.9)
Data are n (%) or mean ± SD.
Participants’ Experiences With Hypoglycemia
Participants’ reported experiences with hypoglycemia are presented in Table 3. Participants disclosed experiencing a mean of 3.6 ± 6.0 SH events in the preceding 12 months and a mean of 15.3 ± 60.4 (median 5, range 1–1,000) SH events in total in their lifetime. Participants noticed hypoglycemia in a variety of ways, most commonly by experiencing symptoms (80.7%), but also by using the functions of their CGM system (alert [63.3%], glucose reading [49.7%], or predictive low glucose warning [48.1%]). About one-fourth of the participants also made use of fingerstick BGM.
TABLE 3 Participants’ Experiences With Hypoglycemia (N = 316)
Hypoglycemia-Related Survey Data Value
Hypoglycemia data
Blood glucose level at which hypoglycemia is being treated, mg/dL
Number of mild hypoglycemic events per week
Number of moderate hypoglycemic events per week
Number of SH events in the previous 12 months
Number of SH events in the lifetime
70.2 ± 21.8
7.4 ± 13.2
5.0 ± 13.3
3.6 ± 6.0
15.3 ± 60.4
How hypoglycemia is usually noticed (multiple answers possible)
Feeling symptoms
CGM low alert
Looking at the CGM reading
CGM predictive low glucose warning
Fingerstick BGM
Other
255 (80.7)
200 (63.3)
157 (49.7)
152 (48.1)
82 (25.9)
5 (1.6)
Number of SH events
In the previous 12 months
CGM users (n = 254)
Non-CGM users (n = 62)
In the lifetime
CGM users (n = 254)
Non-CGM users (n = 62)
In the lifetime, excluding those with >100 lifetime events (n = 3)
CGM users (n = 251)
Non-CGM users (n = 62)
3.1 ± 4.2
5.7 ± 10.6
15.9 ± 66.5
13.2 ± 21.7
10.4 ± 16.7
13.2 ± 21.7
Awareness of hypoglycemia
1 (always aware)
2
3
4
5
6
7 (never aware)
30 (9.5)
77 (24.4)
68 (21.5)
85 (26.9)
43 (13.6)
7 (2.2)
6 (1.9)
Times that hypoglycemia is of most concern (multiple answers possible)
Overnight
During exercise
After exercise
During the day
Other
238 (75.3)
139 (44.0)
105 (33.2)
87 (27.5)
17 (5.4)
Occurrence of the most recent SH event
<1 month ago
1–3 months ago
3 months to <6 months ago
6 months to 1 year ago
1–2 years ago
2–5 years ago
>5 years ago
44 (13.9)
76 (24.1)
51 (16.1)
48 (15.2)
30 (9.5)
22 (7.0)
45 (14.2)
Type of emergency resulting from SH (multiple answers possible)
Change in mental state requiring assistance
Passing out/loss of consciousness or seizure
Calling a paramedic
Emergency department visit(s)
Requiring glucagon
Hospitalizations with at least 1 night spent in the hospital
258 (81.6)
196 (62.0)
195 (61.7)
185 (58.5)
181 (57.3)
115 (36.4)
Emotions experienced in connection with SH (multiple answers possible)
Worry/concern
Anxiety
Stress
Fear
Panic
Weakness
Embarrassment
Anger
Guilt
Sadness
Shame
214 (67.7)
203 (64.2)
202 (63.9)
172 (54.4)
166 (52.5)
159 (50.3)
160 (50.6)
112 (35.4)
106 (33.5)
89 (28.2)
67 (21.2)
Data are mean ± SD or n (%).
When comparing SH events in the previous 12 months for participants who used or did not use CGM, CGM users experienced a mean of 3.1 ± 4.2 (median 1.0, range 0–30) SH events, and non-CGM users experienced a mean of 5.7 ± 10.6 (median 2.5, range 0–58) SH events. Excluding participants who reported having had >100 lifetime SH events (n = 3), CGM users reported a mean of 10.4 ± 17.7 (median 5.0, range 1–100) SH events in their lifetime, and non-CGM users reported a mean of 13.2 ± 21.7 (median 4.5, range 0–100) SH events in their lifetime.
Using the criteria employed by Gold et al. (35), participants’ awareness of hypoglycemia was classified as “normal” (scores of 1–2) or “impaired” (scores of 4–7) on a visual analog scale. “Indeterminant” awareness (score of 3) was not included in either of these categories. Overall, 33.9% of the participants reported having normal awareness of hypoglycemia, and 44.6% reported having impaired awareness of hypoglycemia. Most participants (75.3%) considered the night or the time during or after exercise (77.2%) as the time of most concern about hypoglycemia.
SH events were associated with several emergency situations, including calling an ambulance (61.7%), visiting an emergency department (58.5%), and staying for at least 1 night at the hospital (36.4%). More than half of the participants stated that they had felt either worry, anxiety, stress, fear, panic, weakness, or embarrassment during an SH episode.
Participants’ Experiences With Glucagon Rescue Treatments
Participants’ experiences with glucagon rescue treatments are reported in Table 4. Almost all participants (97.2%) had been prescribed glucagon either at the time of completing the survey or in the past, with 80.4% reporting a current glucagon prescription. Overall, 74.7% of participants had a current and filled glucagon prescription, and 25.3% did not. Participants reported a mean of 1.7 ± 5.3 uses of glucagon to treat SH in the previous 12 months.
TABLE 4 Participants’ Experiences With Glucagon Rescue Treatments (N = 316)
Glucagon-Related Survey Data Value
Current prescription for glucagon
Yes, currently have a glucagon prescription and have filled it
Yes, have been prescribed glucagon in the past but not recently
Yes, currently have a glucagon prescription but have not filled it
No, have never had a glucagon prescription
236 (74.7)
53 (16.8)
18 (5.7)
9 (2.8)
Number of times glucagon was used to treat SH in the previous 12 months 1.7 ± 5.3
Likelihood of having a current and filled glucagon prescription
Impaired awareness of hypoglycemia (scores 4–7; n = 141)
Having a current and filled glucagon prescription
Not having a current and filled glucagon prescription
Not having impaired awareness of hypoglycemia (scores 1–3; n = 175)
Having a current and filled glucagon prescription
Not having a current and filled glucagon prescription
113 (80.1)
28 (19.9)
123 (70.3)
52 (29.7)
Current ownership of a glucagon kit
Yes
No
Unsure
251 (79.4)
62 (19.6)
3 (0.9)
Age of glucagon kit if currently owning one (n = 251)
<1 year old
1–2 years old
>2 years old
114 (45.4)
117 (46.6)
20 (8.0)
Feelings associated with having a glucagon kit at home (n = 316; multiple answers possible)
Safe
Prepared for hypoglycemia
Confident about hypoglycemia
The same as not having it in my home
151 (47.8)
127 (40.2)
90 (28.5)
87 (27.5)
Educated by an HCP about glucagon use
Yes
No
Unsure
263 (83.2)
47 (14.9)
6 (1.9)
Data are n (%) or mean ± SD.
Participants who had impaired awareness of hypoglycemia (scores 4–7, n = 141) were significantly more likely to have a current and filled glucagon prescription than those who did not have impaired awareness of hypoglycemia (scores 1–3, n = 175): 80.1% (113 of 141) versus 70.3% (123 of 175) (Pearson χ2 test P = 0.05). There were no significant differences between those with or without a current and filled glucagon prescription in A1C (P = 0.31) or age (P = 0.19). In addition, having a current and filled glucagon prescription was not associated with a greater likelihood of receiving care from an endocrinologist (P = 0.70).
Overall, 79.4% of the participants (n = 251) reported that they currently owned a glucagon kit, with 92.0% (n = 231) owning a kit <2 years old and 8.0% (n = 20) owning a kit >2 years old. Approximately half of the participants stated that having a glucagon kit in their home made them feel safe, 40.2% felt that they were prepared for hypoglycemia, and 28.5% felt confident about hypoglycemia; however, 27.5% reported that having a glucagon kit in their home made no difference in how they felt about hypoglycemia. HCPs had provided education about glucagon use to 83.2% of the participants, but 16.8% of participants had not received education by an HCP or were unsure about whether they had received such education. Compared with other HCPs, care from an endocrinologist was not associated with better education about glucagon use (P = 0.60).
Barriers to Glucagon Use as a Rescue Treatment
Barriers to glucagon rescue use reported by the participants are included in Table 5. Importantly, the 251 participants who currently owned a glucagon kit had a mean out-of-pocket cost of $73.40 ± $85.80 (median $30, range $0–400) for their kit. When queried, 85.1% of the 316 participants reported that rescue glucagon was administered when needed. In the remaining 14.9% (n = 47) for whom glucagon was not administered when needed, the reasons were diverse, including problems with the reconstitution and administration process (n = 33), inability to locate the kit (n = 23), lack of training (n = 15), an expired kit (n = 14), the rescuing individual being unaware of the existence of the kit (n = 13), and lack of confidence of the rescuing individual (n = 7).
TABLE 5 Barriers to Using Glucagon Rescue Treatment
Barrier-Related Survey Data Value
Out-of-pocket cost for a glucagon kit, $ (n = 251) 73.40 (85.80)
Ability to have available glucagon administered (n = 316)
Able
Unable
269 (85.1)
47 (14.9)
Reasons for inability to have available glucagon administered (n = 47; multiple answers possible)
The rescuing individual was not able to locate the glucagon kit.
The rescuing individual was not trained to use the glucagon kit.
The glucagon was expired.
The rescuing individual was not aware of the glucagon kit.
There was a problem with mixing.
The rescuing individual was not able to use the glucagon kit correctly.
The rescuing individual was not comfortable administering the glucagon.
The rescuing individual who delivered glucagon broke the needle.
The process was too complex.
Other
23 (48.9)
15 (31.9)
14 (29.8)
13 (27.7)
11 (23.4)
10 (21.3)
7 (14.9)
7 (14.9)
5 (10.6)
3 (6.4)
Reasons for not currently having a glucagon prescription (n = 9; multiple answers possible)
My doctor has never discussed glucagon with me.
I carry other supplies instead (glucose tablets, juice, etc.).
My CGM device gives me alerts before my blood glucose gets too low.
I don’t need it.
I am worried about temperature changes affecting the kit.
It costs too much to fill the prescription.
The kit has too many steps to be useful.
There is nobody to administer the glucagon (e.g., living alone).
Those around me would call 911 if I became severely hypoglycemic.
Other
6 (66.7)
6 (66.7)
5 (55.6)
1 (11.1)
1 (11.1)
1 (11.1)
1 (11.1)
1 (11.1)
1 (11.1)
1 (11.1)
Data are n (%).
The most important reasons for not having a glucagon prescription stated by participants who have never been prescribed glucagon (n = 9) were that their doctor had never discussed it with them (n = 6), they usually carried oral carbohydrates (n = 6), and their CGM device alerted them to low blood glucose levels (n = 5).
Participants’ Preferences for Characteristics of Glucagon Rescue Products
Most participants preferred ready-to-use glucagon rescue products, with 47.8% favoring an intranasal spray and 38.0% a premixed autoinjector (Table 6). The remaining 14.2% of the participants preferred injectable glucagon that requires reconstitution before use. Valued characteristics of glucagon rescue treatments included the ease of use of ready-to-use products (76.3%), intuitive modes of administration such as a nasal spray or ready-to-use injection (55.7%), a fast onset of action (53.5%), and storage at room temperature (42.1%).
TABLE 6 Participants’ Preferences for Characteristics of Glucagon Products (N = 316)
Preference-Related Survey Data Value
Preferred method of glucagon administration
Intranasal—spray in the nose
Premixed autoinjector—already reconstituted
Injection—needs to be reconstituted
151 (47.8)
120 (38.0)
45 (14.2)
Preferred characteristics of glucagon rescue treatments (multiple answers possible)
Ease of use (ready-to-use treatment)
The way it is administered (nasal spray, ready-to-use injection)
How fast the glucagon works compared with others on the market
Treatment that can be carried at room temperature (vs. stored in a refrigerator)
Having multiple kits available
241 (76.3)
176 (55.7)
169 (53.5)
133 (42.1)
102 (32.3)
Main improvements to be made in glucagon rescue treatments (multiple answers possible)
Ease of use (premixed dosing)
Cost
Administration type (nasal, syringe)
Storage options
Time to physical recovery
Fewer steps in the instructions
Size of kit
All of the above
Other
198 (62.7)
169 (53.5)
159 (50.3)
122 (38.6)
126 (39.9)
96 (30.4)
99 (31.3)
39 (12.3)
3 (0.9)
Importance of time to plasma glucose recovery in glucagon rescue treatments
Very important
Fairly important
Important
Slightly important
Not at all important
148 (46.8)
84 (26.6)
57 (18.0)
26 (8.2)
1 (0.3)
Data are n (%).
The participants also discussed various ways to improve glucagon rescue products, including greater ease of use (62.7%), reduced cost (53.5%), and more intuitive modes of administration (50.3%). The time to plasma glucose recovery was rated as a “very important” or “fairly important” characteristic of glucagon rescue products by 73.4% of the participants (Table 6).
Discussion
The burden of SH remains a challenge throughout the lives of people with type 1 diabetes and has a substantial impact on their emotional well-being, diabetes management, and health outcomes (17,27). Although previous studies have examined the emotional impact of SH and glucagon rescue treatment on caregivers of individuals with type 1 diabetes (36–38), literature is limited about the views and emotions connected with hypoglycemia and glucagon rescue use of people with type 1 diabetes themselves (19). For this reason, the objective of this research study was to provide an updated real-world perspective on emergency glucagon use of individuals with type 1 diabetes, as available glucagon products have become more diversified with the introduction of second-generation, ready-to-use rescue products.
The surveyed sample of individuals with type 1 diabetes had a mean age of 35.6 ± 8.9 years and a mean diabetes duration of 17.2 ± 12.7 years since diagnosis. Participants experienced a mean of 7.4 mild and 5.0 moderate hypoglycemic events per week and a mean of 3.6 SH events in the past year and 15.3 SH events (median 5 SH events) in their lifetime. Approximately half of the most recent SH events occurred in the 6 months immediately before the survey, emphasizing the relative frequency with which SH episodes occurred in the surveyed sample. The data on SH events in the 12 months before survey completion align with recent surveys of individuals with type 1 diabetes in other countries that reported means of 1.5–4.2 SH events in the previous 12 months (39,40). Importantly, CGM users among the surveyed sample continued to experience SH events, albeit fewer than non-CGM users. The reasons for this observation may be manifold, including lack of functional data because of system warm-up time, early termination, transmitter or receiver failure, or, more commonly, insufficient supply as a result of insurance problems.
Of note, the majority of participants in this study had received care from a paramedic (61.7%) or had an emergency department visit (58.5%), and more than one-third of participants (36.4%) had experienced a hospital stay of at least 1 night as a result of an SH event. These data illustrate the high level of health resource utilization as a direct consequence of SH, which is associated with significant costs to the U.S. health care system.
Importantly, the survey highlighted the emotional burden of SH on those affected, showing a considerable proportion of participants expressing worry or concern (67.7%), fear (54.4%), panic (52.5%), embarrassment (50.6%), and shame (21.2%) in connection with SH events. These findings are reflective of the findings of other recent survey studies on the negative emotional impact of SH on individuals with type 1 diabetes (39–43).
Participants reported being aware of their hypoglycemia episodes primarily by experiencing symptoms and through the use of CGM; however, a substantial proportion (44.6%) reported impaired awareness of hypoglycemia. A high prevalence of hypoglycemia unawareness, even in those using CGM, has previously been reported; yet, the result from this survey was slightly higher than the rates noted in people with type 1 diabetes in other studies (39,40,42,44,45).
Nocturnal hypoglycemia and hypoglycemia during or after exercise were concerns for approximately three-fourths of the participants. Although we did not measure how often participants experienced nocturnal or exercise-related hypoglycemia, current research supports our participants’ concerns, indicating that 28.7–51.4% of participants’ most recent SH events occurred during the night (39,40,42). Both nocturnal hypoglycemia and exercise-related hypoglycemia are associated with immediate clinical consequences (46,47). Moreover, nocturnal hypoglycemia has a long-term impact on glucose counterregulatory mechanisms that may lead to cognitive impairment, reduced hypoglycemia awareness, and autonomic failure (47).
Importantly, this research study showed encouraging results, with 83.2% of the participants receiving education from their HCP pertaining to glucagon rescue use, although receiving care from an endocrinologist compared with other HCPs was not associated with perceived better education about glucagon use. However, this is an increase from the 71.0% of adults with type 1 diabetes who reported being educated by an HCP on this topic in a 2019 U.S. survey (21). Furthermore, almost all of the participants in the current study (97.2%) had been prescribed a glucagon rescue product at the time of the survey or in the past, 80.4% had a current prescription for glucagon, and 74.7% had a current prescription and had filled it. Participants with hypoglycemia unawareness were significantly more likely to have a current and filled glucagon prescription than those who had normal awareness of hypoglycemia. A possible explanation is that those with hypoglycemia unawareness may be more vigilant about having glucagon at their disposal because they are conscious of their own difficulty in recognizing symptoms of an oncoming SH episode. Those with a normal awareness of SH may be reassured that they can recognize symptoms of an oncoming SH episode and take evasive action (e.g., consuming oral carbohydrates) and thus may feel less reliant on glucagon.
The U.S. survey by Haymond et al. (21) noted that 85.2% (n = 225 of 264) of participating adults with type 1 diabetes had been prescribed glucagon and that 58.3% (n = 154 of 264) had a current glucagon prescription (21); 51% of those participants who had experienced an SH event in the past had not been able to have glucagon administered to them when needed, even if the kit was close by. In contrast, 14.9% of the participants in the present survey were unable to have glucagon administered to them when required. Thus, in addition to an increased number of people with type 1 diabetes being prescribed glucagon rescue therapy since 2019, the proportion of people who have been administered rescue glucagon when required has improved substantially. Nonetheless, this survey showed that known obstacles to the appropriate use of rescue glucagon (21,22,27) continue to pose challenges, including problems with the reconstitution and administration process, unawareness of the existence of the kit or inability to locate it, lack of training, kit expiration, and lack of confidence of the rescuing individual.
Participants who had never received a glucagon prescription (n = 9) stated that the main reasons for this were that their doctor had never discussed it with them, they carried oral carbohydrates with them, or they relied on CGM. A global survey report suggests differences among countries regarding the proportion of individuals with type 1 diabetes who were unable to have glucagon administered during an SH event because they did not have a glucagon prescription or it was not filled (39,40,42), ranging from 25.0% in a French cohort (42) to 68.8% in a Japanese cohort (39).
One of the most critical barriers to the use of glucagon rescue products identified in this study was a mean out-of-pocket cost of $73.40 ± $85.80 for glucagon kits. Cost was identified by 53.5% of participants as one of the improvements they would like to see in glucagon rescue products. Financial barriers may therefore contribute to people with type 1 diabetes not filling their glucagon prescriptions or relying on other means of achieving normoglycemia.
Our survey data further demonstrated that most participants preferred ready-to-use glucagon products over first-generation glucagon products that require reconstitution because of the former’s ease of use, ready-to-use administration format, fast onset of action, and storage at room temperature. Participants favored a nasal mode of administration (47.8%) and a premixed autoinjector (38.0%) over reconstituted first-generation glucagon (14.2%). When queried, at least half of the participants indicated that they would also like to see improvements in ease of use (62.7%) and more intuitive modes of administration (50.3%). This finding suggests that a large proportion of the participants may still use first-generation glucagon products that require complex reconstitution and administration steps.
Advice for Primary Care Providers
Despite the increasing use of CGM, SH and fear of SH continue to pose challenges for individuals with type 1 diabetes. Although the typical automated insulin delivery system responds by adjusting the insulin dose when an SH event occurs, this process does not obviate the immediate need for administration of counterregulatory glucagon. Thus, glucagon continues to be a necessary and potentially life-saving component of therapy for type 1 diabetes.
To provide the best possible care for individuals with type 1 diabetes, primary care providers are advised to follow the ADA’s guidelines for prescribing glucagon (3) and to address the following topics with their patients. Ask whether the individual has a supply of glucagon at home. If so, ascertain that the glucagon is not out of date.
If not, determine the barriers to accessing glucagon that the individual may face.
Ask about who in the individual’s support network is trained in glucagon use. Ascertain whether these trusted individuals know where the glucagon is stored and are confident about administering it.
Given that ready-to-use glucagon products are now available, determine whether these individuals need to be retrained or whether additional people can be added to the individual’s circle of trusted individuals to optimize safety.
Limitations
A limitation of this study is that the surveyed sample is not representative of the overall population of people with type 1 diabetes in the United States. The survey sample was predominantly White and female and relatively young, affluent, and well-educated. The study design made use of an online survey format, and recruitment was conducted via digital media, which may have excluded people with limited computer skills or restricted computer or Internet access. The study included only adults with type 1 diabetes who had experienced at least one SH event in their lifetime and who were aware of the glucagon products on the market; therefore, the survey sample may be skewed toward people who were more aware of the available glucagon products than those who had never experienced an SH episode. In addition, the survey was conducted retrospectively, and data were self-reported, which carries the potential for misclassification of hypoglycemic events as well as recall bias, especially for the number of SH events in the participants’ lifetime. It was noted that a small number of participants reported a very high number of SH events in their lifetime (up to 1,000 events), which may have skewed the mean; the median may be more representative of the overall survey sample.
Conclusion
This study highlighted several important findings about the experiences of individuals with type 1 diabetes in the United States with SH and glucagon rescue use. The survey results suggest that the majority of individuals with type 1 diabetes have a current and filled prescription for rescue glucagon and have received education about glucagon rescue use from their HCP. The proportion of individuals with type 1 diabetes who were unable to have glucagon administered to them when required has decreased substantially compared with previous surveys. These data show a promising trend toward increased utilization of rescue glucagon and improved training in its use by HCPs in the past few years.
Despite these positive findings, there are continued challenges, including a high level of costly health care resource utilization directly resulting from SH events, substantial out-of-pocket expenses for glucagon kits, and a substantial prevalence of hypoglycemia unawareness. This study also highlighted the emotional burden of SH for individuals with type 1 diabetes.
Ready-to-use glucagon products do not yet seem to be fully adopted by HCPs and the diabetes community. Greater dissemination of these products among individuals with type 1 diabetes could improve the use of glucagon for SH rescue even more in the future. Further research is needed to understand experiences with and emotional effects of hypoglycemia and glucagon rescue use in both individuals with type 1 diabetes and their caregivers to improve the health and quality of life of people with type 1 diabetes.
Article Information
Acknowledgments
The authors thank the people with type 1 diabetes who contributed to the focus groups and survey data. Medical writing and editorial assistance for this manuscript were provided by Claudia Brockmeyer, DPhil, Katharine Timberlake, DPhil, and Marlena Radomska, BSc, of Oxford PharmaGenesis, Oxford, U.K.
Funding
This study and the preparation of this article were funded by Zealand Pharma.
Duality of Interest
S.A.W. has received consulting fees from Zealand Pharma independent of the submitted work. M.W. is a former employee of Zealand Pharma. No other potential conflicts of interest relevant to this article were reported.
Author Contributions
A.S.H., K.S.C., J.L., J.B., and W.A.W. participated in the conceptualization and design of the study, enrolled the survey participants, and conducted the survey. A.S.H., K.S.C., and H.N. analyzed the survey data. H.N. conducted the correlation analysis. M.W. obtained funding for medical writing and editorial assistance. All of the authors interpreted the data, reviewed the manuscript critically for intellectual content, provided feedback for incorporation, and approved the final version for publication. A.S.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
J.L. is currently affiliated with Ironwood Pharmaceuticals, Boston, MA.
J.B. is currently affiliated with PatientsLikeMe, Boston, MA.
M.W. is currently affiliated with Market Access and RWE Consulting and Contract Services, Winget LLC, Boston, MA.
==== Refs
References
1. Abraham MB, Jones TW, Naranjo D, . ISPAD Clinical Practice Consensus Guidelines 2018: Assessment and management of hypoglycemia in children and adolescents with diabetes. Pediatr Diabetes 2018;19 (Suppl. 27 ):178–192 29869358
2. Agiostratidou G, Anhalt H, Ball D, . Standardizing clinically meaningful outcome measures beyond HbA1c for type 1 diabetes: a consensus report of the American Association of Clinical Endocrinologists, the American Association of Diabetes Educators, the American Diabetes Association, the Endocrine Society, JDRF International, the Leona M. and Harry B. Helmsley Charitable Trust, the Pediatric Endocrine Society, and the T1D Exchange. Diabetes Care 2017;40 :1622–1630 29162582
3. American Diabetes Association Professional Practice Committee. 6. Glycemic targets: Standards of Medical Care in Diabetes—2022. Diabetes Care 2022;45 (Suppl. 1 ):S83–S96 34964868
4. Cryer PE. Mechanisms of hypoglycemia-associated autonomic failure in diabetes. N Engl J Med 2013;369 :362–372 23883381
5. Kahn PA, Liu S, McCoy R, Gabbay RA, Lipska K. Glucagon use by U.S. adults with type 1 and type 2 diabetes. J Diabetes Complications 2021;35 :107882 33593689
6. Reno CM, Litvin M, Clark AL, Fisher SJ. Defective counterregulation and hypoglycemia unawareness in diabetes: mechanisms and emerging treatments. Endocrinol Metab Clin North Am 2013;42 :15–38 23391237
7. Gubitosi-Klug RA, Braffett BH, White NH, .; Diabetes Control and Complications Trial (DCCT)/Epidemiology of Diabetes Interventions and Complications (EDIC) Research Group. Risk of severe hypoglycemia in type 1 diabetes over 30 years of follow-up in the DCCT/EDIC study. Diabetes Care 2017;40 :1010–1016 28550194
8. Khunti K, Davies M, Majeed A, Thorsted BL, Wolden ML, Paul SK. Hypoglycemia and risk of cardiovascular disease and all-cause mortality in insulin-treated people with type 1 and type 2 diabetes: a cohort study. Diabetes Care 2015;38 :316–322 25492401
9. Lu CL, Shen HN, Hu SC, Wang JD, Li CY. A population-based study of all-cause mortality and cardiovascular disease in association with prior history of hypoglycemia among patients with type 1 diabetes. Diabetes Care 2016;39 :1571–1578 27385329
10. American Diabetes Association. Economic costs of diabetes in the U.S. in 2017. Diabetes Care 2018;41 :917–928 29567642
11. Foos V, Varol N, Curtis BH, . Economic impact of severe and non-severe hypoglycemia in patients with type 1 and type 2 diabetes in the United States. J Med Econ 2015;18 :420–432 25629654
12. Hinahara J, Weinzimer SA, Bromley ER, Goss TF, Kendall DM, Hammer M. Dasiglucagon demonstrates reduced costs in the treatment of severe hypoglycemia in a budget impact model. J Manag Care Spec Pharm 2022;28 :461–472 35332789
13. Edelman S, Zhou FL, Preblick R, . Burden of cardiovascular disease in adult patients with type 1 diabetes in the US. Pharmacoecon Open 2020;4 :519–528 31997126
14. Leon BM, Maddox TM. Diabetes and cardiovascular disease: epidemiology, biological mechanisms, treatment recommendations and future research. World J Diabetes 2015;6 :1246–1258 26468341
15. Sussman M, Benner J, Haller MJ, Rewers M, Griffiths R. Estimated lifetime economic burden of type 1 diabetes. Diabetes Technol Ther 2020;22 :121–130 31886730
16. Jeon JY, Kim SR, Kim HJ, . Risk factors of severe hypoglycemia requiring medical assistance and neurological sequelae in patients with diabetes: a case-control study. Medicine (Baltimore) 2016;95 :e5365 27893672
17. Wild D, von Maltzahn R, Brohan E, Christensen T, Clauson P, Gonder-Frederick L. A critical review of the literature on fear of hypoglycemia in diabetes: implications for diabetes management and patient education. Patient Educ Couns 2007;68 :10–15 17582726
18. Willis WD, Diago-Cabezudo JI, Madec-Hily A, Aslam A. Medical resource use, disturbance of daily life and burden of hypoglycemia in insulin-treated patients with diabetes: results from a European online survey. Expert Rev Pharmacoecon Outcomes Res 2013;13 :123–130 23402452
19. Chapman K, Hughes AS, Bispham J, Leon C, Nguyen H, Wolf WA. Emergency glucagon: a focused review of psychosocial experiences of rescue drugs for type 1 diabetes. Curr Diab Rep 2022;22 :189–197 35171447
20. Fendrick AM, He X, Liu D, Buxbaum JD, Mitchell BD. Glucagon prescriptions for diabetes patients after emergency department visits for hypoglycemia. Endocr Pract 2018;24 :861–866 30035620
21. Haymond MW, Liu J, Bispham J, Hickey A, McAuliffe-Fogarty AH. Use of glucagon in patients with type 1 diabetes. Clin Diabetes 2019;37 :162–166 31057222
22. Kedia N. Treatment of severe diabetic hypoglycemia with glucagon: an underutilized therapeutic approach. Diabetes Metab Syndr Obes 2011;4 :337–346 21969805
23. Mitchell BD, He X, Sturdy IM, Cagle AP, Settles JA. Glucagon prescription patterns in patients with either type 1 or 2 diabetes with newly prescribed insulin. Endocr Pract 2016;22 :123–135 26484405
24. Eli Lilly and Company. Glucagon for injection prescribing information. Available from https://pi.lilly.com/us/rglucagon-pi.pdf. Accessed 31 May 2022
25. Novo Nordisk. GlucaGen (glucagon) prescribing information. Available from https://www.novo-pi.com/glucagenhypokit.pdf. Accessed 31 May 2022
26. Settles JA, Gerety GF, Spaepen E, Suico JG, Child CJ. Nasal glucagon delivery is more successful than injectable delivery: a simulated severe hypoglycemia rescue. Endocr Pract 2020;26 :407–415 32293921
27. Valentine V, Newswanger B, Prestrelski S, Andre AD, Garibaldi M. Human factors usability and validation studies of a glucagon autoinjector in a simulated severe hypoglycemia rescue situation. Diabetes Technol Ther 2019;21 :522–530 31219349
28. Eli Lilly and Company. Baqsimi (glucagon) prescribing information. Available from https://pi.lilly.com/us/baqsimi-uspi.pdf. Accessed 31 May 2022
29. Xeris Pharmaceuticals Inc. Gvoke (glucagon) prescribing information. Available from https://gvokeglucagon.com/pdf/gvoke-prescribing-information.pdf. Accessed 31 May 2022
30. Zealand Pharma. Zegalogue (dasiglucagon) prescribing information. Available from https://www.novo-pi.com/zegalogue.pdf. Accessed 31 May 2022
31. Laferrierre K, Liu J, Hughes AS, Wolf W. T1D Exchange registry: current data insights and future impact [Abstract]. Diabetes 2020;69 (Suppl. 1 ):194-LB
32. Leon C, Liu J, Ravelson J, Chapman KSM, Hughes A, Wolf W. T1D Exchange registry: current insights on participant outcomes [Abstract]. Diabetes 2021;70 (Suppl. 1 ):186-LB
33. T1D Exchange. T1D Exchange registry of people with type 1 diabetes in the United States. Available from https://t1dexchange.org/registry. Accessed 25 April 2022
34. Alchemer. Alchemer survey software. Available from https://www.alchemer.com. Accessed 20 May 2021
35. Gold AE, MacLeod KM, Frier BM. Frequency of severe hypoglycemia in patients with type I diabetes with impaired awareness of hypoglycemia. Diabetes Care 1994;17 :697–703 7924780
36. Muradoğlu S, Yeşiltepe Mutlu G, Gökçe T, Can E, Hatun Ş. An evaluation of glucagon injection anxiety and its association with the fear of hypoglycemia among the parents of children with type 1 diabetes. J Clin Res Pediatr Endocrinol 2021;13 :285–292 33491925
37. Patton SR, Dolan LM, Henry R, Powers SW. Parental fear of hypoglycemia: young children treated with continuous subcutaneous insulin infusion. Pediatr Diabetes 2007;8 :362–368 18036061
38. Patton SR, Dolan LM, Smith LB, Thomas IH, Powers SW. Pediatric parenting stress and its relation to depressive symptoms and fear of hypoglycemia in parents of young children with type 1 diabetes mellitus. J Clin Psychol Med Settings 2011;18 :345–352 21643962
39. Iwahori T, Snoek F, Nagai Y, Spaepen E, Mitchell BD, Peyrot M. Conversations and Reactions Around Severe Hypoglycemia (CRASH): Japan results from a global survey of people with T1DM or insulin-treated T2DM and caregivers. Diabetes Ther 2022;13 :517–533 35199292
40. Mojdami D, Mitchell BD, Spaepen E, . Conversations and reactions around severe hypoglycemia study: results of hypoglycemia experiences in Canadian adults with insulin-treated diabetes and their caregivers. Can J Diabetes 2021;45 :236–242 33087294
41. Ampudia-Blasco FJ, Artime E, Díaz S, . Conversations and Reactions Around Severe Hypoglycaemia (CRASH): Spanish results of a global survey of people with type 1 diabetes or insulin-treated type 2 diabetes and caregivers. Endocrinol Diabetes Nutr (Engl Ed) 2021;68 :557–566 34872639
42. Chevalier N, Penfornis A, Riveline JP, . Conversations and Reactions Around Severe Hypoglycemia (CRASH) global survey of people with type 1 diabetes or insulin-treated type 2 diabetes and caregivers: findings from the French cohort. Ann Endocrinol (Paris) 2022;83 :16–26 34871601
43. Mönnig E, Spaepen E, Osumili B, . Conversations and Reactions Around Severe Hypoglycaemia (CRASH): results from the German cohort of a global survey of people with type 1 diabetes or insulin-treated type 2 diabetes and caregivers. Exp Clin Endocrinol Diabetes 2022;130 :145–155 33368091
44. Porcellati F, Di Mauro S, Mazzieri A, . Glucagon as a therapeutic approach to severe hypoglycemia: after 100 years, is it still the antidote of insulin? Biomolecules 2021;11 :1281 34572493
45. Olsen SE, Asvold BO, Frier BM, Aune SE, Hansen LI, Bjørgaas MR. Hypoglycaemia symptoms and impaired awareness of hypoglycaemia in adults with type 1 diabetes: the association with diabetes duration. Diabet Med 2014;31 :1210–1217 24824356
46. Cockcroft EJ, Narendran P, Andrews RC. Exercise-induced hypoglycaemia in type 1 diabetes. Exp Physiol 2020;105 :590–599 31785115
47. Edelman SV, Blose JS. The impact of nocturnal hypoglycemia on clinical and cost-related issues in patients with type 1 and type 2 diabetes. Diabetes Educ 2014;40 :269–279 24695260
|
PMC010xxxxxx/PMC10338281.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
CD220096
10.2337/cd22-0096
Feature Articles
Real-World Persistence, Adherence, Hypoglycemia, and Health Care Resource Utilization in People With Type 2 Diabetes Who Continued With the Second-Generation Basal Insulin Analog Insulin Glargine 300 Units/mL or Switched to a First-Generation Basal Insulin (Insulin Glargine 100 Units/mL or Detemir 100)
https://orcid.org/0000-0002-0760-9332
Edelman Steven 1
Goldman Jennifer 2
Malone Daniel C. 3
Preblick Ronald 4
Munaga Kovida 4
Li Xuan 4
Gill Jasvinder 4
Gangi Sumana 5
1 UC San Diego School of Medicine, San Diego, CA
2 Massachusetts College of Pharmacy and Health Sciences, Boston, MA
3 University of Utah, Salt Lake City, UT
4 Sanofi, Bridgewater, NJ
5 Southern Endocrinology, Rowlett, TX
Corresponding author: Steven Edelman, [email protected]
Summer 2023
28 3 2023
28 3 2023
41 3 425434
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
People with type 2 diabetes receiving a second-generation basal insulin (BI) analog may be switched to a first-generation formulation for financial reasons or changes in health insurance. However, because second-generation BI analogs have more even pharmacokinetic profiles, longer durations of action (>24 vs. ≤24 hours), and more stable action profiles than first-generation BI analogs, such a change may result in suboptimal treatment persistence and/or adherence. This study compared treatment persistence, treatment adherence, rates of hypoglycemia, and health care resource utilization outcomes in people with type 2 diabetes who either continued treatment with the second-generation BI Gla-300 or switched to a first-generation BI. The study showed that continuing with Gla-300 was associated with a lower risk of discontinuing therapy, fewer emergency department visits, and lower hypoglycemia event rates than switching to a first-generation BI.
Sanofi US
==== Body
pmcDiabetes is a major health burden. In 2021, 32.2 million adults (aged 20–79 years) in the United States were estimated to have diabetes, of whom 90% had type 2 diabetes (1). One or more formulations of insulin are used by ∼7.4 million Americans with diabetes (2). Second-generation (longer-acting) basal insulin (BI) analogs (insulin glargine 300 units/mL [Gla-300] and insulin degludec 100 or 200 units/mL [U-100 or U-200]) provide stable and longer-acting pharmacokinetic and pharmacodynamic profiles, with longer durations of action (>24 vs. ≤24 hours) and less within-day and between-day glucose variability compared with first-generation (long-acting) BI analogs (insulin glargine 100 units/mL [Gla-100] and insulin detemir 100 units/mL [IDet]) (3,4). Clinical studies have demonstrated a reduced risk of hypoglycemia in people who received second- versus first-generation BIs, with a comparable reduction in A1C (5–10).
Patients receiving second-generation BIs may need to switch to first-generation formulations for nonclinical reasons such as changes in health insurance coverage (11). However, switching from a second- to a first- generation BI may result in suboptimal persistence (i.e., continuing to take the medication for the prescribed period) and adherence (i.e., correctly following the prescribed medication dosing regimen). Suboptimal persistence and/or adherence can negatively affect health care resource utilization (HRU). To our knowledge, no studies have compared outcomes in people who switched from a second- to a first-generation BI. This retrospective, observational study was conducted to compare treatment persistence and adherence, hypoglycemia, and HRU outcomes in adults with type 2 diabetes who either continued treatment with the second-generation BI Gla-300 or switched to a first-generation BI.
Research Design and Methods
Study Design
This analysis used the US Optum Clinformatics Data Mart with Socio-Economic Status database and included data from adults (aged ≥18 years) with a diagnosis of type 2 diabetes using International Classification of Diseases, 10th revision (ICD-10), codes who were receiving the second-generation BI Gla-300 and either continued treatment with it or switched to a first-generation BI between 1 January 2016 and 30 April 2021, inclusive (the study period). The date of treatment initiation (i.e., switching to a first-generation product or continuing Gla-300) was considered the index date. The baseline period was the 12 months before the index date, and the follow-up period, in which outcomes were measured, was the 12 months after the index date (Figure 1). Eligible participants had at least one pharmacy fill of Gla-300 or a first-generation BI during the identification period (1 January 2017 through 30 April 2020) and had continuous medical and prescription drug coverage and three or more claims for Gla-300 during the baseline period. Participants were excluded if they had a diagnosis of type 1 diabetes, were pregnant, or had gestational diabetes or polycystic ovarian syndrome during the study period. They were also excluded if they had any claims for NPH insulin, Gla-100, or IDet during the baseline period. An infographic summarizing this study is available (Supplementary Figure S1).
FIGURE 1 Study design. BI, basal insulin; Gla-100, insulin glargine 100 units/mL; Gla-300, insulin glargine 300 units/mL; HRU, health care resource utilization; IDet, insulin detemir 100 units/mL; PDC, proportion of days covered; T2D, type 2 diabetes.
Participants were matched on previous number of claims for Gla-300 during the baseline period. For those who switched from Gla-300 to a first-generation BI, the cohort included people who switched to either Gla-100 or IDet during the identification period; the switch date to the first-generation BI was defined as the index date. To avoid selection bias, the cohort who continued therapy with Gla-300 also included people who initially continued therapy with Gla-300 and later switched to a first-generation BI during the identification period; therefore, people included in this cohort had one or more fills for Gla-300. The index date for the cohort who continued treatment with Gla-300 was defined by the fill date of Gla-300 relative to the cohort of those who switched from Gla-300 to a first-generation BI.
Propensity score–matching (PSM) was used with a greedy nearest neighbor matching algorithm without replacement. For PSM, first a caliper (i.e., a measure of the required closeness of the match, defined based on a proportion of the SD of the logit of the propensity score) was identified. The algorithm then selected a participant who continued treatment with Gla-300 and a match who switched to Gla-100 or IDet, whose propensity score was closest to that of the Gla-300 participant within the caliper distance of each other. Matches were chosen one at a time for each Gla-300 participant.
Study Outcomes
The primary outcome was treatment persistence, defined as no discontinuation of the index treatment until the end of the follow-up period. Using the traditional measure of persistence with a fixed gap of “days’ supply” does not accurately reflect use of titratable injectable therapy. Therefore, this study used an alternative methodology as described by Wei et al. (12), which accounts for individual variation in treatment periods. The prescription for the index BI was considered discontinued if it was not refilled within the expected time of medication coverage, which was defined by the 90th percentile of the time for which medication was used. The metric quantity of BI supplied between patients’ first and second fills was calculated, and patients were grouped according to different metric quantities. The 90th percentile of the duration between first and second fill within each metric quantity group was then calculated. Persistence was measured by assigning the estimated allowable time of medication coverage according to the metric quantity of BI received (12). Sensitivity analyses were conducted using the 75th and 95th percentiles of time period of medication use. Participants were considered nonpersistent if they did not have continuous coverage of the index BI during the follow-up period (whether that was through 12 months, plan disenrollment, start of another therapy, or death).
Secondary outcome measures were treatment adherence, HRU, hypoglycemia events, and A1C change from baseline. Treatment adherence was defined as the proportion of days covered and calculated by dividing the total days supplied on the claim by the number of days in the refill interval (assuming all medications were consumed as prescribed), using a cutoff of ≥80% to define adherence and <80% for poor adherence.
All-cause, diabetes-related, and hypoglycemia-related HRU was assessed during the follow-up period and included hospital admissions and emergency department (ED) visits. Hypoglycemia incidence and event rates were calculated during the follow-up period, with hypoglycemia defined by either ICD-10 codes or by laboratory results. Change in A1C from baseline to 12 months was assessed in a subgroup of the propensity score–matched population who had valid A1C values at both baseline and 12 months.
Statistical Analyses
For treatment persistence, a Cox proportional hazards model with baseline imbalances adjusted as covariates was used to compare the risk of treatment discontinuation between the treatment groups, and Kaplan–Meier analyses were performed to compare the time to discontinuation. The percentage of participants with treatment persistence throughout the 12-month follow-up period was calculated along with the mean duration of persistence in days.
For treatment adherence, differences in the proportion of participants covered for ≥80% of days were assessed using a Cox proportional hazard model. Incidence rates for hospitalizations, ED visits, and hypoglycemia were reported as mean number of events and event rates per 100 person-years of follow-up [100 PYFU]). Change in A1C was calculated between baseline and 12 months (270–390 days) after the index date. A1C values closest to baseline and 12 months were used.
Sensitivity Analyses
Sensitivity analyses were conducted to assess any impact on study outcomes of any treatment groups of interest. Sensitivity analysis 1 compared outcomes in those who either continued treatment with Gla-300 or switched to Gla-100 only. Sensitivity analysis 2 was conducted using data from individuals who had at least one follow-up outpatient visit to account for any imbalance in care that may have occurred (e.g., if those who switched therapy required extra physician visits during the transition in therapies). Sensitivity analysis 3 was conducted using data from individuals who had a valid baseline A1C measurement, because a baseline A1C value was not required for the main analysis. For sensitivity analyses 1–3, the PSM algorithm used the same set of variables as the main analyses.
Sensitivity analysis 4 was conducted using data from individuals who had a valid baseline A1C measurement to account for any imbalance in baseline A1C between treatment cohorts. The treatment cohorts in sensitivity analysis 4 underwent PSM incorporating their baseline A1C value.
Data and Resource Availability
The data that support the findings of this study are available from Optum Clinformatics, but restrictions apply to the availability of these data, which were used under license for the current study and therefore are not publicly available. Only aggregated data can be shared with a third party, subject to approval by Optum and under the provisions of a signed agreement between Optum and that third party.
Results
Baseline Characteristics
From an initial sample of 637,260 people with type 2 diabetes who had at least one pharmacy claim of Gla-300 or a first-generation BI during the identification period, 2,760 people who continued therapy with Gla-300 and 1,109 who switched from Gla-300 to either Gla-100 (n = 838; 75.6%) or IDet (n = 271; 24.4%) were identified as eligible for PSM (Supplementary Table S1). After PSM, there were 1,104 participants in each group. The Gla-300 and first-generation BI groups were well balanced. The mean ages of the groups were 67.9 and 67.2 years, respectively, and the proportions of females in each group were 50.3 and 51.9%, respectively (Table 1). The between-group standardized mean difference was >1 for race, year of index treatment initiation, hyperlipidemia, baseline number of oral antidiabetic medications, and proportion of participants with one or more office visits at baseline. These characteristics were adjusted for in models as covariates.
TABLE 1 Baseline Characteristics of the Propensity Score–Matched Population
Propensity Score–Matched Population
Continued Gla-300 (n = 1,104) Switched to First-Generation BI (n = 1,104) SMD*
Age, years 0.066
Mean (SD) 67.90 (10.69) 67.21 (10.34)
Median (Q1–Q3) 70 (62–75) 69 (61–74)
Sex, n (%) 0.032
Male 548 (49.7) 531 (48.1)
Female 556 (50.3) 573 (51.9)
Race/ethnicity, n (%) 0.101
Asian 23 (2.1) 25 (2.3)
African American 194 (17.6) 193 (17.5)
Hispanic 247 (22.4) 206 (18.7)
White 591 (53.5) 629 (57.0)
Other/unknown 49 (4.5) 51 (4.6)
Health plan type, n (%) 0.084
Commercial 263 (23.9) 313 (28.4)
Medicare 841 (76.1) 791 (71.6)
Region, n (%) 0.098
West 200 (18.1) 182 (16.5)
Midwest 119 (10.8) 171 (15.5)
South 708 (64.1) 648 (58.7)
Northeast 75 (6.8) 102 (9.2)
Unknown 2 (0.3) 1(0.1)
Education level, n (%) 0.073
Less than 12th grade 14 (1.3) 11 (1.0)
High school diploma 444 (40.2) 456 (41.3)
Less than bachelor’s degree 549 (49.7) 521 (47.2)
Bachelor’s degree plus 63 (5.7) 78 (7.1)
Unknown 34 (3.2) 38 (3.4)
Home ownership status, n (%) 0.046
Probable homeowner 795 (71.9) 772 (69.9)
Probable renter 100 (9.0) 111 (10.1)
Unknown 210 (19.0) 221 (20.0)
Year of index treatment initiation, n (%) 0.523
2017 161 (14.7) 324 (29.4)
2018 351 (31.8) 342 (31.0)
2019 382 (34.6) 345 (31.2)
2020 210 (19.0) 93 (8.4)
Baseline Gla-300 fills <0.001
Mean (SD) 6.00 (3.25) 6.00 (3.25)
Median (Q1–Q3) 5 (3–8) 5 (3–8)
Baseline common comorbidities and diabetes complications of interest, n (%)
Hypertension 918 (83.1) 932 (84.4) 0.036
Hyperlipidemia 750 (67.9) 800 (72.5) 0.100
Obesity 400 (36.2) 433 (39.2) 0.062
Chronic kidney disease 418 (37.8) 389 (35.2) 0.054
Neuropathy 432 (39.1) 416 (37.7) 0.029
Depression 293 (26.5) 294 (26.6) 0.003
Nephropathy 126 (11.5) 130 (11.8) 0.009
Baseline A1C, continuous 0.014
n 638 592
Mean (SD) 8.17 (4.25) 8.68 (4.53)
Median (Q1–Q3) 7.8 (6.9–9.1) 8.4 (7.3–9.7)
Baseline hypoglycemia, n (%) 82 (7.4) 96 (8.7) 0.047
Baseline number of OADs, n (%) 0.382
0 152 (13.8) 305 (27.6)
1 270 (24.4) 252 (22.8)
2 320 (29.0) 308 (27.9)
≥3 362 (32.9) 239 (21.6)
Using GLP-1 receptor agonist at baseline, n (%) 372 (33.7) 378 (34.2) 0.012
Baseline health care utilization, n (%)
≥1 hospitalization 306 (27.7) 376 (34.1) 0.097
≥1 ED visit 160 (14.5) 162 (14.7) 0.025
≥1 office visit 550 (49.8) 508 (46.0) 0.102
* An SMD >0.1, shown in bold type, is indicative of an imbalance. GLP-1, glucagon-like peptide 1; OAD, oral antidiabetic drug; Q, quartile; SMD, standardized mean difference.
Primary Outcome: Treatment Persistence
During the 12-month follow-up period, a higher proportion of individuals who continued Gla-300 versus those who switched to a first-generation BI were persistent with therapy (64.6 vs. 44.1%; hazard ratio [HR] 0.59, 95% CI 0.52–0.68) (Figure 2). The mean (SD) number of persistent days was 237 (130.7) for Gla-300 and 191 (138.6) for first-generation BIs. Results of sensitivity analyses of the proportion at the 75th percentile and 95th percentile were consistent with those of the main analysis (75th percentile 38.2 vs. 23.8%; 95th percentile 74.6 vs. 54.7%).
FIGURE 2 Persistence, adherence, and hypoglycemia event rates. Cox proportional hazard model. 100 PYFU, 100 person-years of follow-up; BI, basal insulin; CI, confidence interval; Gla-300, insulin glargine 300 units/mL; HR, hazard ratio; OR, odds ratio; PDC, proportion of days covered; RR, rate ratio.
Results from sensitivity analysis 1 (Supplementary Table S2) were consistent with the main analyses, with a greater proportion of individuals who continued Gla-300 being persistent with therapy compared with those who switched to Gla-100. In addition, persistence with treatment was also greater for those who received Gla-300 than for those switching to a first-generation BI in the subgroups who had one or more outpatient visits at baseline (sensitivity analysis 2), a valid baseline A1C value (sensitivity analysis 3), or a valid A1C value included in PSM (sensitivity analysis 4) and continued treatment with Gla-300 compared with those who switched to a first-generation BI (Supplementary Table S2).
Secondary Outcomes
A similar proportion of participants who continued Gla-300 versus those who switched to a first-generation BI were adherent to therapy (34.1 vs. 32.3%; odds ratio 0.91, 95% CI 0.76–1.10) (Figure 2). The mean (SD) number of adherent days during the 12-month follow-up period was 214 (115.1) for participants who continued Gla-300, and 192 (125.65) for those who switched to a first-generation BI. Results of sensitivity analyses 1–4 were consistent with those of the main analysis (Supplementary Table S3).
During follow-up, 122 participants in the Gla-300 group experienced a total of 289 hypoglycemia events, and 162 participants who switched to a first-generation BI experienced a total of 473 hypoglycemia events. The incidence rate was lower for those continuing Gla-300 than for those switching to a first-generation BI (26.2 vs. 42.8 per 100 PYFU). In sensitivity analyses 1, 2, and 4, those who continued treatment with Gla-300 had lower hypoglycemia event rates compared with those who switched to a first-generation BI (Supplementary Table S4); in sensitivity analysis 3, the differences in hypoglycemia rates were comparable between those continuing Gla-300 and those who switched to a first-generation BI (Supplementary Table S4).
ED visit event rates were lower for participants who continued Gla-300 versus those who switched to a first-generation BI (all-cause 100.5 vs. 146.8 per 100 PYFU; diabetes-related 79.3 vs. 116.2 per 100 PYFU; hypoglycemia-related: 3.8 vs. 7.4 per 100 PYFU) (Figure 3). In sensitivity analyses 1–3, those who continued treatment with Gla-300 had lower ED visit event rates compared with those who switched to a first-generation BI (Supplementary Table S5); in sensitivity analysis 4, ED visit rates were comparable between those continuing Gla-300 and those who switched to a first-generation BI (Supplementary Table S5). Event rates for all-cause (53.8 vs. 69.9 per 100 PYFU), diabetes-related (21.3 vs. 28.2 per 100 PYFU), and hypoglycemia-related (1.2 vs. 1.5 per 100 PYFU) hospitalizations were comparable between groups (Figure 4).
FIGURE 3 ED visit event rates. 100 PYFU, 100 person-years of follow-up; BI, basal insulin; CI, confidence interval; Gla-300, insulin glargine 300 units/mL; HR, hazard ratio; RR, rate ratio.
FIGURE 4 Hospitalization event rates. 100 PYFU, 100 person-years of follow-up; BI, basal insulin; CI, confidence interval; Gla-300, insulin glargine 300 units/mL; RR, rate ratio.
Follow-up data for A1C were available for only 228 of 1,104 people (21%) in the Gla-300 group and 205 of 1,104 people (18.6%) in the first-generation BI group. Mean (SD) A1C at 12 months was similar between groups (8.15 [1.69] and 8.18 [1.64]% for those who continued Gla-300 and those who switched to a first-generation BI, respectively).
Reductions in A1C from baseline were smaller for participants who continued Gla-300 versus those who switched to a first-generation BI (difference in least squares means 0.29%). However, mean baseline A1C values were imbalanced between groups: 8.17 and 8.63% in the Gla-300 and first-generation BI groups, respectively. In sensitivity analysis 4, which included baseline A1C in the PSM algorithm, mean baseline A1C was comparable between groups (8.68 and 8.62% in the Gla-300 and first-generation BI groups, respectively). At 12 months, a reduction in A1C of 0.16% was observed in those who continued Gla-300 compared with a reduction of 0.45% in those who switched to a first-generation BI (difference in least squares mean 0.29%, 95% CI 0.285–0.289).
Discussion
The results of this retrospective, real-world, observational study in people with type 2 diabetes suggest that, compared with switching to a first-generation BI, continued use of the second-generation BI Gla-300 was associated with increased persistence with treatment. In addition, lower all-cause, diabetes-related, and hypoglycemia-related ED visits and a reduced number of hypoglycemia events were observed for participants who continued treatment with Gla-300 versus those who switched to a first-generation BI analog. The proportions of participants who were adherent to therapy were comparable between groups, as were rates of all-cause, diabetes-related, and hypoglycemia-related hospitalizations.
The rates of adherence and persistence for insulin therapy observed in this study are within the ranges observed for insulin in other studies, which have reported adherence rates of between 30 and 86% (13) and 1-year persistence rates in the range of 21–66% (14). To our knowledge, no studies have been conducted to compare persistence in people continuing treatment with a second-generation BI versus those who switch to a first-generation BI. In a study using data from the Optum Clinformatics database that compared outcomes for people with type 2 diabetes who switched to Gla-300 versus other BIs, after 6 months of follow-up, discontinuation of treatment was lower for those who received Gla-300 versus other BIs (20.4 vs. 36.4%) (15).
Although treatment persistence improved with continuing Gla-300 versus switching to a first-generation BI, in the current study, adherence to therapy was comparable between groups. A possible explanation for this finding is that, in those who switched to a first-generation BI, the insulin dose and the number of refills were similar to those when they were receiving Gla-300, which may be expected if the switch was not for medical reasons. This would mean that their insulin regimens were similar after switching to a first-generation BI from Gla-300 and would not considerably affect their lifestyle (i.e., they would have been taking doses at the same time, same number of injections, and so forth). Therefore, it would be fair to assume that factors affecting adherence would not change significantly after switching to a first-generation BI from Gla-300. This may have contributed to there being no detectable difference in adherence between the two regimens observed in this study. Additionally, it is possible that adherence rates are good when people first switch to a new insulin or therapy. However, it is likely that discontinuation rates were higher for those who switched to a first-generation BI versus those who continued with Gla-300 because of higher rates of hypoglycemia and ED visits. Increased hypoglycemia and ED visits could have resulted in a reduction in persistence while not affecting adherence because of the more serious nature of these types of events.
In the current study, continued use of Gla-300 was associated with fewer hypoglycemia events compared with switching to a first-generation BI. These findings support those in the literature demonstrating that second-generation BIs are associated with reduced risk of overall documented or severe hypoglycemia and of severe nocturnal hypoglycemia compared with first-generation BIs (4–9,16–19). It is important to note that, in the current study, the incidence of hypoglycemia events was determined either by laboratory values or by an ICD-10 code. Therefore, the true incidence of hypoglycemia was likely underreported. However, because this methodology was consistently used across the different treatment groups, the overall documentation of hypoglycemia would have been balanced between groups; therefore, any between-group differences that were detected were likely real.
Continuing treatment with Gla-300 was associated with fewer all-cause, diabetes-related, and hypoglycemia-related ED visits compared with switching to a first-generation BI (or to Gla-100 only), and hospitalization rates were similar between groups. These findings may have been driven by lower rates of hypoglycemia observed in the Gla-300 group, suggesting that hypoglycemia was treated in the ED and patients were then discharged without requiring hospitalization. Results of a retrospective study revealed that the risks of hospitalization or ED visit and the costs associated with these events were higher in those who experienced hypoglycemia in the 6 months after initiating BI therapy versus those who did not experience hypoglycemia early in their BI treatment (20). The findings of the current study may have implications for the cost of care, as the potential reduction in costs associated with the reduction in ED visits could offset the higher pharmacy costs associated with second-generation BIs compared with first-generation BIs. However, it was not possible to analyze cost data in this study; thus, no firm conclusions can be drawn.
The improved persistence observed with continued use of Gla-300 compared with switching to a first-generation BI also may have been driven by hypoglycemia rates, since hypoglycemia has been linked to poor persistence (20). Fear of hypoglycemia is a major concern for people with diabetes; 55.5% of patients who participated in the questionnaire-based Diabetes Attitudes, Wishes and Needs second study (21) reported that they were “very worried about the risk of hypoglycemia.”
Data for baseline and 12-month follow-up A1C values were available for only 21% of participants who continued Gla-300 and 18.6% of those who switched to a first-generation BI. This limitation in available data resulted in an imbalance in baseline A1C, with a higher baseline A1C in the group who switched to a first- generation BI compared with those who continued Gla-300. Sensitivity analysis 4 was conducted to try to account for this imbalance by matching baseline A1C in the PSM algorithm. The results showed no difference in the change in A1C between baseline and 12 months between those who continued treatment with Gla-300 compared with those who switched to a first-generation BI, although the difference remained greater in the group who switched. It is possible that the greater reduction in the group who switched may have been driven in part by more intensified titration or other support received at the time of switching; those continuing therapy with Gla-300 were maintaining therapy and likely not titrating their doses. Previous studies have demonstrated similar glycemic control between second- and first-generation BIs, with the main difference being a reduction in hypoglycemia observed with second-generation BIs, as observed in the current study (5,7–9).
Limitations
The Optum claims database has a large population with long-term follow-up and breadth of coverage; however, caution is required in interpreting these results because of the nature of retrospective studies. Generalizability of this study may be limited to the populations represented. The lack of randomization may have introduced bias, although confounding factors should have been mitigated by using PSM. Additionally, electronic medical records data are not designed specifically for research purposes, so some data may have been missing, erroneous, or misclassified (22). In the current analyses, baseline A1C values were available for a limited number of participants, and this imbalance likely drove the greater reduction in A1C from baseline observed in the group who switched to first-generation BIs. Persistence with and adherence to therapy can be estimated only from claims data owing to the lack of individualized dosing information. It was also not possible to determine why participants switched to first-generation BIs from Gla-300. Furthermore, it was not possible to determine how individualized A1C targets may have affected glycemic outcomes or whether medication was taken or titrated to the maximum effective dose. Additionally, the database is current only to 2019, and treatment patterns may have since changed.
Conclusion
Compared with those who switched to a first-generation BI, those who continued with Gla-300 had a lower risk of discontinuing therapy; fewer all-cause, diabetes-related, and hypoglycemia-related ED visits; and fewer hypoglycemia events. The lower rates of hypoglycemia observed in the Gla-300 group could potentially have driven the lower rates of ED visits observed in this study, translating into better persistence observed with Gla-300 versus first-generation BIs. These results suggest that improving coverage for and access to second-generation BIs may be beneficial for people with type 2 diabetes who need insulin therapy to control their diabetes.
Article Information
Funding
This study was funded by Sanofi US. Medical writing support was provided by Barrie Anthony, PhD, CMPP, of Evidence Scientific Solutions and funded by Sanofi US.
Duality of Interest
S.E. has served on advisory boards and speakers’ bureaus for AstraZeneca, MannKind, and Xeris and on an advisory board for BrightSight and is a board member for Senseonics and TeamType1. J.Go. serves on speakers’ bureaus for Abbott Diabetes, Amarin, Lilly, Novo Nordisk, and Xeris. D.C.M. is or has been a consultant to Avidity, Novartis, Pharmacyclics, Sarepta, and Seres. R.P. and J.Gi. are employees of and stockholders in Sanofi. K.M. and X.L. are employees of Sanofi. S.G. has been a speaker for Sanofi. No other potential conflicts of interest relevant to this article were reported.
Author Contributions
R.P., K.M., and X.L. contributed to the study design and acquisition and analyses of data. All authors were involved in the interpretation of the data and in drafting and revising the manuscript for important intellectual content, and all approved the publication for submission and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. S.E. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation
Portions of the data in this article were presented at the American Diabetes Association’s 82nd Scientific Sessions in New Orleans, LA, 3–7 June 2022.
This article contains supplementary material online at https://doi.org/10.2337/figshare.22255987.
==== Refs
References
1. International Diabetes Federation. IDF Diabetes Atlas. 10th ed. Brussels, Belgium, International Diabetes Federation, 2021. Available from https://www.diabetesatlas.org. Accessed August 2022
2. Cefalu WT, Dawes DE, Gavlak G, . Insulin Access and Affordability Working Group: conclusions and recommendations. Diabetes Care 2018;41 :1299–1311 29739814
3. Cheng AYY, Patel DK, Reid TS, Wyne K. Differentiating basal insulin preparations: understanding how they work explains why they are different. Adv Ther 2019;36 :1018–1030 30929185
4. Heise T, Hövelmann U, Nosek L, Hermanski L, Bøttcher SG, Haahr H. Comparison of the pharmacokinetic and pharmacodynamic profiles of insulin degludec and insulin glargine. Expert Opin Drug Metab Toxicol 2015;11 :1193–1201 26086190
5. Bolli GB, Riddle MC, Bergenstal RM, .; EDITION 3 Study Investigators. New insulin glargine 300 U/ml compared with glargine 100 U/ml in insulin-naïve people with type 2 diabetes on oral glucose-lowering drugs: a randomized controlled trial (EDITION 3). Diabetes Obes Metab 2015;17 :386–394 25641260
6. Marso SP, McGuire DK, Zinman B, .; DEVOTE Study Group. Efficacy and safety of degludec versus glargine in type 2 diabetes. N Engl J Med 2017;377 :723–732 28605603
7. Meneghini L, Blonde L, Gill J, . Insulin glargine 300 U/mL versus first-generation basal insulin analogues in insulin-naïve adults with type 2 diabetes: 12-month outcomes of ACHIEVE Control, a prospective, randomized, pragmatic real-life clinical trial. Diabetes Obes Metab 2020;22 :1995–2003 32538550
8. Meneghini LF, Sullivan SD, Oster G, . A pragmatic randomized clinical trial of insulin glargine 300 U/mL vs first-generation basal insulin analogues in insulin-naïve adults with type 2 diabetes: 6-month outcomes of the ACHIEVE Control study. Diabetes Obes Metab 2020;22 :2004–2012 32729217
9. Wysham C, Bhargava A, Chaykin L, . Effect of insulin degludec vs insulin glargine U100 on hypoglycemia in patients with type 2 diabetes: the SWITCH 2 randomized clinical trial. JAMA 2017;318 :45–56 28672317
10. Yki-Järvinen H, Bergenstal R, Ziemen M, .; EDITION 2 Study Investigators. New insulin glargine 300 units/mL versus glargine 100 units/mL in people with type 2 diabetes using oral agents and basal insulin: glucose control and hypoglycemia in a 6-month randomized controlled trial (EDITION 2). Diabetes Care 2014;37 :3235–3243 25193531
11. Anderson SL, Trujillo JM, Anderson JE, Tanenberg RJ. Switching basal insulins in type 2 diabetes: practical recommendations for health care providers. Postgrad Med 2018;130 :229–238 29260929
12. Wei W, Pan C, Xie L, Baser O. Real-world insulin treatment persistence among patients with type 2 diabetes. Endocr Pract 2014;20 :52–61 24013990
13. Guerci B, Chanan N, Kaur S, Jasso-Mosqueda JG, Lew E. Lack of treatment persistence and treatment nonadherence as barriers to glycaemic control in patients with type 2 diabetes. Diabetes Ther 2019;10 :437–449 30850934
14. Hamersky CM, Fridman M, Gamble CL, Iyer NN. Injectable antihyperglycemics: a systematic review and critical analysis of the literature on adherence, persistence, and health outcomes. Diabetes Ther 2019;10 :865–890 31054132
15. Zhou FL, Nicholls C, Xie L, Wang Y, Vaidya N, Meneghini LF. Hypoglycaemia and treatment patterns among insulin-treated patients with type 2 diabetes who switched to insulin glargine 300 units/mL versus other basal insulin in a real-world setting. Endocrinol Diabetes Metab 2019;2 :e00073 31294087
16. Javier Escalada F, Halimi S, Senior PA, . Glycaemic control and hypoglycaemia benefits with insulin glargine 300 U/mL extend to people with type 2 diabetes and mild-to-moderate renal impairment. Diabetes Obes Metab 2018;20 :2860–2868 30003642
17. Riddle MC, Bolli GB, Ziemen M, Muehlen-Bartmer I, Bizet F; EDITION 1 Study Investigators. New insulin glargine 300 units/mL versus glargine 100 units/mL in people with type 2 diabetes using basal and mealtime insulin: glucose control and hypoglycemia in a 6-month randomized controlled trial (EDITION 1). Diabetes Care 2014;37 :2755–2762 25078900
18. Ritzel R, Harris SB, Baron H, . A randomized controlled trial comparing efficacy and safety of insulin glargine 300 units/mL versus 100 units/mL in older people with type 2 diabetes: results from the SENIOR study. Diabetes Care 2018;41 :1672–1680 29895556
19. Yale JF, Aroda VR, Charbonnel B, . Glycaemic control and hypoglycaemia risk with insulin glargine 300 U/mL versus glargine 100 U/mL: a patient-level meta-analysis examining older and younger adults with type 2 diabetes. Diabetes Metab 2020;46 :110–118 30366067
20. Dalal MR, Kazemi M, Ye F, Xie L. Hypoglycemia after initiation of basal insulin in patients with type 2 diabetes in the United States: implications for treatment discontinuation and healthcare costs and utilization. Adv Ther 2017;34 :2083–2092 28779282
21. Nicolucci A, Kovacs Burns K, Holt RIG, .; DAWN2 Study Group. Diabetes Attitudes, Wishes and Needs second study (DAWN2™): cross-national benchmarking of diabetes-related psychosocial outcomes for people with diabetes. Diabet Med 2013;30 :767–777 23711019
22. Hernandez-Boussard T, Monda KL, Crespo BC, Riskin D. Real world evidence in cardiovascular medicine: ensuring data validity in electronic health record-based studies. J Am Med Inform Assoc 2019;26 :1189–1194 31414700
|
PMC010xxxxxx/PMC10338283.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
CD230023
10.2337/cd23-0023
Clinical Pharmacology Updates
Special Report: Potential Strategies for Addressing GLP-1 and Dual GLP-1/GIP Receptor Agonist Shortages
https://orcid.org/0000-0001-5124-4484
Whitley Heather P. 1 2
Trujillo Jennifer M. 3
https://orcid.org/0000-0002-4734-7402
Neumiller Joshua J. 4
1 Department of Pharmacy Practice, Auburn University Harrison College of Pharmacy, Auburn, AL
2 Clinical Pharmacy Specialist, Baptist Family Medicine, Baptist Health System, Montgomery, AL
3 Department of Clinical Pharmacy, University of Colorado Skaggs School of Pharmacy and Pharmaceutical Sciences, Aurora, CO
4 Department of Pharmacotherapy, Washington State University, Spokane, WA
Corresponding author: Heather P. Whitley, [email protected]
Summer 2023
07 4 2023
07 4 2023
41 3 467473
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
==== Body
pmcUnexpected drug shortages of the long-acting glucagon-like peptide 1 (GLP-1) receptor agonists dulaglutide and semaglutide and the dual GLP-1/glucose-dependent insulinotropic polypeptide (GIP) receptor agonist tirzepatide emerged in late 2022 and have persisted through the first quarter of 2023 (1). The drug shortage has not affected product doses equally; some doses are more widely available than others (2). For the purposes of this article, we will consider the dual GLP-1/GIP receptor agonist tirzepatide to be included when referring to GLP-1 receptor agonist product shortages. These shortages predominately occurred because of an unexpected increase in demand for GLP-1 receptor agonists without adequate adjustment in production (3). The inadequate supply has created new access barriers for patients previously using or wishing to initiate these products for glycemic management, weight loss, and/or cardiovascular risk reduction. Thus, strategies to mitigate this disruption to patient care are needed.
Disruptions in the supply chain create important challenges for patients and clinicians, forcing providers to seek alternative approaches to overcome what is hoped to be a temporary hurdle. Decreased availability and high cost have also led patients to seek alternative ways to obtain these medications. Many social media sites and online clinics claim to offer commercially available GLP-1 receptor agonist medications or even compounded versions at a discounted price. Although compounded medications are regulated and generally considered safe (4), compounded GLP-1 receptor agonists are unique. Each GLP-1 receptor agonist is only available as a branded medication manufactured by a single company. If the manufacturers are not supplying the compounding pharmacies with the active ingredient, then the source is unknown (5).
Past literature most commonly addresses drug shortages in the hospital setting and recommends a myriad of strategies, including stockpiling, using multiple or alternative suppliers, rationing among patients, substituting with alternative therapies, minimizing medication waste, maximizing utilization as long as possible based on manufacturer expiration date, adjusting formularies, and implementing formulary restrictions (6–9). Although few articles have outlined specific strategies to address this obstacle in the community/ambulatory setting, those available recommend contacting the supplier or another pharmacy, using other generic or brand-name options of the same medication, or switching to an alternative product altogether (10).
Given the lack of generic options within the GLP-1 receptor agonist class and the limited applicability of the other recommended approaches given the unique benefits of these therapies, this article sets out potential strategies that diabetes care teams can consider to overcome barriers related to the shortage. These suggestions are offered in the context of current persistent drug shortages and, while not ideal, are provided in the hope of mitigating undesired negative consequences from an inability of patients with type 2 diabetes to obtain GLP-1 receptor agonist therapies amid this global access challenge.
Considerations for Missed Doses and Alternative Dosing Strategies
Considerations for Missed Doses
Inconsistent use of drugs in the GLP-1 receptor agonist class is challenging because tolerance is developed to the common side effect of gastrointestinal (GI) disturbance with regular use. For this reason, most GLP-1 receptor agonists are initiated at the lowest available dose and titrated slowly, which helps to mitigate GI intolerance. However, it is not uncommon for patients to inadvertently miss doses (11). Manufacturers of GLP-1 receptor agonists provide guidance on the resumption of therapy after a single missed dose, and this guidance varies by product and dosing interval (Table 1) (12–19). However, when a supply of a patient’s GLP-1 receptor agonist is depleted and a refill is not available for an extended period, reinitiating at a lower dose to mitigate GI side effects once therapy can be resumed should be considered. Table 2 summarizes available recommendations and information to guide dose selection when resuming a GLP-1 receptor agonist after a more prolonged lapse in therapy (20,21).
TABLE 1 Manufacturer Recommendations for Missed Doses of GLP-1 Receptor Agonists
Agent Recommended Dosing Interval Manufacturer Recommendations for Missed Doses
Short-acting agents
Exenatide Twice daily Skip missed dose and resume at the next scheduled dose.
Lixisenatide Once daily If a dose is missed, administer within 1 hour prior to next meal.
Long-acting agents
Dulaglutide Once weekly Administer as soon as possible if there are ≥3 days (72 hours) until next scheduled dose.
If <3 days before next scheduled dose, skip the missed dose and administer on the next scheduled day.
Exenatide XR Once weekly Administer as soon as possible if there are ≥3 days (72 hours) until the next scheduled dose.
If <3 days before next scheduled dose, skip the missed dose and administer on the next scheduled day.
Liraglutide Once daily If dose is missed, resume with the next scheduled dose.
Semaglutide (injectable) Once weekly Administer as soon as possible within 5 days after the missed dose.
If >5 days have passed, skip the dose and administer on the next scheduled day.
Semaglutide (oral) Once daily If dose is missed, resume with the next scheduled dose.
Tirzepatide Once weekly Administer as soon as possible within 4 days (96 hours) after the missed dose.
If >4 days have passed, skip the dose and administer on the next scheduled day.
TABLE 2 Considerations for Resuming a GLP-1 Receptor Agonist After a Prolonged Lapse in Therapy
Agent Last Dose Administered Recommendation(s) for Resuming Therapy
Dulaglutide* 1.5 mg once weekly Resume at 1.5 mg once-weekly dose.
Expect comparable tolerability to that experienced prior to dose interruption.
3 or 4.5 mg once weekly Use best judgment if ≥3 doses are missed. It is unknown whether tolerance to the GI adverse events will remain if reinitiated at the higher dose after ≥3 missed doses.
Decision can be informed by patient’s prior GI tolerability.
In consideration of the above, clinicians may consider reinitiating at 1.5 mg once weekly.
Injectable semaglutide† 1 mg once weekly If ≤2 doses are missed, reinitiate at 1 mg once weekly.
If 3–4 doses are missed, reinitiate at 0.5 mg weekly.
If ≥5 doses are missed, reinitiate at 0.25 mg once weekly.
Tirzepatide‡ ≥5 mg once weekly If ≤2 doses are missed, reinitiate at the same dose (provided the dose was adequately tolerated).
If ≥3 doses are missed, reinitiate at 5 mg once weekly.
* Based on manufacturer-provided information (20).
† Based on personal communication with C. Wong on 27 February 2023.
‡ Based on supplementary material published with ref. 21.
Alternative Dosing Options
Until the shortage is resolved, alternative dosing strategies may temporarily help to maintain the desired weekly dose or extend the duration of a given supply. Potential strategies depend on the specific product’s pen device.
Injectable Semaglutide
Semaglutide is available in an adjustable multidose pen that patients dial to a marked dose before injecting. This allows administration of alternative intermediate doses based on the number of “clicks” between each marked dose. Clinically, these intermediate doses are occasionally recommended to provide a more gradual dose titration to improve GI tolerability (22); however, this technique is not supported by the manufacturer (H.P.W., personal communication).
By applying this concept in reverse, a patient could potentially inject a lower dose using a more readily available higher-strength product, by dialing to an alternative number of clicks. If this strategy is used, special attention should be given to the product’s shelf-life once opened, which is limited to 8 weeks (56 days) (Table 3) (13). Additionally, this method of using an intermediate dose may allow patients to extend the duration of a semaglutide pen and continue therapy, albeit with a lower-than-desired weekly dose. It is important to note that the pen concentration designed to deliver 0.25- and 0.5-mg doses recently changed from 2 mg/1.5 mL to 2 mg/3 mL (23). Furthermore, the concentrations of all semaglutide pens vary; consequently, the number of clicks to achieve a given dose also varies. For this reason, using clicks to administer alternative doses has a high potential for confusion and dosing errors; however, it may be an option for continuing semaglutide therapy during the product shortage.
TABLE 3 Alternative Intermediate Doses of Injectable Semaglutide
Semaglutide Pen (Product Concentration) Weekly Semaglutide Intermediate Doses
0.25 and 0.5 mg (2 mg/3 mL)* 0.5 mg 0.4 mg 0.33 mg 0.25 mg 0.12 mg 0.06 mg
1 mg (4 mg/3 mL) 1 mg 0.8 mg 0.66 mg 0.5 mg 0.25 mg 0.12 mg
2 mg (8 mg/3 mL) 2 mg 1.6 mg 1.33 mg 1 mg 0.5 mg 0.25 mg
Number of clicks for desired intermediate dose 74 59 49 37 18 9
Number of available weekly doses using an intermediate dose 4 5 6 8 8† 8†
* New semaglutide pen concentration as of March 2023; doses not confirmed by the manufacturer but rather mathematically calculated by the authors.
† Injected semaglutide is stable for 56 days (8 weeks) once opened and should be discarded thereafter.
Dulaglutide and Tirzepatide
Unlike semaglutide, dulaglutide and tirzepatide are available only in single-use, single-dose pens. Although these agents are available in multiple strengths, each pen delivers a discrete and nonadjustable dose.
Currently, the 0.75- and 1.5-mg strengths of dulaglutide are more readily available than the 3- and 4.5-mg doses. Prescribing the 1.5-mg pen to be injected twice or thrice weekly as an equivalent to the 3- or 4.5-mg dosage, respectively, may be an alternative when higher-strength pens are unavailable. Although some sources recommend against this method because of potentially diminishing the supply of the lower doses (3), this dosing schedule was used in a phase 2 study (24) and thus can be considered a reasonable alternative (25).
A similar approach could be considered with tirzepatide, which is available in strengths of 2.5, 5, 7.5, 10, 12.5, and 15 mg (19). However, the manufacturer of tirzepatide recommends against the practice of additive multiple injections because it has not been studied (26). Outside of the increased burden to the patient of injecting multiple smaller doses to achieve the desired therapeutic dose, this practice may be reasonable when there are supply shortages of only higher-dose strengths, as presently experienced with dulaglutide.
Extended Interval Dosing
Although drug shortages may stimulate hospitals to ration medications among patients through an ethical allocation selection process (7), ambulatory patients may likewise choose to ration their own supply to lengthen the duration of their available GLP-1 receptor agonist product. To do so, patients may report intentionally extending the interval between doses and may inquire about how long they can safely delay their next dose without fully forfeiting efficacy. Because clinical trials and outcome studies are conducted with a specific dosing frequency in mind, the literature does not provide much of an evidence base to support extended dosing schedules. However, considerations for extending the dosing interval may be theorized according to the product’s half-life. It is important to note that these adjustments are theoretical and may not translate into a clinically supported outcome; however, in times of supply shortages, it may be preferred to a cessation of therapy for an extended period.
Interchanging GLP-1 Receptor Agonist Therapies
Selecting a Therapeutic Equivalent Dose for Glucose Lowering
Although there are other reasons for interchanging one GLP-1 receptor agonist for another agent in the class (e.g., enhanced glucose efficacy, better weight loss, improved tolerability, and cost), the present rationale is to sidestep limited access caused by the drug shortage. In time, results from several studies will guide equivalent dosing interchanges within the GLP-1 receptor agonist class (27,28). Several head-to-head studies have been conducted comparing one GLP-1 receptor agonist to another, but no study exists comparing doses among all of the agents in this class. Thus, establishing dosing equivalence is challenging but can be valuable in a real-world clinical setting (29,30).
Indirect comparisons among studies provide insight to developing this practical guidance. Trujillo et al. (31,32) provided comprehensive analyses of GLP-1 receptor agonist investigations using active comparative treatment arms. Since these analyses were published, higher GLP-1 receptor agonist therapeutic doses have gained U.S. Food and Drug Administration (FDA) approval, and new products such as the dual GLP-1/GIP receptor agonist tirzepatide have reached the market; these newer products must also be considered in the spectrum of dosing equivalents within the therapeutic class (19,33,34). Table 4 offers a suggested comparison of the equivalent doses for currently available GLP-1 receptor agonists based on their glycemic impact (31–35). This information can support clinical decision-making, especially during this time of GLP-1 receptor agonist shortages.
TABLE 4 GLP-1 Receptor Agonist Drug Shortages and Suggested Comparative Doses for Treating Type 2 Diabetes
Agent Dosing Route and Interval Comparative Doses
Exenatide SC twice daily 5 μg* 10 μg
Lixisenatide SC daily 10 μg* 20 μg
Liraglutide SC weekly 0.6 mg* 1.2 mg 1.8 mg
Exenatide XR SC weekly 2 mg
Dulaglutide SC weekly 0.75 mga* 1.5 mga 3 mgb† 4.5 mgb†
Semaglutide SC weekly 0.25 mgb* 0.5 mgb 1 mga 2 mga‡
Semaglutide PO daily 3 mg* 7 mg 14 mg
Tirzepatide SC weekly 2.5 mga* 5 mga‡ 7.5 mga 10 mga 12.5 mga 15 mga
According to the FDA’s drug shortage database as of 10 March 2023 (2), patients may have limited or intermittent access in community pharmacies to three agents in varying doses: dulaglutide, injectable semaglutide, and tirzepatide.
a Drug doses that are currently in short supply but still available.
b Drug doses with only limited or intermittent availability.
* Comparative efficacy of starting doses is not known and based on the clinical judgement of authors.
† Based on information from ref. 33.
‡ Based on information from ref. 35. PO, by mouth; SC, subcutaneous.
Methods for Interchanging GLP-1 Receptor Agonists
The nidus prompting the therapeutic interchange guides the method by which a GLP-1 receptor agonist substitution should occur. If the interchange is desired to overcome a supply shortage, initiating the new agent in place of the original product at an equivalent dose is reasonable (Table 4) (29,30). However, an equivalent dose chart should be used as a starting guide. Additional patient-specific factors to consider include the current degree of glucose control and potential need for additional glucose lowering, the length of time the patient has been off the medication, and how the patient tolerated the GLP-1 receptor agonist initially. When switching from a product administered once or twice daily, begin use of the new product the day after discontinuing the original product. When switching from a weekly administered product, begin the new product 7 days after discontinuing the original product.
Managing Prior Authorizations
Interchanging a GLP-1 receptor agonist in short supply with an available one may require prescribing a nonformulary product or one that requires prior authorization. Maintaining a collaborative line of communication between the prescriber and the pharmacy may help to facilitate access to the alternative GLP-1 receptor agonist (36). Open communication may also help prescribers identify available GLP-1 receptor agonists and likewise may prompt community pharmacists to efficiently return prior authorization requests to prescribers. Clear documentation in the medical record explaining the reason, importance, and necessity of the GLP-1 receptor agonist interchange also may expedite access for patients. Designating a staff member to complete prior authorizations will additionally facilitate the process (7). Lastly, it would be reasonable for managed care organizations to provide leniency for the use of alternative GLP-1 receptor agonist therapies when a preferred agent is in short supply.
Avoiding New GLP-1 Receptor Agonist Prescriptions and Using Alternative Antihyperglycemic Agents
During drug shortages, when possible, avoid prescribing medications that are in short supply for a drug-naive patient (3). If comparable GLP-1 receptor agonists are also unavailable, considering the use of another antihyperglycemic treatment may be necessary to control a patient’s glycemic status instead of prescribing a new GLP-1 receptor agonist (9). The alternative therapy should be selected through shared decision-making to identify the most appropriate product based on a given patient’s needs and preferences. Once the GLP-1 receptor agonist shortage resolves, the preferred therapy can be added or interchanged with the temporary agent.
For patients with chronic kidney disease, heart failure, or established atherosclerotic cardiovascular disease (ASCVD) or those at high risk for ASCVD, substituting a sodium–glucose cotransporter 2 (SGLT2) inhibitor may be a preferred choice (37). Products from this class are taken by mouth once daily, have a negligible risk of causing hypoglycemia, and also facilitate weight loss. Additionally, like several of the GLP-1 receptor agonists, SGLT2 inhibitors have benefits that extend beyond glucose control. Many hold extended FDA indications for reducing the risk of major adverse cardiovascular events, cardiovascular death, and/or hospitalization for heart failure or for slowing decline in kidney function. Although not as robust at lowering glucose or weight as some of the long-acting GLP-1 receptor agonists, SGLT2 inhibitors are a reasonable substitute, particularly for patients with type 2 diabetes and compelling conditions.
Because of the overlapping mechanisms of dipeptidyl peptidase 4 (DPP-4) inhibitors, a product from this class may be a reasonable selection for temporary use in place of a GLP-1 receptor agonist. Similar to short-acting GLP-1 receptor agonists, DPP-4 inhibitors provide postprandial glucose reduction with a minimal impact on fasting blood glucose; yet, their A1C benefit is considerably less, particularly when compared with the glycemic efficacy of long-acting GLP-1 receptor agonists. Substituting a DPP-4 inhibitor for a long-acting GLP-1 receptor agonist will leave the patient’s fasting blood glucose component unaddressed and may therefore require the addition of another agent. However, the DPP-4 inhibitors are conveniently dosed by mouth once daily without regard to meals.
Other options are also available as an alternative to GLP-1 receptor agonists if not already in use, including metformin, thiazolidinediones (TZDs), sulfonylureas, or basal insulin. As with any other therapy, risks, benefits, cost, dosing frequency, and method of administration should be considered with the patient to bridge the unavailability of the preferred therapy. It is worthwhile to reevaluate metformin therapy if the patient is not using metformin or the dose is not maximized. Metformin effectively lowers glucose without the risk of hypoglycemia or weight gain. TZDs are another option that offers good glycemic control and a low risk of hypoglycemia, but they are limited by side effects, including weight gain and edema. Sulfonylureas effectively lower A1C but increase the risk of hypoglycemia and facilitate weight gain. Still, these agents are all low-cost and easily administered once or twice daily by mouth. Basal insulin is another option to lower glucose. The dose can be individualized based on blood glucose monitoring. However, like sulfonylureas, basal insulin also increases the risk of hypoglycemia and weight gain.
Patient Education Through the GLP-1 Receptor Agonist Shortage
Regardless of the method selected to overcome the current limited access to GLP-1 receptor agonists, educating patients and setting expectations during the transitional period is time well spent. Caution patients to consider the credibility and reliability of online sources and explain the concerns related to compounded products (4,5).
For interchanged GLP-1 receptor agonists, provide adequate education about the new delivery device or method of administration, given that administration routes, dosing frequencies, and delivery devices are not universal among GLP-1 receptor agonists; the nuances for administration are not intuitive to patients. Using a demonstration device at the time of prescribing and dispensing can help overcome administration challenges at home.
For alternative pharmacotherapies, help to set expectations regarding side effect profiles, onset of benefit, and dosing schedules. Finally, it is prudent to advise patients to self-monitor their blood glucose more frequently during therapy transition to identify and mitigate hyperglycemia and hypoglycemia (38). Review glycemic goals and discuss an action plan for managing glucose excursions to avoid the need for urgent or emergent care.
Conclusions
Drug shortages are inconvenient to patients, providers, and the health care system, and the GLP-1 receptor agonist shortage has been no exception. Adequate adjustments to patients’ medications are necessary to sidestep the risk of hyperglycemia resulting from the unavailability of preferred GLP-1 receptor agonist drugs. Several options are available, although not necessarily ideal. Once the drug shortage is resolved, patients, in collaboration with their providers, may choose to reinitiate their preferred GLP-1 receptor agonist therapy to optimize their diabetes management.
Article Information
Duality of Interest
H.P.W. is an advisor for Abbott Laboratories. J.M.T. is an advisor to Novo Nordisk and Sanofi. J.J.N. is a speaker for Dexcom, a consultant to Bayer, and an advisor to Boehringer Ingelheim, Eli Lilly, and Sanofi. No other potential conflicts of interest relevant to this article were reported.
Author Contributions
H.P.W. researched data and wrote the manuscript. J.M.T. and J.J.N. researched data and reviewed/edited the manuscript. H.P.W. is the guarantor of this work and, as such, had full access to all the information reported and takes responsibility for the integrity of the content.
==== Refs
References
1. Putka S. Diabetes patients also struggle to access GLP-1 agonists: patients are making substitutions, crossing state lines, stopping care because of shortages. Available from https://www.medpagetoday.com/special-reports/features/102773#:∼:text=All%20three%20of%20the%20most,loss%20and%20is%20expected%20to. Accessed 10 March 2023
2. U.S. Food and Drug Administration. FDA drug shortages. Available from https://www.accessdata.fda.gov/scripts/drugshortages/default.cfm. Accessed 10 March 2023
3. Beba H, Ranns C, Hambling C, Diggle J, Brown P. PCDS consensus statement: a strategy for managing the supply shortage of the GLP-1 RAs Ozempic and Trulicity. Diabetes & Primary Care 2022;24 :147–149
4. Mattingly AN. The role of outsourcing facilities in overcoming drug shortages. J Am Pharm Assoc (2003) 2021;61 :e110–e114 32943335
5. Landsverk G. Shortages of a ‘game changer’ weight-loss drug are driving people to buy potentially risky knockoff versions. Available from https://www.insider.com/buy-compounded-semaglutide-online-risks-wegovy-ozempic-2023-1. Accessed 10 March 2023
6. Drug shortages lead to rationing, treatment delay. Am J Nurs 2020;120 :14–15
7. Hantel A, Siegler M, Hlubocky F, Colgan K, Daugherty CK. Prevalence and severity of rationing during drug shortages: a national survey of health system pharmacists. JAMA Intern Med 2019;179 :710–711 30907925
8. Shukar S, Zahoor F, Hayat K, . Drug shortage: causes, impact, and mitigation strategies. Front Pharmacol 2021;12 :693426 34305603
9. Ventola CL. The drug shortage crisis in the United States: causes, impact, and management strategies. P T 2011;36 :740–757, 749–757 22346307
10. Panic G, Yao X, Gregory P, Austin Z. How do community pharmacies in Ontario manage drug shortage problems? Results of an exploratory qualitative study. Can Pharm J 2020;153 :371–377
11. Jimmy B, Jose J. Patient medication adherence: measures in daily practice. Oman Med J 2011;26 :155–159 22043406
12. Novo Nordisk. Victoza (liraglutide) prescribing information. Bagsvaerd, Denmark, Novo Nordisk, July 2022
13. Novo Nordisk. Ozempic (semaglutide) injection prescribing information. Plainsboro, NJ, Novo Nordisk, October 2022
14. Eli Lilly and Company. Trulicity (dulaglutide) injection prescribing information. Indianapolis, IN, Eli Lilly and Company, November 2022
15. AstraZeneca. Bydureon BCise (exenatide extended-release) injectable suspension prescribing information. Wilmington, DE, AstraZeneca Pharmaceuticals, December 2022
16. AstraZeneca. Byetta (exenatide) injection prescribing information. Wilmington, DE, AstraZeneca Pharmaceuticals, December 2022
17. Sanofi. Adlyxin (lixisenatide) injection prescribing information. Bridgewater, NJ, Sanofi, June 2022
18. Novo Nordisk. Rybelsus (semaglutide) tablets prescribing information. Plainsboro, NJ, Novo Nordisk, January 2023
19. Eli Lilly and Company. Mounjaro (tirzepatide) injection prescribing information. Indianapolis, IN, Eli Lilly and Company, September 2022
20. Eli Lilly and Company. If a patient stopped Trulicity (dulaglutide) treatment for a few weeks, should they restart on their previous dose or re-escalate from the 0.75 mg dose? Available from https://www.lillymedical.com/en-us/answers/if-a-patient-stopped-trulicity-dulaglutide-treatment-for-a-few-weeks-should-they-restart-on-their-previous-dose-or-re-escalate-from-the-0-75-mg-dose-152166?hcpToken=A12DSa08bhrd123gg8&channel=AE. Accessed 13 March 2023
21. Frías JP, Davies MJ, Rosenstock J, . Tirzepatide versus semaglutide once weekly in patients with type 2 diabetes. N Engl J Med 2021;385 :503–515 34170647
22. MacFadyen J, Maedley A. Micro-dosing of Ozempic. Available from https://img1.wsimg.com/blobby/go/be2ff530-09ee-47b8-912b-a0144799f206/downloads/Micro-dosing-of-Ozempic.pdf?ver=1586887677774. Accessed 10 March 2023
23. Novo Nordisk. The Ozempic 0.25 mg and 0.5 mg doses will be delivered from a 3 mL pen. Available from https://www.novomedlink.com/content/dam/novonordisk/novomedlink/new/diabetes/products/resources/library/documents/ozempic-information-sheet.pdf. Accessed 10 March 2023
24. Frias JP, Wynne AG, Matyjaszek-Matuszek B, . Efficacy and safety of an expanded dulaglutide dose range: a phase 2, placebo-controlled trial in patients with type 2 diabetes using metformin. Diabetes Obes Metab 2019;21 :2048–2057 31050143
25. Eli Lilly and Company. Can multiple Trulicity (dulaglutide) injections be used to make a larger dose? Available from https://www.lillymedical.com/en-us/answers/can-multiple-trulicity-dulaglutide-injections-be-used-to-make-a-larger-dose-51139?hcpToken=A12DSa08bhrd123gg8&channel=AE. Accessed 10 March 2023
26. Eli Lilly and Company. Can multiple tirzepatide injections be combined to make a larger dose? Available from https://www.lillymedical.com/en-us/answers/can-multiple-tirzepatide-injections-be-combined-to-make-a-larger-dose-165235?hcpToken=A12DSa08bhrd123gg8&channel=AE. Accessed 10 March 2023
27. ClinicalTrials.gov. A study of tirzepatide (LY3298176) in adult participants with type 2 diabetes switching from dulaglutide (SURPASS-SWITCH). Available from https://www.clinicaltrials.gov/ct2/show/NCT05564039. Accessed 10 March 2023
28. ClinicalTrials.gov. A study of tirzepatide (LY3298176) in adults with type 2 diabetes switching from a GLP-1 RA (SURPASS-SWITCH-2). Available from https://www.clinicaltrials.gov/ct2/show/NCT05706506?cond=surpass+switch-2&draw=2&rank=1. Accessed 10 March 2023
29. Almandoz JP, Lingvay I, Morales J, Campos C. Switching between glucagon-like peptide-1 receptor agonists: rationale and practical guidance. Clin Diabetes 2020;38 :390–402 33132510
30. Jain AB, Ali A, Gorgojo Martínez JJ, . Switching between GLP-1 receptor agonists in clinical practice: expert consensus and practical guidance. Int J Clin Pract 2021;75 :e13731 32975890
31. Trujillo JM, Nuffer W, Ellis SL. GLP-1 receptor agonists: a review of head-to-head clinical studies. Ther Adv Endocrinol Metab 2015;6 :19–28 25678953
32. Trujillo JM, Nuffer W, Smith BA. GLP-1 receptor agonists: an updated review of head-to-head clinical studies. Ther Adv Endocrinol Metab 2021;12 :2042018821997320 33767808
33. Frías JP, Bonora E, Nevarez Ruiz L, . Efficacy and safety of dulaglutide 3.0 mg and 4.5 mg versus dulaglutide 1.5 mg in metformin-treated patients with type 2 diabetes in a randomized controlled trial (AWARD-11). Diabetes Care 2021;44 :765–773 33397768
34. Frías JP, Auerbach P, Bajaj HS, . Efficacy and safety of once-weekly semaglutide 2·0 mg versus 1·0 mg in patients with type 2 diabetes (SUSTAIN FORTE): a double-blind, randomised, phase 3B trial. Lancet Diabetes Endocrinol 2021;9 :563–574 34293304
35. Vadher K, Patel H, Mody R, . Efficacy of tirzepatide 5, 10 and 15 mg versus semaglutide 2 mg in patients with type 2 diabetes: an adjusted indirect treatment comparison. Diabetes Obes Metab 2022;24 :1861–1868 35589616
36. Tan YX, Moles RJ, Chaar BB. Medicine shortages in Australia: causes, impact and management strategies in the community setting. Int J Clin Pharm 2016;38 :1133–1141 27383246
37. ElSayed NA, Aleppo G, Aroda VR, .; American Diabetes Association. 9. Pharmacologic approaches to glycemic treatment: Standards of Care in Diabetes—2023. Diabetes Care 2023;46 (Suppl. 1 ):S140–S157 36507650
38. Michigan Collaborative for Type 2 Diabetes. Supporting patients during GLP-1 receptor agonist shortages. Available from https://assets.ctfassets.net/1p0t6jhbvwk9/nfGRbeVaKRxZJ5UaCStVl/dfabf3bb1315a9f7ce03f17c08ea7daa/MCT2D-GLP1-shortages-N02032023.pdf. Accessed 15 March 2023
|
PMC010xxxxxx/PMC10338284.txt
|
==== Front
Clin Diabetes
Clin Diabetes
clinical diabetes
Clinical Diabetes : A Publication of the American Diabetes Association
0891-8929
1945-4953
American Diabetes Association
CD220080
10.2337/cd22-0080
Feature Articles
Association Between Change in A1C and Use of Professional Continuous Glucose Monitoring in Adults With Type 2 Diabetes on Noninsulin Therapies: A Real-World Evidence Study
Nemlekar Poorva M.
Hannah Katia L.
https://orcid.org/0000-0001-7989-9597
Norman Gregory J.
Dexcom, Inc., San Diego, CA
Corresponding author: Gregory J. Norman, [email protected]
Summer 2023
24 1 2023
24 1 2023
41 3 359366
© 2023 by the American Diabetes Association
2023
https://www.diabetesjournals.org/journals/pages/license Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.
This retrospective analysis examined the association between change in A1C and professional continuous glucose monitoring (p-CGM) use in patients with type 2 diabetes and poor glycemic control who were not using insulin. Data from 15,481 eligible patients (p-CGM users n = 707 and p-CGM nonusers n = 14,774) showed a greater decrease in A1C from baseline to the end of follow-up for p-CGM users, and differences favored p-CGM users regardless of whether they started insulin therapy during the follow-up period. These findings suggest that people with type 2 diabetes who have poor glycemic control using multiple noninsulin therapies may benefit from p-CGM, which can reduce A1C over a 6-month period compared with usual care.
==== Body
pmcCurrent treatment guidelines for diabetes suggest a glycemic target of A1C <7% for most people with diabetes (1) and the use of metformin as a first-line therapy with lifestyle changes for those with type 2 diabetes (2). Yet, a significant percentage of people with type 2 diabetes do not achieve their glycemic goals (3–5), which can lead to the development and progression of diabetes complications (6–11). Most people usually require a second-line therapy within a couple of years after treatment initiation. Additionally, there is often resistance among physicians to start patients on insulin therapy (5,12–15). When insulin is introduced, there is often poor patient adherence, which ultimately affects glycemic control (3,16–18).
Currently, once people with type 2 diabetes start insulin therapy, the most common method of glucose monitoring is self-monitoring of blood glucose (SMBG) using a blood glucose meter and test strips. Glucose monitoring is considered an important component of effective diabetes management for people taking insulin and is encouraged to prevent hypoglycemia and hyperglycemia (19). Some people on intensive insulin therapy, defined as multiple daily injections of insulin or the use of an insulin pump, may need to check their blood glucose levels 6–10 times per day. SMBG frequency was found to be correlated with lower A1C for people with type 1 diabetes (20,21).
For people with noninsulin-treated type 2 diabetes, SMBG has been shown to be less helpful for reaching glycemic goals unless the data are being used to make lifestyle changes or manage medications through active, shared decision-making between patients and health care providers (HCPs) (19,22–25). For example, in a meta-analysis by Mannucci et al. (24), the use of structured SMBG regimens with clearly defined timing and frequency of glucose measurements resulted in improved glycemic control compared with unstructured SMBG or no SMBG. The largest improvement in A1C was found for structured SMBG when the data were also used to adjust diabetes medications. However, patients and physicians often do not actively leverage the glucose data from SMBG. Other limitations to using SMBG are that it only provides a static point-in-time glucose value, it requires frequent painful fingersticks, and insurance coverage for test strips is limited for most people who do not take insulin (26).
Continuous glucose monitoring (CGM) systems may be a useful tool to help improve glycemic control prior to starting insulin therapy. People with diabetes, HCPs, and health care team members have started using CGM technology to achieve better glycemic outcomes (27–29). The two currently available type of personal CGM systems (i.e., CGM systems owned by individual patients) are real-time and intermittently scanned CGM systems. However, there are cost and access limitations for CGM for people with type 2 diabetes who are not treated with intensive insulin therapy. Presently, CGM is recommended and covered by insurance plans for people with type 1 diabetes and for those with type 2 diabetes who require intensive insulin therapy. However, noninsulin-treated patients with type 2 diabetes may not have insurance coverage for personal CGM use, and there are no established treatment guidelines for personal CGM use in this population (30).
In addition to personal CGM, professional CGM (p-CGM) systems are devices owned by clinics and provided by HCPs to patients for short-term use (usually a period of 3–10 days) to continuously collect glycemic data. p-CGM sensors transmit data to a receiver, which is often blinded to the patient, and the data are then assessed by the HCP. Hence, p-CGM is often referred to as masked or retrospective CGM (31,32). Newer real-time p-CGM devices can be used in blinded or unblinded modes (33,34). HCPs use p-CGM as a tool for identifying glycemic patterns to make informed decisions about patients’ treatment regimens and lifestyle modifications (29). In addition, use of p-CGM is beneficial in setting appropriate individualized targets for time in range (TIR; the percentage of time spent with glucose of 70–180 mg/dL), as well as identifying nocturnal glucose patterns (33). p-CGM may also be applicable for people who have reservations about committing to long-term personal CGM use, allowing them to try wearing a CGM sensor before purchasing a system and for those who do not have adequate insurance coverage for personal CGM but may still benefit from the period collection of CGM data (34).
Two recent studies investigated the use of p-CGM for type 2 diabetes management regardless of insulin therapy regimen (32,35). A quality improvement project in a team-based primary care setting evaluated care models using 2 weeks of blinded p-CGM, followed by an HCP visit that included shared decision-making regarding lifestyle and medication modifications (35). The use of p-CGM resulted in improved glycemic management (average change in A1C of −0.6%), which was attributed to lifestyle counseling and medication intensification, while the number of medications remained stable. Additionally, retrospective analysis of U.S. health care claims and laboratory data of patients with type 2 diabetes showed that use of p-CGM was associated with improved A1C and decreased growth in total health care spending 1 year after the use of p-CGM compared with 1 year before p-CGM (32).
For people with type 2 diabetes who do not use insulin, a recent pilot study assessed the episodic use of unblinded real-time p-CGM (10 days/month for 3 months) in patients who had previously failed to achieve glycemic goals while on multiple noninsulin therapies. Although 34.1% of patients reached an A1C <7.5% at 12 weeks, overall change in A1C did not differ statistically between groups. A clinically meaningful 10% increase in average TIR within the first week of use suggested that patients were able to make lifestyle changes based on real-time p-CGM data (36).
Although these previous studies of p-CGM have shown some evidence of improved glycemic control in people with type 2 diabetes, the aims of this study were to further examine the real-world potential value of p-CGM for individuals with type 2 diabetes who have poor glycemic control while using two or more noninsulin antidiabetic therapies and to understand medication modifications after p-CGM use.
Research Design and Methods
Study Design
This was a retrospective, observational, database study using de-identified administrative health claims and linked laboratory data from Optum’s Clinformatics Data Mart (CDM) database. The CDM is statistically de-identified under the expert determination method consistent with the Health Insurance Portability and Accountability Act and managed according to Optum customer data use agreements in compliance with Code of Federal Regulations title 45, section 164.514(b) (1)13 (37). CDM administrative claims submitted for payment by HCPs and pharmacies are verified, adjudicated, and de-identified prior to inclusion. These data, including patient-level enrollment information, are derived from claims submitted for all medical and pharmacy health care services with information related to health care costs and resource utilization.
Study Population
The population included adult patients ≥30 years of age who had type 2 diabetes identified using International Classification of Diseases, 9th/10th Revisions (ICD-9/ICD-10), were enrolled in commercial or Medicare Advantage health plans, and had no prior personal CGM or p-CGM use. Patients with poor glycemic control, defined as an A1C value between 7.8 and 10.5%, and using two or more noninsulin therapies were included in the cohort. p-CGM users were identified using Current Procedural Terminology (CPT) codes 95250 and 95251 between 1 January 2018 and 31 October 2020 (the study identification period). Patients without claims for use of p-CGM who had a pharmacy claim for an oral antidiabetic drug (OAD) during the study identification period were selected as the control cohort (nonusers of p-CGM). An index date was set as the earliest observed claim for p-CGM or an OAD.
Included patients were continuously enrolled in a health plan from at least 6 months pre-index (baseline) to at least 6 months post-index date (follow-up) and had no claims for a CGM device (personal or professional) during the baseline period.
Type 2 diabetes diagnoses were confirmed by patients having one or more inpatient hospital or emergency department medical claims or at least two ambulatory medical claims at least 30 days apart with an ICD-9 or ICD-10 diagnosis code for diabetes in any position on the claim during the baseline period. Additionally, patients were required to be using at least two noninsulin medications and not taking insulin (basal or bolus) in the baseline period. Finally, patients were required to have one or more laboratory A1C test results with values between 7.8 and 10.5% during the baseline period (including the index date) and at least one A1C laboratory test value during the follow-up period.
Outcome Measures
The primary outcome measure was change in A1C determined from the average of available A1C values during the baseline and follow-up periods. Change in A1C was computed as the average A1C follow-up value minus the average A1C baseline value, with negative values indicating improvement in A1C. Secondary outcomes were change in the number of medications by class, insulin use in follow-up period, and change in A1C for patients starting insulin and not starting insulin during the follow-up period.
Statistical Analysis
Descriptive statistics, including percentages, means, and SDs, were calculated for patient characteristics and study outcomes and presented by patient cohort categories (i.e., p-CGM users and nonusers). Bivariate differences between the p-CGM user and nonuser groups were tested with independent sample t tests for continuous measures and χ2 tests for categorical measures. Within-cohort comparisons of continuous measures were tested with paired t tests. The difference-in-differences (DiD) estimate was calculated as the difference in change in A1C values between the p-CGM user and nonuser cohorts. The DiD estimate indicates the magnitude and direction of change in outcome between the two groups. The association between p-CGM use and antidiabetic medication changes was tested using a logistic regression model. A χ2 test was used to compare the difference in insulin use in the follow-up period between the two groups. For all analyses, statistical tests were two-tailed, with P ≤0.05 considered statistically significant. Analyses were performed using Instant Health Data software (Panalgo, Boston, MA) and R, v. 3.2.1, software (R Foundation for Statistical Computing, Vienna, Austria).
Results
A total of 15,481 patients were identified during the study identification period, including 707 p-CGM users and 14,774 nonusers. Demographic characteristics were similar between cohorts as shown in Table 1. The majority in both groups were older (≥65 years of age), there were slightly more males, and most were Caucasian and predominantly covered by Medicare Advantage plans. Among patients using a p-CGM, an endocrinologist was the prescriber for 282 patients (39.9%), whereas for 275 patients (38.9%), the prescriber was a family or internal medicine physician or nurse practitioner. The remaining 21.2% of encounters at which p-CGM was prescribed were with some other type of HCP. Thus, ∼40% of patients received their p-CGM through primary care.
TABLE 1 Baseline Demographic Characteristics of the Study Cohort (N = 15,481)
Characteristic p-CGM Users (n = 707) Nonusers (n = 14,774) P
Age, years 66.1 ± 10.8 (30–89) 66.7 ± 10.9 (30–89) 0.13
Age-group, years
30–44
45–54
55–64
≥65
30 (4.2)
72 (10.2)
157 (22.2)
448 (63.4)
541 (3.7)
1,587 (10.7)
3,064 (20.7)
9,582 (64.9) 0.63
Sex
Female
Male
344 (48.7)
363 (51.3)
6,846 (46.3)
7,926 (53.7) 0.24
Race/ethnicity
Asian
Black
Caucasian
Hispanic
47 (7.0)
106 (15.8)
337 (50.2)
182 (27.1)
928 (6.6)
1,954 (13.8)
7,367 (52.1)
3,884 (27.5) 0.49
Geographical region
Midwest
Northeast
South
West
29 (4.1)
129 (18.3)
426 (60.3)
123 (17.4)
1,668 (11.3)
1,421 (9.6)
7,886 (53.4)
3,795 (25.7) <0.0001
Payer type
Commercial
Medicare
200 (28.3)
507 (71.7)
3,971 (26.9)
10,803 (73.1) 0.43
Charlson comorbidity index score 1.65 ± 1.6 1.42 ± 1.5 <0.0001
Blood glucose test strips claims at baseline 0.86 ± 1.3 0.77 ± 1.3 0.08
Data are mean ± SD (range), mean ± SD, or n (%).
The p-CGM group reduced their A1C by a mean 0.83%, from 8.70 to 7.87%, while the nonusers had a reduction of 0.32%, from 8.56 to 8.23%. The DiD estimate was a −0.51% change in A1C and was statistically significant (P <0.0001), as shown in Table 2.
TABLE 2 Baseline, Follow-Up, and Change in A1C for All Patients and Stratified by Insulin Initiation in Follow-Up Period
A1C, % p-CGM Users Nonusers DiD
Baseline Follow-Up Difference* Baseline Follow-Up Difference* Estimate 95% CI P
All patients
(p-CGM users n = 707;
nonusers n = 14,774) 8.70 ± 0.78 7.87 ± 1.15 −0.83 8.56 ± 0.77 8.23 ± 1.21 −0.32 −0.51 −0.62 to −0.40 <0.0001
Patients who started insulin
during follow-up (p-CGM
users n = 140; nonusers
n = 1,450) 8.91 ± 0.80 8.34 ± 1.23 −0.57 8.77 ± 0.82 8.9 ± 1.34 0.13 −0.71 −0.98 to −0.44 <0.0001
Patients who did not start
insulin during follow-up
(p-CGM users n = 567;
nonusers n = 13,324) 8.65 ± 0.76 7.75 ± 1.09 −0.9 8.53 ± 0.76 8.16 ± 1.17 −0.37 −0.53 −0.64 to −0.41 <0.0001
Baseline and follow-up A1C data are mean ± SD.
* Difference is calculated as post-index value minus pre-index value.
As shown in Table 3, 140 p-CGM users (19.8%) started using insulin during the follow-up period, with most (112, or 15.8%) using basal insulin. About 10% of nonusers started insulin during the follow-up period. The A1C change among insulin users in the p-CGM group was −0.57%, from 8.90 to 8.34%, whereas the nonuser group had a slight increase in A1C of 0.13%, from 8.77 to 8.90%. The DiD estimate for change in A1C was −0.71% and statistically significant (Table 2). Similarly, for patients not starting insulin, we found a significant DiD change in A1C of −0.53% for nonusers of insulin during follow-up, reflecting a larger A1C decrease for p-CGM users compared with nonusers (Table 2).
TABLE 3 Medication Use During Baseline and Follow-Up Periods
Drug Class* p-CGM Users† (n = 707) Nonusers† (n = 14,774) OR 95% CI P
Baseline Follow-Up Baseline Follow-Up
Biguanide 486 (68.7) 452 (63.9) 10,631 (72.0) 10,383 (70.3) 0.68 0.54–0.86 0.0011
Sulfonylurea 410 (58.0) 351 (49.7) 9,335 (63.2) 9,401 (63.7) 0.39 0.31–0.49 <0.0001
DPP-4 inhibitor 144 (20.4) 124 (17.5) 3,017 (20.4) 3,163 (21.4) 0.61 0.46–0.80 0.0004
GLP-1 receptor agonist 187 (26.5) 249 (35.2) 1,792 (12.1) 2,113 (14.3) 3.30 2.59–4.21 <0.0001
SGLT2 inhibitor 162 (22.9) 191 (27.0) 2,002 (13.5) 2,362 (16.0) 1.62 1.26–2.09 0.0002
Thiazolidinedione 93 (13.2) 110 (15.6) 1,777 (12.0) 2,071 (14.0) 1.12 0.82–1.54 0.47
Meglitinide 36 (5.1) 53 (7.5) 282 (1.9) 320 (2.2) 4.23 2.66–6.73 <0.0001
α-Glucosidase inhibitor 7 (1.0) 6 (0.9) 137 (0.9) 156 (1.1) 0.54 0.16–1.83 0.32
Insulin
Total insulin‡
Basal insulin§
Bolus insulin§
Mixed insulin§
NA
NA
NA
NA
140 (19.80)
112 (15.84)
45 (6.36)
13 (1.84)
NA
NA
NA
NA
1,450 (9.81)
1,219 (8.25)
289 (1.96)
148 (1) <0.0001
* Table reports medication at therapeutic drug class level. Example medications in each class include biguanide (metformin), sulfonylurea (glipizide), DPP-4 inhibitor (linagliptin), GLP-1 receptor agonist (exenatide), SGLT2 inhibitor (empagliflozin), thiazolidinedione (pioglitazone), meglitinide (repaglinide), and α-glucosidase inhibitor (acarbose). NA, not applicable.
† Data are n (%).
‡ Data are n (%) of total insulin users in the cohort.
§ Insulin users are in more than one group.
Table 3 shows the diabetes medication use by cohorts during the baseline and follow-up periods. The most frequently used medications in both cohorts were biguanides followed by sulfonylureas, glucagon-like peptide 1 (GLP-1) receptor agonists, and dipeptidyl peptidase 4 (DPP-4) inhibitors in the baseline and follow-up periods. Biguanide, sulfonylurea, and DPP-4 inhibitor use decreased slightly from baseline to follow-up in both groups, with odds ratios (ORs) showing that the p-CGM group was less likely to use these medications at follow-up, adjusting for baseline, compared with the nonuser group. GLP-1 receptor agonist, sodium–glucose cotransporter 2 (SGLT2) inhibitor, and meglitinide use increased in both groups, with ORs showing that the p-CGM group was more likely to use these medications at follow-up, adjusting for baseline, compared with the nonuser group. Overall, 447 patients (63.2%) had an additional p-CGM use during the 6-month follow-up period.
Discussion
The findings from this study suggest that there is a glycemic benefit of p-CGM use for adults with type 2 diabetes who are not on insulin therapy and are taking multiple OAD and/or noninsulin injectable medications with poor glycemic control. Use of p-CGM was associated with a clinically meaningful 0.51% reduction in A1C compared with the cohort of nonusers.
These findings align with other studies demonstrating a significant reduction in A1C values for p-CGM users with type 2 diabetes compared with control subjects (32,34,35). Previous studies have shown that p-CGM enables HCPs to make decisions such as adding insulin to patients’ treatment regimens based on recorded sensor data, helping patients reach glycemic targets and promoting positive behavior change (35,36,38). Our findings suggest that p-CGM use is also a means to assess and monitor glycemia for noninsulin-treated patients and can be used as a tool for therapy modification.
Our analysis showed that ∼20% of patients initiated insulin therapy in the follow-up period after p-CGM use, which was more than the 10% who started insulin in the nonuser group, but this finding might be considered lower than expected given that the cohort started with an average A1C >7.5% and may reflect HCP and/or patient inertia regarding initiating insulin therapy (12,14,15).
Even among patients who used p-CGM and did not start insulin therapy, there was significant improvement in A1C. This finding suggests that p-CGM can help with medication management and behavior change that might improve glycemic control and delay the need to start insulin therapy. The more granule daily patterns of glucose excursions that can be viewed through CGM, compared with average glycemic control reflected in A1C values, can help HCPs target diet and physical activity counseling to patients, which in turn could delay or prevent microvascular and macrovascular diabetes complications (8,10,38).
Associations were found between p-CGM use and changes in the use of noninsulin diabetes medications. The most prominent change was a 33% increase in the use of GLP-1 receptor agonists, which seems warranted given the cardiovascular risk-reduction benefits of this drug class and the cohort’s baseline A1C and average age >65 years. There was also a 14% decrease in sulfonylurea use in the p-CGM group, which may have been precipitated by p-CGM detection of previously unknown time spent in the hypoglycemic range.
There are limitations to our study. The number of identified p-CGM users with A1C values was relatively small compared with the nonuser cohort. For this analysis, we included only patients who had A1C values in both the baseline and follow-up time periods. Although the cohorts were comparable with regard to demographic variables and baseline A1C, there was the chance of confounding by indication, through which patients using p-CGM could have been different from nonusers on unmeasured variables that prompted the use of p-CGM.
Although the database provided crucial information about participants’ health status, comorbidities, medications, and use of p-CGM, it did not provide other information such as socioeconomic status or participation in diabetes health coaching programs. We also cannot determine from the data whether and to what extent HCPs and patients engaged in shared decision-making about diabetes and lifestyle management. Similarly, changes in physical activity or other positive behavioral changes and their impact on overall glycemic improvement could not be assessed. Without this information, it is not possible to fully understand how p-CGM may help individuals to improve their diabetes management.
p-CGM use is documented in health care claims with CPT codes 95250 and 95251; therefore, it was not possible to distinguish the specific p-CGM brands or products used. Three p-CGM systems were available in the United States during this study period: the Dexcom G4/G6 Pro, the FreeStyle Libre Pro, and the Medtronic iPro2. Differences among these brands may influence users’ experience and adherence to wearing the devices (34).
A key strength of our analysis was its use of comprehensive, standardized health information contained within a large database. Access to the database allowed us to use quantitative data to verify participant eligibility and identify patients’ treatment regimens during the baseline and follow-up periods. An additional study strength was that the resulting cohorts were ethnically diverse, with nearly half being non-Caucasian, which supports the generalizability of the findings beyond Caucasian individuals.
Conclusion
Although randomized controlled trials are considered the gold standard for demonstrating the effects of therapeutic interventions, results from retrospective, observational studies of large health data systems can inform clinicians and regulatory agencies about the real-world efficacy of treatment interventions. Our findings suggest that p-CGM use in adults on noninsulin therapies is associated with improved glycemic control and thus may be a tool HCPs can use to help patients reach their glycemic goals.
Article Information
Acknowledgment
The authors thank Dr. David Price for his feedback on the data analysis and critical review of the manuscript.
Duality of Interest
The authors are employees of Dexcom, Inc. No other potential conflicts of interest relevant to this article were reported.
Author Contributions
P.M.N. researched the data, conducted data analyses, and wrote the manuscript. K.L.H. and G.J.N. researched the data and reviewed and edited the manuscript. P.M.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation
Portions of this article were presented at the 15th annual Advanced Technologies & Treatments for Diabetes meeting in Barcelona, Spain, 27–30 April 2022.
==== Refs
References
1. American Diabetes Association Professional Practice Committee. 6. Glycemic targets: Standards of Medical Care in Diabetes—2022. Diabetes Care 2022;45 (Suppl. 1 ):S83–S96 34964868
2. American Diabetes Association Professional Practice Committee. 9. Pharmacologic approaches to glycemic treatment: Standards of Medical Care in Diabetes—2022. Diabetes Care 2022;45 (Suppl. 1 ):S125–S143 34964831
3. Meneghini LF, Mauricio D, Orsi E, .; DUNE Investigators. The Diabetes Unmet Need with Basal Insulin Evaluation (DUNE) study in type 2 diabetes: achieving HbA1c targets with basal insulin in a real-world setting. Diabetes Obes Metab 2019;21 :1429–1436 30768845
4. Stone MA, Charpentier G, Doggen K, .; GUIDANCE Study Group. Quality of care of people with type 2 diabetes in eight European countries: findings from the Guideline Adherence to Enhance Care (GUIDANCE) study. Diabetes Care 2013;36 :2628–2638 23628621
5. Khunti K, Wolden ML, Thorsted BL, Andersen M, Davies MJ. Clinical inertia in people with type 2 diabetes: a retrospective cohort study of more than 80,000 people. Diabetes Care 2013;36 :3411–3417 23877982
6. Nathan DM, Cleary PA, Backlund JY, .; Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC) Study Research Group. Intensive diabetes treatment and cardiovascular disease in patients with type 1 diabetes. N Engl J Med 2005;353 :2643–2653 16371630
7. Control and Complications Trial Research Group; Nathan DM, Genuth S, Lachin J, . The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med 1993;329 :977–986 8366922
8. Ismail-Beigi F, Craven T, Banerji MA, .; ACCORD Trial Group. Effect of intensive treatment of hyperglycaemia on microvascular outcomes in type 2 diabetes: an analysis of the ACCORD randomised trial. Lancet 2010;376 :419–430 20594588
9. UK Prospective Diabetes Study (UKPDS) Group. Effect of intensive blood-glucose control with metformin on complications in overweight patients with type 2 diabetes (UKPDS 34). Lancet 1998;352 :854–865 9742977
10. Hayward RA, Reaven PD, Wiitala WL, .; VADT Investigators. Follow-up of glycemic control and cardiovascular outcomes in type 2 diabetes. N Engl J Med 2015;372 :2197–2206 26039600
11. Holman RR, Paul SK, Bethel MA, Matthews DR, Neil HAW. 10-Year follow-up of intensive glucose control in type 2 diabetes. N Engl J Med 2008;359 :1577–1589 18784090
12. Fu AZ, Sheehan JJ. Treatment intensification for patients with type 2 diabetes and poor glycaemic control. Diabetes Obes Metab 2016;18 :892–898 27160505
13. Khunti K, Damci T, Meneghini L, Pan CY; SOLVE Study Group. Study of Once Daily Levemir (SOLVE™): insights into the timing of insulin initiation in people with poorly controlled type 2 diabetes in routine clinical practice. Diabetes Obes Metab 2012;14 :654–661 22443213
14. Mata-Cases M, Franch-Nadal J, Real J, . Therapeutic inertia in patients treated with two or more antidiabetics in primary care: factors predicting intensification of treatment. Diabetes Obes Metab 2018;20 :103–112 28656746
15. Kostev K, Gölz S, Scholz BM, Kaiser M, Pscherer S. Time to insulin initiation in type 2 diabetes patients in 2010/2011 and 2016/2017 in Germany. J Diabetes Sci Technol 2019;13 :1129–1134 30862186
16. Kostev K, Dippel FW, Rathmann W. Glycemic control after initiating basal insulin therapy in patients with type 2 diabetes: a primary care database analysis. Diabetes Metab Syndr Obes 2015;8 :45–48 25609990
17. Pantalone KM, Misra-Hebert AD, Hobbs TM, . Intensification patterns and the probability of HbA1c goal attainment in type 2 diabetes mellitus: real-world evidence for the concept of ‘intensification inertia’. Diabet Med 2020;37 :1114–1124 30653705
18. Canivell S, Mata-Cases M, Real J, . Glycaemic control after treatment intensification in patients with type 2 diabetes uncontrolled on two or more non-insulin antidiabetic drugs in a real-world setting. Diabetes Obes Metab 2019;21 :1373–1380 30756446
19. American Diabetes Association Professional Practice Committee. 7. Diabetes technology: Standards of Medical Care in Diabetes—2022. Diabetes Care 2022;45 (Suppl. 1 ):S97–S112 34964871
20. Ziegler R, Heidtmann B, Hilgard D, Hofer S, Rosenbauer J; DPV-Wiss-Initiative. Frequency of SMBG correlates with HbA1c and acute complications in children and adolescents with type 1 diabetes. Pediatr Diabetes 2011;12 :11–17 20337978
21. Miller KM, Beck RW, Bergenstal RM, .; T1D Exchange Clinic Network. Evidence of a strong association between frequency of self-monitoring of blood glucose and hemoglobin A1c levels in T1D exchange clinic registry participants. Diabetes Care 2013;36 :2009–2014 23378621
22. Grant RW, Huang ES, Wexler DJ, . Patients who self-monitor blood glucose and their unused testing results. Am J Manag Care 2015;21 :e119–e129 25880487
23. Ipp E, Aquino RL, Christenson P. Point: Self-monitoring of blood glucose in type 2 diabetic patients not receiving insulin: the sanguine approach. Diabetes Care 2005;28 :1528–1530 15920087
24. Mannucci E, Antenore A, Giorgino F, Scavini M. Effects of structured versus unstructured self-monitoring of blood glucose on glucose control in patients with non-insulin-treated type 2 diabetes: a meta-analysis of randomized controlled trials. J Diabetes Sci Technol 2018;12 :183–189 28697625
25. Young LA, Buse JB, Weaver MA, .; Monitor Trial Group. Glucose self-monitoring in non-insulin-treated patients with type 2 diabetes in primary care settings: a randomized trial. JAMA Intern Med 2017;177 :920–929 28600913
26. Price D, Walker T. The rationale for continuous glucose monitoring-based diabetes treatment decisions and non-adjunctive continuous glucose monitoring use. Eur Endocrinol 2016;12 :24–30 29632583
27. Sherrill CH, Houpt CT, Dixon EM, Richter SJ. Effect of pharmacist-driven professional continuous glucose monitoring in adults with uncontrolled diabetes. J Manag Care Spec Pharm 2020;26 :600–609 32347180
28. Ulrich H, Bowen M. The clinical utility of professional continuous glucose monitoring by pharmacists for patients with type 2 diabetes. J Am Pharm Assoc 2021;61 :e76–e82
29. Bergenstal RM. Continuous glucose monitoring: transforming diabetes management step by step. Lancet 2018;391 :1334–1336 29459022
30. Beck RW, Bergenstal RM. Continuous glucose monitoring for type 2 diabetes: how does it compare with type 1 diabetes? Diabetes Technol Ther 2022;24 :153–156 34569811
31. Fonseca VA, Grunberger G, Anhalt H, .; Consensus Conference Writing Committee. Continuous glucose monitoring: a consensus conference of the American Association of Clinical Endocrinologists and American College of Endocrinology. Endocr Pract 2016;22 :1008–1021 27214060
32. Sierra JA, Shah M, Gill MS, . Clinical and economic benefits of professional CGM among people with type 2 diabetes in the United States: analysis of claims and lab data. J Med Econ 2018;21 :225–230 28994334
33. American Association of Diabetes Care & Education Specialists, American Association of Nurse Practitioners. Professional Continuous Glucose Monitoring Implementation Playbook. Chicago, IL, American Association of Diabetes Care & Education Specialists, 2020
34. Longo R, Sperling S. Personal versus professional continuous glucose monitoring: when to use which on whom. Diabetes Spectr 2019;32 :183–193 31462872
35. Simonson GD, Bergenstal RM, Johnson ML, Davidson JL, Martens TW. Effect of professional CGM (pCGM) on glucose management in type 2 diabetes patients in primary care. J Diabetes Sci Technol 2021;15 :539–545 33719598
36. Price DA, Deng Q, Kipnes M, Beck SE. Episodic real-time CGM use in adults with type 2 diabetes: results of a pilot randomized controlled trial. Diabetes Ther 2021;12 :2089–2099 34089138
37. U.S. Department of Health and Human Services, Office for Civil Rights. Guidance regarding methods for de-identification of protected health information in accordance with the Health Information Insurance Portability and Accountability Act (HIPAA) privacy rule. Available from https://www.hhs.gov/hipaa/for-professionals/privacy/special-topics/de-identification/index.html. Accessed 4 January 2023
38. Teodoro de Oliveira AO, Bartholomew K, Lavin-Tompkins J, Sperl-Hillen J. Use of continuous glucose monitoring as an educational tool in the primary care setting. Diabetes Spectr 2013;26 :120–123
|
PMC009xxxxxx/PMC9014343.txt
|
==== Front
Vascular
Vascular
spvas
VAS
Vascular
1708-5381
1708-539X
SAGE Publications Sage UK: London, England
10.1177_17085381221087816
10.1177/17085381221087816
Letter to Editor
Correspondence on “aortic enlargement and COVID-19”
https://orcid.org/0000-0003-0078-7897
Mungmunpuntipantip Rujittika 1
Wiwanitkit Viroj 2
1 Private Academic Consultant , Bangkok Thailand
2 Honorary professor, 75138 Dr DY Patil University , Pune, India
Rujittika Mungmunpuntipantip, Private Academic Consultant, 11 Bangkok, 112 Bangkok 10330, Thailand Email: [email protected]
13 4 2022
8 2023
13 4 2022
31 4 828829
© The Author(s) 2022
2022
SAGE Publications
This article is made available via the PMC Open Access Subset for unrestricted re-use and analyses in any form or by any means with acknowledgement of the original source. These permissions are granted for the duration of the COVID-19 pandemic or until permissions are revoked in writing. Upon expiration of these permissions, PMC is granted a perpetual license to make this article available via PMC and Europe PMC, consistent with existing copyright protections.
typesetterts10
==== Body
pmcDear Editor, we would like to share ideas on the publication “An interesting finding: What is the relation between aortic enlargement and COVID-19? 1 ” In the present study, medical records and thorax tomographies of patients were analyzed. Bitargil et al. mentioned that “the mean aortic diameter of COVID-19 patients is larger than non–COVID-19 patients with similar comorbidities referred to a pandemic …… COVID-19 was the leading factor. 1 ” We agree that COVID-19 might have some clinical effects on vascular system. If COVID-19 results in enlargement of aorta, the pathological process should be acute process that the change can be detected in a short period of infection. The possible pathological process might be an inflammatory process affect heart and large vessel. 2
The present study might show that there is a significant increased diameter of aorta in COVID-19 case. The diameter enlargement might or might not be associated with COVID-19. The important concern is on the pre-COVID-19 vascular status of the patients. In this study, there is no data on pre–COVID-19 vascular/health status of the patients. As Bitargil et al. noted, there are several possible confounding factors that might affect the aorta. Some patients might already have the vascular problem prior to COVID-19. In addition to the underlying personal illness, there is possibility of other concurrent medical problem in COVID-19 patient that might also cause enlargement of aorta. For example, dengue is a possible concurrent medical problem 3 and dengue can cause enlargement of aorta. 4
Regarding the investigation technique, thorax tomography, the diameter of the aorta grows with age and male gender. 5 To distinguish pathologic atherosclerotic alterations in the ascending aorta, gender-specific and age-adjusted aortic diameter are required. 5 Finally, it should be noted that the interpersonal variability in interpretation of aorta diameter from tomography image is high 5 and if there are many tomography radiologists in the present report by Bitargil et al., it is necessary to assess interpersonal variability. These points should be discussed and if further studies are conducted, it is necessary to recognize these basic considerations.
ORCID iD
Rujittika Mungmunpuntipantip https://orcid.org/0000-0003-0078-7897
The author(s) disclosed the following conflicting interest for the research, authorship, and/or publication of this article: authors cannot pay for any charge and ask for full waiving for this correspondence.
Funding: The author(s) received no financial support for the research, authorship, and/or publication of this article.
Authors’ contributions: RM 50 % and VW 50 %: (1a) Substantial contributions to study conception and design, (1b) Substantial contributions to acquisition of data, (1c) Substantial contributions to analysis and interpretation of data, (2) Drafting the article or revising it critically for important intellectual content, and (3) Final approval of the version of the article to be published.
==== Refs
References
1 Bitargil M Demir T Çetin HK , et al. An interesting finding: what is the relation between aortic enlargement and COVID-19? Vascular 2022:17085381211068228. doi: 10.1177/17085381211068228. Online ahead of print.
2 Siddiqi HK Libby P Ridker PM . COVID-19 - a vascular disease. Trends Cardiovasc Med 2021; 31 (1 ): 1–5.33068723
3 Nunez-Avellaneda D Villagómez FR Villegas-Pineda JC , et al. Evidence of coinfections between SARS-CoV-2 and select Arboviruses in Guerrero, Mexico, 2020-2021. Am J Trop Med Hyg 2022; 106 (3 ): 896–899.35073512
4 Tahir H Daruwalla V Hayat S . Myocarditis leading to severe dilated cardiomyopathy in a patient with dengue fever. Case Rep Cardiol 2015; 2015 : 319312.25802766
5 Mao SS Ahmadi N Shah B , et al. Normal thoracic aorta diameter on cardiac computed tomography in healthy asymptomatic adult; impact of age and gender. Acad Radiol 2008; 15 (7 ): 827–834.18572117
|
PMC009xxxxxx/PMC9676717.txt
|
==== Front
Eye (Lond)
Eye (Lond)
Eye
0950-222X
1476-5454
Nature Publishing Group UK London
36402856
2317
10.1038/s41433-022-02317-7
Article
Initial experiences of cataract & lens surgery in 1269 patients in outpatient clean rooms using a portable laminar air flow device
http://orcid.org/0000-0003-3147-3452
Patel Radhika Pooja [email protected]
12
While Benjamin 13
Smith Alaric 1
Deutsch John 1
Scotcher Stephen 1
Morphis Georgios 13
Williams Geraint P. 13
Madge Simon N. 13
1 grid.413816.9 0000 0004 0398 5909 Hereford County Hospital, Stonebow Road, Hereford, HR1 2ER UK
2 grid.439257.e 0000 0000 8726 5837 Moorfields Eye Hospital, 162 City Road, London, EC1V 2PD UK
3 The Wye Clinic, 35 Edgar Street, Hereford, HR4 9JP UK
19 11 2022
8 2023
37 11 22122215
20 1 2022
4 10 2022
14 11 2022
© The Author(s), under exclusive licence to The Royal College of Ophthalmologists 2022. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Background
In 2020, routine cataract surgery was halted in most countries due to the COVID-19 pandemic in order to reduce transmission. With a consequent lack of theatre space, we developed a safe cataract pathway in outpatient department clean rooms to minimize patient exposure and time spent in hospital using a sterile laminar air flow device. We describe our initial experiences of restarting elective cataract surgery in the UK outpatient setting, outside of the operating theatre environment.
Methods
This was a prospective consecutive study of our clinical practice. A sterile air zone unit, the Toul Meditech Operio Mobile device, was used to create a sterile surgical site in three separate outpatient clean rooms from May 2020 to December 2021 in different geographical locations within Herefordshire, UK. Observations of the time spent in the department and a formal patient satisfaction survey were carried out for the initial 100 patients. All patients were followed up to assess development of post-operative complications.
Results
1269 patients were included in the study. No patients sustained post-operative infection (n = 0/1269, 0%). For the initial 100 patients, the average time spent within the department was 74.3 min (unilateral cases, range 45–115 min) and 93.1 min (bilateral, 55–135 min). Patient satisfaction was high.
Conclusion
Initial results demonstrate a safe, efficient and effective cataract surgery pathway with high patient satisfaction by converting outpatient clean rooms into ophthalmic operating theatres using the Toul Meditech Operio Mobile.
Subject terms
Surgery
Technology
issue-copyright-statement© The Royal College of Ophthalmologists 2023
==== Body
pmcIntroduction
The Coronavirus disease 2019 (COVID-19) global pandemic has had an unprecedented effect on healthcare across the world. In an effort to reduce transmission of the highly contagious SARS-Cov2 virus responsible for the disease, all NHS elective and non-urgent care was cancelled or postponed [1] in 2020. Over the 12-week period between March and May 2020, globally an estimated 28,000,000 routine surgeries were cancelled. In the UK this equated to a weekly cancellation rate of roughly 43,000 surgeries [2]. Deferring these surgeries has had its own consequences, resulting in worsening of patients’ conditions and adding risk to the eventual surgery [3]. With the duration of the COVID pandemic uncertain, there was understandably a desire to restart routine surgery as safely as possible.
We aim to describe our innovative experience of restarting elective cataract surgery in the UK during the COVID-19 pandemic using an air-flow device. Following the recommendations of the Royal College of Ophthalmologists, we describe a pathway that allowed us to minimise the total time patients spend in the hospital, reduce contact with staff and eliminate the use of a waiting area [4]. We used mobile paging devices, which allowed patients to leave the department after insertion of a mydriatic device. Converting an outpatient clean room into an operating theatre allowed us to perform cataract surgery in the outpatient setting at a time where theatre space was limited. In order to do this, we used a mobile sterile air zone unit, the Toul Meditech Operio Mobile (shown in Supplementary Fig. 1). Bacteria, skin scales and lint particulate matter from the air in the operating room are thought to be a source of infection in sterile surgery [5]. Modern operating theatres have ventilation systems that are designed to reduce this risk. The Toul Meditech Operio Mobile device produces a filtered ultra-clean air flow that can be directed over the surgical site and sterile instruments to maintain sterility [6]. Air is cleaned by passing it through a high efficiency particulate air (HEPA) filtration system that has been shown to filter particulates of sizes less than 0.18 µm. This creates a zone of clear air measuring 20 inches in width, 47 inches in length and 15 inches in height within which the operating field is positioned. The SARS-Cov2 virus is known to have a size of approximately 0.1 µm and therefore the Toul Meditech Operio Mobile device has also been recommended as a possible solution to reduce the transmission of COVID-19 in an operating environment but so far there is paucity of data on its use in cataract surgery. These papers aim to describe our initial experiences and service evaluation of the Toul Meditech Operio Mobile device in cataract surgery.
Materials and methods
Initially, the first 100 patients undergoing cataract surgery between May and July 2020 under five senior surgeons were studied in detail. In order to provide accurate data on the time spent in the department for consultant-delivered routine cataract surgery, patients undergoing combined procedures (e.g., phaco + istents, oculoplastics) and all patients on such mixed lists, and those undertaken by trainee surgeons were excluded.
All patients were contacted via telephone pre-operatively to discuss the risks of COVID-19 by a consultant surgeon. At the time of booking, all patients were asked COVID-19 screening questions. All patients had initially completed stage one of the consent process with a doctor at the clinic/listing stage.
All staff were required to have their temperature taken daily and to wear surgical masks at all times while in the department (prior to UK government directive mandating universal wearing).
An outpatient style flow through the department was created to ensure patients were in the hospital for the shortest possible time with staggered arrival times. In line with NHS England directives [4], as the patients were all treated as outpatients, patients were not required to undergo a two-week isolation period prior to surgery, nor did they undergo PCR swabbing for SARS-Cov2. On arrival, they had their temperature checked and COVID-19 screening questions were again asked. Once they had been booked into the department and initial checks were completed, a Mydriasert mydriatic device (Thea Pharmaceuticals) was inserted into the inferior fornix of the relevant eye(s) to be operated and the patient was asked to wait outside the department, often in their car, until pupil dilation was complete. Mydriasert is an insoluble ophthalmic insert for mydriasis that gradually releases its active ingredients of 0.25 mg tropicamide and 5.4 mg phenylephrine over up to a two-hour period, however, sufficient dilatation is typically achieved in 40 min [7]. The patients were then paged using a cleanable series of remote buzzer/pager devices to come back to the department to complete consent with the surgeon and then taken directly into surgery. This eliminated the use of a waiting area, as well as reducing time spent in the department.
The Toul Meditech Operio Mobile medical device was used in the clean room in order to filter the air and provide a sterile area within which to carry out cataract surgery, in line with Royal College of Ophthalmologists guidelines [8]. A representative from Toul advised on the positioning of the device. The risk of infection was further reduced by ensuring the door was kept closed for the duration of the surgery and that there was minimal talking while operating, with all staff wearing masks. Instruments were laid out in the clean room under the Toul Meditech Operio Mobile filter rather than in a separate room. The number of people in the room was minimised using one surgeon, one surgical assistant, one scrub nurse and usually one theatre support worker. Phacoemulsification was carried out using the Bausch & Lomb Stellaris, using 2.2 mm and 2.85 mm incisions, with copious dispersive viscoelastic on the cornea during surgery. All surgery was carried out with two-stage povidone iodine cleaning, careful draping and intracameral antibiotics.
The use of the clean room for surgery, in conjunction with the Toul Meditech Operio Mobile, was agreed with the local infection control team.
Patients’ various arrival and departure times were all collected prospectively. The initial 100 patients were contacted by telephone two weeks after cataract surgery to answer a short questionnaire regarding any post-operative complications, development of COVID symptoms and satisfaction (using a Likert scale). Departmental records regarding attendance with post-operative endophthalmitis were also interrogated.
After the initial 100 patients, another 215 consecutive patients underwent surgery in the same clean room, with all patients undergoing follow-up with local optometrists. The same equipment was then procured to reproduce the operating environment in two other geographic locations in Herefordshire, within outpatient clean rooms elsewhere. Using the same protocols as above, a further 954 consecutive patients underwent routine phacoemulsification surgery, with all using the Bausch & Lomb Stellaris and 2.2 mm incisions. All of these 954 patients were followed up in person by the operating surgeon in every case.
Results
In total, 1269 patients were included in the study.
The initial 100 cases
The first 100 patients to undergo treatment with this pathway were studied in detail. Of these, the average age was 74.48 (range 39–89) with 51% male (n = 51) (Table 1). The axial length of the operated eye ranged between 20.40 and 28.00 mm with an average of 23.73 mm (Table 1). A range of patients with varying surgical difficulty was included with 22 % diabetic (n = 22), 10% glaucoma (n = 10), 1% with previous uveitis (n = 1) and 1% with pseudoexfoliation syndrome (n = 1). Vision blue was used in 10%, (n = 10), adjuncts for pupil expansion (iris hooks, Malyugin ring) in 0% (n = 0) and capsular tension rings in 1%, (n = 1). 4% (n = 4) of patients had other co-morbidities such as age-related macular degeneration, 3% (n = 3) had previous vitrectomies, and 5% (n = 5) were high myopes.Table 1 Demographic table and comorbidities of initial 100 patients showing wide range of surgical difficulty.
Number of patients % of patients
Gender
Male 51 51%
Female 49 49%
Age (years)
31–40 1 1%
41–50 2 2%
51–60 6 6%
61–70 17 17%
71–80 48 48%
81–90 26 26%
Axial length
20.01–21.99 7 7%
22.00–23.99 60 60%
24.00–25.99 25 25%
26.00–27.99 7 7%
>28.00 1 1%
Co-morbidity
Diabetic 22 22%
Glaucoma 10 10%
Dense cataract 10 10%
High myope 5 5%
Small pupil 5 5%
Age-related macular degeneration 4 4%
Only eye 3 3%
Previous vitrectomy 3 3%
Corneal dystrophy 2 2%
Previous Uveitis 1 1%
Pseudoexfoliation 1 1%
Surgical Adjuncts used
Vision blue 10 10%
Capsular tension ring 1 1%
Malyugin Ring 0 0%
Iris Hooks 0 0%
Of the initial 100 cases, the majority received topical anesthesia (81%, n = 81) with one surgeon opting for anesthesia via sub-Tenon injection (19%, n = 19). Similarly, the majority (79%) were performed with a 2.2 mm incision (79%, n = 79) with one surgeon extending incisions to 2.85 mm (20%, n = 20) and 1% (n = 1) at 3 mm section length. No intra-operative complications were noted. Patients spent on average 74.3 min (range 45–115) in the department (Table 2). This was broken down into the initial booking-in time and insertion of Mydriasert that took 9.2 min on average (range 5–15) and the time for examining the patient, completing consent, performing the operation and explaining the post-operative drops and care (65.1 min (range 40–100)).Table 2 Time spent in the department in minutes of unilateral and bilateral cataracts broken down into the initial booking-in time, examining the patient, performing the operation and explaining the post-operative drops and care.
Unilateral cataract Average time in minutes Range Bilateral cataracts Average time in minutes Range
Time taken to book in and insert Mydriasert 9.2 5–15 8.5 5–15
Time for consent, surgery and post-op care 65.1 40–100 84.7 60–120
Total average time in department 74.3 93.1
One surgeon undertook bilateral cataract surgeries, which were analysed separately (Table 2). The average time spent in the department for bilateral surgeries was 93.1 min (range 55–135) with the initial insertion of dilator lasting 8.5 min on average (range 5–15) and time for consent, bilateral surgery and post-op care 84.7 min on average (range 50–120).
The results of the post-operative questionnaire at 2 weeks showed 0/100 (0%) of patients developed post-operative complications. No patient reported a reduction in vision or pain consistent with endophthalmitis and there were no visits to our emergency ophthalmology service with suspected or confirmed endophthalmitis in this period. No patients had developed any symptoms of COVID-19 at 2 weeks postoperatively. No staff developed any symptoms of COVID-19. The Likert–scale used for patient satisfaction showed that most patients were very satisfied with the cataract service (average 9.54 (range 4–10)). All patients felt the service ran efficiently and appreciated the experience of being the only patient in the department at the time.
Remaining 1169 cases
Of the further 215 consecutive patients, who underwent surgery in the same clean room with the same equipment, 0/215 patients (0%) developed postoperative endophthalmitis. Of the further 954 consecutive patients, who then underwent routine phacoemulsification using the same equipment and protocols in two other clean rooms in Herefordshire, 0/954 patients (0%) developed postoperative endophthalmitis. In total, therefore, 0/1269 (0.0%) consecutive patients developed postoperative endophthalmitis. No patient or staff member developed symptoms of COVID-19 infection from any location.
Discussion
This is the first service evaluation of the safety of cataract surgery using the Toul Meditech Operio Mobile device in an outpatient clean room, outside of an operating theatre environment. Of the 1269 consecutive patients, no cases of intraocular infection occurred and no patients or staff developed COVID-19. Our preliminary data show that cataract surgery appears to be safe and efficient in a clean room using the Toul Meditech Operio Mobile unit creating a sterile field. For the initial 100 patients, the mean time spent in the outpatient setting was significantly less than Royal College of Ophthalmologist guidelines (90 min) [4] and patients were very satisfied with the service.
The use of the Toul Meditech Operio Mobile device in surgery in general is not novel and it has been licensed for use in intravitreal injections as well as cataract surgery [6, 9–12]. Despite this, there is very little data investigating its use in clinical settings of cataract surgery. Laminar air flow models have been investigated since the 1960s and have been shown to significantly reduce the incidence of devastating surgical site infections [13]. A study undertaken in Hudiksvall Hospital in August 2015 investigated whether the use of the Toul Meditech Operio Mobile device during general surgery reduced the air-borne bacterial colony forming units (CFU) near the surgical site and over instruments [14]. They investigated two hernia surgeries and one colon surgery and measured the CFU levels and particulates in the air by using a CFU sampler and sterile agar plates. A maximum of 7 people were present in the room and ensured good door discipline similar to our operating procedure [15]. They found a significant difference in CFU (mean value 1) in the critical zone covered by the device when compared to the ambient air (mean value 29.5) used as a reference producing air over 50 times cleaner for the device.
The use of Toul Meditech Operio Mobile for safety during intravitreal injections has also been assessed at a Swedish eye department [16]. They assessed the air CFU levels at the surgical site during 23 injections and found a significant difference with a mean CFU of 1 at the critical zone versus a mean of 48 in the ambient air. Similarly, the use of the device in a simulated setting for intravitreal injections showed a safer procedural environment with reduction in particulate matter in the ambient air [12] and a case series of 39 small joint arthroplasty surgeries showed no significant difference in the rate of infections, intraoperative or post-operative complications over a 12 month follow up [17]. More recently Osher et al used the Toul Meditech Operio Mobile to detect the number of particulate matter and lint fibres in the sterile surgical field and found a statistically significant reduction in these particles with 0% lint fibres falling onto the sterile field [18]. They report relative paucity of data of the use of the device in clinical practice for cataract surgery. Our study is the first to investigate its use in cataract surgery at this scale.
The Royal College of Ophthalmologists produces guidelines to ensure minimum standards in ventilation in operating theatres, however, ventilation systems used can vary in airflow patterns and cleaning efficiency may not be uniform. The filters used may become obstructed and need regular maintenance, or be blocked by surgical equipment [18]. The Toul Meditech Operio Mobile device creates a small uniform sterile field that we have shown to be safe. Its portable and mobile nature circumvents these disadvantages of larger ventilation systems, allowing the safe conversion of an outpatient clinic rooms into a sterile operating theatre.
We found productivity to be equivalent to a standard operating theatre, with the ability to complete the same number of cases on a list and the National Health Service Getting It Right First Time (GIRFT) standard of one cataract per 30 min is easily achievable even with only one Toul Meditech Operio Mobile device. The suppliers often advise the purchase of two machines, which therefore allows for sterile laying up for the next case by a third party, which should then further improve productivity. With the set-up used within the NHS setting described, training continued in much the same way as in a standard operating theatre, with no extra difficulties experienced. The ability to further increase productivity could be an important factor to relieve some of the pressures of recovery post pandemic.
Our results are consistent with the studies discussed above showing no post-operative complications or infections in a consecutive group of 1269 patients undergoing routine phacoemulsification surgery. No patients and staff developed any symptoms of COVID-19. The incidence of endophthalmitis in the developed world is low at 0.01–0.08% and 0.14% in the UK determined from the British Ophthalmological Surveillance Unit (BOSU) study [19]. We acknowledge that, as postoperative endophthalmitis is a rare event, these data do not prove the safety of the system described, nevertheless the data do show a level of postoperative infection comparable to existing national standards. Further data from future BOSU studies with a larger sample size (exact numbers to be determined by statistical analysis beyond the scope of this paper) may help provide further confirmation of safety. Our initial experiences with the Toul Meditech Operio Mobile appear to show that it offers a safe, air-filtered environment for cataract surgery, within the outpatient setting described facilitating an excellent, patient-centred and efficient use of resources and time.
Summary
What was known before
Due to the lack of theatre space during the COVID-19 pandemic and growing demands to the cataract surgery service, there is a need to develop safe and efficient surgical pathways to meet this demand. Air-flow devices have been used to create sterile surgical environments in specialities such as orthopaedics as well as in ophthalmology for intravitreal injections but there is limited data for its use in cataract surgery.
What this study adds
We describe the first use of a novel laminar airflow device in cataract surgery, allowing safe and efficient use of an outpatient cataract pathway for the first time. The findings from our study would allow many units in this country and worldwide to continue to provide safe cataract surgery while adapting and expanding their practice.
Supplementary information
Supplementary Figure - Image of TOUL device
Legend for Supplementary Figure 1
Supplementary information
The online version contains supplementary material available at 10.1038/s41433-022-02317-7.
Author contributions
RP, BW and SM- conceptualisation, methodology, analysis of data. All authors (RP, BW, SM, AS, JD, SS, GM, GW) contributed to data collection, development of manuscript and approved final manuscript.
Data availability
All data generated or analysed during this study are included in this published article (and its supplementary information files).
Competing interests
No conflicts of interest. GPW is a consultant for Thea Pharmaceuticals. SNM is a consultant for Thea Pharmaceuticals and Alcon UK. No other financial interests or relationships to disclose.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
==== Refs
References
1. Tran K Cimon K Severn M Pessoa-Silva CL Conly J Aerosol generating procedures and risk of transmission of acute respiratory infections to healthcare workers: A systematic review PLoS One 2012 7 35797 10.1371/journal.pone.0035797
2. COVIDSurgCollaborative. Elective surgery cancellations due to the COVID-19 pandemic: global predictive modelling to inform surgical recovery plans Br J Surg 2020 395 1022 10.1002/bjs.11746
3. Myles PS Maswime S Mitigating the risks of surgery during the COVID-19 pandemic Lancet 2020 396 2 3 10.1016/S0140-6736(20)31256-3 32479826
4. The Royal College of Ophthalmologists. Guidance on the Resumption of Cataract Services during COVID.; 2020. https://www.rcophth.ac.uk/wp-content/uploads/2021/01/Resumption-of-Cataract-Services-COVID-August-2020-2.pdf (last accessed 30th October 2022)
5. Lidwell OM Lowbury EJL Whyte W Blowers R Stanley SJ Lowe D Airborne contamination of wounds in joint replacement operations: the relationship to sepsis rates J Hosp Infect 1983 4 111 31 10.1016/0195-6701(83)90041-5 6195220
6. Sossai D Dagnino G Sanguineti F Franchin F Mobile laminar air flow screen for additional operating room ventilation: Reduction of intraoperative bacterial contamination during total knee arthroplasty J Orthop Traumatol 2011 12 207 11 10.1007/s10195-011-0168-5 22072304
7. Shah A Johal S Lee N Mydriasert pupillary dilation for cataract surgery: An economic and clinical study Cataract and refractive surgery BMC Ophthalmol 2015 15 56 10.1186/s12886-015-0042-y 26036871
8. Guidance OS Theatre facilities and equipment. R Coll Ophthalmol. Published online 2018. https://www.rcophth.ac.uk/wp-content/uploads/2021/01/Theatre-facilities-equipment-Copy.pdf (last accessed 30th October 2022)
9. Friberg S Ardnor B Lundholm R Friberg B The addition of a mobile ultra-clean exponential laminar airflow screen to conventional operating room ventilation reduces bacterial contamination to operating box levels J Hosp Infect 2003 55 92 97 10.1016/S0195-6701(03)00143-9 14529632
10. Nilsson KG Lundholm R Friberg S Assessment of horizontal laminar air flow instrument table for additional ultraclean space during surgery J Hosp Infect 2010 76 243 6 10.1016/j.jhin.2010.05.016 20846746
11. von Vogelsang AC Förander P Arvidsson M Löwenhielm P Effect of mobile laminar airflow units on airborne bacterial contamination during neurosurgical procedures J Hosp Infect 2018 99 271 8 10.1016/j.jhin.2018.03.024 29580895
12. Lapid-Gortzak R Traversari R van der Linden JW Lesnik Oberstein SY Lapid O Schlingemann RO Mobile ultra-clean unidirectional airflow screen reduces air contamination in a simulated setting for intra-vitreal injection Int Ophthalmol 2017 37 131 7 10.1007/s10792-016-0236-1 27138593
13. McHugh SM Hill ADK Humphreys H Laminar airflow and the prevention of surgical site infection. More harm than good? Surgeon 2015 13 52 8 10.1016/j.surge.2014.10.003 25453272
14. Toul Meditech. The use of the air zone unit Operio and the instrument table SteriStay over critical zone and instruments to reduce air-borne contamination at hernia and colon cancer surgery. Published online 2015.
15. Smith EB Raphael IJ Maltenfort MG Honsawek S Dolan K Younkins EA The effect of laminar air flow and door openings on operating room contamination J Arthroplast 2013 28 1482 5 10.1016/j.arth.2013.06.012
16. Toul Meditech. The use of Operio at intravitreal injections. Published online 2015. https://www.toulmeditech.com/us/clinical#hide45 (last accessed 30th October 2022).
17. Nisar A Shah Z Pendse A Chakrabarti I Day case total joint arthroplasty in the hand: Results in a district general hospital J Hand Surg Eur Vol 2009 34 367 70 10.1177/1753193408102117 19321527
18. Osher RH Figueiredo GB Schneider JG Kratholm J Purifying air over the operating field with a new mobile laminar airflow device to reduce the possibility of airborne contamination J Cataract Refract Surg 2021 47 1327 32 10.1097/j.jcrs.0000000000000613 34156771
19. Kamalarajah S Silvestri G Sharma N Khan A Foot B Ling R Surveillance of endophthalmitis following cataract surgery in the UK Eye 2004 18 580 7 10.1038/sj.eye.6700645 15184923
|
PMC009xxxxxx/PMC9734330.txt
|
==== Front
Eye (Lond)
Eye (Lond)
Eye
0950-222X
1476-5454
Nature Publishing Group UK London
36481958
2341
10.1038/s41433-022-02341-7
Article
A lack of an association between COVID-19 vaccination and corneal graft rejection: results of a large multi-country population based study
http://orcid.org/0000-0002-3590-0856
Roberts Harry W. [email protected]
12
Wilkins Mark R. 13
http://orcid.org/0000-0002-3504-3145
Malik Mohsan 1
Talachi-Langroudi Melody 3
Myerscough James 45
http://orcid.org/0000-0002-6419-6941
Pellegrini Marco 56
http://orcid.org/0000-0001-5654-3942
Yu Angeli Christy 56
http://orcid.org/0000-0003-3635-3521
Busin Massimo 56
1 grid.436474.6 0000 0000 9168 0080 Corneal and External Diseases Unit, Moorfields Eye Hospital NHS Foundation Trust, London, UK
2 West of England Eye Unit, Royal Devon University Healthcare NHS Foundation Trust, Exeter, UK
3 grid.83440.3b 0000000121901201 UCL Institute of Ophthalmology, London, UK
4 grid.412711.0 0000 0004 0417 1042 Southend University Hospital, Southend, UK
5 Department of Ophthalmology, Ospedali Privati Forlì “Villa Igea”, Forlì, Italy
6 grid.8484.0 0000 0004 1757 2064 Department of Translational Medicine, University of Ferrara, Ferrara, Italy
8 12 2022
8 2023
37 11 23162319
2 6 2022
29 11 2022
29 11 2022
© The Author(s), under exclusive licence to The Royal College of Ophthalmologists 2022. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Purpose
The aim of the study was to present the rates of corneal transplant rejection from 2018 to 2022 at both Moorfields Eye Hospital UK, and Ospedali Privati Forli (OPF) “Villa Igea”, Italy and evaluate the purported association between COVID-19 vaccination and rejection.
Methods
We performed a retrospective review of rejection cases presenting to the two units. Monthly rates were correlated against regional vaccination programme rates. At OPF, conditional Poisson regression model was employed to estimate the incidence risk ratio (IRR) of graft rejection following COVID-19 vaccination risk period compared with the control period.
Results
Between January 2018 and March 2022, there were 471 (Moorfields), 95 (OPF) episodes of rejection. From the start of vaccination programme in the UK in late January 2021, the median number of graft rejections per month at Moorfields was 6 (range: 5–9), which was not significantly different to post-lockdown, pre-vaccination programme (March 2020–January 2021), p = 0.367. At OPF, the median rates of rejection before and after initiation of the vaccination programme were not significantly different (p = 0.124). No significant increase in incidence rate of rejection in the risk period following COVID-19 vaccination was found (IRR = 0.53, p = 0.71).
Conclusion
No notable increase in rates of transplant rejection was noted in year 2021 when COVID-19 vaccination was broadly implemented. The apparent temporal relationship between COVID-19 vaccination and corneal graft rejection highlighted in several case reports may not represent a causative association.
Subject terms
Corneal diseases
Risk factors
issue-copyright-statement© The Royal College of Ophthalmologists 2023
==== Body
pmcIntroduction
COVID-19 caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has, as of May 2022, resulted in excess of 15 million deaths worldwide, notwithstanding the morbidity burden caused by long term effects of infection and curtailment of personal freedoms imposed on many populations by their national governments in public health measures to restrict the replication of the virus [1].
However, the development of a number of vaccines has led to a reduction in the morbidity and mortality of this novel infection and in many countries have had significant benefits on alleviating pressures on healthcare systems and reducing imposed limitations on personal freedoms. Phase Ill clinical trials from various manufacturers have confirmed the high efficacy of vaccines against serious COVID-19 infection with low incidence rates of major adverse events [2, 3]. However, the relatively limited sample sizes and follow-up durations of phase III clinical trials inherently restrict the ability to detect rare and serious adverse vaccine-associated outcomes.
Since the inception of various nation-based vaccination programmes, ophthalmologists have identified and sought to publish cases of corneal transplant rejection with a temporal association to the SARS-CoV-2 vaccine and many such case reports have been published to date [4–16]. However, case reports are low level evidence which can be subject to observer, reporting and publication bias. As physicians, we have the duty to offer our patients sound medical advice based on the best available evidence and to not unnecessarily contribute to vaccine hesitancy, if unwarranted.
In order to evaluate whether national SARS-CoV-2 vaccine programmes have had an impact on the incidence of immunological rejection of corneal transplants we conducted an observational retrospective cohort study to investigate the rates of graft rejection presenting to the Emergency Department at Moorfields Eye Hospital, City Road, London, UK and Ospedali Privati Forlì “Villa Igea”, Italy both before and after the national SARS-CoV-2 vaccination programme. Moorfields Eye is responsible for over 20% of all corneal graft surgery nationally (A Rahman, Eye Bank Manager, Moorfields Lion Eye Bank, email communication, July, 2020) and provides specialised tertiary level ophthalmic care to the greater London area including running 24-h 7 days a week walk-in ophthalmic emergency department (ED), which prior to the pandemic would welcome on average over 1900 attendances per week. On the other hand, Ospedali Privati Forlì “Villa Igea” is tertiary care eye centre seeing 200–300 cornea cases weekly and performing over 10% of corneal transplants performed in Italy (D Ponzin, Fondazione Banca degli Occhi del Veneto Onlus, personal communication, March 2022).
Methods
This multicentre retrospective cohort study was approved as a Clinical Audit report by the Clinical Audit Committee at Moorfields Eye Hospital (London, UK) and Institutional review board/ ethics Committee approval was obtained from the Comitato Etico Ospedali Privati Forlì (Forlì, Italy). The study was performed in accordance with the tenets of the Declaration of Helsinki.
We performed a retrospective review of cases presenting to the cornea and external disease service at Moorfields Eye Hospital, City Road and Ospedali Privati Forlì “Villa Igea” (OPF) from January 2018 to March 2022. Inclusion criteria included all patients 18 years or older with a clinical diagnosis of graft rejection. At Moorfields, the case notes of all included cases were then reviewed by two independent observers experienced in cornea in order to perform a final confirmation of the diagnosis with the benefit of medical notes from subsequent follow up in the corneal service. All cases presenting at Ospedali Privati Forlì “Villa Igea” were seen by a cornea specialist. Graft rejection was defined as patients who developed an epithelial rejection line, subepithelial or stromal infiltrates, keratic precipitates or anterior chamber cell reaction with or without clinically apparent increase in stromal thickness or clarity.
Data analysis
All data had been collected prospectively and entered into the patient’s electronic medical records from the Moorfields Electronic Patient Record (EPR) System (OpenEyes, Apperta Foundation CIC, Sunderland, UK), which mandates recording.
Using the Moorfields dataset, we statistically compared new case presentations using non-parametric independent group Mann–Whitney U pre- and post- lockdown (March 2020), and pre- and post-vaccination roll out (February 2021). Similarly, data from the electronic database of OPF were also analysed. Vaccine status was ascertained among patients who developed corneal graft rejection. Conditional Poisson regression analysis was used to evaluate whether the association exists between COVID-19 vaccination and corneal graft rejection. Incidence risk ratio (IRR) of corneal graft rejection was calculated to compare the COVID-19 vaccination risk period, defined as the interval between vaccination and 60 days from the last dose with the control period defined as the observation period excluding risk period.
Furthermore, we obtained cumulative regional vaccination statistics from the Public Health England open-source data set to undertake a regression analysis to ascertain the effect of the UK COVID-19 vaccine programme on new case presentations [17]. A p value less than 0.05 was considered clinically significant.
Results
Moorfields Eye Hospital
Between January 2018 and March 2022, there were 471 corneal graft rejection episodes with a median rate of 9 patients per month (Range 3–18) at Moorfields Eye Hospital. 62% were male (n = 292), and the average age was 56 (range 18–97).
In the 26 months prior to national lockdown from January 2018 to February 2020 the median number of corneal graft rejections per month was 12 (range 8–18), which was significantly different to post lock down (March 2020–March 2022, median 6 cases per month, p = 0.001). From the start of vaccination programme in the United Kingdom late January 2021, the median number of corneal graft rejections per month was 6 (Range 5–9), which was not significantly different to post-lockdown, pre-vaccination programme (March 2020–January 2021), p = 0.367.
In total, 44 million received the first dose, 41 million people have had the second dose, and 32 million have received the third (booster) dose of the COVID-19 vaccine in England (as of March 2022). The cumulative percentage uptake of the COVID-19 vaccine in London was reported to be 70%, 65%, 46% for first, second and third dose respectively (Fig. 1). Regression analysis did not demonstrate a significant relationship between regional cumulative percentage vaccination uptake (first, second or third dose) and the number of corneal graft rejection episodes per month following vaccination roll out (r2 = 0.09, p = 0.667).Fig. 1 The number of presentations of immunological graft rejection at the Emergency Department, Moorfields Eye Hospital, City Road, London compared with the SARS-CoV-2 vaccine uptake in London.
Blue line - number of immune mediated graft rejection in Emergency Department, Moorfields Eye Hospital, City Road, London. Orange line - cumulative percentage uptake of COVID vaccination in London (first dose). Grey line - cumulative percentage uptake of COVID vaccination in London (second dose). Orange line - cumulative percentage uptake of COVID vaccination in London (third dose).
Ospedali Privati Forlì
During the same time period, 95 episodes of corneal graft rejection were diagnosed at OPF with a median rate of 2 per month. Of the 95 cases, 82 (86%) patients had received COVID-19 vaccination, compared to 85% of the overall population. The median rates of rejection before and after initiation of the vaccination programme were not significantly different (p = 0.124). Figure 2 shows the uptake of the COVID-19 vaccines in Emilia-Romagna, Italy and frequency of rejection episodes diagnosed over the same time period. No notable increase was noted when COVID-19 vaccination was broadly implemented.Fig. 2 The number of presentations of immunological graft rejection at Ospedali Privati Forlì “Villa Igea” compared with the SARS-CoV-2 vaccine uptake in Emilia-Romagna, Italy.
Blue line - number of immune mediated graft rejections in Forli, Italy. Orange line - cumulative percentage uptake of COVID vaccination in Emilia-Romagna, Italy (first dose). Orange line - cumulative percentage uptake of COVID vaccination in Emilia-Romagna, Italy (second dose).
Using conditional Poisson regression analysis of rejection episodes between January 2018 and March 2022, we found no significant increase in incidence rate of rejection between COVID-19 vaccination and 60 days from the last vaccine dose (IRR = 0.53, p = 0.71).
Discussion
Before the national lockdown in March 2020, Moorfields ED saw an average of 12 corneal graft rejections a month. From the start of the UK national vaccination programme the average number was 6 rejections per month. Regression analysis found no significant effect on vaccines on the number of presentation of rejections (r2 = 0.09, p = 0.667). Similarly, no increase in rejection cases were observed at OPF, Italy. If SARS-CoV-2 vaccines were associated with even a slight association of risk of corneal graft rejection, then one would expect that by vaccinating the majority of the population in the space of a few months there would have been an increase in the number of rejection episodes presenting at both centres. Given that the risk of rejection persists throughout the lifetime of the corneal graft even without an identifiable trigger, background incidence can account for rejection episodes that expectedly occur at any time over any given period following keratoplasty. We, therefore, suggest that based on our routinely collected longitudinal data using standard case definitions and methods of ascertainment, our data do not lend support to an association between COVID-19 vaccination and corneal graft rejection. This is further supported as this finding is replicated across two large European centres, each of which provide a large proportion of their national corneal transplantation workload.
Our results are in keeping with the only other two publications we are aware of which have evaluated the risk of rejection associated with vaccines. The Corneal Preservation Time Study group found that vaccines within the previous three months were not a significant factor associated with corneal graft rejection in Descemet stripping automated endothelial keratoplasty [18]. Equally, a Wills Eye Hospital prospective case–control study evaluating trigger factors for penetrating keratoplasty rejection did not find an association with recent vaccinations in their 22 rejection patients compared with controls (immunisation exposure equally prevalent in both groups) [19]. Both of the above studies occurred prior to COVID vaccinations.
There are a number of methodological issues with a study of this nature. We examined the monthly rates of corneal transplant rejections but did not characterise these clinically. At Moorfields, the vaccine status of corneal graft rejection patients presenting during this time frame was not rigorously recorded, but at OPF the conditional Poisson regression analysis did not find a significant increase in incidence rate of rejection between COVID-19 vaccination and 60 days from the last vaccine dose.
At Moorfields we observed a reduction in the rate of rejection which occurred in keeping with the first national lockdown. It is well recorded that healthcare seeking behaviours were affected by the pandemic. Face-to-face attendances at Moorfields ED did reduce by over 50% in the early phase of the first national lockdown in March–April 2020 [20]. However, the case mix and severity of conditions also shifted. Prior to the pandemic, blepharitis was the most common cause for presentation, after the onset of the pandemic it became acute anterior uveitis [20]. Hence to some extent, the reduction in attendances would have been skewed against more minor conditions, and may have had less of an effect on severe conditions, although it is well documented that rates of presentations of rhegmatogenous retinal detachments reduced during the first lockdown [20, 21].
A reduction in recent corneal transplant rates may have had an impact on rates of corneal graft rejection as 50% of corneal rejection episodes may occur during the first twelve months after surgery [18]. On average, approximately 4,000 corneal transplants are performed in the UK each year, however, this would have been significantly less in the 12 months prior to the vaccination programme. There was a 92% reduction in corneal elective work at Moorfields Eye Hospital during the first national lockdown [22]. Services were likewise significantly reduced in OPF Italy [23]. In OPF patients were not advised to increase their topical steroid dose at the time of vaccine. At Moorfields, where there are more than ten corneal consultants, management may have been more heterogenous. Nevertheless, there is no available evidence to support increasing topical steroids around the time of vaccination, this practice is speculative.
Our real world data reflects overall trends during a unique time period with inherent limitations on the use of more complex analytic methods requiring more granular information to account for potential confounding variables such as care at other sites, self quarantining, self treatment, untreated rejections, etc. Both Moorfields and OPF continued ophthalmic services for urgent care throughout the pandemic (indeed Moorfields continued to offer a 24/7 emergency service while other smaller London units closed). While both centres observed a sharp decline in healthcare utilisation in March–April 2020, there was a recovery of numbers after this period.
Despite the clear limitations of this study, we advocate that if there was a true association between the SARS-CoV-2 vaccine and corneal graft rejection then the effects of vaccinating nearly the entire population over a period of a few months would have impacted in a visible way upon the number of presentations.
In the future, it may be more helpful to perform a matched case-control study through large transplant registries to identify in a more robust way whether there may be any association not detected by this study in both centres.
In summary, national vaccination programmes in the UK and Italy for SARS-CoV-2 has not seen an associated increase in the incidence of corneal graft rejection which may be expected if there were an association.
Summary
What was known before
In the last 2 years many case reports or case series have questioned an association between SARS-CoV-2 vaccine and corneal transplant rejection.
What this study adds
This paper demonstrates that 2 major corneal centres, Moorfields in London UK and Ospedali Privati Forli “Villa Igea”, Italy did not see an increase in the rate of rejection presentations during the national vaccination campaigns, strongly indicating a lack of evidence for any association.
Author contributions
All authors have made substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work; AND Drafting the work or revising it critically for important intellectual content; AND Final approval of the version to be published.
Data availability
Data available upon written request to the corresponding author.
Competing interests
HR has undertaken paid consultancy work for Alcon Inc (Fort Worth, TX, USA) in the past 36 months and has received honoraria from Thea Pharmaceuticals Ltd (Keele, UK). The other authors have no financial interest to disclose.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
==== Refs
References
1. World Health Organization Technical Advisory Group for COVID-19 Mortality Assessment. https://www.who.int/news/item/05-05-2022-14.9-million-excess-deaths-were-associated-with-the-covid-19-pandemic-in-2020-and-2021. Accessed May 20, 2022.
2. Voysey M Clemens SAC Madhi SA Weckx LY Folegatti PM Aley PK Articles safety and efficacy of the ChAdOx1 nCoV-19 vaccine (AZD1222) against SARS-CoV-2: an interim analysis of four randomised controlled trials in Brazil, South Africa, and the UK Lancet 2021 397 99 111 10.1016/S0140-6736(20)32661-1 33306989
3. Polack FP Thomas SJ Kitchin N Absalon J Gurtman A Lockhart S Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine N. Engl J Med 2020 383 2603 15 10.1056/NEJMoa2034577 33301246
4. Balidis M, Mikropoulos D, Gatzioufas Z, de Politis PB, Sidiropoulos G, Vassiliadis V. Acute corneal graft rejection after anti-severe acute respiratory syndrome-coronavirus-2 vaccination: A report of four cases. Eur J Ophthalmol. 2021:11206721211064033. 10.1177/11206721211064033. Online ahead of print.
5. Abousy M Bohm K Prescott C Bonsack JM Rowhani-Farid A Eghrari AO Bilateral EK rejection after COVID-19 vaccine Eye Contact Lens 2021 47 625 8 10.1097/ICL.0000000000000840 34516456
6. Yu S Ritterband DC Mehta I Acute corneal transplant rejection after severe acute respiratory syndrome coronavirus 2 mRNA-1273 vaccination Cornea 2022 41 257 9 10.1097/ICO.0000000000002886 34690266
7. Parmar DP Garde PV Shah SM Bhole PK Acute graft rejection in a high-risk corneal transplant following COVID-19 vaccination: A case report Indian J Ophthalmol 2021 69 3757 8 10.4103/ijo.IJO_2515_21 34827040
8. Phylactou M Li J-PO Larkin DFP Characteristics of endothelial corneal transplant rejection following immunisation with SARS-CoV-2 messenger RNA vaccine Br J Ophthalmol 2021 105 893 6 10.1136/bjophthalmol-2021-319338 33910885
9. Rallis KI, Ting DSJ, Said DG, Dua HS. Corneal graft rejection following COVID-19 vaccine. Eye. 2021. ePub ahead of print.
10. Nioi M d’Aloja E Fossarello M Napoli PE Dual corneal-graft rejection after mRNA vaccine (BNT162b2) for COVID-19 during the first six months of follow-up: case report, state of the art and ethical concerns Vaccines (Basel) 2021 9 1274 10.3390/vaccines9111274 34835205
11. Molero-Senosiain M, Houben I, Savant S, Savant V. Five cases of corneal graft rejection after recent COVID-19 vaccinations and a review of the literature. Cornea. 2022;41:669–72.
12. Simão MF Kwitko S Corneal graft rejection after inactivated SARS-CoV-2 vaccine: case report Cornea 2022 41 502 4 10.1097/ICO.0000000000002970 34907940
13. Wasser LM Roditi E Zadok D Berkowitz L Weill Y Keratoplasty rejection after the BNT162b2 messenger RNA vaccine Cornea 2021 40 1070 2 10.1097/ICO.0000000000002761 34029238
14. Shah AP Dzhaber D Kenyon KR Riaz KM Ouano DP Koo EH Acute corneal transplant rejection after COVID-19 vaccination Cornea 2022 41 121 4 10.1097/ICO.0000000000002878 34620770
15. Rajagopal R Priyanka TM Stromal rejection in penetrating keratoplasty following COVID-19 vector vaccine (Covishield)—A case report and review of literature Indian J Ophthalmol 2022 70 319 21 10.4103/ijo.IJO_2539_21 34937268
16. Ravichandran S Natarajan R Corneal graft rejection after COVID-19 vaccination Indian J Ophthalmol 2021 69 1953 4 10.4103/ijo.IJO_1028_21 34146066
17. Vaccinations in United Kingdom (2021). httpscoronavirus.data.gov.ukdetailsvaccinations.
18. Stulting RD Lass JH Terry MA Benetz BA Cohen NJ Ayala AR Factors associated with graft rejection in the cornea preservation time study Am J Ophthalmol 2018 196 197 207 10.1016/j.ajo.2018.10.005 30308200
19. Miedziak AI Tambasco F Lucas Glass TC Rapuano CJ laibsan PR Cohen E Evaluation of triggers for corneal graft rejection. Ophthalmic surgery, lasers and imaging Retina 1999 30 133 9
20. Wickham L Hay G Hamilton R Wooding J Tossounis H da Cruz L The impact of COVID policies on acute ophthalmology services-experiences from Moorfields Eye Hospital NHS Foundation Trust Eye 2020 34 1189 92 10.1038/s41433-020-0957-2 32405045
21. Akram H Dowlut MS Karia N Chandra A Emergency retinal detachment surgery during Covid-19 pandemic: a national survey and local review Eye 2021 35 2889 90 10.1038/s41433-020-01187-1 32963312
22. Din N Phylactou M Fajardo-Sanchez J Watson M Ahmad S The impact of COVID-19 on acute and elective corneal surgery at Moorfields Eye Hospital London Clin Ophthalmol 2021 15 1639 45 10.2147/OPTH.S302576 33986586
23. Busin M Yu AC Ponzin D Coping with COVID-19: an italian perspective on corneal surgery and eye banking in the time of a pandemic and beyond Ophthalmology 2020 127 e68 9 10.1016/j.ophtha.2020.04.031 32376056
|
PMC009xxxxxx/PMC9735073.txt
|
==== Front
Eye (Lond)
Eye (Lond)
Eye
0950-222X
1476-5454
Nature Publishing Group UK London
36460858
2313
10.1038/s41433-022-02313-x
Review Article
Transforming ophthalmology in the digital century—new care models with added value for patients
Faes Livia 1
Maloca Peter M. 2345
Hatz Katja 67
Wolfensberger Thomas J. 8
Munk Marion R. 910
Sim Dawn A. 11121314
http://orcid.org/0000-0002-9868-154X
Bachmann Lucas M. [email protected]
1516
Schmid Martin K. 17
1 grid.439257.e 0000 0000 8726 5837 Moorfields Eye Hospital, 162 City Rd, London, EC1V 2PD UK
2 grid.436474.6 0000 0000 9168 0080 Moorfields Eye Hospital NHS Foundation Trust, 162 City Road, London, EC1V 2PD UK
3 grid.508836.0 Institute of Molecular and Clinical Ophthalmology (IOB), Basel, Switzerland
4 grid.6612.3 0000 0004 1937 0642 OCTlab, University Basel, Mittlere Strasse 91, CH-4056 Basel, Switzerland
5 Hirslanden St. Anna im Bahnhof Luzern, Lucerne, Switzerland
6 Vista Eye Clinic Binningen, Hauptstrasse 55, CH-4102 Binningen, Switzerland
7 grid.6612.3 0000 0004 1937 0642 Faculty of Medicine, University of Basel, Basel, Switzerland
8 grid.9851.5 0000 0001 2165 4204 Jules Gonin Eye Hospital, University of Lausanne, Lausanne, Switzerland
9 grid.411656.1 0000 0004 0479 0855 Ophthalmology, Inselspital, University Hospital Bern, Bern, Switzerland
10 grid.16753.36 0000 0001 2299 3507 Northwestern University, Feinberg School of Medicine, Chicago, IL USA
11 grid.436474.6 0000 0000 9168 0080 Moorfields Ophthalmic Reading Centre and Artificial Intelligence Lab, Moorfields Eye Hospital NHS Foundation Trust, London, UK
12 grid.436474.6 0000 0000 9168 0080 Medical Retina Service, Moorfields Eye Hospital NHS Foundation Trust, London, UK
13 grid.83440.3b 0000000121901201 Institute of Ophthalmology, University College London, London, UK
14 grid.451056.3 0000 0001 2116 3923 NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, England
15 grid.483560.c Medignition AG, Engelstrasse 6, 8004 Zurich, Switzerland
16 grid.7400.3 0000 0004 1937 0650 University of Zurich, CH-8091 Zurich, Switzerland
17 grid.413354.4 0000 0000 8587 8621 Eye Clinic, Lucerne Cantonal Hospital LUKS, 6000 16 Lucerne, Switzerland
3 12 2022
8 2023
37 11 21722175
9 8 2022
31 10 2022
10 11 2022
© The Author(s), under exclusive licence to The Royal College of Ophthalmologists 2022. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Ophthalmology faces many challenges in providing effective and meaningful eye care to an ever-increasing group of people. Even health systems that have so far been able to cope with the quantitative patient increase, due to their funding and the availability of highly qualified professionals, and improvements in practice routine efficiency, will be pushed to their limits. Further pressure on care will also be caused by new active substances for the largest group of patients with AMD, the so-called dry form. Treatment availability for this so far untreated group will increase the volume of patients 2–3 times. Without the adaptation of the care structures, this quantitative and qualitative expansion in therapy will inevitably lead to an undersupply.There is increasing scientific evidence that significant efficiency gains in the care of chronic diseases can be achieved through better networking of stakeholders in the healthcare system and greater patient involvement. Digitalization can make an important contribution here. Many technological solutions have been developed in recent years and the time is now ready to exploit this potential. The exceptional setting during the SARS-CoV-2 pandemic has shown many that new technology is available safely, quickly, and effectively. The emergency has catalyzed innovation processes and shown for post-pandemic time after that we are equipped to tackle the challenges in ophthalmic healthcare - ultimately for the benefit of patients and society.
Subject terms
Scientific community
Health care
issue-copyright-statement© The Royal College of Ophthalmologists 2023
==== Body
pmcDigitalization accelerates change
Digitalization has radically changed many areas of our lives. Astonished teenagers listen with eyes wide to the stories of cassette recorders, booking air travel on the telephone, looking through library card index for research, or visiting the bank teller to withdraw money for the week. Growing up in a world of Spotify, travel portals, AirBnB, Amazon, Google, Neobanks and multiplayer online games, these tales seem like stories from an old, distant world, even though they were the norm a mere few decades ago.
Conversely, such disruptive and radical changes have not yet occurred in the medical domain [1]. Despite the availability of technology for the purpose of delivering digital first solutions such as electronic medical records, algorithmically supported image analysis and remote diagnosis with telemedicine, change in healthcare has been slower, more cautious, and unequally distributed compared to industries such as retail, commerce, or banking [2, 3].
Digitalization in medicine
Many areas of medical care have remained essentially unchanged for half a century. Medical practice remains to a large extent institution-bound and doctor-centred [4]. The patiens [lat] - the patiently sufferer, goes to the place of medical knowledge - the medical centre - to obtain advice on the existence and the course of (his) disease. While digitalization has led to comprehensive customer-centricity in many areas, such examples are rare in medicine. While doctor-centred institution-bound care will remain important throughout digital innovation in medicine, it will be crucial to determine which aspects of care can be decentralized to the benefit of relieving the pressure of healthcare systems, patients, and medical personnel. Approaches to greater patient-centricity and more decentralized care can be observed in chronic diseases such as bronchial asthma, chronic obstructive pulmonary disease, heart failure and diabetes mellitus (summarized in [5]). For some, rarer diseases, so-called home care approaches have become established, in which patients are visited at home by specialized nurses and treated with infusions, for example [6].
Digitalization in ophthalmology — hurdles and opportunities
In ophthalmology, new approaches making benefit of innovative digital solutions have been developed, particularly in Northern Europe, the UK and Australia, partly in national programmes, in the care of patients with chronic retinal diseases such as diabetic macular oedema (DMO) and age-related macular degeneration (AMD) [7–21]. Both conditions are chronic and lead to severe visual impairment if left untreated. If detected, they can be controlled albeit with complex, variable, frequent and prolonged treatments regimes - the regular application of the active substance directly into the eye several times a year. The strain on the patient and their carers to follow such a therapy for a longer period of time is considerable [22]. Unfortunately, we still observe too many therapy discontinuations and only some of the reasons are known including low vision at baseline and extent of co-morbidity [23]. This occurs despite overwhelming evidence that therapy interruptions lead to a lack of care, which results in vision loss and a large national disease burden and in many regions treatment discontinuation is not detected and acknowledged. However, it may be difficult for ophthalmologists to perceive this problem in their daily practice.
Many initiatives, including telemedicine services or home-monitoring programmes aimed at introducing digital innovations into ophthalmic care to combat treatment disruptions are still in their infancy, despite their medical and economic importance. The obstacles to implementing innovation are complex and country specific. In a survey of global experts in retinal diseases on their views on the introduction of digital health applications, many were rather sceptical. The lack of reimbursement for these kinds of services was the main reason cited for not offering tele-ophthalmology, a barrier that was removed during the SARS-CoV-2 pandemic [1].
A lack of pressure to innovate and a lack of incentive systems, regulatory concerns such as data security and unsuitable tariff structures for mapping innovative care systems in existing remuneration models and financial losses on the part of the physician are other frequently cited factors. Again, the emergence of the SARS-CoV-2 pandemic and the associated lockdown abruptly changed this in many countries [24].
In most countries, outpatient ophthalmology services were largely disrupted if not completely halted within days. Patients feared infection in hospital, medical staff were withdrawn to treat Covid-19 patients and regular services were suspended for safety reasons. In the UK, where the impact of the pandemic on ophthalmic care was analyzed in detail, 80% of outpatient consultation capacity was suddenly stopped for several months. In the USA, but also in Switzerland, the processes of follow-up examinations, for example in patients with diabetic retinal diseases, had to be drastically adapted for social distancing purposes. In some cases, vision tests and imaging diagnostics were abandoned and only treatment was given. Furthermore, many patients with chronic eye diseases, who would have needed treatment chose to remain at home and thus risked - and sometimes experienced - permanent vision loss.
The SARS-CoV-2 pandemic as a catalyst for innovation
In this crisis, the ophthalmologic routine was shaken up and there was suddenly room for new ideas. There was pressure to innovate, and the delivery of care now suddenly had to reach the patient rather than the other way around. Across the globe, initiatives developed to ensure minimal necessary care using home measurements, telemedicine services and telephone consultations [25].
Out of the need to innovate, new types of care structures quickly emerged, which were also able to demonstrate their benefits in scientific studies. New paradigms were created out of pure necessity of the pandemic situation where the patient is at the centre of care. This focus leads to a more decentralized medicine, a stronger networking of different specialists in patient care and - through the inclusion of home measurements - a new quality of data for the monitoring of disease progression. From this interplay, a new vision of ophthalmic care emerges which even today has a great rationale. Table 1 provides a selection of monitoring tests available to assess the visual system at home.Table 1 A selection of monitoring tests available for home evaluation of ophthalmologic parameters.
Clinical parameter Test User
Visual acuity
Eye Test Patient
OptOK Patient
Snellen Patient
HOTV Acuity Patient
Eye Test Free Patient
Check My Eyes Patient
iSnellen Patient
Eye Test Dr Patient
Eye Chart Patient
iExam Patient
Optician Patient
logMAR Tumbling-E VA charts Family screeners
Sightbook Patient
Checkup Vision Assessment (Digisight) Patient
Telephone assisted five-staged evaluation tool Interviewer
Ridgevue Vision application Lay person, Patient
Odysight Patient
Metamorphopsia/Hyperacuity
Alleye Patient
ForeseeHome AMD Monitor Patient
Home Vision Monitor / mVTX Patient
MacuFix Patien
PreView preferential hyperacuity perimetry Patient
M-CHARTS Patient
3D display device Patient
Amsler grid Patient
Retinal morphology
Self-Examination Low-Cost Full-Field Optical Coherence Tomography (SELFF-OCT) Patient
notal home optical coherence tomography Patient
Intraocular pressure Proview eye pressure monitor Patient
iCare® Home Patient
Visual Field Test Melbourne Rapid Field Patient
visualFields Easy Patient
Eyecatcher Patient
The new supply reality
The care of AMD and DMO is facing major challenges due to socio-demographic change. According to the latest figures, the number of patients with AMD will rise from around 200 to 290 million worldwide by 2040, and the number of patients with diabetes from around 400 to 650 million [26, 27]. In Switzerland for example, the number of patients with AMD and DMO treated with intravitreal injections is currently estimated at approximately 30,000. Around 350 ophthalmologists currently administer around 170 thousand injections per year. In recent years, the market for intravitreal injections has grown by about 17 percentage points per year [28]. Due to demographic changes, the AMD market will grow significantly over the next two decades.
If care remains the same, treatment bottlenecks are to be expected only from demographic shifts. Moreover, further pressure on care will soon be caused by drugs targeting at the much larger group of patients with the so-called dry form of AMD. Two pivotal phase 3 trials on the treatment of advanced dry AMD were recently finalized and are currently submitted to the regulatory authorities [29]. This will increase the volume of patients many times over.
Pharmaceutical companies are already advanced in developing innovative therapy options with reduced number of annual treatments. In addition, two novel therapy concepts for AMD and DMO are about to enter the market, which can reduce the therapy intensity to half or even one third [30–32]. Although these innovations will bring initial relief to the care system, care efficiency must also increase and attractive, innovative reimbursement models must be implemented [1].
Individualization of treatment — Personalized Healthcare
Ideally, treatment is directed to the specific treatment needs of a patient. If the condition is stable and no treatment is needed at the medical centre, there should be no in person consultation. The so-called treat and extend strategy is already in use in many patients and combines a clinical follow-up with the treatment visit. But due to possible appointment delays and deterioration in the fellow untreated eye, an additional safety net is necessary, especially for patients with long treatment intervals. If treatment is necessary, it should be possible to provide it flexibly. In principle, this ideal type of care is within reach because all the necessary organizational and technological requirements are already in place. Ironically, one is almost inclined to say, it was the pandemic crisis that opened our eyes to this. It highlights again that social and medical transition cannot be planned; sometimes it progresses slowly, sometimes quickly based on crisis and need [33].
An important piece of the puzzle of this care is the patient’s simple self-measurement of the course of the disease. The data from these measurements are automatically categorized (green, yellow, red) and sent to the attending physician or the clinic via telemedicine platforms and the patient receives instant feedback about his status. If the data show abnormalities in the sense of a worsening of the disease, the clinic contacts the patient and plans further care or vice versa and needs to be agreed with the patient initially. Currently, two mobile applications that test a specific visual function and can be used on smartphones or tablets. Clinical studies have shown that therapy management through self-measurement with home monitoring assessing visual function is able to identify the group of patients who need treatment [11, 12, 15, 17, 20, 24]. Very recently, also a mobile imaging device (Optical Coherence Tomography (OCT)) has become available for home measurement by the patient, providing real-life structural retina data [16]. A few home OCT devices have already been tested for their applicability in clinical studies [18, 34]. Ultimately, a home measurement system creates an additional opportunity of patient doctor interaction and a channel to address fears and uncertainties regarding the therapy.
Relief from physician-centred care structures
Such an innovative and patient-centred care system could be supplemented by the inclusion of other non-medical stakeholders who care for the patient regarding visual function if patient number pressure requires it. It is readily conceivable that regular imaging examinations – which are typically performed in hospitals or by specialists in private practice and provide the indication to treat according to the guidelines – could be performed in a decentralized system at the optician’s or pharmacist’s premises under the supervision of ophthalmologists and in conformity with the legal system.
If patients are severely restricted in their mobility, a team of non-medical specialists can also carry out a basic ophthalmological examination with a telemedical assessment on site within a home care framework. Such a service has been available in Switzerland for about two years and is reimbursed by health insurances [35]. Patients and doctors can derive a lot of benefit from this additional service. For patients, the hurdle to accessing medical care is noticeably reduced and the attending physicians can create the necessary organizational prerequisites before the doctor’s visit to ensure an efficient examination and treatment due to the information they receive from the referring service.
Image analysis can also be performed by specially trained non-medical professionals, supported by automated image analysis outside the hospital in a reading centre. Reading centres and other specialists with the help of retina experts are the ones who currently develop AI based systems which should be able to decide if there is disease activity and make treatment suggestions, e.g., treatment intervals [10, 14, 21, 36–38]. In the UK, a national system organized in this way to detect diabetics with retinal diseases requiring treatment has led to a marked reduction in legal blindness and an improvement in care without the need to increase medical staff [19]. Key of these approaches is the hybrid model, combining advanced imaging technology with an initial automated assessment, followed by a second human virtual re-examination if needed.
Reduction of opportunity costs for patients and their relatives
The organization of treatment for AMD and DMO also has resource-binding consequences for the patient, carer and relatives. However, reliable study data on this subject are hard to find [22]. Going to a medical consultation and treatment often requires accompanying persons who will stay away from work during this time, driving services and planning. In the best case, these efforts are limited to times when the use of medical services is indispensable. The introduction of a telemedicine service at the renowned Moorfields Eye Hospital in London during the lockdown in 2020 reduced the number of patient journeys by 1.4 million kilometres in the emergency department alone. It saved patients and their carers 6.4 years of travel time and avoided the equivalent of 51,000 litres of CO2 emissions from petrol (Dr Dawn Sim, personal communication).
Outlook
Ophthalmology faces many challenges in providing effective and meaningful eye care to an ever-increasing group of people. Even the health systems that have so far been able to cope will face similar pressures of a global aging demographic and need for highly qualified professionals. The digital era offers new innovative approaches to decentralized, personalize, and democratize eye care for patients whilst allowing the healthcare workforce to practice at the top of their license. The overwhelming challenges for ophthalmologists in the coming years to cope with the large number of additional patients in the practice require a concerted interaction of new therapeutic approaches, technological innovations but also regulatory support and incentive systems so that the change can be initiated in time. To this end, it is also necessary to provide training for professionals on the new technologies and possibilities so that sufficient competence is built up for the implementation of new care structures.
Author contributions
LF, DAS, and LMB initiated the project and wrote a first draft of the text. All authors made textual additions and provided critical intellectual input. All authors approved the content of the final manuscript.
Data availability
No research data are provided in this manuscript.
Competing interests
The authors declare no competing interests.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
==== Refs
References
1. Faes L Rosenblatt A Schwartz R Touhami S Ventura CV Chatziralli IP Overcoming barriers of retinal care delivery during a pandemic-attitudes and drivers for the implementation of digital health: a global expert survey Br J Ophthalmol 2021 105 1738 43 10.1136/bjophthalmol-2020-316882 33067360
2. Herzlinger RE. Why innovation in health care is so hard. Harv Bus Rev. 2006. https://hbr.org/2006/05/why-innovation-in-health-care-is-so-hard
3. Hjelm NM Benefits and drawbacks of telemedicine J Telemed Telecare 2005 11 60 70 10.1258/1357633053499886 15829049
4. Wasan KM Berry L Kalra J Physician centric healthcare: is it time for a paradigm shift? J R Soc Med 2017 110 295 6 10.1177/0141076817707337 28463065
5. Noah B Keller MS Mosadeghi S Stein L Johl S Delshad S Impact of remote patient monitoring on clinical outcomes: an updated meta-analysis of randomized controlled trials NPJ Digit Med 2018 1 20172 10.1038/s41746-017-0002-4 31304346
6. Polinski JM Kowal MK Gagnon M Brennan TA Shrank WH Home infusion: Safe, clinically effective, patient preferred, and cost saving Health 2017 5 68 80 10.1016/j.hjdsi.2016.04.004
7. Abramoff MD Lavin PT Birch M Shah N Folk JC Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices NPJ Digit Med 2018 1 39 10.1038/s41746-018-0040-6 31304320
8. Chopra R Wagner SK Keane PA Optical coherence tomography in the 2020s-outside the eye clinic Eye 2021 35 236 43 10.1038/s41433-020-01263-6 33168975
9. Choremis J Chow DR Use of telemedicine in screening for diabetic retinopathy Can J Ophthalmol 2003 38 575 9 10.1016/S0008-4182(03)80111-4 14740799
10. Faes L Fu DJ Huemer J Kern C Wagner SK Fasolo S A virtual-clinic pathway for patients referred from a national diabetes eye screening programme reduces service demands whilst maintaining quality of care Eye 2021 35 2260 9 10.1038/s41433-020-01240-z 33128024
11. Faes L Islam M Bachmann LM Lienhard KR Schmid MK Sim DA False alarms and the positive predictive value of smartphone-based hyperacuity home monitoring for the progression of macular disease: a prospective cohort study Eye 2021 35 3035 40 10.1038/s41433-020-01356-2 33414531
12. Gross N Bachmann LM Islam M Faes L Schmid MK Thiel MA Visual outcomes and treatment adherence of patients with macular pathology using a mobile hyperacuity home-monitoring app: a matched-pair analysis BMJ Open 2021 11 e056940 10.1136/bmjopen-2021-056940 34949632
13. Group AHSRChew EY Clemons TE Bressler SB Elman MJ Danis RP Randomized trial of a home monitoring system for early detection of choroidal neovascularization home monitoring of the Eye (HOME) study Ophthalmology 2014 121 535 44 10.1016/j.ophtha.2013.10.027 24211172
14. Gupta A Cavallerano J Sun JK Silva PS Evidence for Telemedicine for Diabetic Retinal Disease Semin Ophthalmol 2017 32 22 8 10.1080/08820538.2016.1228403 27748634
15. Islam M, Sansome S, Das R, Lukic M, Chong Teo KY, Tan G, et al. Smartphone-based remote monitoring of vision in macular disease enables early detection of worsening pathology and need for intravitreal therapy. BMJ Health Care Inform. 2021;28:1–7.
16. Keenan TDL, Goldstein M, Goldenberg D, Zur D, Shulman S, Loewenstein A. Prospective, longitudinal pilot study: daily self-imaging with patient-operated home OCT in neovascular age-related macular degeneration. Ophthalmol Sci. 2021;1. 10.1016/j.xops.2021.100034.
17. Korot E, Pontikos N, Drawnel FM, Jaber A, Fu DJ, Zhang G, et al. Enablers and barriers to deployment of smartphone-based home vision monitoring in clinical practice settings. JAMA Ophthalmol. 2022;140:153–60.
18. Maloca P Hasler PW Barthelmes D Arnold P Matthias M Scholl HPN Safety and feasibility of a novel sparse optical coherence tomography device for patient-delivered retina home monitoring Transl Vis Sci Technol 2018 7 8 10.1167/tvst.7.4.8 30050725
19. Scanlon PH The English National Screening Programme for diabetic retinopathy 2003-2016 Acta Diabetol 2017 54 515 25 10.1007/s00592-017-0974-1 28224275
20. Schmid MK Thiel MA Lienhard K Schlingemann RO Faes L Bachmann LM Reliability and diagnostic performance of a novel mobile app for hyperacuity self-monitoring in patients with age-related macular degeneration Eye 2019 33 1584 9 10.1038/s41433-019-0455-6 31043690
21. Shi L Wu H Dong J Jiang K Lu X Shi J Telemedicine for detecting diabetic retinopathy: a systematic review and meta-analysis Br J Ophthalmol 2015 99 823 31 10.1136/bjophthalmol-2014-305631 25563767
22. Vukicevic M Heraghty J Cummins R Gopinath B Mitchell P Caregiver perceptions about the impact of caring for patients with wet age-related macular degeneration Eye 2016 30 413 21 10.1038/eye.2015.235 26611848
23. Westborg I Rosso A Risk factors for discontinuation of treatment for neovascular age-related macular degeneration Ophthalmic Epidemiol 2018 25 176 82 10.1080/09286586.2017.1397701 29131696
24. Teo KYC Bachmann LM Sim D Lee SY Tan A Wong TY Patterns and characteristics of a clinical implementation of a self-monitoring program for retina diseases during the COVID-19 pandemic Ophthalmol Retin 2021 5 1245 53 10.1016/j.oret.2021.02.005
25. Patel S, Hamdan S, Donahue S. Optimising telemedicine in ophthalmology during the COVID-19 pandemic. J Telemed Telecare. 2022;28:498–501.
26. Ogurtsova K da Rocha Fernandes JD Huang Y Linnenkamp U Guariguata L Cho NH IDF Diabetes Atlas: global estimates for the prevalence of diabetes for 2015 and 2040 Diabetes Res Clin Pract 2017 128 40 50 10.1016/j.diabres.2017.03.024 28437734
27. Wong WL Su X Li X Cheung CM Klein R Cheng CY Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: a systematic review and meta-analysis Lancet Glob Health 2014 2 e106 16 10.1016/S2214-109X(13)70145-1 25104651
28. Chopra R, Preston GC, Keenan TDL, Mulholland P, Patel PJ, Balaskas K, et al. Intravitreal injections: past trends and future projections within a UK tertiary hospital. Eye. 2022;36:1373–8.
29. Apellis Pharmaceuticals I. Study to compare the efficacy and safety of intravitreal APL-2 therapy with sham injections in patients with geographic atrophy (GA) secondary to age-related macular degeneration. 2021. https://www.clinicaltrialsgov/ct2/show/NCT03525600.
30. Heier JS, Khanani AM, Quezada Ruiz C, Basu K, Ferrone PJ, Brittain C, et al. Efficacy, durability, and safety of intravitreal faricimab up to every 16 weeks for neovascular age-related macular degeneration (TENAYA and LUCERNE): two randomised, double-masked, phase 3, non-inferiority trials. Lancet. 2022;399:729–40.
31. Holekamp NM, Campochiaro PA, Chang MA, Miller D, Pieramici D, Adamis AP, et al. Archway randomized phase 3 trial of the port delivery system with ranibizumab for neovascular age-related macular degeneration. Ophthalmology. 2022;129:295–307.
32. Wykoff CC, Abreu F, Adamis AP, Basu K, Eichenbaum DA, Haskova Z, et al. Efficacy, durability, and safety of intravitreal faricimab with extended dosing up to every 16 weeks in patients with diabetic macular oedema (YOSEMITE and RHINE): two randomised, double-masked, phase 3 trials. Lancet. 2022;399:741–55.
33. Kuhn TS. The Structure of Scientific Revolutions. Chicago: University of Chicago Press; 1970.
34. Ho AC, Heier JS, Holekamp NM, Garfinkel RA, Ladd B, Awh CC, et al. Real-world performance of a self-operated home monitoring system for early detection of neovascular age-related macular degeneration. J Clin Med. 2021;10:1355.
35. Geschwindner H. 2021. https://www.gerontologieblog.ch/augenmobil-bringt-die-augenheilkunde-ins-pflegezentrum/.
36. Jones L Bryan SR Miranda MA Crabb DP Kotecha A Example of monitoring measurements in a virtual eye clinic using ‘big data’ Br J Ophthalmol 2018 102 911 5 10.1136/bjophthalmol-2017-310440 29074497
37. Kern C Fu DJ Kortuem K Huemer J Barker D Davis A Implementation of a cloud-based referral platform in ophthalmology: making telemedicine services a reality in eye care Br J Ophthalmol 2020 104 312 7 10.1136/bjophthalmol-2019-314161 31320383
38. Saleem SM Pasquale LR Sidoti PA Tsai JC Virtual ophthalmology: telemedicine in a COVID-19 era Am J Ophthalmol 2020 216 237 42 10.1016/j.ajo.2020.04.029 32360862
|
PMC009xxxxxx/PMC9745701.txt
|
==== Front
Eye (Lond)
Eye (Lond)
Eye
0950-222X
1476-5454
Nature Publishing Group UK London
36513858
2338
10.1038/s41433-022-02338-2
Article
Clinical efficacy and safety of intravitreal fluocinolone acetonide implant for the treatment of chronic diabetic macular oedema: five-year real-world results
http://orcid.org/0000-0003-3984-6014
Dobler Emilie [email protected]
Mohammed Bashar Raouf
Chavan Randhir
Lip Peck Lin
Mitra Arijit
Mushtaq Bushra
grid.414513.6 0000 0004 0399 8996 Birmingham and Midland Eye Centre, Birmingham, UK
13 12 2022
8 2023
37 11 23102315
13 6 2022
24 10 2022
28 11 2022
© The Author(s), under exclusive licence to The Royal College of Ophthalmologists 2022. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
Background/Aim
To report 5-year real-world efficacy and safety data following the treatment of chronic diabetic macular oedema (DMO) with the intravitreal 0.19 mg fluocinolone acetonide implant(ILUVIEN).
Methods
Retrospective cohort study of 31 eyes treated with ILUVIEN for chronic DMO at a tertiary centre in Birmingham (UK). Best corrected visual acuity (BCVA) and central retinal thickness (CRT) were recorded at baseline, and then at 1-,2-,3-, and 5-years. Safety was assessed based on intraocular pressure (IOP) -lowering medication, surgery, and other complications.
Results
BCVA significantly improved 1-year post-ILUVIEN (+4.2 letters, p < 0.05) and gradually reverted to baseline levels over the 5-year period of follow-up (+0.2 letters at year-5). A significant and sustained CRT reduction was observed throughout the 5-years. The proportion of eyes on IOP-lowering medication increased from 16% at baseline, to 70% at 5-years (p < 0.001) with eyes on a mean of 1.3 medications. Laser trabeculoplasty (n = 2), cyclodiode laser (n = 1), and trabeculoplasty and trabeculotomy (n = 1, in the same eye; 3.2%) were required for uncontrolled IOP. Other complications included endophthalmitis (n = 1) and vitreous haemorrhage (n = 1). 58% of eyes required additional intravitreal injections, with a mean 29.2 months to first injection. We observed a 69% reduction in treatment burden following treatment with ILUVIEN implant.
Conclusions
Our real-world results confirm the efficacy of the ILUVIEN implant over 5 years, with two-thirds of eyes having improved or stable visual acuity 5 years after ILUVIEN, and an overall sustained improvement in anatomical outcome. Although the rate of IOP-lowering medications use was higher than previously reported, the rate of incisional IOP-lowering surgery and other complications remained low and in keeping with rates reported in larger studies.
Subject terms
Diabetes complications
Drug therapy
Retinal diseases
issue-copyright-statement© The Royal College of Ophthalmologists 2023
==== Body
pmcIntroduction
Diabetic macular oedema (DMO) is a leading cause of visual loss in young adults in developed countries. It is a multifactorial condition, which directly impairs central vision, affecting 12% of Type 1 and 28% of Type 2 diabetics within 9 years of diagnosis [1]. Its treatment used to be limited to focal/grid macular laser photocoagulation and sub-Tenon or intravitreal short-acting corticosteroid injections, such as the off-label use of triamcinolone acetonide [2, 3]. In recent years, anti-VEGF agents, such as Ranibizumab and Aflibercept, have transformed the treatment of DMO and become first-line options [4]. Nevertheless, for an estimated 40% of patients, response to anti-VEGF remains sub-optimal [5]. In such cases, intravitreal corticosteroids remain a valuable alternative pharmacological option by targeting alternative pathways to VEGF, in particular sustained-release implants of dexamethasone (Ozurdex®, Allergan Inc., Irvine, California) and fluocinolone acetonide (ILUVIEN®, Alimera Sciences Ltd.; Alpharetta GA, USA) [6–9].
ILUVIEN provides a slow-release preparation of fluocinolone acetonide 0.19 mg and is approved by regulators in the UK for the treatment of chronic DMO that is insufficiently responsive to alternative therapies in eyes with a pseudophakic lens and it offers the advantage of prolonged clinical effects lasting for up to three years [10, 11]. Its effectiveness and safety have been well established in several clinical trials, with the most common adverse effect reported being cataract formation and elevated intraocular pressure (IOP) [12, 13]. Several reports of real-world outcomes of chronic DMO treatment with ILUVIEN have been published with 3 years of follow-up reported [14–17]. To our knowledge, this is the first long-term report of real-world safety and efficacy of intravitreal ILUVIEN implant over a 5-year follow-up period in a cohort of 31 eyes at a tertiary ophthalmology centre in Birmingham, United Kingdom.
Materials and Methods
Study design
This a retrospective study of a cohort of patients who have been treated for chronic DMO with an intravitreal ILUVIEN implant (fluocinolone acetonide 0.19 mg) over a three-year period (2014–2016) at the Birmingham and Midland Eye Centre (UK). Clinical records were used to identify patients meeting these criteria and n = 60 treated eyes were identified as part of this real-world cohort. No ethics committee approval was required, as this data was collected retrospectively for departmental clinical effectiveness purposes. This analysis was conducted in accordance with the Declaration of Helsinki and the UK’s Data Protection Act. Patients gave informed consent for all investigations and treatments.
Due to the retrospective nature of this study, the COVID-19 pandemic caused some disruption to the timing of 5-year follow-up visits for patient treated in 2015. Eyes were included in the analysis only if they had documented Best-Corrected Visual Acuity (BCVA) and an Optical Coherence Tomography (OCT) scan both at baseline and at the 5-year follow-up visit (accepted from a minimum of 4.5 years following ILUVIEN implant). Data was collected from case notes, clinical letters, and Topcon OCT (3D OCT-2000; Topcon Corporation, Tokyo, Japan). BCVA was measured at the last clinic visit prior to ILUVIEN implant injection, and patients were excluded if there was any other intravitreal injection or macular laser in the interval between the last measured BCVA and ILUVIEN implant.
In order to observe the safety profile of ILUVIEN in a real-world setting, patients were included in this study regardless of any prior history of IOP steroid response, past or current IOP-lowering medication, or any diagnosis of OHT. The clinical decision to use ILUVIEN in those situations rested with the treating clinician and the patient. ILUVIEN product characteristics specify that it is contraindicated in the presence of pre-existing glaucoma.
Individual clinical decisions regarding the use and timing of any rescue intravitreal injections, macular laser, and IOP-lowering therapy following ILUVIEN rested with the clinician’s judgement and the patient’s informed choices.
Twenty-nine eyes were excluded as they missed 5-year follow-up data due to death (n = 21), discharge from clinic (n = 2), and loss to follow-up (n = 6). Therefore, 31 eyes were included in this analysis, which belonged to 25 individual patients (6 patients had both eyes included in this analysis).
BCVA was converted from Snellen visual acuity score to Early Treatment Diabetic Retinopathy Study letter score using the formula described by Gregori et al., in order to facilitate statistical analyses [18].
Study endpoints
Baseline demographic data was collected, including age, sex, ethnicity, diabetes mellitus type, duration of DMO, prior treatment (macular laser, vitrectomy, intravitreal corticosteroids, intravitreal anti-VEGF), and baseline IOP-lowering medication.
Primary outcome measures were the change from baseline in BCVA and Central Retinal Thickness (CRT) five years after starting treatment with the ILUVIEN implant.
Secondary outcome measures included the change in BCVA and CRT at 1-, 2- and 3-years post-ILUVIEN, number and type of complications following ILUVIEN, IOP-lowering treatments (number of IOP-lowering medications at 5-year follow-up visit, selective laser trabeculoplasty or cyclodiode laser treatment, incisional surgery), number and type of intravitreal injections or implants within 5 years post-ILUVIEN, and number of retinal laser photocoagulation treatments.
Statistical analyses
Statistical analyses were performed using Wilcox’s signed rank paired t-test, with a level of p < 0.05 being accepted as statistically significant. Centre values are reported as mean ± standard deviation.
Results
The mean follow-up period was 1867 ( ±122) days, which is equivalent to 5 years and 6 weeks. Two eyes (from a single patient) were missing 2- and 3- year follow-up data, and one eye was missing 1-year BCVA measurement, but as they all had adequate baseline and 5-year follow-up data, they were included in this analysis.
Baseline characteristics
Baseline characteristics are presented in Table 1. As per the UK national guidelines for treatment with ILUVIEN, all eyes were pseudophakic and had some form of prior treatment for DMO at baseline, with 97% having received prior anti-VEGF therapy, 58% having received prior intravitreal corticosteroids (intravitreal triamcinolone or intravitreal Ozurdex implant) and 68% having received prior macular laser photocoagulation. The mean interval between the last intravitreal injection and/or macular laser and ILUVIEN implant was 213 ( ± 289) days, with the shortest recorded interval being 54 days. No eye received Ozurdex within 6 months prior to ILUVIEN. No eye had a pre-existing diagnosis of glaucoma at baseline. Five eyes were on IOP-lowering medication at baseline, due to a diagnosis of ocular hypertension (n = 3) or previous steroid-related IOP elevation (n = 2).Table 1 Baseline characteristics of 31 eyes included in this analysis.
Baseline characteristics Eyes (n = 31 from 25 patients)
Age in years, mean ± SD 67 ± 8.0
Gender, n (%) Male 12 (39%)
Female 19 (61%)
Ethnicity, n (%) Asian 13 (42%)
White 9 (29%)
Afro-caribbean 7 (23 %)
Mixed 2 (6%)
Diabetes type, n (%) Type 1 3 (10%)
Type 2 28 (90%)
DMO duration in years, mean ± SD 5.9 ± 3.5
BCVA (ETDRS letters) BCVA mean ± SD 48.1 ± 16.2
Patients with <60 letters, n (%) 22 (71%)
Patients with ≥60 letters, n (%) 9 (29%)
Central retinal thickness, μm (mean ± SD) 477.1 ± 159.5
Prior treatment, n (%) Vitrectomy 4 (13%)
Macular/focal/grid laser 21 (68%)
Any intravitreal therapy 30 (97%)
Mean number of treatments ± SD 9.7 ± 6.3
Any anti-VEGF 30 (97%)
Mean number of treatments ± SD 8.2 ± 5.6
Bevacizumab 24 (77%)
Mean number of treatments ± SD 4.3 ± 2.5
Ranibizumab 24 (77%)
Mean number of treatments ± SD 6.0 ± 3.4
Any intravitreal corticosteroid 18 (58%)
Mean number of treatments ± SD 3.0 ± 2.8
Ozurdex intravitreal implant 2 (6%)
Mean number of treatments ± SD 1.0 ± 0.0
Triamcinolone acetonide intravitreal injection 16 (52%)
Mean number of treatments ± SD 3.3 ± 2.9
On IOP-lowering medication, n (%) 5 (16%)
BCVA
BCVA at baseline ranged from 0 to 76 letters, with a mean of 48.1 ( ±16) letters, which improved to 52.3 ( ±17) letters after one-year (a gain of +4.2 letters) before gradually reducing back down to 48.3 ( ±23) letters at 5 years. Compared to baseline, the difference in BCVA at 1-year was statistically significant (p < 0.05), but it was not statistically significant at 2-, 3-, and 5-year follow-up.
The change in BCVA at baseline and at one, two, three, and five-year follow-up visits is shown in Fig. 1.Fig. 1 Functional and anatomical outcomes over five years following ILUVIEN.
Change in BCVA (ETDRS letters) and CRT (μm) over 5 years following treatment with Fluocinolone Acetonide (FAc) ILUVIEN implant. *p < 0.05.
Five years after treatment with ILUVIEN, 13 eyes had an improved BCVA of 5 letters or more compared to baseline, 8 eyes had a similar BCVA as baseline (+/− 4 letters from baseline), and 10 eyes had a worse BCVA with a loss of 5 letters or more compared to baseline. This means that 68% of eyes had a similar or improved BCVA after 5 years when compared to baseline.
The proportion of eyes achieving a BCVA of 60 letters or more (6/18 Snellen equivalent) increased from 29% at baseline to 42% at 1-year, before reducing to 39% at 2- and 3-year, and 35% at 5-year post-ILUVIEN.
CRT
Baseline CRT ranged from 222 μm to 835 μm, with a mean of 477.1 μm ( ±160), which improved to 323.7 μm ( ±117) after 1 year (a 32% reduction), and remained stable thereafter, with a mean CRT of 310.2 μm ( ±116) after 5 years. The difference in CRT compared to baseline was statistically significant (p < 0.001) for all time points. Changes in CRT over 5 years following ILUVIEN are shown in Fig. 1.
Patients with a thicker baseline CRT (≥400 μm) had a more pronounced decrease in CRT after 1 year (−234.7 μm), which was maintained after 5 years (−257 μm), whereas there was no significant change in CRT in the group with thinner baseline CRT (<400 μm) at any timepoint. However, this was not reflected in the BCVA changes in those two groups. The group with thin baseline CRT had a statistically significant increase in BCVA at 2-years (+10.3 letters, p < 0.05), and the group with thick baseline CRT had a statistically significant increase in BCVA at 1-year (+5.7 letters, p < 0.05), although both groups had no significant change in BCVA at 5 years compared to baseline. Results are summarised in Table 2.Table 2 Change in CRT (μm) compared to baseline for two categories of baseline CRT (<400 μm and ≥400 μm).
Number of eyes Change from baseline (in um) after
1 year 2 years 3 years 5 years
CRT ≥ 400 um CRT 20 −234.7* −225.4* −239.5* −257*
BCVA 5.7* 1.4 2 0.3
CRT < 400 um CRT 11 −5.5 39.9 52 −2.9
BCVA 2 10.3* 3.3 0.2
*p < 0.05.
Rescue therapy
After 5 years post-ILUVIEN, 42% of patients remained free of any rescue intravitreal injection. Eighteen eyes required rescue intravitreal therapy over 5 years, with a mean time to first rescue injection of 29.2 ± 14 months. Sixteen eyes received rescue anti-VEGF therapy (mean 6.4 ± 4.8 injections over 5 years), two eyes received an Ozurdex implant (mean 1.0 ± 0.0 implants over 5 years) and five eyes received a repeat ILUVIEN implant. Two eyes received intravitreal triamcinolone injections, which were given peri-operatively for epiretinal membrane surgery and retinal detachment surgery.
Eyes that did not require any rescue intravitreal injections were found to have received less macular laser at baseline than eyes that received rescue intravitreal injections (46% vs. 83%, respectively).
Two eyes received PRP laser and three eyes received macular laser over 5 years post-ILUVIEN.
Repeat ILUVIEN implant
Five eyes received one repeat ILUVIEN implant, with a mean time to repeat ILUVIEN of 38 ± 4 months. One of these five eyes suffered a rhegmatogenous retinal detachment affecting the macula 20 months after the initial ILUVIEN implant (eye number 18), which was surgically repaired, and ILUVIEN implant was removed during the vitrectomy, before receiving a second implant 41 months after the first one. This eye had a predictably poorer outcome. Mean change in BCVA from baseline is summarised in Table 3. No eye received more than one repeat ILUVIEN during this study’s 5-year follow-up period.Table 3 Change in BCVA (ETDRS letters) compared to baseline in eyes receiving no further intravitreal injections, eyes receiving repeat ILUVIEN implant (results shown including and excluding one eye which had a retinal detachment 20 months post-ILUVIEN), and eyes receiving other rescue intravitreal injections (anti-VEGF and/or Ozurdex implant).
Rescue intravitreal injection Number of eyes Mean change in BCVA compared to baseline (ETDRS letters)
1 year 2 years 3 years 5 years
None 13 +6 +5.1 +2 +3.8
ILUVIEN Including all eyes 5 +4.6 +6.4 +4.2 −5.8
Excluding 1 eye with retinal detachment at 20 months 4 +5.8 +14.5 +8 +1.5
Anti-VEGF and/or Ozurdex implant 13 +2.5 +1.1 +0.9 −1
Treatment burden
Eyes required a mean of 2.5 intravitreal injections per year prior to ILUVIEN, vs. 0.78 intravitreal injections per year in the 5 years post-ILUVIEN, representing a reduction in treatment burden of 69%.
Safety
IOP-related events
Five eyes (16%) were on IOP-lowering drops at baseline, versus 22 eyes (70%) on IOP-lowering drops at the 5-year follow-up visit. Eyes received an average of 0.2 ( ± 0.6) IOP-lowering topical medications at baseline, versus 1.3 ( ±1.1) IOP-lowering medications at the 5-year follow-up visit.
No eye had a prior diagnosis of glaucoma. The 5 eyes on IOP-lowering medication at baseline had either a diagnosis of OHT (n = 3) or prior steroid-response (n = 2). One of these eyes had poorly controlled IOP following ILUVIEN and required SLT and incisional surgery; the other four eyes continued to have well-controlled IOP on topical medication following ILUVIEN.
Over the five years following ILUVIEN, two eyes had selective laser trabeculoplasty (SLT) only, one eye had cyclodiode laser, and one eye had both SLT and incisional glaucoma surgery. As detailed above, the eye requiring SLT and incisional surgery had a prior history of ocular hypertension (OHT). The other three eyes requiring SLT or cyclodiode laser had no prior history of OHT or glaucoma, and one received a diagnosis of steroid-induced OHT, while the other two were diagnosed with OHT and glaucoma in both eyes several years after ILUVIEN and were not deemed to be steroid-induced by their glaucoma specialist.
Eyes receiving repeat ILUVIEN (n = 5) were not found to be at any significantly increased risk of IOP-related complications, with eyes receiving a mean of 1.5 IOP-lowering medications after year-5 and 1 eye receiving SLT.
Other complications
One eye developed rubeotic glaucoma (unilateral, occurred 4 years and 2 months after treatment with ILIUVIEN), which was managed with panretinal photocoagulation and topical IOP-lowering medication.
Other significant complications included one case of endophthalmitis presenting 3 days post-ILUVIEN implant (confirmed by vitreous tap), which was treated with intensive intravitreal antibiotic therapy and made a good recovery, with a 5-year BCVA of 61 letters (vs. 55 letters at baseline); one case of vitreous haemorrhage presenting 6 days post-ILUVIEN; and one case of rhegmatogenous retinal detachment 20 months post-ILUVIEN. One patient required epiretinal membrane surgery and vitrectomy 32 months post-ILUVIEN.
Safety-related outcomes are summarised in Table 4.Table 4 Summary of IOP-related outcomes and other significant complications occurring within 5 years following ILUVIEN.
Adverse events Eyes (n = 31) Further details
IOP-related events On IOP-lowering medication at baseline, n (%) 5 (16%)
On IOP-lowering medication after 5 years, n (%) 22 (70%)
Number of IOP-lowering agents after 5 years, mean ± SD 1.3 ± 1.1
SLT laser only, n (%) 2 (6.5%)
Cyclodiode laser only, n (%) 1 (3%)
Incisional glaucoma surgery, n (%) 1 (3%) Trabeculotomy
Other complications Endophthalmitis 1 (3%) 3 days post-ILUVIEN
Vitreous haemorrhage 1 (3%) 6 days post-ILUVIEN
Rhegmatogenous retinal detachment 1 (3%) 20 months post-ILUVIEN
Epiretinal membrane surgery 1 (3%) 32 months post-ILUVIEN
Rubeotic glaucoma 1 (3%) 49 months post-ILUVIEN
Discussion
This is the first report of real-world outcomes of patients over 5 years following treatment with intravitreal ILUVIEN implant for chronic DMO and it confirms the safety and efficacy of ILUVIEN demonstrated in FAME and PALADIN trials, as well as other real-world studies. Although our cohort had a lower baseline BCVA than in FAME and PALADIN studies (48.1 vs. 53.3 and 61.3 respectively), we still observed a statistically significant BCVA gain of +4.2 letters one year after ILUVIEN, which is in keeping with BCVA gains reported those studies (+4.4 letters in FAME study low-dose group after 2 years, and +3.71 in PALADIN study after 1 year), as well as in real-life studies such as the Medisoft audit study and IRISS study (+3.6 letters and +3.7 letters respectively after 1 year) [12–15]. Gains in BCVA were observed over 3 years post-ILUVIEN, although there was a gradual return to baseline BCVA at year 5, which is in keeping with the estimated duration of action of ILUVIEN of up to 3 years. The FAME, PALADIN, and Medisoft audit studies all demonstrated a sustained improvement in BCVA over three years, whereas in our small cohort we observed a peak improvement at 1-year, followed by a gradual decline towards baseline [13, 15, 19]. This could be in part due to the fact that our cohort had a higher mean baseline CRT than in FAME and Paladin studies (477 μm vs. 461 μm and 386 μm, respectively), a factor which has been shown to be associated with DMO persistence or earlier recurrence [20]. Anatomically, there was a significant improvement in CRT observed after 1 year and sustained throughout the 5-year follow-up period. Interestingly, this did not translate into sustained BCVA gains, a phenomenon which has been reported in several studies and may be attributable to other factors, such as neural and glial cell loss, disorganisation of the inner retinal layers, and macular ischaemia associated with DMO [21–23]. Further studies investigating different anatomical characteristics on OCT other than CRT and their predictive value on functional outcomes would be required, in order to better understand the differing functional responses to treatment and to better tailor individual treatment plans for different patients.
The proportion of eyes on topical IOP-lowering medication at the 5-year endpoint (70% after 5 years vs. 16% at baseline) was significantly higher than that reported in other studies: in the PALADIN study 22% of eyes were on IOP-lowering medication at year-3; in the FAME study 23.9% of eyes required treatment-emergent IOP-lowering medication over 3 years; in the IRISS study 23.3% of eyes required treatment-emergent IOP-lowering medication over 3 years; and in the Medisoft Audit study 29.7% of eye required treatment-emergent IOP-lowering medication over 2 years [13–15, 19]. Nevertheless, most eyes had well-controlled IOP on topical treatment alone, with the proportion of eyes receiving trabeculoplasty alone (6.5%) and incisional IOP-lowering surgery (3.2%) over 5 years post-ILUVIEN being in keeping with rates reported in other studies (1.3% and 4.8%, respectively in FAME study over 3 years) [19]. This suggests that the majority of eyes have well-controlled IOP on topical medication alone [19]. Therefore, the higher proportion of patients we observed on IOP-lowering eye drops may not be an accurate surrogate measure for the true rate of persistent OHT and glaucoma. In a real-world busy clinical practice, patients may not be as closely monitored as in clinical trials, particularly during the covid-19 pandemic period, which limited face-to-face assessments and may have impacted the regular monitoring required for patients receiving sustained-release intravitreal corticosteroids. Clinicians may therefore have a more cautious approach, using a lower threshold to start IOP-lowering therapy than in clinical trials. We can also hypothesise that there may be less of an emphasis in real-world practice to stop IOP-lowering drops in a timely manner, even when IOP has been well-controlled for several months, and while collecting the data for this study we did observe patients remaining on IOP-lowering treatment with well-controlled IOP for several years, without an attempt to stop treatment. This may represent sub-optimal clinical practice and, moving forward will be the object of an internal departmental review to improve the care of patients receiving sustained-release corticosteroid implants. The high proportion of patients already on IOP-lowering eye drops at baseline (16%) also sets this real-world study apart from the FAME (patients with any history of glaucoma or OHT were excluded) and PALADIN studies (9.6% of patients were on IOP-lowering medication at baseline) and may have an impact on the proportion of patients on IOP-lowering drops after 5 years [13]. Nevertheless, our findings reinforce the idea that patients receiving sustained-release intravitreal corticosteroid preparations require ongoing regular monitoring of IOP. Further research would be required to investigate the true impact of ILUVIEN on IOP and data on serial IOP measurements, optic disc cupping, and visual fields may be more informative in establishing the true adverse effects of ILUVIEN.
This study demonstrates a significant reduction in the number of intravitreal treatments required following ILUVIEN, with a remarkable 69% reduction in treatment burden and nearly half of eyes remaining free of intravitreal injections for 5 years. This is comparable with the 70.5% reduction in treatment frequency reported in the PALADIN study over 3 years following treatment with ILUVIEN [13]. The reduction in treatment burden we observed translates to nearly 9 fewer injections per eye on average over the 5 years following treatment with ILUVIEN; a very positive outcome for both patients, with a reduced risk of possible injection-related infection and discomfort, and providers, with a reduction in the ever-growing demand for intravitreal injections. This may be an important consideration for ophthalmology service providers proactively planning the delivery of diabetic eye disease care, which includes the use of long-term therapies, in accordance with the Royal College of Ophthalmologists Way Forward report. Further research would be needed to establish the optimum type and timing of rescue interventions following treatment with ILUVIEN.
Our study’s main strength is its long duration of follow-up in a real-world setting, including patients with previous OHT, previous vitrectomy, and where patients were reinjected with a second ILUVIEN. Potential limitations of this study include a small sample size, its retrospective nature, the lack of comparator arm, and use of IOP-lowering medication and laser/surgical intervention as a surrogate measure for IOP-related adverse events. The use of rescue treatments following ILUVIEN is also a potential confounding factor, but this study shows the long-term outcomes of real-life patients treated with ILUVIEN, for whom rescue intravitreal injections and laser treatment are commonly used adjunctive treatments.
Conclusion
This real-life study suggests that intravitreal ILUVIEN fluocinolone acetonide 0.19 mg sustained-release implant is a safe and effective treatment option for the treatment of chronic DMO in patients with a pseudophakic lens. We observed a significant improvement in both functional and anatomical outcomes one year after treatment, and after 5 years around two-thirds of eyes had the same or better visual acuity than at baseline, with a sustained reduction in CRT. The most commonly observed adverse effect was IOP elevation, which we found was higher than reported in other studies, although this may be due to confounding factors and the rate of serious adverse events remains low and in keeping with published literature. Larger studies are required to corroborate these findings.
Summary
What was known before
ILUVIEN is effective for the treatment of chronic diabetic macular oedema in pseudophakic eyes with effects lasting up to 3 years.
Main adverse effects include cataract formation in phakic eyes and intraocular pressure elevation.
What this study adds
This real-world study confirms the efficacy of the ILUVIEN implant over 5 years, with two-thirds of eyes having improved or stable visual acuity 5 years after ILUVIEN, and an overall sustained improvement in anatomical outcome.
Intraocular pressure elevation is a common adverse effect of ILUVIEN, but appears to be well controlled on topical therapy in this real-world setting, which includes a variety of patients, and also after repeated treatment with the ILUVIEN implant. Over 5 years following treatment with the ILUVIEN implant, the rate of serious adverse events, such as incisional IOP-lowering surgery, remains low and in keeping with rates reported in clinical trials.
This study demonstrates a 69% reduction in intravitreal treatment frequency following treatment with the ILUVIEN implant in a real-world setting.
Author contributions
ED contributed to the study’s planning, data collection, data analysis, creation of tables and figures, and writing of the manuscript. BM contributed to the study’s conceptualisation and planning, data analysis, and critically reviewing and editing the manuscript. BRM, RC, PLL, and AM critically reviewed and edited the manuscript.
Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.
Competing interests
E Dobler—none. BR Mohammed—received educational travel sponsorship from Bayer and Novartis and attended educational meeting sponsored by Alimera. R Chavan–received speaker fees and travel grants from Novartis, Bayer, and Allergan. PL Lip—none. A Mitra—consultant for Roche, Novartis, Alimera Sciences, and Allergan. B Mushtaq - advisory board member and consultant for Novartis, Bayer, Allergan, and Alimera sciences.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
==== Refs
References
1. Romero-Aroca P Managing diabetic macular edema: The leading cause of diabetes blindness World J Diabetes 2011 2 98 10.4239/wjd.v2.i6.98 21860693
2. Photocoagulation for Diabetic Macular Edema. Early treatment diabetic retinopathy study report number 1 Early Treatment Diabetic Retinopathy Study Research Group. Arch Ophthalmol. 1985;103:1796–806.
3. Heng L Comyn O Peto T Tadros C Ng E Sivaprasad S Diabetic retinopathy: pathogenesis, clinical grading, management and future developments Diabet Med 2013 30 640 50 10.1111/dme.12089 23205608
4. Wells JA Glassman AR Ayala AR Diabetic Retinopathy Clinical Research Network, Aflibercept, bevacizumab, or ranibizumab for diabetic macular edema N. Engl J Med 2015 372 1193 203 10.1056/NEJMoa1414264 25692915
5. Gonzalez V Campbell J Holekamp N Kiss S Loewenstein A Augustin A Early and long-term responses to anti-vascular endothelial growth factor therapy in diabetic macular edema: analysis of protocol I data Am J Ophthalmol 2016 172 72 79 10.1016/j.ajo.2016.09.012 27644589
6. Boyer D Yoon Y Belfort R Bandello F Maturi R Augustin A Three-year, randomized, sham-controlled trial of Dexamethasone Intravitreal implant in patients with diabetic macular edema Ophthalmology 2014 121 1904 14 10.1016/j.ophtha.2014.04.024 24907062
7. Amoaku W Saker S Stewart E A review of therapies for diabetic macular oedema and rationale for combination therapy Eye 2015 29 1115 30 10.1038/eye.2015.110 26113500
8. Ozurdex Summary of Product Characteristics. https://www.medicines.org.uk/emc/medicine/23422 (Accessed 5 May 2022)
9. ILUVIEN Summary of Product Characteristics. https://www.medicines.org.uk/emc/medicine/27636 (Accessed 5 May 2022)
10. National Institute for Health and Care Excellence. Fluocinolone acetonide intravitreal implant for treating chronic diabetic macular oedema after an inadequate response to prior therapy (NICE Guideline Number TA301). 2013. https://www.nice.org.uk/guidance/ta301
11. Campochiaro P Nguyen Q Hafiz G Bloom S Brown D Busquets M Aqueous levels of fluocinolone acetonide after administration of fluocinolone acetonide inserts or fluocinolone acetonide implants Ophthalmology 2013 120 583 7 10.1016/j.ophtha.2012.09.014 23218184
12. Campochiaro P Brown D Pearson A Ciulla T Boyer D Holz F Long-term benefit of sustained-delivery fluocinolone acetonide vitreous inserts for diabetic macular edema Ophthalmology 2011 118 626 35 10.1016/j.ophtha.2010.12.028 21459216
13. Singer M, Sheth V, Mansour S, Coughlin B, Gonzalez V, Three-year safety and efficacy of the 0.19-mg fluocinolone acetonide intravitreal implant for diabetic Macular Edema. Ophthalmology (2022). 10.1016/j.ophtha.2022.01.015
14. Chakravarthy U Taylor SR Koch FH Castro de Sousa JP Bailey C ILUVIEN Registry Safety Study (IRISS) Investigators Group. Changes in intraocular pressure after intravitreal fluocinolone acetonide (ILUVIEN): real-world experience in three European countries Br J Ophthalmol 2019 103 1072 7 10.1136/bjophthalmol-2018-312284 30242062
15. Bailey C Chakravarthy U Lotery A Menon G Talks J Extended real-world experience with the ILUVIEN® (fluocinolone acetonide) implant in the United Kingdom: 3-year results from the Medisoft® audit study Eye 2021 36 1012 8 10.1038/s41433-021-01542-w 33972705
16. Mushtaq B Bhatnagar A Palmer H Real-world outcomes in diabetic macular Edema for the 0.2 μg/Day fluocinolone acetonide implant: case series from the Midlands, UK. Clinical Ophthalmology 2021 15 2935 43
17. Alfaqawi F Lip P Elsherbiny S Chavan R Mitra A Mushtaq B Report of 12-months efficacy and safety of intravitreal fluocinolone acetonide implant for the treatment of chronic diabetic macular oedema: a real-world result in the United Kingdom Eye 2017 31 650 6 10.1038/eye.2016.301 28106887
18. Gregori N Feuer W Rosenfeld P Novel method for analyzing snellen visual acuity measurements Retina 2010 30 1046 50 10.1097/IAE.0b013e3181d87e04 20559157
19. Campochiaro P Brown D Pearson A Chen S Boyer D Ruiz-Moreno J Sustained delivery fluocinolone acetonide vitreous inserts provide benefit for at least 3 years in patients with diabetic Macular Edema Ophthalmology 2012 119 2125 32 10.1016/j.ophtha.2012.04.030 22727177
20. Cicinelli M Rabiolo A Zollet P Capone L Lattanzio R Bandello F Persistent or recurrent diabetic Macular Edema after Fluocinolone Acetonide 0.19 mg implant: risk factors and management Am J Ophthalmol 2020 215 14 24 10.1016/j.ajo.2020.03.016 32209341
21. Browning DJ Glassman AR Aiello LP Beck RW Brown DM Fong DS Relationship between optical coherence tomography-measured central retinal thickness and visual acuity in diabetic macular edema Ophthalmology 2007 114 525 36 10.1016/j.ophtha.2006.06.052 17123615
22. Deák G Schmidt-Erfurth U Jampol L Correlation of central retinal thickness and visual acuity in diabetic Macular Edema JAMA Ophthalmol 2018 136 1215 10.1001/jamaophthalmol.2018.3848 30193350
23. Gerendas B Prager S Deak G Simader C Lammer J Waldstein S Predictive imaging biomarkers relevant for functional and anatomical outcomes during ranibizumab therapy of diabetic macular oedema Br J Ophthalmol 2017 102 195 203 10.1136/bjophthalmol-2017-310483 28724636
|
PMC009xxxxxx/PMC9809901.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12517
10.1289/EHP12517
Response to Letter
Response to “Comment on ‘A Permutation Test-Based Approach to Strengthening Inference on the Effects of Environmental Mixtures: Comparison between Single-Index Analytic Methods’”
https://orcid.org/0000-0003-2606-2050
Day Drew B. 1
Sathyanarayana Sheela 1 2 3
LeWinn Kaja Z. 4
Karr Catherine J. 2 3 5
Mason W. Alex 6
Szpiro Adam A. 7
1 Center for Child Health, Behavior, and Development, Seattle Children’s Research Institute, Seattle, Washington, USA
2 Department of Pediatrics, University of Washington, Seattle, Washington, USA
3 Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, USA
4 Department of Psychiatry and Behavioral Sciences, University of California, San Francisco, San Francisco, California, USA
5 Department of Epidemiology, University of Washington, Seattle, Washington, USA
6 Department of Preventive Medicine, University of Tennessee Health Sciences Center, Memphis, Tennessee, USA
7 Department of Biostatistics, University of Washington, Seattle, Washington, USA
Address correspondence to Drew B. Day, Center for Child Health, Behavior, and Development, Seattle Children’s Research Institute, M/S Cure-3, P.O. Box 5371, Seattle, WA 98145 USA. Telephone: (206) 884-1798. Email: [email protected]
03 1 2023
1 2023
131 1 01800230 11 2022
13 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP12404
==== Body
pmcWe appreciate the careful consideration by Keil et al. of our paper1 comparing single-index mixture exposure models in simulations, including a novel permutation test for weighted quantile sum regression (WQSr) and quantile g-computation (QGC). We agree that mean absolute error (MAE) and mean absolute percent error (MAPE) assess accuracy and not bias. Therefore, we should have stated that the component-specific coefficients estimated by QGC were far less accurate that those of the WQSr models in our simulations. We appreciate the opportunity to make this correction.
However, we differ with Keil et al.’s characterization of QGC being more accurate than WQSr in their simulation scenario with 1 signal and 13 noise components. We reproduced this simulation using the permutation test version of WQSr instead of the training/test split version, which we recommended against using in our paper due to low power. Signal component MAPE was 11.6% for both WQSr and QGC, and MAE for all components was 0.018 for WQSr and 0.026 for QGC, indicating that WQSr is more accurate overall.
Keil et al. noted that WQSr is well-suited to the condition of unidirectional mixture effects used in our simulations.1 It remains unclear how it will perform when run in both directions for the case of bidirectional mixture effects. Similarly, performance of QGC has not been characterized with complex bidirectional mixture effects having more than two signal components.
Keil et al. argued our treatment of failed WQSr iterations (i.e., those with no detectable signal in the positive direction) underestimated error. Imputing zero for failed WQSr iterations accurately represents the result but may underestimate error for versions of WQSr with high failure rates, such as repeated holdout (RH). We adopted this approach partly to ensure a fair evaluation of the RH versions of WQSr, although we recommended avoiding these models because of their high failure rates. We recommended the permutation test WQSr model, for which ≤1% of simulations failed, so our conclusions would not be affected by alternative approaches.
In our alcohol/caffeine mixture analogy, Keil et al. noted that the exposures counteract each other and that QGC can provide that information in addition to direction-specific effects. However, it is not straightforward to characterize beneficial and adverse effects using QGC in complex situations with more than two components, such as a mixture of endocrine-disrupting chemicals with pro- and antiandrogenic mechanisms. It would be difficult to ascertain the beneficial and adverse effects of such a mixture with QGC.
WQSr and QGC are both appealing choices depending on the aims of a given analysis. QGC is well-suited to estimating an overall (i.e., positive plus negative) mixture effect. It also provides estimates of individual component coefficients, although our simulations suggest these may be inaccurate. WQSr is well suited to providing separate estimates of the beneficial and adverse (i.e., positive and negative) effects of the mixture. We believe that in many contexts it is critical to separately detect and characterize beneficial and adverse mixture effects, which is why we recommended WQSr with a permutation test to maximize power.
Acknowledgments
This research was supported by National Institutes of Health grants 1UG3OD023271-01 and 4UH3OD023271-03. Additional funding for the Conditions Affecting Neurocognitive Development and Learning in Early Childhood (CANDLE) study was provided by the Urban Child Institute.
==== Refs
Reference
1. Day DB, Sathyanarayana S, LeWinn KZ, Karr CJ, Mason WA, Szpiro AA. 2022. Permutation test-based approach to strengthening inference on the effects of environmental mixtures: comparison between single-index analytic methods. Environ Health Perspect 130 (8 ):87010, PMID: , 10.1289/EHP10570.36040702
|
PMC009xxxxxx/PMC9809902.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12404
10.1289/EHP12404
Letter to the Editor
Comment on “A Permutation Test-Based Approach to Strengthening Inference on the Effects of Environmental Mixtures: Comparison between Single-Index Analytic Methods”
https://orcid.org/0000-0002-0955-6107
Keil Alexander P. 1
Buckley Jessie P. 2 3
O’Brien Katie M. 4
Ferguson Kelly K. 4
White Alexandra J. 4
1 Occupational and Environmental Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health (NIH), Department of Health and Human Services (DHHS), Bethesda, Maryland, USA
2 Department of Environmental Health and Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
3 Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
4 Epidemiology Branch, National Institute of Environmental Health Sciences, NIH, DHHS, Research Triangle Park, North Carolina, USA
Address correspondence to Alexander P. Keil, 9609 Medical Center Dr., Rockville, MD 20892 USA. Email: [email protected]
03 1 2023
1 2023
131 1 01800108 11 2022
13 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
All authors declare they have no actual or potential competing financial interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP10570
==== Body
pmcRecently, Day et al.1 compared weighted quantile sum regression (WQSr) with quantile-based g-computation (QGC) using simulations and a worked example.2 They wrote that “mixture component-specific coefficients estimated by [QGC] were far more biased than those of any of the WQSr models.”1 Their results do not support this claim because bias was not assessed in isolation.
Instead, Day et al. assessed mean absolute percent error (MAPE), a measure of accuracy that combines bias and variability. To assess bias, we repeated one of their simulations (correlated mixture, β1=0.2) and examined mean percent bias (MPB: average bias/truth) of the component-specific coefficients. MPB of their weighted quantile sum regression bootstrap sample permutation test (WQSBSPT) approach was 2.5–8 times higher than that for QGC (Table 1). MPB for the “mixture effect” was 80 times higher for WQSBSPT than QGC. Thus, the results support a countermanding claim: QGC was far less biased than WQSBSPT. We consequently disagree that their results suggest “caution when interpreting [QGC] coefficients.”1
Table 1 Comparison of mean percent bias for QGC and WQSBSPT.
Mean percent bias (%)
(200 samples of n=500)
Estimand QGC WQSBSPT
Component coefficient, “high” effect size 2 −5
Component coefficient, “low” effect size 6 48
Overall effect size 0.1 8
Note: Mean percent bias for two contrasting approaches (QGC without bootstrapping and weighted quantile sum regression using WQSBSPT) to estimate the effects of a mixture using the data simulation methods of Day et al.1 with a “correlated mixture” developed with the empirical covariance matrix reported in the appendix of their paper. QGC, quantile-based g-computation without bootstrapping; WQSBSPT, weighted quantile sum regression bootstrap sample permutation test.
Day et al.’s simulations assume unidirectional causal exposures, an ideal setting for maximizing WQSr accuracy. A reanalysis using our previously published simulation2 with 1 causal exposure and 13 noise exposures yielded a 3-fold better component coefficient MAPE for QGC than WQSBS-Split (12% and 36%, respectively). Similar to Day et al.’s simulations, this simulation assumes no counteracting exposures. Thus, accuracy results do not generalize across different plausible scenarios.
We also note a fundamental flaw in some simulations. When WQSr methods failed to return a result, Day et al. imputed β^1=0. When analyzing data simulated with β1=0, the authors therefore imputed an estimate with perfect accuracy (no bias, no variability), which exaggerated WQSr performance. Because 86% of Weighted quantile sum regression, random subset, repeated holdout (WQSRS-RH) fits failed to return a result when β1=0, their WQSRS-RH results are not credible. Other WQSr results were also impacted.
Finally, we respond to the authors’ coffee/alcohol analogy. The analogy says that the effects of coffee and alcohol would cancel out one another in a QGC regression such that QGC would erroneously report no joint effect of substance use. If the joint effect of coffee and alcohol on a health outcome of interest is truly null, QGC will tell us so2 and also estimate the independent effects of alcohol and coffee. That is no error; it is crucial to understand when therapeutic agents counteract hazardous agents in a mixture. Alternatively, WQSr will yield biased independent effects of alcohol and coffee and no joint effect.3
Acknowledgments
This research was supported in part by grants from the National Institute of Environmental Health Sciences/National Institutes of Health (R01ES029531 and R01ES030078) (J.P. Buckley) and from the institute’s Intramural Research Program (Z01ES044005).
==== Refs
References
1. Day DB, Sathyanarayana S, LeWinn KZ, Karr CJ, Mason WA, Szpiro AAA. 2022. A permutation test-based approach to strengthening inference on the effects of environmental mixtures: comparison between single-index analytic methods. Environ Health Perspect 130 (8 ):87010, PMID: , 10.1289/EHP10570.36040702
2. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 128 (4 ):47004, PMID: , 10.1289/EHP5838.32255670
3. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2021. Response to “comment on ‘a quantile-based g-computation approach to addressing the effects of exposure mixtures.’” Environ Health Perspect 129 (3 ):38002, PMID: , 10.1289/EHP8820.33688745
|
PMC009xxxxxx/PMC9811992.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36598238
EHP10611
10.1289/EHP10611
Research
Long-Term Exposure to Air Pollution and the Occurrence of Metabolic Syndrome and Its Components in Taiwan
https://orcid.org/0000-0001-7848-1383
Chen Yi-Chuan 1
Chin Wei-Shan 2 3
Pan Shih-Chun 1
Wu Chih-Da 1 4
Guo Yue-Liang Leon 1 5 6
1 National Institute of Environmental Health Sciences, National Health Research Institute, Miaoli, Taiwan
2 School of Nursing, College of Medicine, National Taiwan University (NTU), Taipei, Taiwan
3 Department of Nursing, NTU Hospital, Taipei, Taiwan
4 Department of Geomatics, National Cheng Kung University, Tainan, Taiwan
5 Environmental and Occupational Medicine, College of Medicine, NTU and NTU Hospital, Taipei, Taiwan
6 Graduate Institute of Environmental and Occupational Health Science, College of Public Health, NTU, Taipei, Taiwan
Address correspondence to Yue-Liang Leon Guo, Environmental and Occupational Medicine, College of Medicine, National Taiwan University (NTU) and NTU Hospital, Room 339, No. 17 Xu-Zhou Rd., Taipei, 100, Taiwan. Email: [email protected]
04 1 2023
1 2023
131 1 01700111 11 2021
19 11 2022
05 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Metabolic syndrome (MetS), a major contributor to cardiovascular and metabolic diseases, has been linked with exposure to air pollution. However, the relationship between air pollutants and the five components of MetS [abdominal obesity, elevated triglyceride, decreased high-density lipoprotein cholesterol (HDL-C), elevated blood pressure, and elevated fasting blood glucose levels], has not been clearly described.
Objective:
We examined the association between long-term exposure to air pollutants and the occurrence of MetS and its components by using a longitudinal cohort in Taiwan.
Methods:
The MJ Health Research Foundation is a medical institute that conducts regular physical examinations. The development of MetS, based on a health examination and the medical history of an MJ cohort of 93,771 participants who were enrolled between 2006 and 2016 and had two or more examinations, was compared with estimated exposure to air pollutants in the year prior to health examination. The exposure levels to fine particulate matter [PM with an aerodynamic diameter of ≤2.5μm (PM2.5)] and nitrogen dioxide (NO2) in the participants’ residential areas were estimated using a hybrid Kriging/land-use regression (LUR) model executed using the XGBoost algorithm and a hybrid Kriging/LUR model, respectively. Cox regression with time-dependent covariates was conducted to estimate the effects of annual air pollutant exposure on the risk of MetS and its components.
Results:
During the average follow-up period of 3.4 y, the incidence of MetS was 38.1/1,000 person-years. After mutual adjustment and adjustments for potential covariates, the results indicated that every 10-μg/m3 increase in annual PM2.5 concentration was associated with an increased risk of abdominal obesity [adjusted hazard ratio (aHR)=1.07; 95% confidence interval (CI): 1.01, 1.14], hypertriglyceridemia (aHR=1.17; 95% CI: 1.11, 1.23), low HDL-C (aHR=1.09; 95% CI: 1.02, 1.17), hypertension (aHR=1.15; 95% CI: 1.09, 1.21), and elevated fasting blood glucose (aHR=1.15; 95% CI: 1.10, 1.20). Furthermore, PM2.5 and NO2 may increase the risk of developing MetS among people who already “have” some components of MetS.
Discussion:
Our findings suggest that in apparently healthy adults undergoing physical examination, exposure to PM2.5 and NO2 might be associated with the occurrence of MetS and its components. https://doi.org/10.1289/EHP10611
Supplemental Material is available online (https://doi.org/10.1289/EHP10611).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
In the 2017 Global Burden of Disease Study, cardiovascular disease (CVD) and type 2 diabetes mellitus (T2DM) accounted for 31.8% (17.8 million) and 1.8% (1 million), respectively, of all-cause global deaths for that year.1 Metabolic syndrome (MetS), a cluster of modifiable components—namely abdominal obesity, insulin resistance, dyslipidemia, and elevated blood pressure (BP)—is regarded as an indicator of CVD and T2DM development and as a contributor to all-cause mortality.2,3 Therefore, the identification of MetS can help in the prevention of the onset of the aforementioned diseases.
Traditional risk factors for MetS include older age, the male sex, low socioeconomic status, and poor lifestyle habits.2 In addition, increasing bodies of epidemiological evidence from Asia,4–6 Europe,7–9 and the United States10,11 demonstrate that ambient air pollution may contribute to an increased risk of MetS and its components. A meta-analysis of cohort studies revealed that every 5-μg/m3 annual increment of fine particulate matter [PM with an aerodynamic diameter of ≤2.5μm (PM2.5)] was associated with a 4% higher risk of MetS.12 Overall, the population-attributable risk of MetS associated with long-term PM2.5 exposure was estimated to be 12.28%.12
To control air pollution and prevent its destructive effect on health, legislation has been promulgated in several countries, including in the United States,13 the European Union member states,14 China,15 and Taiwan.16 Notably, a quasi-experimental study revealed that the adverse effects of PM2.5 on dyslipidemia was mitigated after the implementation of the Air Pollution Prevention and Control Action Plan in China.17
However, longitudinal studies evaluating the effects of gradually decreasing concentrations of ambient air pollutants are limited. Studies using a time-dependent Cox regression analysis revealed that PM2.5 exposure was associated with a higher risk of MetS and its components.11,18 Similarly, traffic-related air pollutants were reported to be associated with MetS-related outcomes, such as T2DM.19 Therefore, exploring the long-term health effects of nitrogen dioxide (NO2), an indicator of traffic-related emissions,20 is also valuable. Accordingly, the purpose of this study was to execute a time-dependent Cox regression analysis to assess the effects of long-term exposure to PM2.5 and NO2 on the incidence of MetS and its components in a cohort selected from the MJ Health Research Foundation.
Methods
Study Population
The study population was a cohort selected from the database of the MJ Health Research Foundation in Taiwan, which has been built up as a cohort to collect individuals’ characteristics, life style, and health status through a health screening program to help researchers investigate the relationships between chronic diseases and modifiable risk factors.21–23 The MJ Health Database is a longitudinal, population-based health research cohort, comprising data (questionnaire, physical examination, and blood tests) from apparently healthy individuals seeking physical examination services at a private health care firm in Taiwan (MJ Health Management Institution).22 Individuals were enrolled beginning in 1994 and data is collected on an ongoing basis with, at present, no termination date. Many participants contributed data from multiple visits. At each visit, behavioral and lifestyle information was obtained via questionnaire, and anthropometric and biological data were obtained via the physical examination. Informed consent was collected before physical examination, and only health data with authorization from the MJ participants were available for research purposes.23 The study protocol was approved by the research ethics committee of the National Taiwan University Hospital (No. 202002093RIND).
Approximately half of the MJ participants paid for themselves or family members to have the health examination, and the other half received the examination as part of their employment benefits. Participants’ records were censored if they did not persistently take the physical examination, or if their companies (for those participating as part of their employment benefits) did not renew the contract with MJ and therefore subsequent data collection was not possible.
Our selection pool comprised 171,545 participants who provided deidentified health examination records for the period between 2006 and 2016, were ≥20 years of age at the time of enrollment, did not have CVD, and had a registered address in Taiwan. To observe the occurrence of MetS and its components, we excluded 65,411 participants who took only one health examination for inappropriateness to the follow-up study design. Furthermore, because air quality monitoring stations are not as dense in eastern Taiwan as they are in the western regions of the island, 1,626 participants who lived in the east (∼0.6% of the cohort) were excluded. Only the air pollutants of the participant’s first address were included for the analysis to avoid potential mixing of early and delayed effects. Thus, for participants who changed address, health records before they moved were used for analysis. After excluding records after a changed address, those who had only one health examination record, or those who lived in eastern Taiwan, we categorized the remaining participants into six groups according to their baseline condition as follows: a) without abdominal obesity, b) with normal triglyceride (TG) levels, c) with normal high-density lipoprotein cholesterol (HDL-C) levels, d) with normal BP levels, e) with normal fasting blood glucose (FBG) levels, and f) without MetS. Criteria for each of these groups is described in the section “Definition of MetS and Its Components.” Thus, the final cohort comprised 77,862, 76,101, 74,488, 69,871, 59,681, and 76,349 participants in the above six groups. Such categorization allowed the participants to be included in more than one group (Table S1). Figure 1 illustrates the flowchart of the data selection process.
Figure 1. Flowchart of data selection from the MJ database from 2006 to 2016. The MJ database is a longitudinal, population-based health research cohort, comprising data (questionnaire, physical examination, and blood tests) from apparently healthy individuals seeking physical examination services at a private health care firm in Taiwan (MJ Health Management Institution).22 Many participants contributed data from multiple visits. At each visit, behavioral and lifestyle information was obtained via questionnaire, and anthropometric and biological data were obtained via the physical examination.23 Only the air pollutants of the participant’s first address were included for the analysis to avoid potential mixing of early and delayed effects. For participants who changed address, only health records before they moved were used for analysis. Therefore, in the first exclusion process, despite 99,475 records being removed, no participant was removed.
Figure 1 is a flowchart with two steps. Step 1: Starting with an initial 171,545 number of participants and 509,706 health examination records, when participants enrolled in the MJ cohort from 2006 to 2016: Age greater than or equal to 20 years, did not have cardiovascular diseases, registered address in Taiwan where, under “address changed during the follow up,” there are 0 number of participants and 99,475 health examination records, under “had only 1 health examination record,” there are 65,411 number of participants and 65,411 health examination records, and under “lived in eastern Taiwan,” there are 1,628 number of participants and 5,250 health examination records. Step 2: Under initially without abdominal obesity, there are 77,962 participants and 255,921 health examination records; under initially without elevated triglycerides, there are 76,101 participants and 248,227 health examination records; under initially without decreased high-density lipoprotein cholesterol, there are 74,488 participants and 242,152 health examination records; under initially without elevated blood pressure, there are 69,871 participants and 228,114 health examination records; under initially without elevated fasting food glucose, there are 59,681 participants and 195,577 health examination records; and under initially without metabolic syndrome, there are 76,349 participants and 250,664 health examination records.
Exposure Assessment
According to the “Air Quality Annual Report of R.O.C. (Taiwan), 2015,”24 yearly distributions of PM2.5 and NO2 gradually decreased from 33.5 to 22.1 μg/m3 and from 18.75 to 14.21 ppb, respectively, from 2006 to 2016. For PM2.5, the southwestern region is situated on the leeside of the mountains and under the effect of monsoonal flow, where ambient PM combines with local anthropogenic emissions from industries, and consequently, PM2.5 concentrations increase from the northern to southern part of western Taiwan.25 For NO2, primary emission from vehicles, there were ∼21.5 million registered motor vehicles in Taiwan, 39.6%, 27.2%, 30.6%, and 2.6% in northern, central, southern, and eastern Taiwan, respectively.26 Compared with similar areas in these four regions, NO2 concentrations were higher in northern and southern regions, followed by central and eastern regions (Table S2). The regions correspond to administrative units and are considered to have slight cultural differences. Roughly, the southern region is tropical, including five cities/counties, and the central and northern are subtropical, including five and six cities/counties, respectively. Western Taiwan is essentially the combination of the northern, central, and southern regions (Figure S1).
The average annual concentrations of PM2.5 and NO2 for the year prior to the health examination were recorded at the township level according to each participant’s address and used as surrogates for long-term exposure. The original daily concentrations of pollutants were measured continuously and reported hourly—PM2.5 by the β-ray attenuation method and tapered element oscillating microbalance technology, and NO2 by chemiluminescence24—from 73 fixed air quality monitoring stations, included 68 from western and 5 from eastern Taiwan.27 Given that residents of the eastern coast represented a very small (0.6%) proportion of MJ examinees and that the monitoring stations were not as dense in eastern Taiwan, the decision was made to exclude eastern coast participants. Estimations of PM2.5 and NO2 from 2005 to 2015 were modeled from the 68 monitoring stations from the western coast of Taiwan. Modeling was performed as previously described.27,28 Daily concentrations of PM2.5 and NO2 at monitoring stations were obtained from the Taiwan Environmental Protection Administration air quality database and were then aggregated into annual averages for modeling. Interpolated PM2.5 and NO2 values were generated via a leave-one-out ordinary Kriging model, as explanatory variables in the stepwise land-use regression (LUR).27,28 For PM2.5, predictors included geospatial variables [i.e., distance to the nearest airport, forest, farmland, and Normalized Difference Vegetation Index (NDVI) within circular buffers], meteorological variables (i.e., temperature, relative humidity, wind speed, wind direction, precipitation, and ultraviolet index), and copollutants [i.e., sulfur dioxide (SO2), ozone (O3), and NO2]. At a 50×50m grid resolution, the hybrid Kriging/LUR with the XGBoost algorithm model showed the adjusted R2 of 10-fold cross-validation was 0.93.27 For NO2, predictors included geospatial variables (i.e., agriculture, forest, transportation, water, building, public facilities, recreation, mining or salt production, industrial parks, incinerator chimneys and powerplants, Chinese restaurants, temples, funeral facilities, crematoria, and NDVI), meteorological variable (i.e., temperature), and copollutants [i.e., SO2, O3, and PM with an aerodynamic diameter of ≤10μm (PM10)]. The developed hybrid Kriging/LUR model with a 250×250m grid resolution showed the adjusted R2 of 10-fold cross-validation was 0.88.28
Definition of MetS and Its Components
The definition of MetS applied in this study was based on a joint scientific statement from the International Diabetes Federation and the American Heart Association/National Heart, Lung, and Blood Institute.29 The criteria for the clinical diagnosis of MetS were the presence of three or more of the following components for Asian individuals: a) having a waist circumference of ≥90cm in men and ≥80cm in women, b) having a TG level of ≥150mg/dL or receiving drug treatment for hypertriglyceridemia, c) having an HDL-C level of <40mg/dL (1.0 mmol/L) in males and <50mg/dL (1.3 mmol/L) in females or receiving drug treatment for decreased HDL-C, d) having a systolic BP level of ≥130 or diastolic BP level of ≥85 mmHg or receiving drug treatment for hypertension, and e) having an FBG level of ≥100mg/dL or receiving drug treatment for hyperglycemia.29 The same criteria were used individually to classify change in each component (negative to positive) for the other cohorts as well.
In health examinations by MJ Health Research Foundation, measurements of waist circumference and BP were standardized according to the recommendations by the Health Promotion Administration, Ministry of Health and Welfare, Taiwan. The measurements for FBG, TG, and HDL-C were conducted using a Toshiba C-8000.30
In the present study, time of incident MetS and its components was defined as the first detection of such conditions by examination, which included interview and blood collection. Participants who had moved from their first address, were lost to follow-up (i.e., did not continue to participate in the health examination), or had not developed MetS or its components by 31 December 2016 were regarded as right-censored data in our study.
Covariates
MetS has been associated with age, smoking, heavy carbohydrate intake, and physical inactivity and negatively associated with moderate alcohol intake (women only), education level,2 and marital satisfaction (in women).31 In addition, it has been related to poor sleep quality,32 and either short sleep or long sleep duration.33 We included these variables as covariates for analysis, except for carbohydrate intake which was unavailable in MJ data set. We thus included fried food consumption and processed food consumption as covariates in the present study.
Individual characteristics included as covariates were age, sex, baseline body mass index (BMI), marital status, education level, sleeping time per day, smoking habits, alcohol drinking habits, fried and processed food consumption, and regular exercise. Body weight and height were examined by experienced nurses during examination. BMI was calculated by dividing body weight (in kilograms) by squared height (in meters squared), whereas the other covariates were collected at each health examination via self-administered questionnaire.23 In addition, because of established cultural differences, as well as differences in concentrations of PM2.5 and NO2 among northern, central, and southern Taiwan, region of residence was included as a covariate as well.
Marital status was classified as single/divorced/separated/widowed and married/cohabiting; and education level was categorized as junior high school and below, general and vocational high school, college, and master’s degree and above. Sleeping time per day was categorized into the following groups: <6, 6–8, and >8h. Smoking habits were classified as never smoking/former smoking, secondhand smoke exposure, and frequent smoking/daily smoking; alcohol drinking habits were divided into never drinking/former drinking, occasional drinking, and frequent drinking/daily drinking. Fried and processed food consumption was classified as none, little, or ≤1 portion/wk; 2–3 portions/wk; and ≥4 portions/wk. Regular exercise was classified as none, little or <1h weekly; 1–4 h weekly or once per 2–3 d; ≥5h weekly or daily. Moreover, in sensitivity analyses, age was divided into the following categories: ≤44, 45–64, and ≥65y, and baseline BMI was divided into <18.5, 18.5–24, and ≥24 kg/m2.
Statistical Analysis
In previous longitudinal studies with lower to higher annual PM2.5 concentrations, the effects of increases in PM2.5 on MetS and its components changed from statistically nonsignificant8,34,35 to significant detrimental.4,36–41 Because the effect of ambient air pollutants on health may have decreased over time as a result of decreasing pollutant concentrations during the follow-up period, we performed a time-dependent Cox regression analysis to estimate the effects of long-term exposure to PM2.5 and NO2 on the incidence of abdominal obesity, elevated TG, reduced HDL-C, elevated BP, elevated FBG, and MetS. The terms elevated and reduced refer to values above and below the standard reference range as defined in the “Definition of MetS and Its Components” section. The advantage of a time-dependent Cox regression analysis is that it allows hazard ratios (HRs) to be separated into distinct time windows,42 making it suitable for this study given that concentrations of ambient air pollutants in Taiwan have gradually decreased.24 In the present study, the time-varying average of PM2.5 and NO2 are on the yearly scale. Spearman correlation coefficients were used to examine the correlation between PM2.5 and NO2.
The study models were adjusted for the aforementioned covariates, except for baseline BMI, with a time-varying method as well. Missing information on covariates were represented by the previous value available for each participant, contributing an additional ∼3% of the valid data for analysis.
To assess the robustness of the outcomes, we performed sensitivity analyses, and adjusted for additional covariates. First, we included not only characteristics’ covariates but also baseline waist circumference, TG, HDL-C, BP, and FBG for additional adjustment in each cohort under the consideration that an individual’s baseline status may affect the incidence of MetS and its components. Although some articles included BMI as a covariate in the model,18,43 BMI could be associated with air pollution as well; thus, the model without baseline BMI was examined (Table S3). Second, in previous studies, the effects of noise (which is highly associated with traffic-related nitrogen oxides) on MetS and its components were inconsistent.10,35 In addition, sleeping time has been found to be associated with traffic noise,44 so adjusting for sleeping time with NO2 might have been an overadjustment. Considering the lack of traffic noise information, we examined the effects without sleeping time in our study (Table S4). Third, owing to the varied prevalence of MetS across age and sex,2 we performed sensitivity analyses by age and sex stratification. The potential modification effects were examined by adding interaction terms of age, sex, and the air pollutants into the time-dependent Cox regression models.
The area of Taiwan is ∼36,000 km2,45 with a population of 23.5 million people.46 To illustrate the population density and the distribution of our study participants, Figure 2 was created using the QGIS Desktop software (version 3.22; Open Source Geospatial Foundation). All the statistical analyses were performed using SAS software (version 9.4; SAS Institute Inc.). We considered p<0.05 as indicating statistical significance for a two-tailed test.
Figure 2. Geographical distributions of population density in Taiwan (N=23,539,816), and the participants of the metabolic syndrome and its components cohorts (N=93,771). The area of Taiwan is ∼36,000 km2,44 with a population of 23.5 million people.45 (A) Population density. (B) General distribution of enrolled participants; each dot represents one participant. The small, outlined areas represent townships. This figure was created using the QGIS Desktop software (version 3.22; Open Source Geospatial Foundation).
Figures 2A and 2B are maps of Taiwan. Figure 2A depicts the geographical distribution of population density in Taiwan in December 2016. The density of inhabitants per kilometer squared is divided into five categories, namely, 5 to 165, 165 to 452, 452 to 954, 954 to 3,025, and 3,025 to 39,203. Figure 2B depicts the geographical distribution of enrolled participants in the metabolic syndrome and its component cohorts from 2006 to 2016. Each dot represents one participant. A scale depicts the distance in kilometers, ranging from 0 to 50 in increments of 25.
Results
Figure 2 shows the geographical distributions of the overall study participants (N=93,771), which could be similar to the population density in Taiwan (N=23,539,816). Table 1 presents the baseline characteristics of the study population, which was divided into six cohorts: without abdominal obesity, with normal TG, with normal HDL-C, with normal BP, with normal FBG, and without MetS. Most of the participants (66.1%–74.9%) were ≤44 years of age at the time of enrollment, and the male and female distributions in the population were similar. In terms of BMI, 58.0%–65.9% of the participants were within the normal range (18.5–24 kg/m2). More than 60% of the participants were married or cohabiting and had a college-level education or higher. Approximately 80% of the participants were never smokers or former smokers, and 84.3%–87.1% of them were never alcohol drinkers or former drinkers. Furthermore, >70% of the participants slept 6–8 h/d, and nearly half did no, little or <1h weekly regular exercise. Regarding fried and processed food consumption, ∼30% and 60% of the participants consumed none and little or ≤1 portion per week, respectively. Most of the participants (68.7%–71.9%) lived in northern Taiwan. Overall, there were 93,771 participants enrolled in the six cohorts: 5.1% (n=4,766) in only one cohort, and 36.9% (n=34,572) in all six cohorts (Table S1). The annual concentration of PM2.5 was revealed to be mildly correlated with NO2 in the six cohorts in whole Taiwan, with Spearman correlation coefficients ranging from 0.249 to 0.291 (all p<0.001; Table S5).
Table 1 Baseline characteristics [n (%) or mean±SD] of the MJ Health Database study population between 2006 and 2016, Taiwan.
Variable Cohort without abdominal obesity (N=77,862) Cohort with normal TG (N=76,101) Cohort with normal HDL-C
(N=74,488) Cohort with normal BP (N=69,871) Cohort with normal FBG (N=59,681) Cohort without MetS (N=76,349)
Already diagnosed with MetS
No 69,970 (93.6) 68,633 (93.5) 67,163 (90.2) 63,061 (94.1) 54,903 (96.0) 76,349 (100)
Yes 4,765 (6.4) 4,771 (6.5) 7,266 (9.8) 3,930 (5.9) 2,287 (4.0) 0 (0)
Missing 3,127 2,697 59 2,880 2,490 0
Enrolled age (y) 40.2±11.6 40.3±12.0 41.1±12.1 39.0±10.6 38.6±11.0 40.1±11.6
≤44 53,821 (69.1) 52,354 (68.8) 49,209 (66.1) 51,388 (73.5) 44,694 (74.9) 53,006 (69.4)
45–64 21,206 (27.2) 20,488 (26.9) 21,821 (29.3) 16,963 (24.3) 13,324 (22.3) 20,594 (27.0)
≥65 2,835 (3.6) 3,259 (4.3) 3,458 (4.6) 1,520 (2.2) 1,663 (2.8) 2,749 (3.6)
Sex
Male 37,055 (47.6) 33,623 (44.2) 37,469 (50.3) 31,064 (44.5) 25,145 (42.1) 35,573 (46.6)
Female 40,807 (52.4) 42,478 (55.8) 37,019 (49.7) 38,807 (55.5) 34,536 (57.9) 40,766 (53.4)
Body mass index (kg/m2) 22.1±2.7 22.6±3.4 22.9±3.5 22.5±3.3 22.4±3.3 22.4±3.1
<18.5 6,891 (8.9) 6,773 (8.9) 6,130 (8.2) 6,461 (9.2) 6,033 (10.1) 6,518 (8.5)
18.5–24 51,311 (65.9) 46,206 (60.7) 43,187 (58.0) 42,900 (61.4) 37,041 (62.1) 47,820 (62.6)
≥24 19,643 (25.2) 23,103 (30.4) 25,152 (33.8) 20,504 (29.3) 16,593 (27.8) 22,004 (28.8)
Missinga 17 19 19 6 14 7
Marital status
Single/divorced/separated/widowed 26,155 (33.8) 26,324 (34.8) 24,659 (33.3) 24,031 (34.6) 22,003 (37.1) 25,670 (33.8)
Married/cohabitating 51,227 (66.2) 49,307 (65.2) 49,457 (66.7) 45,411 (65.4) 37,315 (62.9) 50,325 (66.2)
Missinga 480 470 372 429 363 354
Education level
Junior high school and below 8,748 (11.3) 9,373 (12.4) 9,616 (12.9) 6,405 (9.2) 5,736 (9.6) 8,696 (11.4)
General and vocational high school 14,822 (19.1) 14,237 (18.8) 13,809 (18.6) 13,816 (18.9) 11,057 (18.6) 14,312 (18.8)
College 43,004 (55.4) 41,637 (54.9) 40,188 (54.1) 39,950 (57.4) 34,220 (57.5) 42,185 (55.4)
Master’s degree and above 11,001 (14.2) 10,572 (14.0) 10,682 (14.4) 10,063 (14.5) 8,450 (14.2) 10,975 (14.4)
Missinga 287 282 193 267 218 181
Smoking habits
Never smoking/former smoking 60,882 (78.4) 60,732 (80.1) 58,107 (78.2) 54,083 (77.6) 46,807 (78.7) 60,056 (78.8)
Secondhand smoke exposure 3,290 (4.2) 3,411 (4.5) 3,196 (4.3) 3,022 (4.3) 2,623 (4.4) 3,294 (4.3)
Frequent smoking/daily smoking 13,447 (17.3) 11,722 (15.5) 13,035 (17.5) 12,545 (18.0) 10,060 (16.9) 12,857 (16.9)
Missinga 243 236 150 221 191 142
Alcohol drinking habitsb
Never drinking/former drinking 66,801 (86.3) 66,154 (87.4) 62,822 (84.7) 60,452 (87.0) 51,990 (87.6) 65,718 (86.4)
Occasional drinking 7,316 (9.4) 6,750 (8.9) 7,667 (10.3) 6,484 (9.3) 5,231 (8.8) 7,213 (9.5)
Frequent drinking/daily drinking 3,322 (4.3) 2,774 (3.7) 3,677 (5.0) 2,553 (3.7) 2,148 (3.6) 3,108 (4.1)
Missinga 423 423 322 382 312 310
Sleeping time per day (h)
<6 16,751 (21.6) 16,946 (22.3) 16,706 (22.5) 15,027 (21.6) 12,939 (21.8) 16,713 (21.9)
6–8 56,114 (72.3) 54,328 (71.6) 53,112 (71.5) 50,300 (72.2) 42,810 (72.0) 54,801 (71.9)
>8 4,731 (6.1) 4,560 (6.0) 4,491 (6.0) 4,298 (6.2) 3,733 (6.3) 4,647 (6.1)
Missinga 266 267 179 246 199 161
Regular exercise
None, little or <1h weekly 35,252 (49.6) 34,312 (49.7) 33,498 (49.3) 33,239 (52.1) 28,042 (51.4) 34,779 (49.9)
1–4 h weekly or once per 2–3 d 26,581 (37.4) 25,678 (37.2) 24,962 (36.8) 23,412 (36.7) 20,249 (37.1) 26,001 (37.3)
≥5h weekly or daily 9,195 (12.9) 9,047 (13.1) 9,428 (13.9) 7,126 (11.2) 6,219 (11.4) 8,985 (12.9)
Missinga 6,834 7,064 6,600 6,094 5,171 6,584
Fried food consumption per week (portions)
None, little or ≤1 22,525 (29.0) 21,849 (28.8) 21,370 (28.8) 19,233 (27.6) 16,376 (27.5) 21,904 (28.8)
2–3 40,811 (52.6) 39,706 (52.4) 38,849 (52.3) 36,876 (53.0) 31,493 (52.9) 40,015 (52.5)
≥4 14,252 (18.4) 14,267 (18.8) 14,072 (18.9) 13,504 (19.4) 11,610 (19.5) 14,251 (18.7)
Missinga 274 279 197 258 202 179
Processed food consumption per week (portions)
None, little or ≤1 48,818 (62.9) 47,402 (62.5) 45,996 (61.9) 43,233 (62.1) 36,854 (62.0) 47,669 (62.6)
2–3 25,588 (33.0) 25,255 (33.3) 25,064 (33.7) 23,411 (33.6) 20,093 (33.8) 25,364 (33.3)
≥4 3,175 (4.1) 3,161 (4.2) 3,226 (4.3) 2,964 (4.3) 2,529 (4.3) 3,150 (4.1)
Missinga 281 283 202 263 205 184
Region
Northc 54,142 (69.5) 54,215 (71.2) 51,167 (68.7) 50,234 (71.9) 41,235 (69.1) 53,446 (70.0)
Centrald 10,178 (13.1) 9,553 (12.6) 10,352 (13.9) 8,868 (12.7) 8,722 (14.6) 10,153 (13.3)
Southe 13,542 (17.4) 12,333 (16.2) 12,969 (17.4) 10,769 (15.4) 9,724 (16.3) 12,750 (16.7)
Note: The criteria for the clinical diagnosis of MetS were the presence of any three or more of the following components for Asian individuals a) having a waist circumference of ≥90cm in men and ≥80cm in women, b) having a TG level of ≥150mg/dL or receiving drug treatment for hypertriglyceridemia, c) having an HDL-C level of <40mg/dL (1.0 mmol/L) in males and <50mg/dL (1.3 mmol/L) in females or receiving drug treatment for decreased HDL-C, d) having a systolic BP level of ≥130 or diastolic BP level of ≥85 mmHg or receiving drug treatment for hypertension, and e) having an FBG level of ≥100mg/dL or receiving drug treatment for hyperglycemia.29 The same criteria were used individually to classify change in each component (event to happen) for the other cohorts as well. BP, blood pressure; FBG, fasting blood glucose; HDL-C, high-density lipoprotein cholesterol; MetS, metabolic syndrome; SD, standard deviation; TG, triglyceride.
a Missing information on covariates were initially represented by the previous value available of each participant. Participants without available value for representation were not eligible for data analysis depending on the covariates included in the models in Table 3 and Tables S3 and S4.
b Never drinking/former drinking: teetotaler, abstainer, or drank less than once weekly; occasional drinking: drank once or twice weekly; frequent drinking/daily drinking: drank more than three times weekly.
c The northern region included Taipei, New Taipei, Keelung, Hsinchu, and Taoyuan Cities and Hsinchu County.
d The central region included Taichung City and Miaoli, Changhua, Nantou, and Yunlin Counties.
e The southern region included Kaohsiung, Tainan, and Chiayi Cities and Chiayi and Pingtung Counties.
Table 2 presents the results obtained for the three regions (northern, central, and southern Taiwan). Participants living in the southern region had the highest annual exposure concentration of PM2.5, and those in the northern and southern regions had the highest exposure concentration of NO2. During the exposure period of 2005 to 2015, the data shows a gradual decreasing trend, with fluctuation in the annual average concentrations of PM2.5 and NO2 (Figure S2; the complete data set is included as Excel Table S1). Over the period of 2005–2015, based on the residences of the participants in the cohorts of MetS and its components, PM2.5 decreased by 5.93–7.20 μg/m3 and NO2 decreased by 4.04–5.33 ppb, respectively. For our longitudinal analysis (Table 2), we included 77,862 participants who were initially free of abdominal obesity; the average incidence of abdominal obesity was 31.95/1,000 person-years over a mean±standard deviation (SD) follow-up period of 3.5±2.3y, and the incidence was lowest in southern Taiwan (23.11/1,000 person-years). Among the participants with normal TG at baseline, the average incidence of elevated TG was 50.01/1,000 person-years, ranging from 49.46 (south) to 52.16 (central). For those with initially normal HDL-C, the average incidence of reduced HDL-C was 28.89/1,000 person-years, ranging from 23.22 (north) to 44.32 (south). The average incidence of elevated BP in the normal BP cohort was 56.81, ranging from 55.25 (south) to 63.31 (central). The average incidence of elevated FBG in the normal FBG cohort was 112.77/1,000 person-years, with the lowest rate being in central Taiwan (88.09) and the highest being in the north (123.12). Among the cohort without MetS, MetS occurred after 3.4±2.3y of follow-up, with an average incidence of 38.07/1,000 person-years and the lowest rate occurring in the southern region (36.39) and the highest rate occurring in the central region (40.73). A total of 84.0% (n=78,786) participants did not have incident any component of MetS by the end of follow-up, whereas 0.1% (n=58) had incident all the five components (Table S1).
Table 2 Summary statistics of concentrations of air pollutants (2005–2015) and incidence of metabolic syndrome and its components in the MJ Health Research cohort by region between 2006 and 2016, Taiwan.
Statistics Cohort without abdominal obesity (N=77,862) Cohort with normal TG (N=76,101) Cohort with normal HDL-C (N=74,488) Cohort with normal BP (N=69,871) Cohort with normal FBG (N=59,681) Cohort without MetS (N=76,349)
1-y average concentration for the year before health check-up (mean±SD)
PM2.5 (μg/m3)
North 27.67±4.65 27.71±4.66 27.49±4.72 27.74±4.67 27.93±4.61 27.61±4.68
Central 33.97±3.81 34.00±3.84 33.98±3.81 34.06±3.82 34.19±3.77 33.99±3.82
South 42.77±4.03 42.82±4.05 42.73±4.05 42.86±4.00 42.89±3.99 42.76±4.04
Total 31.30±7.33 31.05±7.22 31.18±7.37 30.94±7.13 31.48±7.13 31.13±7.28
NO2 (ppb)
North 20.41±5.06 20.48±5.08 20.35±5.13 20.48±5.05 20.63±5.03 20.42±5.11
Central 15.86±2.85 15.82±2.89 15.79±2.89 16.00±2.81 15.99±2.87 15.81±2.87
South 20.35±4.69 20.38±4.72 20.24±4.80 20.60±4.55 20.54±4.66 20.35±4.69
Total 19.78±5.00 19.86±5.04 19.65±5.07 19.91±4.98 19.86±4.98 19.76±5.04
Participants with event [n (%)]a 8,632 (11.1%) 12,643 (16.6%) 7,266 (9.8%) 13,055 (18.7%) 20,541 (34.5%) 9,898 (13.0%)
Follow-up period (y) 3.5±2.3 3.3±2.2 3.4±2.2 3.0±2.0 3.1±2.1 3.4±2.3
Incidence rate (per 1,000 person-years)b
North 34.23 49.74 23.22 55.93 123.12 37.94
Central 32.28 52.16 36.16 63.31 88.09 40.73
South 23.11 49.46 44.32 55.25 95.31 36.39
Total 31.95 50.01 28.89 56.81 112.77 38.07
Note: BP, blood pressure; FBG, fasting blood glucose; HDL-C, high-density lipoprotein cholesterol; MetS, metabolic syndrome; NO2, nitrogen dioxide; PM2.5, fine particulate matter (PM with an aerodynamic diameter of ≤2.5μm); SD, standard deviation; TG, triglyceride.
a Event referred to occurrence of MetS and components. For cohort without abdominal obesity, event meant having a waist circumference of ≥90cm in men and ≥80cm in women. For cohort with normal TG, event meant having a TG level of ≥150mg/dL or receiving drug treatment for hypertriglyceridemia. For cohort with normal HDL-C, event meant having an HDL-C level of <40mg/dL (1.0 mmol/L) in males and <50mg/dL (1.3 mmol/L) in females or receiving drug treatment for decreased HDL-C. For cohort with normal BP, event meant having a systolic BP level of ≥130 or diastolic BP level of ≥85 mmHg or receiving drug treatment for hypertension. For, cohort with normal FBG, event meant having an FBG level of ≥100mg/dL or receiving drug treatment for hyperglycemia. For cohort without MetS, event meant the presence of any three or more of the above components.
b Incidence rate was calculated by dividing the number of events by total per 1,000 person-years for each cohort.
Table 3 presents the results of our time-dependent Cox regression analysis performed to assess the effects of two pollutants simultaneously on the HRs of developing abdominal obesity, elevated TG, reduced HDL-C, elevated BP, elevated FBG, and MetS. In Model 1—with adjustments for age, sex, marital status, education level, sleeping time per day, smoking habits, alcohol drinking habits, and fried and processed food consumption—every 10-μg/m3 increase in PM2.5 concentration in the year prior to the health examination significantly enhanced the HRs for all five components of MetS (i.e., abdominal obesity; TG, BP, or FBG above reference range; and HDL-C below reference range), and 10-ppb increase in NO2 concentration enhanced the HR for FBG above reference range. In Model 2, which included an additional adjustment for regular exercise, the results were similar to Model 1. In Model 3, which additionally adjusted for baseline status (waist circumference, TG, HDL-C, systolic and diastolic BP, and FBG for each cohort) and baseline BMI, and the association of PM2.5 remained significant; however, the association of NO2 with FBG became statistically nonsignificant. Every 10-μg/m3 increase in PM2.5 was associated with an increased risk for abdominal obesity [adjusted hazard ratio (aHR)=1.07; 95% confidence interval (CI): 1.01, 1.14], elevated TG (aHR=1.17; 95% CI: 1.11, 1.23), reduced HDL-C (aHR=1.09; 95% CI: 1.02, 1.17), elevated BP (aHR=1.15; 95% CI: 1.09, 1.21), and elevated FBG (aHR=1.15; 95% CI: 1.10, 1.20). As for the incidence of MetS, we present the results of participants who had none or some (1–2) components of MetS at baseline. For the participants initially without any component of MetS, we found no association between exposure to higher PM2.5 (aHR=0.99; 95% CI: 0.89, 1.12) and NO2 (aHR=0.88; 95% CI: 0.76, 1.03) and increased risk of MetS incidence. For those with one component of MetS, a 10-μg/m3 increase in PM2.5 was associated with a 12% greater risk of MetS (aHR=1.12; 95% CI: 1.04, 1.20). For those had two components of MetS, every 10-μg/m3 increase in PM2.5 and every 10-ppb increase in NO2 was associated with a 14% (aHR=1.14; 95% CI: 1.07, 1.22) and a 10% (aHR=1.10; 95% CI: 1.03, 1.18) increased risk for MetS incidence, respectively.
Table 3 Associations [aHR (95% CI)] of PM2.5 and NO2 with metabolic syndrome and its components among participants of the MJ Health Research cohort in Taiwan between 2006 and 2016.
MetS and its components Model 1a Model 2b Model 3c
PM2.5 NO2 PM2.5 NO2 PM2.5 NO2
Development of abdominal obesity in the no abdominal obesity cohort (n=76,960; e=8,555) (n=70,729; e=7,705) (n=70,589; e=7,698)
1.15 (1.01, 1.12) 1.06 (0.99, 1.12) 1.08 (1.02, 1.15) 1.01 (0.95, 1.07) 1.07 (1.01, 1.14) 0.97 (0.92, 1.03)
Development of reduced HDL-C in the normal HDL-C cohort (n=75,210; e=12,542) (n=68,733; e=11,373) (n=68,727; e=11,372)
1.16 (1.10, 1.22) 1.03 (0.99, 1.08) 1.17 (1.11, 1.23) 1.02 (0.97, 1.07) 1.17 (1.11, 1.23) 0.99 (0.94, 1.03)
Development of elevated TG in the normal TG cohort (n=73,707; e=7,210) (n=67,592; e=6,490) (n=67,587; e=6,489)
1.10 (1.02, 1.18) 1.01 (0.95, 1.07) 1.12 (1.04, 1.20) 1.00 (0.94, 1.07) 1.09 (1.02, 1.17) 0.94 (0.88, 1.01)
Development of elevated BP in the normal BP cohort (n=69,075; e=12,965) (n=63,514; e=11,772) (n=63,510; e=11,771)
1.18 (1.12, 1.24) 1.03 (0.98, 1.08) 1.18 (1.12, 1.24) 1.01 (0.97, 1.06) 1.15 (1.09, 1.21) 1.04 (0.99, 1.10)
Development of elevated FBG in the normal FBG cohort (n=59,681; e=20,549) (n=54,274; e=18,615) (n=54,269; e=18,614)
1.11 (1.06, 1.15) 1.07 (1.03, 1.11) 1.11 (1.06, 1.16) 1.06 (1.02, 1.10) 1.15 (1.10, 1.20) 1.03 (0.99, 1.07)
Development of MetS based on the number of MetS components exhibited at baseline (n=75,596; e=9,818) (n=69,460; e=8,850) (n=69,456; e=8,848)
0 (n=34,285; e=897) (n=31,788; e=824) (n=31,787; e=824)
0.99 (0.89, 1.11) 0.87 (0.75, 1.01) 0.99 (0.89, 1.11) 0.87 (0.75, 1.01) 0.99 (0.89, 1.12) 0.88 (0.76, 1.03)
1 (n=25,277; e=3,175) (n=23,137; e=2,858) (n=23,134; e=2,856)
1.12 (1.04, 1.20) 1.02 (0.94, 1.11) 1.12 (1.04, 1.21) 1.02 (0.94, 1.11) 1.12 (1.04, 1.20) 1.01 (0.93, 1.10)
2 (n=16,034; e=5,746) (n=14,535; e=5,168) (n=14,535; e=5,168)
1.12 (1.05, 1.19) 1.13 (1.06, 1.21) 1.12 (1.05, 1.19) 1.14 (1.06, 1.21) 1.14 (1.07, 1.22) 1.10 (1.03, 1.18)
Note: All estimates were calculated for every 10-μg/m3 increase in PM2.5 and every 10-ppb increase in NO2 in the annual average concentrations, two-pollutant model, determined using time-dependent Cox regression. Missing information on covariates were initially represented by the previous value available of each participant. Participants without available value for representation were not eligible for data analysis depending on the covariates in the models, leading to different eligible numbers of participants in different models. The terms elevated and reduced refer to above and below normal reference range, respectively. aHR, adjusted hazard ratio; CI, confidence interval; BP, blood pressure; e, number of participants with the incident outcome; FBG, fasting blood glucose; HDL-C, high-density lipoprotein cholesterol; MetS, metabolic syndrome; n, number of participants without missing variables in each model; NO2, nitrogen dioxide; PM2.5, fine particulate matter (PM with an aerodynamic diameter of ≤2.5μm); TG, triglyceride.
a Adjusted by age, sex, marital status (single/divorced/separation/widowed, married/cohabitating), education level (junior high school and below, general and vocational high school, college, master’s degree and above), smoking habits (never smoking/former smoking, secondhand smoke exposure, frequent smoking/daily smoking), alcohol drinking habits (never drinking/former drinking, occasional drinking, frequent drinking/daily drinking), sleeping time per day (<6, 6–8, >8h), fried food consumption (none, little or ≤1 portion weekly, 2–3 portions weekly, ≥4 portions weekly), processed food consumption (none, little or ≤1 portion weekly, 2–3 portions weekly, ≥4 portions weekly), and region (north, central, south).
b Additionally adjusted for the covariates from Model 1 and exercise (none, little or <1h weekly, 1–4 h weekly or once per 2–3 d; ≥5h weekly or daily).
c Additionally adjusted for the covariates from Model 2, baseline body mass index (<18.5, 18.5–24, ≥24 kg/m2), and the initial status—baseline waist circumference for abdominal obesity cohort, baseline TG for elevated TG cohort, baseline HDL-C for reduced HDL-C cohort, baseline systolic BP and baseline diastolic BP for elevated BP cohort, baseline FBG for elevated FBG cohort.
Tables S3 and S4 present models without baseline BMI and without sleeping time, respectively. The associations between PM2.5, NO2, MetS, and its components were similar to the findings in Table 3.
Figure S3 presents the potential modification effects of age and sex by adding the interaction terms in stratified analyses. Such findings were similar to our major results in Table 3. For sex, positive associations between PM2.5, elevated TG, reduced HDL-C, elevated BP, elevated FBG, and from initially having one or more components of MetS were found in both male and female participants. In addition, the association between PM2.5 and abdominal obesity was significant in male (aHR=1.15; 95% CI: 1.07, 1.23) but not in female (aHR=0.98; 95% CI: 0.91, 1.06, pInteraction=0.000); the association of PM2.5 with elevated BP was more prominent in female (aHR=1.22; 95% CI: 1.14, 1.30) than male (aHR=1.12; 95% CI: 1.06, 1.18, pInteraction=0.003). For age, positive associations between PM2.5, elevated TG, elevated BP, and elevated FBG were found in all age subgroups (≤44, 45–64, ≥65 years of age). In the abdominal obesity cohort, participants ≤44 years of age had higher risk (aHR=1.10; 95% CI: 1.03, 1.18) when exposed to increased PM2.5 than participants ≥65 years of age (aHR=0.93; 95% CI: 0.80, 1.07, pInteraction=0.017). No significant modification effect of age or sex subgroup in the associations between NO2, MetS, and its components was observed. The complete data set is included as Excel Table S2.
Discussion
Primary Findings
By using a time-dependent Cox regression analysis to consider the gradually changing concentrations of PM2.5 and NO2 in cohorts of 59,681–77,862 participants from the MJ Health Database with a follow-up period of 3.0–3.5 y, we observed associations between long-term PM2.5 and NO2 exposure and incidence of abdominal obesity, elevated TG, reduced HDL-C, elevated BP, elevated FBG, and MetS. An additional exposure to 10 μg/m3 of PM2.5 annually was associated with an increased risk of abdominal obesity (7%), hypertriglyceridemia (17%), reduced HDL-C (9%), hypertension (15%), and elevated FBG (15%). The associations between PM2.5 and incident MetS and its components remained robust when additionally adjusting for either baseline BMI (Table S3) or sleeping time (Table S4). In the sensitivity analyses, the association between PM2.5 and abdominal obesity was significant in male and ≤44-y-old participants, whereas the association of PM2.5 with elevated BP was more prominent in female participants. For age subgroups (≤44, 45–64, ≥65y), no interactions were observed between PM2.5, elevated TG, elevated BP, and elevated FBG (Figure S3). In addition, the effects of long-term PM2.5 and NO2 exposure were particularly pronounced in participants who had some components of MetS at baseline (abdominal obesity, high TG levels, low HDL-C levels, high BP levels, and high FBG levels). For participants who had had one or two components of MetS, every 10-μg/m3 increase in PM2.5 was associated with 12% and 14% risk of MetS incidence, respectively. For those who already had two components of MetS, every 10-ppb increase in NO2 was associated with a 10% risk of MetS incidence.
Related Literature
In 14 May 2012, Taiwan tightened the air quality standards of 24-h and yearly average PM2.5 concentrations to 35 and 15 μg/m3, respectively, as well as 24-h and yearly average NO2 concentrations to 100 and 30 ppb, respectively.16 In the present study, which included exposure data for the period from 2005 to 2015, although the annual average concentration of PM2.5 (based on each participant’s address) was ∼30 μg/m3, the concentration decreased gradually over time, as did that of NO2. Therefore, the changes in air pollutant concentrations must be carefully considered when examining their effects on MetS and its components, and time-dependent Cox regression analysis may be appropriate for the present study.
In other countries with comparatively lower annual PM2.5 concentrations (median≤20 μg/m3), longitudinal studies using linear regression, logistic regression, or generalized estimating equations have revealed that the effects of increases in PM2.5 exposure on MetS and its components were statistically nonsignificant, namely, in Germany8,35 and Southern California.34 However, when PM2.5 exposure was used as a time-dependent variable, an annual median value of 26.7 μg/m3 was demonstrated to be associated with higher risks of abdominal obesity, hypertriglyceridemia, reduced HDL-C, hypertension, hyperglycemia, and MetS from the National Health Insurance Service-National Health Screening Cohort in Korea,18 which is consistent with our findings. Moreover, people living in regions with high annual concentrations of PM2.5 (median ≥55 μg /m3) in China have been reported to be at higher risk of abdominal obesity from 31 China provinces,4 reduced HDL-C,36 and MetS41 from the Henan Rural Cohort study, elevated BP among rural and urban regions,37–39 and elevated FBG among an elderly population.40
In comparatively lower annual concentrations (mean<20 ppb) of NO2 (an indicator of traffic-related emissions), exposure was reported to be associated with elevated total cholesterol and low-density lipoprotein but not with reduced HDL-C, elevated FBG among young adults from the Southern California Children’s Health Study,34 or with MetS from the population-based survey in Augsburg, Germany.35 In studies where annual average concentrations were higher (20–30 ppb), NO2 exposure was associated with increased BP among older adults in Taiwan5 but not with elevated FBG among Taiwanese5 and Mexican Americans.47 When NO2 annual average concentration was >30 ppb, an increased risk of MetS was reported in Germany8 and China.41 However, association between NO2 concentrations and hypertension was inconsistent.38,39 In our study, the effect of NO2 on elevated FBG remained significant (Table 3, Model 1 and 2) before baseline status were added for additional adjustment. Therefore, we could not exclude the possibility of an existing impact of NO2 on baseline BMI and FBG before the study enrollment. The inconsistency in findings between previous studies and the present study could be partly attributable to the differences in NO2 concentrations, the study methodology, and analysis.
Possible Mechanisms
Exposure to PM was associated with DNA hypomethylation, and PM-induced reactive oxygen species and elevated cytokine levels has been reported to elicit systemic inflammation in murine and human studies.48 In healthy young adults, high exposure to PM2.5 was found to relate to changes in DNA methylation in genes involved in glucose and lipid metabolism, inflammation, oxidative stress, platelet activation, and cell survival and apoptosis.49 In a study with nonsmoking participants, PM2.5 exposure was found to be associated with autonomic nervous system imbalance, as well as with impaired endothelial function 24 h after exposure. Both were regarded to be relevant to hypertension.50 As for the potential mechanisms of the observed effects of NO2, inhaled NO2 oxidized antioxidants within the epithelial lining fluid and triggered extracellular damage and oxidative stress,51 which can inhibit glucose metabolism in rodents.52 Korean6 and German53 populations exposed to ambient NO2 concentrations were found to have impaired glucose metabolism, an important component of MetS. These mechanisms may support our findings.
Limitations and Strengths
Although we used a prospective cohort and rigorous exposure assessment of air pollutants, our study has some limitations. First, in this study, exposure assessment was based on the estimated concentration in the township the participant gave as their residence in the questionnaire. Personal habits, work exposure, commute, indoor/outdoor differences, and microenvironments render potential deviations in these measurements from the true exposure. However, these deviations tend to be in a random misclassification manner, thus causing the observed relationship between outcomes and exposure to be biased toward the null hypothesis. On the other hand, the outcomes measurements are also subject to measurement error. Given that those health care workers performing the health examination were unaware of the study hypothesis, bias is unlikely. Despite the efforts of standardization in questionnaire, equipment, and measurements,30,54 misclassification could not totally be avoided. Again, these misclassifications likely weaken the observed association. Therefore, the observed relationship likely underestimates the true effects. Second, the participants who changed address during the follow-up were excluded because their exposure could not be attributed to an exact level within the time period. This potentially induced selection bias to the study. However, moving has not been reported as a risk or protective factor for MetS. Third, the surrounding vegetation (greenness) in the participants’ residential areas was not directly included in the study models, neither was participants’ occupation. Some researchers have reported a negative association between MetS-related components and residential surrounding greenness,55–57 but the protective effect was inconsistent with that in urban regions.35,58 Exposure to annual concentrations of PM2.5 and NO2 was estimated at the township level, and green space was regarded as an explanatory predictor in the modeling procedure.27,28 In addition, data on real exposure to greenness for urban residents were unavailable. Accordingly, greenness was not adjusted for in this study. On the other hand, although some occupations have been reported to be associated with incident MetS,59 the MJ health questionnaire changed options of the occupational items during our research period, and the categories could not objectively present workers’ intensity of labor. Therefore, we did not include occupation as a covariate. Fourth, the incidence was based on the month of the participants took the health examination. Therefore, a time lag between the actual incidence of MetS and its components and the health examination existed, and the short-term effects of PM2.5 and NO2 exposure on MetS and its components were not achievable.
Regardless of these limitations, in Table 3, our study used time-dependent analysis to examine the health effects of PM2.5 and NO2 exposure, and found that exposure to PM2.5 was associated with the occurrence the components of MetS. The sensitivity analyses by age and sex stratifications showed that the association between PM2.5 and abdominal obesity was significant in the male and ≤44-y-old groups, and the association between PM2.5 and elevated BP was more prominent in the female group. Furthermore, the results of incident MetS indicates people who already had components of MetS could be vulnerable to the development of MetS when exposed to increased PM2.5 and NO2 concentrations. Our results generally support the hypothesis that long-term exposure to PM2.5 and NO2 are associated with increased risk of MetS and its components. Strategies aimed at improving air quality and using personal protective equipment might reduce the risks, especially for people who already have some components of MetS.
Conclusions
Our findings suggest that exposure to high levels of PM2.5 (∼15 μg/m3 above Taiwan’s current air quality standards)16 was associated with increased risk of abdominal obesity, hypertriglyceridemia, reduced HDL-C, elevated BP, and elevated FBG. Moreover, we also observed that exposure to PM2.5 and NO2 were associated with the risk of developing MetS among people who already had some components of MetS. Additional studies are required to confirm the consistency of the effects of PM2.5 and NO2 exposure at different exposure ranges.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
Part of the data used in this research was authorized by and received from the MJ Health Research Foundation (authorization code: MJHRF2018011A). Any interpretation or conclusion described in this paper does not represent the views of the MJ Health Research Foundation.
This study was supported by the National Health Research Institutes (NHRI-110-EMGP09).
==== Refs
References
1. GBD 2017 Causes of Death Collaborators. 2018. Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 392 (10159 ):1736–1788, PMID: , 10.1016/S0140-6736(18)32203-7.30496103
2. Cornier MA, Dabelea D, Hernandez TL, Lindstrom RC, Steig AJ, Stob NR, et al. 2008. The metabolic syndrome. Endocr Rev 29 (7 ):777–822, PMID: , 10.1210/er.2008-0024.18971485
3. Isomaa B, Almgren P, Tuomi T, Forsén B, Lahti K, Nissén M, et al. 2001. Cardiovascular morbidity and mortality associated with the metabolic syndrome. Diabetes Care 24 (4 ):683–689, PMID: , 10.2337/diacare.24.4.683.11315831
4. Cao S, Guo Q, Xue T, Wang B, Wang L, Duan X, et al. 2021. Long-term exposure to ambient PM2.5 increase obesity risk in Chinese adults: a cross-sectional study based on a nationwide survey in China. Sci Total Environ 778 :145812, PMID: , 10.1016/j.scitotenv.2021.145812.33721648
5. Chuang KJ, Yan YH, Chiu SY, Cheng TJ. 2011. Long-term air pollution exposure and risk factors for cardiovascular diseases among the elderly in Taiwan. Occup Environ Med 68 (1 ):64–68, PMID: , 10.1136/oem.2009.052704.20833756
6. Hwang MJ, Kim JH, Koo YS, Yun HY, Cheong HK. 2020. Impacts of ambient air pollution on glucose metabolism in Korean adults: a Korea National Health and Nutrition Examination Survey study. Environ Health 19 (1 ):70, PMID: , 10.1186/s12940-020-00623-9.32552747
7. Cai Y, Hansell AL, Blangiardo M, Burton PR, BioSHaRE, de Hoogh K, et al. 2017. Long-term exposure to road traffic noise, ambient air pollution, and cardiovascular risk factors in the HUNT and lifelines cohorts. Eur Heart J 38 (29 ):2290–2296, PMID: , 10.1093/eurheartj/ehx263.28575405
8. Matthiessen C, Lucht S, Hennig F, Ohlwein S, Jakobs H, Jöckel KH, et al. 2018. Long-term exposure to airborne particulate matter and NO2 and prevalent and incident metabolic syndrome—results from the Heinz Nixdorf Recall Study. Environ Int 116 :74–82, PMID: , 10.1016/j.envint.2018.02.035.29653402
9. Zhang S, Mwiberi S, Pickford R, Breitner S, Huth C, Koenig W, et al. 2021. Longitudinal associations between ambient air pollution and insulin sensitivity: results from the KORA cohort study. Lancet Planet Health 5 (1 ):e39–e49, PMID: , 10.1016/S2542-5196(20)30275-8.33421408
10. Yu Y, Paul K, Arah OA, Mayeda ER, Wu J, Lee E, et al. 2020. Air pollution, noise exposure, and metabolic syndrome—a cohort study in elderly Mexican-Americans in Sacramento area. Environ Int 134 :105269, PMID: , 10.1016/j.envint.2019.105269.31778933
11. Wallwork RS, Colicino E, Zhong J, Kloog I, Coull BA, Vokonas P, et al. 2017. Ambient fine particulate matter, outdoor temperature, and risk of metabolic syndrome. Am J Epidemiol 185 (1 ):30–39, PMID: , 10.1093/aje/kww157.27927620
12. Ning J, Zhang Y, Hu H, Hu W, Li L, Pang Y, et al. 2021. Association between ambient particulate matter exposure and metabolic syndrome risk: a systematic review and meta-analysis. Sci Total Environ 782 :146855, PMID: , 10.1016/j.scitotenv.2021.146855.33839664
13. U.S. Environmental Protection Agency. 2021. Evolution of the Clean Air Act. https://www.epa.gov/clean-air-act-overview/evolution-clean-air-act#intro [accessed 30 July 2021].
14. European Union. 2008. Directive 2008/50/EC of the European Parliament and of the Council of 21 May 2008 on ambient air quality and cleaner air for Europe. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A02008L0050-20150918 [accessed 30 July 2021].
15. Huang J, Pan X, Guo X, Li G. 2018. Health impact of China’s Air Pollution Prevention and Control Action Plan: an analysis of national air quality monitoring and mortality data. Lancet Planet Health 2 (7 ):e313–e323, PMID: , 10.1016/S2542-5196(18)30141-4.30074894
16. Taiwan Air Quality Monitoring Network. 2020. Air quality standards. https://airtw.epa.gov.tw/ENG/Information/Standard/Rules.aspx [accessed 30 July 2021].
17. Li J, Yao Y, Xie W, Wang B, Guan T, Han Y, et al. 2021. Association of long-term exposure to PM2.5 with blood lipids in the Chinese population: findings from a longitudinal quasi-experiment. Environ Int 151 :106454, PMID: , 10.1016/j.envint.2021.106454.33676285
18. Lee S, Park H, Kim S, Lee EK, Lee J, Hong YS, et al. 2019. Fine particulate matter and incidence of metabolic syndrome in non-CVD patients: a nationwide population-based cohort study. Int J Hyg Environ Health 222 (3 ):533–540, PMID: , 10.1016/j.ijheh.2019.01.010.30797734
19. Li Y, Xu L, Shan Z, Teng W, Han C. 2019. Association between air pollution and type 2 diabetes: an updated review of the literature. Ther Adv Endocrinol Metab 10 :2042018819897046, PMID: , 10.1177/2042018819897046.31903180
20. World Health Organization. 2003. Health Aspects of Air Pollution with Particulate Matter, Ozone and Nitrogen Dioxide: Report on a WHO Working Group, Bonn, Germany 13–15 January 2003. https://apps.who.int/iris/handle/10665/107478 [accessed 30 July 2021].
21. MJ Health Research Foundation. 2015. MJ Health Survey Database, MJ BioData [Data file, in Chinese]. http://www.mjhrf.org [accessed 31 July 2020].
22. Wu X, Tsai SP, Tsao CK, Chiu ML, Tsai MK, Lu PJ, et al. 2017. Cohort profile: the Taiwan MJ Cohort: half a million Chinese with repeated health surveillance data. Int J Epidemiol 46 (6 ):1744–1744g, PMID: , 10.1093/ije/dyw282.28204597
23. MJ Health Research Foundation. 2016. The Introduction of MJ Health Database. Technical Report No. MJHRF-TR-01. http://www.mjhrf.org/file/file/report/MJHRF-TR-01%20MJ%20Health%20Database.pdf [accessed 30 March 2022].
24. Environmental Protection Administration, Executive Yuan, R.O.C. (Taiwan). Air Quality Annual Report of R.O.C. (Taiwan), 2015 [in Chinese]. https://www.epa.gov.tw/DisplayFile.aspx?FileID=920B8BD6AA7922CA [accessed 30 March 2022].
25. Cheng FY, Hsu CH. 2019. Long-term variations in PM2.5 concentrations under changing meteorological conditions in Taiwan. Sci Rep 9 (1 ):6635, PMID: , 10.1038/s41598-019-43104-x.31036848
26. Directorate General of Highways, R.O.C. (Taiwan). 2019. Annual statistics of registered motor vehicles in 2017. https://www.thb.gov.tw/sites/ch/modules/StatisticsSummary/StatisticsSummary-List?node=47df19cb-4615-4f2e-b322-6bcceef70406&y=2017&id=cdb43826-93b2-41ff-a68d-a5d41573b747 [accessed 30 March 2022].
27. Wong PY, Lee HY, Chen YC, Zeng YT, Chern YR, Chen NT, et al. 2021. Using a land use regression model with machine learning to estimate ground level PM2.5. Environ Pollut 277 :116846, PMID: , 10.1016/j.envpol.2021.116846.33735646
28. Chen TH, Hsu YC, Zeng YT, Lung SCC, Su HJ, Chao HJ, et al. 2020. A hybrid kriging/land-use regression model with Asian culture-specific sources to assess NO2 spatial-temporal variations. Environ Pollut 259 :113875, PMID: , 10.1016/j.envpol.2019.113875.31918142
29. Alberti KGMM, Eckel RH, Grundy SM, Zimmet PZ, Cleeman JI, Donato KA, et al. 2009. Harmonizing the metabolic syndrome: a joint interim statement of the International Diabetes Federation Task Force on Epidemiology and Prevention; National Heart, Lung, and Blood Institute; American Heart Association; World Heart Federation; International Atherosclerosis Society; and International Association for the Study of Obesity. Circulation 120 (16 ):1640–1645, PMID: , 10.1161/CIRCULATIONAHA.109.192644.19805654
30. Wang ML. 2016. MJ Health Screening Equipment Use and Replacement Records. Technical Report No. MJHRF-TR-06. http://www.mjhrf.org/file/file/report/MJHRF-TR-06%20Screening%20Equipment.pdf [accessed 30 March 2022].
31. Troxel WM, Matthews KA, Gallo LC, Kuller LH. 2005. Marital quality and occurrence of the metabolic syndrome in women. Arch Intern Med 165 (9 ):1022–1027, PMID: , 10.1001/archinte.165.9.1022.15883241
32. Lian Y, Yuan Q, Wang G, Tang F. 2019. Association between sleep quality and metabolic syndrome: a systematic review and meta-analysis. Psychiatry Res 274 :66–74, PMID: , 10.1016/j.psychres.2019.01.096.30780064
33. Che T, Yan C, Tian D, Zhang X, Liu X, Wu Z. 2021. The association between sleep and metabolic syndrome: a systematic review and meta-analysis. Front Endocrinol (Lausanne) 12 :773646, PMID: , 10.3389/fendo.2021.773646.34867820
34. Kim JS, Chen Z, Alderete TL, Toledo-Corral C, Lurmann F, Berhane K, et al. 2019. Associations of air pollution, obesity and cardiometabolic health in young adults: the Meta-AIR study. Environ Int 133 (pt A ):105180, PMID: , 10.1016/j.envint.2019.105180.31622905
35. Voss S, Schneider A, Huth C, Wolf K, Markevych I, Schwettmann L, et al. 2021. Long-term exposure to air pollution, road traffic noise, residential greenness, and prevalent and incident metabolic syndrome: results from the population-based KORA F4/FF4 cohort in Augsburg, Germany. Environ Int 147 :106364, PMID: , 10.1016/j.envint.2020.106364.33421766
36. Mao S, Chen G, Liu F, Li N, Wang C, Liu Y, et al. 2020. Long-term effects of ambient air pollutants to blood lipids and dyslipidemias in a Chinese rural population. Environ Pollut 256 :113403, PMID: , 10.1016/j.envpol.2019.113403.31711721
37. Cao H, Li B, Liu K, Pan L, Cui Z, Zhao W, et al. 2021. Association of long-term exposure to ambient particulate pollution with stage 1 hypertension defined by the 2017 ACC/AHA Hypertension Guideline and cardiovascular disease: the CHCN-BTH cohort study. Environ Res 199 :111356, PMID: , 10.1016/j.envres.2021.111356.34048743
38. Du J, Shao B, Gao Y, Wei Z, Zhang Y, Li H, et al. 2021. Associations of long-term exposure to air pollution with blood pressure and homocysteine among adults in beijing, China: a cross-sectional study. Environ Res 197 :111202, PMID: , 10.1016/j.envres.2021.111202.33894236
39. Li N, Chen G, Liu F, Mao S, Liu Y, Liu S, et al. 2020. Associations between long-term exposure to air pollution and blood pressure and effect modifications by behavioral factors. Environ Res 182 :109109, PMID: , 10.1016/j.envres.2019.109109.32069739
40. Zhang Y, Li T, Ma R, Yin Z, Wang J, He MZ, et al. 2020. Long-term exposure to ambient fine particulate matter and fasting blood glucose level in a Chinese elderly cohort. Sci Total Environ 717 :137191, PMID: , 10.1016/j.scitotenv.2020.137191.32062280
41. Hou J, Liu X, Tu R, Dong X, Zhai Z, Mao Z, et al. 2020. Long-term exposure to ambient air pollution attenuated the association of physical activity with metabolic syndrome in rural Chinese adults: a cross-sectional study. Environ Int 136 :105459, PMID: , 10.1016/j.envint.2020.105459.31931348
42. Dekker FW, de Mutsert R, van Dijk PC, Zoccali C, Jager KJ. 2008. Survival analysis: time-dependent effects and time-varying risk factors. Kidney Int 74 (8 ):994–997, PMID: , 10.1038/ki.2008.328.18633346
43. Park SK, Auchincloss AH, O’Neill MS, Prineas R, Correa JC, Keeler J, et al. 2010. Particulate air pollution, metabolic syndrome, and heart rate variability: the Multi-Ethnic Study of Atherosclerosis (MESA). Environ Health Perspect 118 (10 ):1406–1411, PMID: , 10.1289/ehp.0901778.20529761
44. Evandt J, Oftedal B, Hjertager Krog N, Nafstad P, Schwarze PE, Marit Aasvang G. 2017. A population-based study on nighttime road traffic noise and insomnia. Sleep 40 (2 ), PMID: .28364487
45. Ministry of the Interior, R.O.C. (Taiwan). 2017. https://ws.moi.gov.tw/001/Upload/400/relfile/0/4405/48349492-6f8c-453ba9d1-4a8f0593b979/year_en.html [accessed 19 December 2022].
46. Ministry of Digital Affairs, R.O.C. (Taiwan). 2022. Population density at the township level of Taiwan in 2016. https://data.gov.tw/dataset/8410 [accessed 19 December 2022].
47. Chen Z, Salam MT, Toledo-Corral C, Watanabe RM, Xiang AH, Buchanan TA, et al. 2016. Ambient air pollutants have adverse effects on insulin and glucose homeostasis in Mexican Americans. Diabetes Care 39 (4 ):547–554, PMID: , 10.2337/dc15-1795.26868440
48. Clementi E, Talusan A, Vaidyanathan S, Veerappan A, Mikhail M, Ostrofsky D, et al. 2019. Metabolic syndrome and air pollution: a narrative review of their cardiopulmonary effects. Toxics 7 (1 ):6, PMID: , 10.3390/toxics7010006.30704059
49. Li H, Chen R, Cai J, Cui X, Huang N, Kan H. 2018. Short-term exposure to fine particulate air pollution and genome-wide DNA methylation: a randomized, double-blind, crossover trial. Environ Int 120 :130–136, PMID: , 10.1016/j.envint.2018.07.041.30081103
50. Brook RD, Urch B, Dvonch JT, Bard RL, Speck M, Keeler G, et al. 2009. Insights into the mechanisms and mediators of the effects of air pollution exposure on blood pressure and vascular function in healthy humans. Hypertension 54 (3 ):659–667, PMID: , 10.1161/HYPERTENSIONAHA.109.130237.19620518
51. Petit PC, Fine DH, Vásquez GB, Gamero L, Slaughter MS, Dasse KA. 2017. The pathophysiology of nitrogen dioxide during inhaled nitric oxide therapy. ASAIO J 63 (1 ):7–13, PMID: , 10.1097/MAT.0000000000000425.27556146
52. Salmon AB. 2012. Oxidative stress in the etiology of age-associated decline in glucose metabolism. Longev Healthspan 1 (1 ):7, PMID: , 10.1186/2046-2395-1-7.24764512
53. Teichert T, Vossoughi M, Vierkötter A, Sugiri D, Schikowski T, Schulte T, et al. 2013. Association between traffic-related air pollution, subclinical inflammation and impaired glucose metabolism: results from the SALIA study. PLoS One 8 (12 ):e83042, PMID: , 10.1371/journal.pone.0083042.24340078
54. Chuang YC. 2016. MJ Health data-cleaning procedure. MJ Health Research Foundation Technical report, MJHRF-TR-04. http://www.mjhrf.org/main/page/resource/en/#resource08 [accessed 30 March 2022].
55. de Keijzer C, Basagaña X, Tonne C, Valentín A, Alonso J, Antó JM, et al. 2019. Long-term exposure to greenspace and metabolic syndrome: a Whitehall II study. Environ Pollut 255 (pt 2 ):113231, PMID: , 10.1016/j.envpol.2019.113231.31561037
56. Twohig-Bennett C, Jones A. 2018. The health benefits of the great outdoors: a systematic review and meta-analysis of greenspace exposure and health outcomes. Environ Res 166 :628–637, PMID: , 10.1016/j.envres.2018.06.030.29982151
57. Yang BY, Liu KK, Markevych I, Knibbs LD, Bloom MS, Dharmage SC, et al. 2020. Association between residential greenness and metabolic syndrome in Chinese adults. Environ Int 135 :105388, PMID: , 10.1016/j.envint.2019.105388.31837524
58. Huang B, Xiao T, Grekousis G, Zhao H, He J, Dong G, et al. 2021. Greenness-air pollution-physical activity-hypertension association among middle-aged and older adults: evidence from urban and rural China. Environ Res 195 :110836, PMID: , 10.1016/j.envres.2021.110836.33549617
59. van Zon SKR, Amick BC III, de Jong T, Brouwer S, Bültmann U. 2020. Occupational distribution of metabolic syndrome prevalence and incidence differs by sex and is not explained by age and health behavior: results from 75 000 Dutch workers from 40 occupational groups. BMJ Open Diabetes Res Care 8 (1 ):e001436, PMID: , 10.1136/bmjdrc-2020-001436.32636219
|
PMC009xxxxxx/PMC9812022.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36598457
EHP10391
10.1289/EHP10391
Research
Long-Term Air Pollution, Genetic Susceptibility, and the Risk of Depression and Anxiety: A Prospective Study in the UK Biobank Cohort
https://orcid.org/0000-0001-6506-6084
Gao Xu 1
Jiang Meijie 1
https://orcid.org/0000-0001-9001-784X
Huang Ninghao 2
Guo Xinbiao 1
Huang Tao 2
1 Department of Occupational and Environmental Health Sciences, School of Public Health, Peking University, Beijing, China
2 Department of Epidemiology and Biostatistics, School of Public Health, Peking University, Beijing, China
Address correspondence to Xu Gao, Department of Occupational and Environmental Health Sciences, School of Public Health, Peking University, No. 38 Xueyuan Rd., Haidian District, Beijing, China 100191. Telephone: 86-136-2190-8907. Email: [email protected]
04 1 2023
1 2023
131 1 01700225 9 2021
10 11 2022
14 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Depression and anxiety are two mental disorders that are often comorbid. However, the associations of long-term air pollution exposure with depression and anxiety remain inconclusive.
Objective:
We conducted a cross-sectional and prospective study to examine the associations of ambient exposure to particulate matter (PM) with a diameter of ≤2.5μm (PM2.5), ≤10μm (PM10), and 2.5–10μm (PMcoarse), nitrogen oxides (NOx), and nitrogen dioxide (NO2) with the risk of depression and anxiety in the UK Biobank.
Methods:
This study included 398,241 participants from the UK Biobank, 128,456 of whom participated the 7-y online mental health survey. A total of 345,876 individuals were free of depression and anxiety at baseline; of those, 16,185 developed incident mental disorders during a median of 8.7 y of follow-up. Depression and anxiety were assessed using hospital admission records and mental health questionnaires. Associations of air pollution with prevalent and incident mental disorders were examined using logistic regression and Cox regression models, respectively.
Results:
Elevated levels of the five air pollutants were associated with higher odds of mental disorders at baseline. Levels of four pollutants but not PMcoarse were also associated with higher odds and risks of mental disorders during follow-up; specifically, hazard ratios [HR, 95% confidence interval (CI)] of an interquartile range increase in PM2.5, PM10, NOx, and NO2 for incident mental disorders were 1.03 (95% CI: 1.01, 1.05), 1.06 (95% CI: 1.04, 1.08), 1.03 (95% CI: 1.01, 1.05), and 1.06 (95% CI: 1.04, 1.09), respectively. An air pollution index reflecting combined effects of pollutants also demonstrated a positive association with the risk of mental disorders. HR (95% CI) of incident mental disorders were 1.11 (95% CI: 1.05, 1.18) in the highest quintile group in comparison with the lowest quintile of the air pollution index. We further observed that the associations between air pollution and mental disorders differed by a genetic risk score based on single nucleotide polymorphisms previously associated with genetic susceptibility to mental disorders in the UK Biobank cohort.
Discussion:
To our knowledge, this research is one of the largest cohort studies that demonstrates an association between mental health disorders and exposure to long-term air pollution, which could be further enhanced by genetic predisposition. https://doi.org/10.1289/EHP10391
Supplemental Material is available online (https://doi.org/10.1289/EHP10391).
The authors have no conflict of interest to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Depression and anxiety are common mental health disorders that often co-occur and can broadly impact the longevity by promoting aging-related conditions.1,2 With the rapidly increasing prevalence of depression and anxiety, as well as their unclear etiology, the task of identifying modifiable risk factors for mental disorders and developing relevant prevention strategies has drawn increasing attention.3,4 Recently, mounting numbers of epidemiological studies have examined the associations between exposure to ambient air pollutants and the risk of depression and anxiety.5–8 However, although the short-term impact of air pollution on mental disorders, especially depression, has been widely studied, results of previous population-based studies on the long-term impact of air pollution exposure remain controversial due to heterogeneities of study designs.5,6,8 Several studies were cross-sectional or crossover analyses with limited time of follow-up at the city level and focused on only particulate matter (PM) with a diameter of ≤2.5μm (PM2.5) and/or ≤10μm (PM10).8–10 Other studies implemented insurance data without personal assessment using questionnaires,11 which may underestimate the influence of air pollution on mental health because people with mild syndromes are not inclined to seek insurance services. A study by Newbury et al. employed the data of mental health service use as the surrogate for the prevalence of mental disorders and specifically examined the relapse of disorders, but similar underestimation of mild syndromes might exist.12 Therefore, it is worthwhile to study the associations of long-term exposure to ambient air pollution with depression and anxiety in a large independent study with a prospective design and comprehensive assessments of depression and anxiety.
Notably, previous studies addressing the associations between air pollution and the risk of depression or anxiety also mainly focused on the individual effect of each air pollutant.5–8 Considering that ambient air pollution is a mixture of gaseous pollutants and particles,13 exploring their synergistic effects of on mental disorders is critical to uncover the holistic nature of air pollution on mental health in real-world settings, which, to our knowledge, have not been assessed in a prospective cohort study. Additionally, mental health disorders have been commonly recognized to be influenced by genetic and environmental factors.14,15 Recently, evidence has suggested that the gene–environment interplay could influence the development of health disorders,16 but the knowledge on whether the genetic susceptibility could modify the association between the combined exposure to multiple air pollutants and mental disorders is scarce.
Hence, we investigated the associations of five major air pollutants, including PM2.5, PM10, PM with diameters of 2.5–10μm (PMcoarse), and nitrogen oxides (NOx and NO2), with the prevalence and incidence of depression and anxiety with a prospective design. This study was based on data of ∼0.4 million UK Biobank participants enrolled in a nationwide population-based cohort study in the UK with a median follow-up of 8.7 y. Leveraging the detailed records of air pollution and genetic variations of UK Biobank, we constructed an air pollution index to examine to what extent exposure to multiple air pollutants was associated with a greater burden of depression and/or anxiety and assessed the possible gene–air pollution interplay.
Methods
Study Design and Population
Study design of UK Biobank was reported in detail previously.17 Briefly, UK Biobank is an ongoing prospective study with 502,536 participants recruited in 2006–2010 at the age of 37–73 y old (baseline survey) with multiple follow-ups. At baseline survey, participants were asked to attend 22 examination centers located in the UK to provide information on their lifestyle and health, and their biological samples were collected. As shown in Figure S1, in this study, we included 398,241 participants with available data on mental health status, air pollution, and genetic variants at baseline to analyze the association between air pollution and prevalent mental disorders. Two subsets were subsequently used to conduct further analyses; accordingly, from 2016 to 2017, a subset with 128,456 participants reported their detailed mental health status via an online 7-y survey and were used to evaluate the prospective associations between baseline air pollution and the detailed syndromes of depression or anxiety at the 7-y survey. A second subset with 345,876 participants who were free of depression or anxiety at baseline was employed to assess the association between air pollution and the incidence of mental disorders diagnosed by the 7-y survey and/or hospital records (see details below). UK Biobank research has received the approval from the North West Multicenter Research Ethical Committee. Written informed consents were provided by all participants.
Assessment of Depression and Anxiety
The assessments of depression and anxiety were conducted using both linked hospital admission records (UK Biobank Data-Fields: 41202 and 41204) and mental health questionnaires at baseline and follow-up. Participants were considered as depression or anxiety positive if either record was positive. For hospital records, participants were classified as cases if they had either an ICD-9/10 primary or secondary diagnosis for depression (ICD-9: 311; ICD-10: F32–F33) and/or anxiety (ICD-9: 300; ICD-10: F40–F41). Mental disorder was defined as participants with depression and/or anxiety. The end date of follow-up was 31 December 2018. For participants who were free of mental disorders at baseline, especially for those included in both subset analyses (n=116,039), their follow-up time was estimated based on the dates of hospital records of incident mental disorder, the dates of the 7-y survey with a positive classification of depression/anxiety, or the date of censoring, whichever occurred first.
For questionnaire screening, at baseline, depression and anxiety symptoms were assessed only by Patient Health Questionnaire (PHQ)-4 questionnaire.18 Participants were required to rate, on a four-point Likert scale from 0 (not at all) to 3 (nearly every day), their response to four items: a) “frequency of depressed mood in the last 2 weeks” (UK Biobank Data-Field: 2050), b) “frequency of unenthusiasm/disinterest in the last 2 weeks” (UK Biobank Data-Field: 2060), c) “frequency of tenseness/restlessness in the last 2 weeks” (UK Biobank Data-Field: 2070), and d) “frequency of tiredness/lethargy in the last 2 weeks” (UK Biobank Data-Field: 2080). Total score ranged from 0 to 12, and a score of ≥6 was considered emotional disorder positive. A total score of ≥3 for items 1 and 2 was considered as positive for depression, and a total score of ≥3 for items 3 and 4 was considered as positive for anxiety based on the reported criteria.18
At the 7-y survey, mental health status was assessed using PHQ-9 (an updated PHQ-4)19 and Generalized Anxiety Disorder (GAD)-7 questionnaires20 between 2016 and 2017 with the same four-point Likert scale from 0 (not at all) to 3 (nearly every day). PHQ-9 consists of 9 items for thoughts and feelings in the last 2 wk: a) “recent thoughts of suicide or self-harm” (Suicidal ideation; UK Biobank Data-Field: 20513), b) “trouble falling or staying asleep, or sleeping too much” (Sleeping problems; UK Biobank Data-Field: 20517), c) “recent changes in speed/amount of moving or speaking” (Psychomotor changes; UK Biobank Data-Field: 20518), d) “recent feelings of inadequacy” (Feelings of inadequacy; UK Biobank Data-Field: 20507). e) “recent feelings of tiredness or low energy” (Fatigue; UK Biobank Data-Field: 20519), f) “recent feelings of depression” (Depressed mood; UK Biobank Data-Field: 20510), g) “recent trouble concentrating on things” (Cognitive problems; UK Biobank Data-Field: 20508), h) “recent poor appetite or overeating” (Appetite changes; UK Biobank Data-Field: 20511), and i) “recent lack of interest or pleasure in doing things” (Anhedonia; UK Biobank Data-Field: 20514). GAD-7 consists of seven items for thoughts and feelings in the last 2 wk: a) “recent inability to stop or control worrying” (Worrying control; UK Biobank Data-Field: 20509), b) “recent restlessness” (Restlessness; UK Biobank Data-Field: 20516), c) “recent trouble relaxing” (Lack of relaxation; UK Biobank Data-Field: 20515), d) “recent easy annoyance or irritability” (Irritability; UK Biobank Data-Field: 20505), e) “recent worrying too much about different things” (Generalized worrying; UK Biobank Data-Field: 20520), f) “recent feelings of foreboding” (Foreboding; UK Biobank Data-Field: 20512), and g) “recent feelings of nervousness or anxiety” (Anxiety feeling; UK Biobank Data-Field: 20506). Any item with a score of ≥1 was considered as positive for this symptom. A PHQ-9 or GAD-7 total score of ≥10 was considered as depression or anxiety symptoms positive according to the corresponding criteria,19,20 respectively.
Exposure Assessments
As previously described,21 the 1-y moving average concentrations of PM2.5, PMcoarse , PM10, NO2, and NOx of UK Biobank were calculated based on a Land Use Regression (LUR) model developed by the ESCAPE project. LUR models calculate the annual moving average concentrations of air pollutants using the predictor variables retrieved from the GIS variables, including land use, traffic, and topography by a 100m×100m resolution. Participants’ ambient air pollution concentrations were then assigned according to their residential coordinates in the 100m×100m grid cells. Previous reports using leave-one-out cross-validation demonstrated good model performance for PM2.5, PM10, NO2 and NOx (R2 of cross-validation=77%, 88%, 87%, and 88%, respectively) and a moderate performance for PMcoarse (cross-validation R2=57%) in the southeast England area, where a majority of the participants came from (London/Oxford). Therefore, participants living in northern England and Scotland (n=41,313; Figure S1) were eliminated from the PM analyses. Details of the ESCAPE LUR models have been described previously.22,23 The LUR estimates of PM were valid within a 400-km area from Greater London for participants living in this area. The 1-y moving average ambient concentrations of PM2.5 , PMcoarse , and NOx were collected in 2010 only, but because 1-y moving average concentrations of NO2 and PM10 were available for several years (2005, 2006, 2007, and 2010 for NO2 and 2007 and 2010 for PM10 ), their averaged values were included in our analysis.
Measurements of Covariates
To control for potential bias resulting from covariates that were reported to be related to mental health or air pollution,6,11,24,25 we included the following, reported at baseline: age (year; continuous), sex (male/female; categorical), body mass index (BMI, kg/m2; continuous), race (based on UK Biobank Data-Field 1657 “self-reported ethnic group,” categorized into White, Black, Asian, and Other; Other included those reported “White and Black mixed” and “other ethnic group” to the question), smoking status (current/former/never; categorical), healthy alcohol intake status (yes/no; categorical), healthy physical activity status (yes/no; categorical), years of education (≤10y and >10y; categorical), forced respiratory volume in the first second (FEV1, liter; continuous), forced vital capacity (FVC, liter; continuous), Townsend deprivation index (continuous), live in urban area (yes/no; categorical), and prevalent hypertension, coronary heart disease (CHD), and diabetes (yes/no; categorical). Height and weight were measured by nurses during the baseline assessment visit, and BMI was calculated by dividing weight (kilograms) by the square of height (meters). Healthy alcohol intake was defined as: <28g/day for male and <14g/day for female. Healthy physical activity status was defined as: ≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed (moderate + vigorous) activity. The activity was assessed using the Metabolic Equivalent Task (MET) minutes based on the short International Physical Activity Questionnaire (IPAQ).26 FEV1 and FVC were measured at baseline via spirometry. Townsend deprivation index ranged from −6.3 to 10.2 and was constructed based on the percentages of four key variables in the UK Biobank according to previously reported algorithm,27 unemployment, overcrowded household, non–car ownership, and non–home ownership. The percentages for each area were based on postcodes and census data at the year of baseline visit, with each participant assigned percentages corresponding to the postcode of their home dwelling but not based on individual-level factors. The percentages of unemployment and overcrowded household were normalized by ln (percentage value+1) because the two were skewed. The four percentages were then z-scored and summarized as the Townsend deprivation index. A higher index indicates a higher level of deprivation. The history of hypertension, CHD, and diabetes was based on self-reported information and medical records. Because all covariates were missing <2% (Table 1), we imputed continuous variables with sex-specific mean values and implemented a missing indicator approach for the categorical variables.28
Table 1 Characteristics of 398,241 participants from the UK Biobank at baseline and the subgroups of participants with 7-y survey data and participants free of mental disorders at baseline.
Characteristica All
(n=398,241) Participants with 7-y survey data
(n=128,456) Participants free of mental disorders at baseline
(n=345,876)
Age (y) 56.55 (8.07) 55.93 (7.74) 56.83 (8.04)
BMI (kg/m2) 27.38 (4.76) 26.77 (4.54) 27.22 (4.58)
FEV1 (L) 2.83 (0.80) 2.92 (0.77) 2.85 (0.80)
FVC (L) 3.74 (1.06) 3.84 (1.03) 3.76 (1.06)
Sex
Male 184,198 (46.3%) 56,692 (44.1%) 162,869 (47.1%)
Female 214,043 (53.7%) 71,764 (55.9%) 183,007 (52.9%)
Race
White 379,373 (95.3%) 124,925 (97.3%) 331,922 (96.0%)
Black 5,421 (1.3%) 1,345 (1.0%) 4,092 (1.2%)
Asian 7,641 (1.9%) 1,296 (1.0%) 5,527 (1.6%)
Other 5,806 (1.5%) 890 (0.7%) 4,335 (1.2%)
Smoking status
Current smoker 40,282 (10.1%) 9,020 (7.0%) 31,625 (9.2%)
Former smoker 139,671 (35.2%) 45,511 (35.5%) 122,804 (35.6%)
Never smoker 217,167 (54.7%) 73,725 (57.5%) 190,501 (55.2%)
Missing 1,121 200 946
Healthy alcohol intakeb
Yes 198,922 (50.0%) 68,602 (53.4%) 176,951 (51.2%)
No 199,064 (50.0%) 59,818 (46.6%) 168,759 (48.8%)
Missing 255 36 166
Healthy physical activityc
Yes 280,132 (71.5%) 92,811 (72.6%) 249,022 (73.0%)
No 111,686 (28.5%) 34,955 (27.4%) 92,140 (27.0%)
Missing 6,423 690 4,714
Years of education
>10y 263,065 (66.1%) 101,925 (79.4%) 232,515 (67.2%)
≤10y 135,176 (33.9%) 26,531 (20.6%) 113,361 (32.8%)
Townsend deprivation indexd −1.43 (2.99) −1.75 (2.79) −1.58 (2.90)
Live in an urban area
Yes 336,257 (85.2%) 106,236 (83.5%) 290,304 (84.7%)
No 58,300 (14.8%) 20,986 (16.5%) 52,502 (15.3%)
Missing 3,684 1,234 3,070
Hypertensione
Yes 220,764 (55.4%) 65,925 (51.3%) 193,121 (55.8%)
No 177,477 (44.6%) 62,531 (48.7%) 152,755 (44.2%)
Coronary heart diseasef
Yes 22,136 (5.6%) 4,519 (3.5%) 17,348 (5.0%)
No 376,105 (94.4%) 123,937 (96.5%) 328,528 (95.0%)
Diabetesg
Yes 20,288 (5.1%) 4,261 (3.3%) 15,946 (4.6%)
No 377,953 (94.9%) 124,195 (96.7) 329,930 (95.4%)
Mental health status at baseline
PHQ-4 score 1.60 (2.10) 1.36 (1.84) 5.58 (2.49)
Mental disorders
Yes 52,365 (13.2%) 12,417 (9.7%) —
No 345,876 (86.8%) 116,039 (90.3%) —
Depression
Yes 25,715 (6.5%) 5,368 (4.2%) —
No 372,526 (93.5%) 123,088 (95.8%) —
Anxiety
Yes 41,783 (10.5%) 10,032 (7.8%) —
No 356,458 (89.5%) 118,424 (92.2%) —
Mental health status at 7-y survey
PHQ-9 score — 2.70 (3.66) —
GAD-7 score — 2.07 (3.33) —
Mental disorders
Yes — 9,443 (7.4%) —
No — 119,013 (92.6) —
Depression
Yes — 7,100 (5.5%) —
No — 121,356 (94.5%) —
Anxiety
Yes — 5,287 (4.1%) —
No — 123,169 (95.9%) —
Note: BMI, Body mass index; FEV1 , forced respiratory volume in the first second; FVC, forced vital capacity; GAD-7, General Anxiety Disorder-7 questionnaire; min, minutes; PHQ-4, Patient Health Questionnaire-4 questionnaire; PHQ-9, Patient Health Questionnaire-9 questionnaire; —, no data.
a Mean values (standard deviation) for continuous variables and n (%) for categorical variables.
b Healthy alcohol intake: male: <28g/day; female: <14g/day.
c Healthy physical activity: ≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed (moderate + vigorous) activity.
d Data missing in 453 (0.11%) participants. This index is composite score based on four key variables: unemployment, overcrowded household, non–car ownership, and non–home ownership (detail see methods).
e Hypertension diagnosed by doctor according to self-reported information and/or medical records.
f Coronary heart disease diagnosed by doctor according to self-reported information and/or medical records.
g Diabetes diagnosed by doctor according to self-reported information and/or medical records.
Genetic Risk Scores for Mental Disorders
Details of the genotyping, data imputation, and quality control in UK Biobank have been previously reported.17,29 Briefly, ∼10% of participants were genotyped using the Applied Biosystems UK BiLEVE Axiom Array by Affymetrix (807,411 sites), with the remaining participants being genotyped using the Applied Biosystems UK Biobank Axiom Array (825,927 sites), both of which were specifically designed for the UK Biobank with 95% shared sites. Phasing and imputation of the remaining 5% unique single-nucleotide polymorphisms (SNPs) were conducted using SHAPEIT3 and IMPUTE2 with both the merged UK10K and 1000 Genomes Phase 3 reference panel and the Haplotype Reference Consortium (HRC) reference panel (i.e., array imputation). SNPs after the array imputation were considered for building a genetic risk score (GRS). We directly used the imputed genetic data without further quality control.29 Based on the availability of imputed SNPs, we created a GRS for mental disorders using 37/44 depression-related SNPs and 9 out of 10 unique anxiety-related SNPs based on the two largest genome-wide association studies (GWASs).30,31 Each SNP was recoded as 0, 1, or 2 according to the number of risk alleles, and missing SNP values of individuals were imputed by corresponding mean values (i.e., individual imputation). To avoid the bias from nonlinearity, this GRS based on multiple genes was calculated and weighted by summing up after multiplying the logarithmic odds ratios (ORs) of each included SNP reported by the selected GWASs on depression or anxiety with the number of risk alleles and then divided by the total effect size. The GRS ranged from 0.85 to 2.70. The detailed SNPs are reported in Table 2.
Table 2 Characteristics of genetic variants associated with depression and anxiety in UK Biobank.
Trait SNP CHR Position p-Value Allele 1 Allele 2 OR SE Frequency
Depression rs1432639 1 72813218 4.6×10–15 A C 1.04 0.005 0.63
rs12129573 1 73768366 4.0×10–12 A C 1.04 0.005 0.37
rs2389016 1 80799329 1.0×10–8 T C 1.03 0.0053 0.28
rs4261101 1 90796053 1.0×10–8 A G 0.97 0.005 0.37
rs9427672 1 197754741 3.1×10–8 A G 0.97 0.0058 0.24
rs11682175 2 57987593 4.7×10–9 T C 0.97 0.0048 0.52
rs1226412 2 157111313 2.4×10–8 T C 1.03 0.0059 0.79
rs7430565 3 158107180 2.9×10–9 A G 0.97 0.0048 0.58
rs34215985 4 42047778 3.1×10–9 C G 0.96 0.0063 0.24
rs11135349 5 164523472 1.1×10–9 A C 0.97 0.0048 0.48
rs4869056 5 166992078 6.8×10–9 A G 0.97 0.005 0.63
rs9402472 6 99566521 2.8×10–8 A G 1.03 0.0059 0.24
rs10950398 7 12264871 2.6×10–8 A G 1.03 0.0049 0.41
rs12666117 7 109105611 1.4×10–8 A G 1.03 0.0048 0.47
rs1354115 9 2983774 2.4×10–8 A C 1.03 0.0049 0.62
rs10959913 9 11544964 5.1×10–9 T G 1.03 0.0057 0.76
rs7856424 9 119733595 8.5×10–9 T C 0.97 0.0053 0.29
rs7029033 9 126682068 2.7×10–8 T C 1.05 0.0093 0.07
rs61867293 10 106563924 7.0×10–10 T C 0.96 0.0061 0.2
rs1806153 11 31850105 1.2×10–9 T G 1.04 0.0059 0.22
rs4074723 12 23947737 3.1×10–8 A C 0.97 0.0049 0.41
rs4143229 13 44327799 2.5×10–8 A C 0.95 0.0091 0.92
rs12552 13 53625781 6.1×10–19 A G 1.04 0.0048 0.44
rs4904738 14 42179732 2.6×10–9 T C 0.97 0.0049 0.57
rs915057 14 64686207 7.6×10–10 A G 0.97 0.0049 0.42
rs10149470 14 104017953 3.1×10–9 A G 0.97 0.0049 0.49
rs8025231 15 37648402 2.4×10–12 A C 0.97 0.0048 0.57
rs8063603 16 6310645 6.9×10–9 A G 0.97 0.0053 0.65
rs7198928 16 7666402 1.0×10–8 T C 1.03 0.005 0.62
rs7200826 16 13066833 2.4×10–8 T C 1.03 0.0055 0.25
rs11643192 16 72214276 3.4×10–8 A C 1.03 0.0049 0.41
rs17727765 17 27576962 8.5×10–9 T C 0.95 0.0088 0.92
rs62099069 18 36883737 1.3×10–8 A T 0.97 0.0049 0.42
rs11663393 18 50614732 1.6×10–8 A G 1.03 0.0049 0.45
rs1833288 18 52517906 2.6×10–8 A G 1.03 0.0054 0.72
rs12958048 18 53101598 3.6×10–11 A G 1.03 0.0051 0.33
rs5758265 22 41617897 7.6×10–9 A G 1.03 0.0054 0.28
Anxiety rs79928194 2 233649290 1.26×10−6 T C 0.862 0.0515 0.905
rs342422 5 83470986 1.28×10−6 A G 0.92 0.036 0.53
rs2451828 5 7748796 7.37×10−7 T C 1.34 0.156 0.019
rs6462203 7 3676002 1.09×10−7 A C 0.901 0.0345 0.265
rs16916239 8 87643741 8.96×10−7 A G 0.903 0.037 0.783
rs113209956 9 2511193 6.36×10−8 T C 0.828 0.057 0.085
rs1458103 11 81047274 6.19×10−8 A C 0.898 0.0345 0.741
rs11855560 15 41024303 6.96×10−7 T C 1.089 0.0365 0.469
rs6030245 20 41070559 5.06×10−7 T C 1.12 0.049 0.795
Note: The 37 risk alleles of depression were selected based on Wray et al.’s study,30 and the nine risk alleles of anxiety were selected based on Meier et al.’s study.31 The ORs, p-values, and frequencies are based on selected GWASs. CHR, chromosome; GWAS, genome-wide association study; OR, odds ratio; SNP, single-nucleotide polymorphism.
Construction of Air Pollution Index
As a post hoc analysis, to account for the combined effects of multiple air pollutants on mental disorders, we created an air pollution index by excluding PMcoarse due to its weak associations with mental disorders using a modified algorithm reported by a previous study21 (for details, see the “Results” section). This index is a continuous weighted score by adding concentrations of the air pollutants, weighted by the effect estimates (β coefficients) on the odds of mental disorders at baseline or the risk of incident mental disorders at the follow-up in the crude model, which adjusted for age and sex only. The equation of the index was: Air pollution index=(β[PM2.5] ×PM2.5 + β[PM10] × PM10 + β[NO2]× NO2 + β[NOx] × NOx)/(β[PM2.5] + β[PM10] + β[NO2]+β[NOx]) × 4, and was also categorized into five groups by quintiles.
To avoid upweighting the positive associations retrieved from this study, because the coefficients for the construction of the index were from the same cohort, we validated it with two approaches. First, we conducted a 10-fold cross-validation analysis that was conducted in a previous air pollution-related study based on the UK Biobank.32,33 UK Biobank data were randomly divided into 10 batches. One of the batches was testing data, and the other nine were the training data in each run. Logistic and Cox regression models were applied to obtain regression coefficients to build the air pollution index in the training data, and the constructed index was then tested in the testing data. After being repeated 10 times, the ORs and HRs were then polled using a fixed-effect meta-analysis to obtain the comprehensive ORs and HRs.34 Similar estimates suggested that this air pollution index was robust in the present study (Table S1). Furthermore, we generated a quantile score without taking the prior weights of each air pollutant into consideration. Ambient concentration within the fourth quartile of each air pollutant except PMcorase was defined as “extreme” exposure to corresponding pollutants. This score was then determined as the sum of “extreme” exposed air pollutants. This score and the air pollution index were highly correlated with each other (Table S2), suggesting that the bias from up-weighting the positive associations in the construction of air pollution index was quite limited.
Statistical Analyses
Sociodemographic and lifestyle factors and mental health outcomes for all participants, participants with the mental health survey data at the 7-y survey, and participants who were free of depression or anxiety were summarized using descriptive statistics.
We first examined the associations of the five air pollutants with the total PHQ-4 score using mixed-effect linear regression models and the odds of mental disorders, depression, and anxiety at baseline using logistic regression models to evaluate the cross-sectional association between air pollution and mental disorders. We adjusted for age, sex, BMI, race, smoking status, healthy alcohol intake status, healthy physical activity status, years of education, FEV1 , FVC, Townsend deprivation index, residence in an urban area, and prevalent hypertension, CHD, and diabetes, and we controlled for the examination center as a random effect in the model to account for the potential clustering bias from health examinations. Yielded estimates were reported as changes per one interquartile range (IQR) increase in the ambient concentration of each air pollutant. Two forms of air pollution index for baseline analysis were built accordingly, and the corresponding dose–response curves of air pollutants and the index with PHQ-4 score were further assessed by restricted cubic spline regression models controlling for previously described covariates. The 5th, 50th, and 95th percentiles of each air pollutant and air pollution index were selected as knots.
Then, in the subset with 7-y survey data, we examined the prospective associations between the baseline air pollution levels and the odds of mental disorders and corresponding symptoms of each disorder (logistic regression) and the scores of PHQ-9 and GAD-7 (linear regression) at the 7-y survey. Models adjusted for covariates described in the previous section and the status of mental disorders (yes/no) at baseline.
Furthermore, the Cox proportional hazards model was used to assess the prospective associations between baseline air pollution levels and the incidence of mental disorders during the follow-up among participants free of depression and anxiety at baseline. Corresponding air pollution index for the joint association was constructed accordingly, and its dose–response relationships with the risk of incident mental disorders, depression, and anxiety were also assessed by restricted cubic spline regression. The 5th, 50th, and 95th percentiles of the air pollution index were selected as knots.
Finally, to evaluate whether the genetic predisposition of mental disorders may modify the association of air pollution index with the odds of mental disorders at baseline and the risk of mental disorders at the follow-up, we tested the gene–air pollution interaction by employing the interaction terms of the air pollution index with the GRS of mental disorders in the models for the prevalence and incidences of depression and anxiety. In the case that interaction effect did not meet criteria for statistical significance, we generated a category variable based on the quintiles of air pollution index, and dichotomized GRS (by median) was further generated to show the joint association of both factors with mental disorders. For all primary analyses, we used the E-values and the lower confidence interval to indicate the robustness to unmeasured confounding.35 This E-value could test the robustness of an association between exposure and outcome and to evaluate evidence for causation. The unmeasured confounders needed to be associated with both air pollution and mental disorders with a risk ratio of the E-value to fully explain the observed associations between air pollutants and mental disorders.
Additionally, three sensitivity analyses were performed to validate the robustness of our primary findings. We first tested the estimates of air pollutants in relation to incident mental disorders in a model adjusting for PM2.5, PM10, NOx, and NO2 to examine the copollutant cofounding. Then, we examined the associations between air pollutants and the concurrent incidence of depression and anxiety. Last, we conducted another analysis using the primary model in participants who self-reported at the baseline survey that they had been living at the same baseline address for more than 5 y to evaluate whether the mobilization may affect the primary findings.
SAS version 9.4 TS1M7 (SAS Institute Inc.) was used to clean data and conduct analyses. A two-sided p-value <0.05 was considered statistically significant.
Results
Participants’ Characteristics and Distributions of Air Pollutants
Characteristics of the study population and the subset of participants free of mental disorders at baseline were similar with respect to most variables but slightly different in the subset of people with 7-y survey data (Table 1). For the whole population, participants’ age [mean±standard deviation (SD)] was 56.6±8.1 y and ∼95% (n=379,373) were White. About 35% (n=217,167) and 55% (n=139,671) of participants were former and never smokers, respectively. A majority of participants had healthy physical activity (70.4%, n=280,132), >10y of education (66.1%, n=263,065), and lived in an urban area (84.4%, n=336,257). Half reported a healthy daily intake of alcohol (n=198,922). About 55% (n=220,764), 6% (n=22,136), and 5% (n=20,288) had prevalent hypertension, CHD, or diabetes diagnosed by doctors, respectively, according to self-reported information and medical records. The average PHQ-4 score was 1.60±2.10, and about 13% (n=52,365), 7% (n=25,715), and 11% (n=41,783) were with prevalent mental disorders, depression, and anxiety, respectively, according to the PHQ-4 score definition or hospital records. The subset with 7-y survey data had a higher proportion of people with >10y of education (79.4%, n=101,925) than did the total cohort or subset free of mental disorders at baseline; about 10% (n=12,417), 4% (n=5,368), and 8% (n=10,032) of the participants in this subgroup had prevalent mental disorders, depression, and anxiety, respectively. The 345,876 participants that were free of mental disorders at baseline were followed for a median time of 8.7 y, and 16,185 of them (4.68%) developed mental disorders (ndepression=10,854; nanxiety=8,257).
Average concentrations of air pollutants at baseline for the total population were 9.97±1.06 μg/m3 (IQR=1.27) for PM2.5, 6.42±0.90 μg/m3 (IQR=0.79) for PMcoarse, 19.24±2.02 μg/m3 (IQR=2.33) for PM10, 43.68±15.54 μg/m3 (IQR=16.53) for NOx, and 29.01±9.17 μg/m3 (IQR=10.87) for NO2. Ambient concentrations of all air pollutants were highly correlated with each other (Table 3).
Table 3 Distributions and correlation matrix of the ambient concentrations of five air pollutants for 398,241 participants at baseline (Pearson correlation).
Pearson correlation coefficients
Air pollutants Mean (SD) Min Max 25th 75th IQR PM2.5 PMcoarse PM10 NO2 NOx
PM2.5 (μg/m³) 9.97 (1.06) 8.17 21.31 9.27 10.54 1.27 1 — — — —
PMcoarse (μg/m³) 6.42 (0.90) 5.57 12.82 5.84 6.63 0.79 0.216 1 — — —
PM10 (μg/m³) 19.24 (2.02) 5.92 30.08 18.03 20.36 2.33 0.623 0.513 1 — —
NOx (μg/m³) 43.68 (15.54) 19.74 265.94 33.90 50.43 16.53 0.849 0.177 0.626 1 —
NO2 (μg/m³) 29.01 (9.17) 8.86 125.13 22.75 33.62 10.87 0.736 0.233 0.779 0.747 1
Note: —, no data; IQR, interquartile range; max, maximum; min, minimum; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with diameter of≤2.5 μm; PMcoarse, particulate matter with diameters of 2.5–10μm; PM10, particulate matter with diameter of≤10μm; SD, standard deviation.
Associations of Air Pollution with the Odds of Mental Disorders at Baseline and 7-y Survey
As shown in Table 4, after adjusting for potential covariates, the ambient levels of four air pollutants except PMcoarse showed strong associations with the PHQ-4 score and the odds of mental disorders. Particularly, an IQR increment of PM2.5 level was associated with a 0.051-unit [standard error (SE)=0.005] increase in PHQ-4 score, and with the ORs (95% CI) of 1.07 (95% CI: 1.06, 1.09), 1.08 (95% CI: 1.07, 1.10), and 1.07 (95% CI: 1.05, 1.08) for mental disorders, depression, and anxiety, respectively. To further estimate the combined associations of the five air pollutants with mental health at baseline, we constructed an air pollution index based on the coefficients of the five pollutants for the odds of mental disorders at baseline retrieved from a crude model adjusting for sex and age only to avoid overadjustment bias in the primary analysis models that controlled for all potential covariates. Both continuous and categorical forms of the index were associated with a higher PHQ-4 score and odds of mental disorders in the fully adjusted model. For instance, the ORs of mental disorders were 1.04 (95% CI: 1.00, 1.08), 1.11 (95% CI: 1.07, 1.14), 1.15 (95% CI: 1.11, 1.19), and 1.16 (95% CI: 1.12, 1.20) for second–fifth quintile groups, respectively, in comparison with the first quintile of the air pollution index (Table 4). Except PMcoarse, the other four air pollutants and corresponding air pollution index demonstrated monotonic increasing dose–response relationships with the PHQ-4 score (Figure 1).
Table 4 Associations of air pollutants (per IQR increase) and air pollution index with the PHQ-4 Score and odds of mental disorders for 398,241 participants at baseline.
PHQ-4 Score Mental disordersa Depression Anxiety
Air pollutants Coefficients (SE) p-Value E-value (CI) ncase/ntotal OR (95% CI) p-Value E-Value (CI) ncase/ntotal OR (95% CI) p-value E-value (CI) Ncase/Ntotal Odds ratio (95% CI) p-value E-value (CI)
PM2.5 0.051 (0.005) <0.0001 1.17 (1.15) 52,365/398,241 1.07 (1.06, 1.09) <0.0001 1.34 (1.31) 25,715/398,241 1.08 (1.07, 1.10) <0.0001 1.37 (1.34) 41,783/398,241 1.07 (1.05–1.08) <0.0001 1.34 (1.28)
PMcoarse 0.008 (0.003) <0.0001 1.06 (1.03) 52,365/398,241 1.01 (1.00, 1.02) 0.18 1.11 (1.00) 25,715/398,241 1.02 (1.00, 1.03) 0.018 1.16 (1.00) 41,783/398,241 1.00 (0.99–1.01) 0.56 1.00 (1.00)
PM10 0.034 (0.004) <0.0001 1.14 (1.12) 52,365/398,241 1.03 (1.01, 1.04) <0.0001 1.21 (1.11) 25,715/398,241 1.05 (1.03, 1.07) <0.0001 1.28 (1.21) 41,783/398,241 1.02 (1.00–1.03) 0.010 1.16 (1.00)
NOx 0.046 (0.004) <0.0001 1.16 (1.15) 52,365/398,241 1.05 (1.04, 1.06) <0.0001 1.28 (1.24) 25,715/398,241 1.06 (1.05, 1.08) <0.0001 1.31 (1.28) 41,783/398,241 1.04 (1.03–1.06) <0.0001 1.24 (1.21)
NO2 0.044 (0.004) <0.0001 1.16 (1.14) 52,365/398,241 1.05 (1.03, 1.06) <0.0001 1.28 (1.21) 25,715/398,241 1.08 (1.06, 1.09) <0.0001 1.37 (1.31) 41,783/398,241 1.03 (1.02–1.05) <0.0001 1.21 (1.16)
Air pollution indexb Coefficients (SE) p-Value E-value (CI) ncase/ntotal OR (95% CI) p-Value E-value (CI) ncase/ntotal OR (95% CI) p-value E-value (CI) Ncase/Ntotal Odds ratio (95% CI) p-value E-value (CI)
Continuous 0.046 (0.004) <0.0001 1.16 (1.15) 52,365/398,241 1.05 (1.04, 1.06) <0.0001 1.28 (1.24) 25,715/398,241 1.07 (1.06, 1.09) <0.0001 1.34 (1.31) 41,783/398,241 1.05 (1.04–1.06) <0.0001 1.28 (1.24)
Q1 Ref — — 8,098/79,649 Ref — — 3,581/79,649 Ref — — 6,492/79,649 Ref — —
Q2 0.024 (0.010) 0.024 1.10 (1.05) 9,188/79,647 1.04 (1.00, 1.08) 0.034 1.24 (1.00) 4,221/79,647 1.04 (0.99, 1.09) 0.12 1.24 (1.00) 7,392/79,647 1.04 (1.00–1.08) 0.042 1.24 (1.00)
Q3 0.067 (0.011) <0.0001 1.20 (1.16) 10,335/79,648 1.11 (1.07, 1.14) <0.0001 1.46 (1.34) 4,981/79,648 1.14 (1.09, 1.20) <0.0001 1.54 (1.40) 8,253/79,648 1.09 (1.05–1.13) <0.0001 1.40 (1.28)
Q4 0.089 (0.011) <0.0001 1.24 (1.20) 11,584/79,649 1.15 (1.11, 1.19) <0.0001 1.57 (1.46) 5,829/79,649 1.19 (1.13, 1.25) <0.0001 1.67 (1.51) 9,249/79,649 1.13 (1.09–1.18) <0.0001 1.61 (1.40)
Q5 0.127 (0.011) <0.0001 1.30 (1.27) 13,160/79,648 1.16 (1.12, 1.20) <0.0001 1.59 (1.49) 7,103/79,648 1.24 (1.18, 1.30) <0.0001 1.79 (1.64) 10,397/79,648 1.14 (1.10–1.19) <0.0001 1.54 (1.43)
Note: Associations of air pollution with PHQ-4 score were tested with mixed-effect linear regression models and associations with the odds of mental disorders, depression, and anxiety at baseline were tested with logistic regression models. Model adjusted for age, sex, BMI, race (White, Black, Asian, and other), smoking status (current/former/never), healthy alcohol intake status (male: <28g/day; female: <14g/day), healthy physical activity status [≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/week mixed (moderate + vigorous) activity], years of education (<10y), FEV1, FVC, Townsend deprivation index, live in urban area (yes/no), and prevalent hypertension, CHD, and diabetes (yes/no). The examination center was controlled for as a random effect. Estimates were reported by one IQR increase in each air pollutant. IQRs of PM2.5=1.27μg/m³, PMcoarse=0.79μg/m³, PM10=2.33μg/m³, NOx=16.53μg/m³, NO2=10.87μg/m³. —, no data; BMI, body mass index; CHD, coronary heart disease; CI, confidence interval; FEV1 FVC, forced vital capacity; IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; OR, odds ratio; PHQ-4, Patient Health Questionnaire-4 questionnaire; PM2.5, particulate matter with diameters ≤2.5μm; PMcoarse, particulate matter with diameters 2.5–10μm; PM10, particulate matter diameter of ≤10μm; Q1–Q5, first to fifth quintiles; Ref, reference; SE, standard error.
a Mental disorders, with depression and/or anxiety, based on PHQ-4 questionnaires and hospital records.
b To estimate the joint effect of multiple air pollutants on each outcome, an air pollution index was estimated using the coefficients of PM2.5, PMcoarse, PM10, NOx, and NO2 for the odds of mental disorders retrieved from the crude logistic model which adjusted for age and sex only.
Figure 1. Graphs of the best fitting models for relationships of air pollutants and air pollution index with PHQ-4 score for 398,241 participants at baseline. Solid line: Point estimation; Dash line: Confidence limits; Dots: Knots (5th, 50th, and 95th percentiles). The restricted cubic spline regression model adjusted for age, sex, BMI, race (White, Black, Asian, and other), smoking status (current/former/never), healthy alcohol intake status (male: <28g/day; female: <14g/day), healthy physical activity status [≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed (moderate + vigorous) activity], years of education (<10 y), FEV1, FVC, Townsend deprivation index, live in urban area (yes/no), and prevalent hypertension, CHD, and diabetes (yes/no). Point estimates and corresponding confidence intervals were shown in Table S10. Note: BMI, body mass index; CHD, coronary heart disease; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; min, minutes; NOx, nitrogen oxides; NO2, nitrogen dioxide; PHQ-4, Patient Health Questionnaire-4 questionnaire; PM2.5, particulate matter with diameter of≤2.5μm; PMcoarse, particulate matter with diameters of 2.5–10μm; PM10, particulate matter with diameter of≤10μm.
Figure 1 is a set of six line graphs, plotting change in patient health questionnaire-4 score, ranging from 0.0 to 1.2 in increments of 0.2; 0.0 to 1.2 in increments of 0.2; 0.0 to 1.2 in increments of 0.2; 0.0 to 1.2 in increments of 0.2; 0.0 to 1.2 in increments of 0.2; and negative 0.3 to 0.3 in increments of 0.1 (y-axis) across particulate matter begin subscript 2.5 end subscript (microgram per meter cubed), ranging from 0 to 18 in increments of 2; particulate matter begin subscript coarse end subscript (microgram per meter cubed), ranging from 0 to 18 in increments of 2; particulate matter begin subscript 10 end subscript (microgram per meter cubed), ranging from 0 to 40 in increments of 5; Nitrogen oxide (microgram per meter cubed), ranging from 0 to 90 in increments of 10; Nitrogen dioxide (microgram per meter cubed), ranging from 0 to 60 in increments of 10; and Air pollution index, ranging as negative 3 standard deviation, negative 2 standard deviation, negative 1 standard deviation, mean, 1 standard deviation, 2 standard deviation, and 3 standard deviation (x-axis) for knots, including fifth, fiftieth, and ninety-fifth percentiles.
A subset of 128,456 participants took the 7-y online mental health survey; therefore, we examined the prospective associations of ambient air pollution with the odds of mental disorders and symptoms of depression and anxiety at the 7-y survey. Figure 2 (Table S3) shows that baseline PM2.5, PM10, NOx, and NO2 were positively associated with most depression and anxiety symptoms, odds of mental disorders, and the scores (total, PHQ-9, and GAD-7 scores) of mental health status at the 7-y survey. The baseline concentration of PMcoarse was not associated with any symptoms or scores. This finding suggests that long-term exposure to PM2.5, PM10, NOx , and NO2 was associated with the prevalence of mental disorders.
Figure 2. Prospective associations of baseline levels of air pollutants with odds of mental disorders and mental health scores at 7-y survey for the 128,456 participants with available data. Dots: Point estimate; Error bar: 95% confidence limits; Dash line: Reference line. Upper part is the odds ratios of logistic regression; lower part is the coefficients of linear regression. Dots and error bars colored in black are statistically significant, otherwise are colored in gray. Associations of air pollution with mental health scores were tested with mixed-effect linear regression models and associations with the odds of mental disorders, depression, and anxiety at 7-y survey were tested with logistic regression models. Point estimates and corresponding confidence intervals were shown in Table S3. Models adjusted for age, sex, BMI, race (White, Black, Asian, and other), smoking status (current/former/never), healthy alcohol intake status (male: <28g/day; female: <14g/day), healthy physical activity status (≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed [(moderate + vigorous) activity], years of education (<10y), FEV1, FVC, Townsend deprivation index, live in urban area (yes/no), and prevalent hypertension, CHD, diabetes (yes/no), and PHQ-4 score at baseline. The examination center was controlled for as a random effect. Note: BMI, body mass index; CHD, coronary heart disease; CI, confidence interval; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; min, minutes; GAD-7, General Anxiety Disorder-7; min, minutes; NOx, nitrogen oxides; NO2, nitrogen dioxide; PHQ-4, Patient Health Questionnaire-4 questionnaire; PHQ-9, Patient Health Questionnaire-9 questionnaire; PM2.5, particulate matter with diameter of≤2.5μm; PMcoarse, particulate matter with diameters of 2.5–10μm; PM10, particulate matter with diameter of≤10μm.
Figure 2 is a set of two forest plots. On the top, the forest plot, plotting Odds of anxiety, odds of depression, odds of mental disorders, anxiety feeling, foreboding, generalized worrying, irritability, lack of relaxation, restlessness, worrying control, anhedonia, appetite changes, cognitive problems, depressed mood, fatigue, feelings of inadequacy, psychomotor changes, sleeping problems, and suicidal ideation (left y-axis) and mental disorders, anxiety symptoms, depression symptoms (right y-axis) across odds ratio (95 percent confidence intervals), ranging from 0.95 to 1.10 in increments of 0.05 (x-axis) for particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript coarse end subscript, particulate matter begin subscript 10 end subscript, nitrogen oxide, and nitrogen dioxide. At the bottom, a forest plot, plotting generalized anxiety disorder-7 score (anxiety), patient health questionnaire-9 score (depression), and total score (mental) (left y-axis) and mental health scores (right y-axis) across Regression coefficients, ranging from 0.0 to 1.5 in increments of 0.5; 0.0 to 0.3 in increments of 0.1; 0.0 to 0.6 in increments of 0.2; 0.000 to 0.100 in increments of 0.025; and 0.00 to 0.15 in increments of 0.05 (x-axis) for particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript coarse end subscript, particulate matter begin subscript 10 end subscript, nitrogen oxide, and nitrogen dioxide.
Associations of Air Pollution with the Incidence of Mental Disorders
The fully adjusted Cox regression model showed the associations of baseline air pollution levels with the incidence of mental disorders for the 345,876 participants free of mental disorders at baseline during a median follow-up of 8.7 y (Table 5). PM10 and NO2 showed strong associations with the risk of incident mental disorders. An IQR increment of PM10 level at baseline was associated with 6% higher risk of mental disorders (95% CI: 1.04, 1.08), 5% higher risk of depression (95% CI: 1.02, 1.07), 7% higher risk of anxiety (95% CI: 1.04, 1.10). An IQR increase in NO2 was associated with similar risks of mental disorders and each disorder but was associated with relatively lower risks with respect to PM2.5 and NOx. PMcoarse level was not associated with any of the health outcomes. The air pollution index was positively associated with the risk of mental disorders. For instance, HRs (95% CI) of the risk of mental disorders were 1.06 (95% CI: 1.00, 1.12), 1.07 (95% CI: 1.01, 1.13), 1.10 (95% CI: 1.04, 1.16), and 1.11 (95% CI: 1.05, 1.18) for second–fifth quintiles, respectively, in comparison with the lowest quintile of the air pollution index (Table 5). Dose–response relationships between the air pollution index and the risk of mental disorders demonstrated monotonically increasing patterns (Figure 3).
Table 5 Associations of baseline levels of air pollutants and air pollution index with incident mental disorders for 345,876 participants free of mental disorders over a median 8.7 y of follow-up.
Incident mental disordersa Incident depression Incident anxiety
Air pollutants ncase/ntotal Hazard ratio
(95% CI) p-Value E-value (CI) ncase/ntotal Hazard ratio
(95% CI) p-Value E-Value
(CI) ncase/ntotal Hazard ratio
(95% CI) p-value E-value
(CI)
PM2.5 16,185/345,876 1.03 (1.01, 1.05) 0.015 1.21 (1.11) 10,854/345,876 1.01 (1.00, 1.04) 0.047 1.11 (1.00) 8,257/345,876 1.03 (1.00–1.06) 0.031 1.21 (1.00)
PMcoarse 16,185/345,876 0.99 (0.98, 1.01) 0.53 1.11 (1.00) 10,854/345,876 0.99 (0.98, 1.01) 0.40 1.11 (1.00) 8,257/345,876 1.00 (0.98–1.02) 0.78 1.00 (1.00)
PM10 16,185/345,876 1.06 (1.04, 1.08) <0.0001 1.31 (1.24) 10,854/345,876 1.05 (1.02, 1.07) 0.0009 1.28 (1.17) 8,257/345,876 1.07 (1.04–1.10) <0.0001 1.34 (1.24)
NOx 16,185/345,876 1.03 (1.01, 1.05) 0.0010 1.21 (1.11) 10,854/345,876 1.03 (1.00, 1.05) 0.031 1.21 (1.00) 8,257/345,876 1.04 (1.01–1.06) 0.0047 1.24 (1.11)
NO2 16,185/345,876 1.06 (1.04, 1.09) <0.0001 1.31 (1.24) 10,854/345,876 1.05 (1.02, 1.08) 0.0002 1.28 (1.16) 8,257/345,876 1.07 (1.04–1.10) <0.0001 1.34 (1.24)
Air pollution
indexb ncase/ntotal Hazard ratio
(95% CI) p-Value E-value (CI) ncase/ntotal Hazard ratio
(95% CI) p-Value E-value
(CI) ncase/ntotal Hazard ratio
(95% CI) p-value E-value
(CI)
Continuous 16,185/345,876 1.04 (1.03, 1.06) <0.0001 1.24 (1.21) 10,854/345,876 1.03 (1.01, 1.06) 0.0030 1.21 (1.11) 8,257/345,876 1.05 (1.03–1.08) <0.0001 1.28 (1.21)
Q1 2,868/69,175 Ref — — 1,900/69,175 Ref — — 1,477/69,175 Ref — —
Q2 3,218/69,175 1.06 (1.00, 1.12) 0.046 1.31 (1.00) 2,134/69,175 1.05 (0.98, 1.12) 0.14 1.28 (1.00) 1,659/69,175 1.07 (0.99–1.16) 0.07 1.34 (1.00)
Q3 3,243/69,176 1.07 (1.01, 1.13) 0.016 1.34 (1.11) 2,178/69,176 1.06 (1.00, 1.13) 0.09 1.31 (1.00) 1,651/69,176 1.08 (1.00–1.16) 0.05 1.37 (1.00)
Q4 3,384/69,175 1.10 (1.04, 1.16) 0.0005 1.43 (1.24) 2,315/69,175 1.07 (1.00, 1.14) 0.046 1.34 (1.00) 1,707/69,175 1.12 (1.04–1.21) 0.004 1.49 (1.24)
Q5 3,472/69,175 1.11 (1.05, 1.18) 0.0002 1.46 (1.28) 2,327/69,175 1.10 (1.03, 1.17) 0.0067 1.43 (1.21) 1,763/69,175 1.15 (1.06–1.24) 0.0009 1.57 (1.31)
Note: Associations of air pollution with the incident mental disorders, depression, and anxiety during the follow-up were tested with Cox proportional hazards models. Model adjusted for age, sex, BMI, race (White, Black, Asian, and other), smoking status (current/former/never), healthy alcohol intake status (male: <28g/day; female: <14g/day), healthy physical activity status [≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed (moderate + vigorous) activity], years of education (<10y), FEV1, FVC, Townsend deprivation index, live in urban area (yes/no), and prevalent hypertension, CHD, and diabetes (yes/no). The examination center was controlled for as a random effect. Estimates were reported by one IQR increase in each air pollutant. IQRs of PM2.5=1.27μg/m³, PMcoarse=0.79μg/m³, PM10=2.33μg/m³, NOx=16.53μg/m³, and NO2=10.87μg/m³. —, no data; BMI, body mass index; CHD, coronary heart disease; CI, confidence interval; IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; PHQ-4, Patient Health Questionnaire-4 questionnaire; PM2.5, fine particles with diameters of ≤2.5μm; PMcoarse, particulate matter with diameters of 2.5–10μm; PM10, particulate matter with diameters ≤10μm; Q1−Q5, first to fifth quintiles; Ref, reference; SE, standard error.
a Incident mental disorders, with incident depression and/or anxiety based on 7-y mental health survey and hospital records.
b To estimate the joint effect of multiple air pollutants on each outcome, an air pollution index was estimated based on all the coefficients of PM2.5, PM10, NOx, and NO2 for incident mental disorders retrieved from the crude Cox models adjusting for age and sex only.
Figure 3. Graphs of the best fitting models for relationships between air pollution index at baseline and incident mental disorders at follow-up for 345,876 mental disorder-free participants. Solid line: Point estimation; Black dash line: Confidence limits; Green dash line: Reference line; Dots: Knots (5th, 50th, and 95th percentiles). The restricted cubic spline regression model adjusted for age, sex, BMI, race (White, Black, Asian, and other), smoking status (current/former/never), healthy alcohol intake status (male: <28g/day; female: <14g/day), healthy physical activity status [≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed (moderate + vigorous) activity], years of education (<10y), FEV1, FVC, Townsend deprivation index, live in urban area (yes/no), and prevalent hypertension, CHD, and diabetes (yes/no). Point estimates and corresponding confidence intervals were shown in Table S11. Note: BMI, body mass index; CHD, coronary heart disease; CI, confidence interval; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; min, minutes; NOx, nitrogen oxides; NO2, nitrogen dioxide; PM2.5, particulate matter with diameter of≤2.5 μm; PMcoarse, particulate matter with diameters of 2.5–10μm; PM10, particulate matter with diameter of≤10μm; SD, standard deviation.
Figure 3 is a set of three line graphs, plotting Hazard ratio (mental disorders), ranging from 0.75 to 1.30 in increments of 0.05; hazard ratio (depression), ranging from 0.75 to 1.30 in increments of 0.05; and hazard ratio (anxiety), ranging from 0.75 to 1.30 in increments of 0.05 (y-axis) across air pollution index, ranging as negative 3 standard deviation, negative 2 standard deviation, negative 1 standard deviation, mean, 1 standard deviation, 2 standard deviation, and 3 standard deviation (x-axis) for knots, including fifth, fiftieth, and ninety-fifth percentiles, point estimation, confidence limits, and reference line.
Sensitivity Analyses
In the model that simultaneously included the four pollutants (except PMcoarse), NOx and NO2 remained associated with incident mental disorders, whereas PM2.5 became null (p=0.06) and PM10 demonstrated attenuated estimates (Table S4). Furthermore, considering that depression or anxiety may exacerbate mental health impairment and thus cause the other, we assessed the association of air pollution with concomitant depression and anxiety at follow-up. A total of 2,926 participants had both incident depression and anxiety during the follow-up. PM2.5 (HR=1.02; 95% CI: 1.00, 1.04, p=0.040), PM10 . (HR=1.04; 95% CI: 1.01, 1.09, p=0.016), NOx (HR=1.03; 95% CI: 1.01, 1.07, p=0.009), and NO2. (HR=1.06; 95% CI: 1.01, 1.10, p=0.0007) were statistically associated with the concomitant incidence, and the highest quintile of the air pollution index was associated with a 13% higher risk of the concomitant incidence than the lowest but was not significant (Table S5, pQ5 of air pollution index=0.05, 95% CI: 1.00, 1.28). Another sensitivity analysis of the 321,090 participants living at the address for more than 5 y also demonstrated robust associations of the air pollution index with prevalent mental disorders (OR=1.05; 95% CI: 1.04, 1.07, p<0.0001) and the incidence of mental disorders (HR=1.03; 95% CI: 1.01, 1.06, p=0.0015), which suggests that the mobility of participants may affect our main findings minimally (Table S6).
Impact of Genetic Susceptibility on the Associations between Air Pollution and Mental Disorders
We tested whether the genetic background could modify the associations between air pollution and mental disorders. The GRS we generated was associated with the PHQ-4 score (Coefficient=0.024, SE=0.003), odds of mental disorders at baseline (OR=1.03; 95% CI: 1.02, 1.04), and the incident mental disorders during the follow-up (HR=1.04; 95% CI: 1.02, 1.06) (Table S7). However, the interaction between GRS and the air pollution index was not significantly associated with prevalence of mental disorders at baseline (Coefficient of interaction term=−0.0001, SE=0.0050, p-interaction=0.84) or their incidence (Coefficient of interaction term=0.011, SE=0.0085, p-interaction=0.19) during the follow-up. We also noted that estimates of the air pollution index were essentially unchanged in mutual adjustment models (Table S8). These findings suggested that both the GRS and air pollution index could be independently associated with the prevalence and risk of mental disorders.
Given that no interaction was observed, we tested the joint effect of both air pollution and genetic susceptibility using the quintiles of air pollution index and the dichotomized GRS. We observed that participants with higher GRS and air pollution index had higher odds of mental disorders and risk of mental disorders (Figure 4; Table S9). Participants with higher GRS and the fifth quintile of air pollution index had ∼20% higher odds of mental disorders at baseline (OR=1.20; 95% CI: 1.15, 1.26) and about 15% greater risk of incident mental disorders during the follow-up (HR=1.15; 95% CI: 1.06, 1.23) than participants with the low GRS and first quintile of the air pollution index.
Figure 4. Joint associations of genetic risk and air pollution levels on the odds of mental disorders for total 398,241 participants and at baseline and incident mental disorders for 345,876 mental disorder-free participants at follow-up. Dots: Point estimate; Error bar: 95% confidence limits; Dash line: Reference line. Associations of air pollution with the odds of mental disorders, depression, and anxiety at baseline were tested with logistic regression models, and associations with the incident mental disorders, depression, and anxiety during the follow-up were tested with Cox proportional hazards models. Models adjusted age, sex, BMI, race (White, Black, Asian, and other), smoking status (current/former/never), healthy alcohol intake status (male: <28g/day; female: <14g/day), healthy physical activity status [≥150 min/wk moderate or ≥75 min/wk vigorous or 150 min/wk mixed (moderate + vigorous) activity], years of education (<10y), FEV1, FVC, Townsend deprivation index, live in urban area (yes/no), and prevalent hypertension, CHD, and diabetes (yes/no). The examination center was controlled for as a random effect. Point estimates and corresponding confidence intervals were shown in Table S9. Note: BMI, body mass index; CHD, coronary heart disease; FEV1, forced expiratory volume in 1 second; FVC, forced vital capacity; min, minutes; High, high genetic risk score; Low, low genetic risk score; Q1–Q5, air pollution index quintiles 1st–5th.
Figures 4(1) and (2) are error bar graphs titled Odds of mental disorders at baseline and Risk of mental disorders at follow-up, plotting odds of mental disorders, ranging from 0.9 to 1.3 in increments of 0.1 and hazard ratio of incident mental disorders, ranging from 0.9 to 1.3 in increments of 0.1 (y-axis) across categories (binary genetic risk and air pollution index quintiles, ranging as Low and quintile 1, high and quintile 1, low and quintile 2, high and quintile 2, low and quintile 3, high and quintile 3, low and quintile 4, high and quintile 4, low and quintile 5, and high and quintile 5 (x-axis), respectively.
Discussion
In this prospective study using the UK Biobank cohort, we elucidated the individual and joint associations of five major air pollutants with depression and anxiety, the two most common mental disorders. We identified that baseline exposures to PM2.5, PM10, NOx, and NO2 were not only associated with the prevalent mental disorders at baseline but were also associated with the risk of mental disorders during a median follow-up of 8.7 y. Combined exposure of the four pollutants represented by an air pollution index was robustly associated with elevated risks of incident mental disorders, depression, and anxiety during the follow-up separately. For the results of primary analyses demonstrated in Table 4 and Table 5, the relatively large E-values indicate that they were robust. We found that such adverse effects of air pollution could be enhanced by the genetic susceptibility to mental disorders despite the fact that no robust interaction between the air pollution index and the GRS was observed.
To the best of our knowledge, this work is the one of the largest population-based cohort studies that provides evidence on the association between air pollution and mental disorders, especially among older adults. Our study suggests that an IQR increase in long-term exposure to PM2.5 was associated with more than 8% increases in the odds of depression at baseline and 1% increases in its risks. The IQR of PM2.5 was 1.27μg/m³ in our study, which is higher than the findings of a systematic review and meta-analysis by Braithwaite et al. reporting 10% higher pooled odds of depression per 10 ug/m3 PM2.5.8 A potential explanation is the approaches to access depression of the selected publications in this review; we noted that the majority of studies used data of self-reported medical records and/or emergency visits, which only captured the information of severe or intense depression, because individuals with light or moderate depression syndromes may ignore their mental status and may not visit doctors. Moreover, the yielded risk of incident depression in our participants free of mental disorders at baseline could be ∼8% per 10μg/m³ increment in PM2.5 level, which was much lower than the findings of Liu et al.’s meta-analysis (18%).5 This difference could be explained by the characteristics of the population of the UK Biobank, which is a volunteer cohort, and participants are likely healthier with fewer self-reported health disorders and live in less socioeconomically deprived areas than the general population.36 Both may limit the effect of air pollution on incident mental disorders because healthier participants and a better community could provide better health resilience to mitigate health risks.37 Reported associations between PM10 and depression have been similar in previous studies,24 which is in line with our findings. Furthermore, our study found relatively stronger long-term positive associations of NOx and NO2 with depression than PM2.5, which was consistent with the study of Bakolis et al. showing that NO2 and NOx had relatively stronger associations with common mental disorders than PM2.5.10 In comparison with depression, the impact of air pollution on anxiety has been much less frequently reported. Our study also reported the associations of long-term exposure not only to PM2.5 but also to PM10, NOx, and NO2 with anxiety. Except the study of Power et al., in which the authors found a positive impact of PM2.5 on prevalent anxiety,38 most of the small number of existing studies only explored the associations of short-term exposure to air pollutants with anxiety and yielded null findings.7
In our study, estimates of the associations of NOx. and NO2 with mental disorders were higher than that of PM2.5 even in the model controlling for all pollutants simultaneously (Table S4), which is in line with previous findings10,12 and suggests that PM2.5 may not be the largest contributors to the development of mental disorders, though NO2 shares some sources with PM2.5. Such robust effects of long-term NOx and NO2. exposure were in agreement with the associations of short-term exposure of NO2 with depression in a recent meta-analysis.6 PMcoarse was the only pollutant that demonstrated a nonlinear association with the mental health PHQ-4 score at baseline. It was not associated with the incidence of mental disorders or most depression and anxiety symptoms during the follow-up, which agrees with the previous statement on the health effects of PMcoarse,39 suggesting that the health implications of PMcoarse remain less characterized than those for PM2.5. Due to such heterogeneities of the associations between various air pollutants and mental health, we attempted to depict the combined associations of relevant air pollutants by constructing the air pollution index. It was associated with the risk of mental disorders and the comorbidity of depression and anxiety and further suggested that it could be featured as a more comprehensive measure of the effects of multiple air pollutants on mental health. A similar algorithm has been implemented in previous epidemiological studies not only including environmental exposures,21,40 but also covering lifestyle and nutrition.41 Such a simple score may make the results of epidemiological studies much easier for nonprofessionals to interpret and understand.
Biological mechanisms underlying the association between ambient air pollution exposure and mental disorder remain inconclusive. Inflammation and oxidative stress are two commonly suggested biological pathways. Neuroinflammation caused by air pollution may deplete the brain serotonin, dysregulate the hypothalamus-pituitary-adrenal axis, and interfere with the production of neurons in the dentate gyrus of the hippocampus.42 Oxidative stress further contributes to the degradation of dopaminergic neurons, which is likely involved in the neuropathology of depression and additionally intimately impairs mitochondrial function, and enhances excitotoxicity and nitric oxide toxicity.43 Moreover, neuroimaging-based epidemiological studies revealed that long-term exposure to air pollution may be associated with changes in brain structures and functioning in white matter, cortical gray matter, and basal ganglia,25 which may intensify the progression of mental disorders. Additionally, the indirect impact of air pollution on human health status, including residential dissatisfaction,44,45 poor self-perceived health and life quality,46 frailty,47 as well as increased risk of major chronic diseases, such as cardiovascular diseases, respiratory diseases, dementia, and cancers,16,48,49 may be taken into account because they could trigger the incidence of depression and anxiety by bringing increasing psychological burden, especially for older adults. A more intriguing aspect is that we found that the combined associations of major air pollutants with mental health could be moderately enhanced by a GRS based on previous established SNPs of mental disorders. A plausible explanation of the observed lack of interaction between the air pollution index and GRS could be that parts of the SNPs we used to build the GRS were related to the development of neuro systems (e.g., RBFOX1 and PTPRT) or regulation of intercellular signaling in the brain (e.g., PDE4B),50–52 which may not directly interact with neuroinflammation or oxidative stress that may result from long-term exposure to air pollution. Nevertheless, given the risk that loci of mental health may soon be revealed by emerging genome analyses in the future, other genetic variants that play critical roles in modifying the risks of mental health resulting from air pollution may be found.
Our study has several strengths, including the large sample size, the detailed and repeated measurements of mental health status, and data available on a wide range of potential covariates and genetic variants. Several limitations are worthwhile to note when interpreting the findings. First, as a volunteer cohort, participants in the UK Biobank cohort were likely healthier than the general population, which may limit the impact of air pollution, because relatively healthier participants may suffer from less psychosocial pressure. Early-life experience of mental health problems may also alter a person’s tendency to live in bad neighborhoods, which often have worse air quality.53 Future studies with more detailed records of mental health status since childhood are warranted to explore this research question. Furthermore, the measurement bias of air pollution must be acknowledged. Our air pollution data were mostly only a single measurement of the annual average outdoor air pollution level in 2010 without any further measures during the follow-up. The baseline visit of the UK Biobank was conducted from 2006 to 2010, which prohibited our further exploration on the lag or short-term (<1 month) effects of air pollution on depression and anxiety. Also, because people, especially older adults, spent most of their time indoors,54 individual exposure to each air pollutant may differ from that indicated by the outdoor ambient concentrations we examined. Additionally, the impacts of the components of PM2.5 and other major air pollutants such as ozone and sulfur dioxide on depression and anxiety are worth investigation in future studies. For the air pollution index, although it was internally validated in our study as introduced in the “Methods” section, future studies are required to use this approach in other independent cohorts. Additionally, given only about one-third participants joined in the 7-y survey and we observed a larger proportion of people with mental disorders in this subset than the whole population without any form of mental disorders at baseline, the real-world burden of depression and anxiety during the follow-up could be worse than that based on the current information. Last, because the precise timing of the onset of mental disorders was lacking, both conditions together may weaken the observed associations of air pollutants with the risk of mental disorders in our study by causing potential underestimation of incident mental disorders and overestimation of censoring time.
In conclusion, our study is one of the largest cohort studies investigating the associations of long-term exposure to major air pollutants with the risk of depression and anxiety in ∼0.4 million people. Our findings established that air quality may be a modifiable factor that could be targeted to reduce the mental health burden in modern society. Beyond this, our study not only underpinned the combined associations of air pollutants with mental disorders using a novel air pollution index, but also highlighted a potential dose–response in the development of depression and anxiety under long-term air pollution exposure. More efforts are warranted to reveal the whole landscape of the impact of environmental exposures to additional air pollutants on mental health and to explore the underlying biological mechanisms in a multidisciplinary approach.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
Data are available in a public, open-access repository per application (https://www.ukbiobank.ac.uk/). This research has been conducted using the UK Biobank Resource under Application Number 44430. The UK Biobank data are available on application to the UK Biobank.
X.G. was supported by grants from the National Key Research and Development Program of China (No. 2022YFC3702704), the China Center for Disease Control Key Laboratory of Environment and Population Health (2022-CKL-03), and Peking University (BMU2021YJ044). The authors thank C. Chen for the language assistance.
==== Refs
References
1. Verhoeven JE, Révész D, van Oppen P, Epel ES, Wolkowitz OM, Penninx BW. 2015. Anxiety disorders and accelerated cellular ageing. Br J Psychiatry 206 (5 ):371–378, PMID: , 10.1192/bjp.bp.114.151027.25657360
2. Révész D, Verhoeven JE, Milaneschi Y, Penninx BW. 2016. Depressive and anxiety disorders and short leukocyte telomere length: mediating effects of metabolic stress and lifestyle factors. Psychol Med 46 (11 ):2337–2349, PMID: , 10.1017/S0033291716000891.27266474
3. Cuijpers P, Beekman AT, Reynolds CF 3rd. 2012. Preventing depression: a global priority. JAMA 307 (10 ):1033–1034, PMID: , 10.1001/jama.2012.271.22416097
4. Beaudreau SA, O’Hara R. 2008. Late-life anxiety and cognitive impairment: a review. Am J Geriatr Psychiatry 16 (10 ):790–803, PMID: , 10.1097/JGP.0b013e31817945c3.18827225
5. Liu Q, Wang W, Gu X, Deng F, Wang X, Lin H, Guo X. 2021. Association between particulate matter air pollution and risk of depression and suicide: a systematic review and meta-analysis. Environ Sci Pollut Res Int 28 (8 ):9029–9049, PMID: , 10.1007/s11356-021-12357-3.33481201
6. Fan SJ, Heinrich J, Bloom MS, Zhao TY, Shi TX, Feng WR, et al. 2020. Ambient air pollution and depression: a systematic review with meta-analysis up to 2019. Sci Total Environ 701 :134721, PMID: , 10.1016/j.scitotenv.2019.134721.31715478
7. Dickerson AS, Wu AC, Liew Z, Weisskopf M. 2020. A scoping review of non-occupational exposures to environmental pollutants and adult depression, anxiety, and suicide. Curr Environ Health Rep 7 (3 ):256–271, PMID: , 10.1007/s40572-020-00280-7.32533365
8. Braithwaite I, Zhang S, Kirkbride JB, Osborn DPJ, Hayes JF. 2019. Air pollution (particulate matter) exposure and associations with depression, anxiety, bipolar, psychosis and suicide risk: a systematic review and meta-analysis. Environ Health Perspect 127 (12 ):126002, PMID: , 10.1289/EHP4595.31850801
9. Petrowski K, Bührer S, Strauß B, Decker O, Brähler E. 2021. Examining air pollution (PM10), mental health and well-being in a representative German sample. Scientific reports 11 (1 ):18436, PMID: , 10.1038/s41598-021-93773-w.34531408
10. Bakolis I, Hammoud R, Stewart R, Beevers S, Dajnak D, MacCrimmon S, Broadbent M. 2021. Mental health consequences of urban air pollution: prospective population-based longitudinal survey. Soc Psychiatry Psychiatr Epidemiol 56 (9 ):1587–1599, PMID: , 10.1007/s00127-020-01966-x.33097984
11. Khan A, Plana-Ripoll O, Antonsen S, Brandt J, Geels C, Landecker H, et al. 2019. Environmental pollution is associated with increased risk of psychiatric disorders in the US and Denmark. PLoS Biol 17 (8 ):e3000353, PMID: , 10.1371/journal.pbio.3000353.31430271
12. Newbury JB, Stewart R, Fisher HL, Beevers S, Dajnak D, Broadbent M, et al. 2021. Association between air pollution exposure and mental health service use among individuals with first presentations of psychotic and mood disorders: retrospective cohort study. Br J Psychiatry 219 (6 ):678–685, PMID: , 10.1192/bjp.2021.119.35048872
13. Kampa M, Castanas E. 2008. Human health effects of air pollution. Environ Pollut 151 (2 ): 362–367, PMID: , 10.1016/j.envpol.2007.06.012.17646040
14. Tsuang MT, Bar JL, Stone WS, Faraone SV. 2004. Gene-environment interactions in mental disorders. World Psychiatry 3 (2 ):73–83, PMID: .16633461
15. Robert Plomin, DeFries JC, Knopik VS, Neiderhiser JM. 2013. Behavioral Genetics (6th edition). New York, NY: Worth Publishers. 10.1007/s10519-013-9598-6.
16. Peters A, Nawrot TS, Baccarelli AA. 2021. Hallmarks of environmental insults. Cell 184 (6 ):1455–1468, PMID: , 10.1016/j.cell.2021.01.043.33657411
17. Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. 2015. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med 12 (3 ):e1001779, PMID: , 10.1371/journal.pmed.1001779.25826379
18. Stanhope J. 2016. Patient Health Questionnaire-4. Occup Med (Lond) 66 (9 ):760–761, PMID: , 10.1093/occmed/kqw165.27994085
19. Kroenke K, Spitzer RL, Williams JB. 2001. The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med 16 (9 ):606–613, PMID: , 10.1046/j.1525-1497.2001.016009606.x.11556941
20. Spitzer RL, Kroenke K, Williams JB, Löwe B. 2006. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med 166 (10 ):1092–1097, PMID: , 10.1001/archinte.166.10.1092.16717171
21. Wang M, Zhou T, Song Y, Li X, Ma H, Hu Y, et al. 2021. Joint exposure to various ambient air pollutants and incident heart failure: a prospective analysis in UK Biobank. Eur Heart J 42 (16 ):1582–1591, PMID: , 10.1093/eurheartj/ehaa1031.33527989
22. Eeftens M, Beelen R, de Hoogh K, Bellander T, Cesaroni G, Cirach M, et al. 2012. Development of Land Use Regression models for PM2.5, PM2.5 absorbance, PM10 and PMcoarse in 20 European study areas; results of the ESCAPE project. Environ Sci Technol 46 (20 ):11195–11205, PMID: , 10.1021/es301948k.22963366
23. Beelen R, Hoek G, Vienneau D, Eeftens M, Dimakopoulou K, Pedeli X, et al. 2013. Development of NO2 and NOx land use regression models for estimating air pollution exposure in 36 study areas in Europe–The ESCAPE project. Atmos Environ 72 :10–23, 10.1016/j.atmosenv.2013.02.037.
24. Vert C, Sánchez-Benavides G, Martínez D, Gotsens X, Gramunt N, Cirach M, et al. 2017. Effect of long-term exposure to air pollution on anxiety and depression in adults: a cross-sectional study. Int J Hyg Environ Health 220 (6 ):1074–1080, PMID: , 10.1016/j.ijheh.2017.06.009.28705430
25. de Prado Bert P, Mercader EMH, Pujol J, Sunyer J, Mortamais M. 2018. The effects of air pollution on the brain: a review of studies interfacing environmental epidemiology and neuroimaging. Curr Environ Health Rep 5 (3 ):351–364, PMID: , 10.1007/s40572-018-0209-9.30008171
26. Craig CL, Marshall AL, Sjöström M, Bauman AE, Booth ML, Ainsworth BE, et al. 2003. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc 35 (8 ):1381–1395, PMID: , 10.1249/01.MSS.0000078924.61453.FB.12900694
27. Townsend P, Phillimore P, Beattie A. 1988. Health and Deprivation: Inequality and the North: New York, NY: Routledge.
28. Cummings P. 2013. Missing data and multiple imputation. JAMA Pediatr 167 (7 ):656–661, PMID: , 10.1001/jamapediatrics.2013.1329.23699969
29. Bycroft C, Freeman C, Petkova D, et al. 2017. Genome-wide genetic data on–500,000 UK Biobank participants. bioRxiv:166298.
30. Meier SM, Trontti K, Purves KL, Als TD, Grove J, Laine M, et al. 2019. Genetic variants associated with anxiety and stress-related disorders: a genome-wide association study and mouse-model study. JAMA Psychiatry 76 (9 ):924–932, PMID: , 10.1001/jamapsychiatry.2019.1119.31116379
31. Wray NR, Ripke S, Mattheisen M, Trzaskowski M, Byrne EM, Abdellaoui A, et al. 2018. Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nat Genet 50 (5 ):668–681, PMID: , 10.1038/s41588-018-0090-3.29700475
32. Simon RM, Subramanian J, Li MC, Menezes S. 2011. Using cross-validation to evaluate predictive accuracy of survival risk classifiers based on high-dimensional data. Brief Bioinform 12 (3 ):203–214, PMID: , 10.1093/bib/bbr001.21324971
33. Fan M, Sun D, Zhou T, Heianza Y, Lv J, Li L, et al. 2020. Sleep patterns, genetic susceptibility, and incident cardiovascular disease: a prospective study of 385 292 UK biobank participants. Eur Heart J 41 (11 ):1182–1189, PMID: , 10.1093/eurheartj/ehz849.31848595
34. Nikolakopoulou A, Mavridis D, Salanti G. 2014. How to interpret meta-analysis models: fixed effect and random effects meta-analyses. Evid Based Ment Health 17 (2 ):64, PMID: , 10.1136/eb-2014-101794.24778439
35. VanderWeele TJ, Ding P. 2017. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med 167 (4 ):268–274, PMID: , 10.7326/M16-2607.28693043
36. Fry A, Littlejohns TJ, Sudlow C, Doherty N, Adamska L, Sprosen T, et al. 2017. Comparison of sociodemographic and health-related characteristics of UK Biobank participants with those of the general population. Am J Epidemiol 186 (9 ):1026–1034, PMID: , 10.1093/aje/kwx246.28641372
37. Wulff K, Donato D, Lurie N. 2015. What is health resilience and how can we build it? Annu Rev Public Health 36 :361–374, PMID: , 10.1146/annurev-publhealth-031914-122829.25581148
38. Power MC, Kioumourtzoglou MA, Hart JE, Okereke OI, Laden F, Weisskopf MG. 2015. The relation between past exposure to fine particulate air pollution and prevalent anxiety: observational cohort study. BMJ 350 :h1111, PMID: , 10.1136/bmj.h1111.25810495
39. Adar SD, Filigrana PA, Clements N, Peel JL. 2014. Ambient coarse particulate matter and human health: a systematic review and meta-analysis. Curr Environ Health Rep 1 (3) :258–274, PMID: , 10.1007/s40572-014-0022-z.25152864
40. Park SK, Tao Y, Meeker JD, Harlow SD, Mukherjee B. 2014. Environmental risk score as a new tool to examine multi-pollutants in epidemiologic research: an example from the NHANES study using serum lipid levels. PLoS One 9 (6 ):e98632, PMID: , 10.1371/journal.pone.0098632.24901996
41. Arouca A, Moreno LA, Gonzalez-Gil EM, Marcos A, Widhalm K, Molnár D, et al. 2019. Diet as moderator in the association of adiposity with inflammatory biomarkers among adolescents in the HELENA study. Eur J Nutr 58 (5 ):1947–1960, PMID: , 10.1007/s00394-018-1749-3.29948222
42. Mithaiwala MN, Santana-Coelho D, Porter GA, O’Connor JC. 2021. Neuroinflammation and the kynurenine pathway in CNS disease: molecular mechanisms and therapeutic implications. Cells 10 (6 ):1548, PMID: , 10.3390/cells10061548.34205235
43. Bhatt S, Nagappa AN, Patil CR. 2020. Role of oxidative stress in depression. Drug Discov Today 25 (7 ):1270–1276, PMID: , 10.1016/j.drudis.2020.05.001.32404275
44. Liu S, Ouyang Z, Chong AM, Wang H. 2018. Neighborhood environment, residential satisfaction, and depressive symptoms among older adults in residential care homes. Int J Aging Hum Dev 87 (3 ):268–288, PMID: , 10.1177/0091415017730812.28942658
45. von Lindern E, Hartig T, Lercher P. 2016. Traffic-related exposures, constrained restoration, and health in the residential context. Health Place 39 :92–100, PMID: , 10.1016/j.healthplace.2015.12.003.26995669
46. Nguyen HA, Anderson CAM, Miracle CM, Rifkin DE. 2017. The association between depression, perceived health status, and quality of life among individuals with chronic kidney disease: an analysis of the National Health and Nutrition Examination Survey 2011–2012. Nephron 136 (2 ):127–135, PMID: , 10.1159/000455750.28249290
47. Vaughan L, Corbin AL, Goveas JS. 2015. Depression and frailty in later life: a systematic review. Clin Interv Aging 10 :1947–1958, PMID: , 10.2147/CIA.S69632.26719681
48. Dominski FH, Lorenzetti Branco JH, Buonanno G, Stabile L, Gameiro da Silva M, Andrade A. 2021. Effects of air pollution on health: a mapping review of systematic reviews and meta-analyses. Environ Res 201 :111487, PMID: , 10.1016/j.envres.2021.111487.34116013
49. Gao X, Coull B, Lin X, Vokonas P, Spiro A, Hou L, Schwartz J. 2021. Short-term air pollution, cognitive performance, and nonsteroidal anti-inflammatory drug use in the Veterans Affairs Normative Aging Study. Nat Aging 1 (5 ):430–437, PMID: , 10.1038/s43587-021-00060-4.34841262
50. Tibbo AJ, Baillie GS. 2020. Phosphodiesterase 4B: master regulator of brain signaling. Cells 9 (5 ):1254, PMID: , 10.3390/cells9051254.32438615
51. Lee JR. 2015. Protein tyrosine phosphatase PTPRT as a regulator of synaptic formation and neuronal development. BMB Rep 48 (5 ):249–255, PMID: , 10.5483/BMBRep.2015.48.5.037.25748173
52. Wamsley B, Jaglin XH, Favuzzi E, Quattrocolo G, Nigro MJ, Yusuf N, et al. 2018. Rbfox1 mediates cell-type-specific splicing in cortical interneurons. Neuron 100 (4 ):846–59.e7, PMID: , 10.1016/j.neuron.2018.09.026.30318414
53. Fuller R, Landrigan PJ, Balakrishnan K, Bathan G, Bose-O’Reilly S, Brauer M, et al. 2022. Pollution and health: a progress update. Lancet Planet Health 6 (6 ):e535–e547, PMID: , 10.1016/S2542-5196(22)00090-0.35594895
54. World Health Organization. 2010. WHO Guidelines for Indoor Air Quality: Selected Pollutants. Copenhagen, Denmark: Regional Office for Europe, World Health Organization.
|
PMC009xxxxxx/PMC9819217.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36607286
EHP10745
10.1289/EHP10745
Research
Long-Term Exposure to Transportation Noise and Ischemic Heart Disease: A Pooled Analysis of Nine Scandinavian Cohorts
https://orcid.org/0000-0002-5170-9728
Pyko Andrei 1 2
Roswall Nina 3
Ögren Mikael 4 5
Oudin Anna 6 7
Rosengren Annika 8 9
Eriksson Charlotta 1 2
Segersson David 10 11
Rizzuto Debora 12 13
Andersson Eva M. 4 5
Aasvang Gunn Marit 14
Engström Gunnar 15
Gudjonsdottir Hrafnhildur 16 17
Jørgensen Jeanette T. 18
Selander Jenny 1
Christensen Jesper H. 19
Brandt Jørgen 19 20
Leander Karin 1
Overvad Kim 21 22
Eneroth Kristina 23
Mattisson Kristoffer 24
Barregard Lars 4 5
Stockfelt Leo 4 5
Albin Maria 1 2 24
Simonsen Mette K. 25
Tiittanen Pekka 26
Molnar Peter 4 5
Ljungman Petter 1 27
Solvang Jensen Steen 19
Gustafsson Susanna 28
Lanki Timo 26 29 30
Lim Youn-Hee 18
Andersen Zorana J. 18
Sørensen Mette 3 31
Pershagen Göran 1 2
1 Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden
2 Center for Occupational and Environmental Medicine, Region Stockholm, Stockholm, Sweden
3 Danish Cancer Society Research Centre, Copenhagen, Denmark
4 Occupational and Environmental Medicine, School of Public Health and Community Medicine, Institute of Medicine, University of Gothenburg, Gothenburg, Sweden
5 Department of Occupational and Environmental Medicine, Sahlgrenska University Hospital, Gothenburg, Sweden
6 Planetary Health, Lund University, Lund, Sweden
7 Sustainable Health, Department of Public Health and Clinical Medicine, Umeå University, Sweden
8 Molecular and Clinical Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden
9 Sahlgrenska University Hospital, Region Västra Götaland, Gothenburg, Sweden
10 Swedish Meteorological and Hydrological Institute, Norrköping, Sweden
11 Department of Environmental Science, Stockholm University, Stockholm, Sweden
12 Aging Research Center, Department of Neurobiology, Care Sciences and Society, Karolinska Institutet and Stockholm University, Stockholm, Sweden
13 Stockholm Gerontology Research Center, Stockholm, Sweden
14 Department of Environmental Health, Norwegian Institute of Public Health, Oslo, Norway
15 Department of Clinical Science, Lund University, Malmö, Sweden
16 Centre for Epidemiology and Community Medicine, Region Stockholm, Stockholm, Sweden
17 Department of Global Public Health, Karolinska Institutet, Stockholm, Sweden
18 Department of Public Health, University of Copenhagen, Copenhagen, Denmark
19 Department of Environmental Science, Aarhus University, Roskilde, Denmark
20 iClimate – Interdisciplinary Centre for Climate Change, Aarhus University, Roskilde, Denmark
21 Department of Public Health, Aarhus University, Aarhus, Denmark
22 Department of Cardiology, Aalborg University Hospital, Aalborg, Denmark
23 Environment and Health Administration, Stockholm, Sweden
24 Division of Occupational and Environmental Medicine, Lund University, Lund, Sweden
25 Department of Neurology, The Parker Institute, Frederiksberg Hospital, Capital Region, Frederiksberg, Denmark
26 Department of Health Security, Finnish Institute for Health and Welfare (THL), Kuopio, Finland
27 Department of Cardiology, Danderyd Hospital, Stockholm, Sweden
28 Environment Department, City of Malmö, Malmö, Sweden
29 School of Medicine, University of Eastern Finland, Kuopio, Finland
30 Department of Environmental and Biological Sciences, University of Eastern Finland, Kuopio, Finland
31 Department of Natural Science and Environment, Roskilde University, Roskilde, Denmark
Address correspondence to Andrei Pyko, Institute of Environmental Medicine, Karolinska Institutet, Nobels väg 13, 171 77 Stockholm, Sweden. Email: [email protected]
6 1 2023
1 2023
131 1 01700307 12 2021
09 12 2022
09 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Transportation noise may induce cardiovascular disease, but the public health implications are unclear.
Objectives:
The study aimed to assess exposure–response relationships for different transportation noise sources and ischemic heart disease (IHD), including subtypes.
Methods:
Pooled analyses were performed of nine cohorts from Denmark and Sweden, together including 132,801 subjects. Time-weighted long-term exposure to road, railway, and aircraft noise, as well as air pollution, was estimated based on residential histories. Hazard ratios (HRs) were calculated using Cox proportional hazards models following adjustment for lifestyle and socioeconomic risk factors.
Results:
A total of 22,459 incident cases of IHD were identified during follow-up from national patient and mortality registers, including 7,682 cases of myocardial infarction. The adjusted HR for IHD was 1.03 [95% confidence interval (CI) 1.00, 1.05] per 10 dB Lden for both road and railway noise exposure during 5 y prior to the event. Higher risks were indicated for IHD excluding angina pectoris cases, with HRs of 1.06 (95% CI: 1.03, 1.08) and 1.05 (95% CI: 1.01, 1.08) per 10 dB Lden for road and railway noise, respectively. Corresponding HRs for myocardial infarction were 1.02 (95% CI: 0.99, 1.05) and 1.04 (95% CI: 0.99, 1.08). Increased risks were observed for aircraft noise but without clear exposure–response relations. A threshold at around 55 dB Lden was suggested in the exposure–response relation for road traffic noise and IHD.
Discussion:
Exposure to road, railway, and aircraft noise in the prior 5 y was associated with an increased risk of IHD, particularly after exclusion of angina pectoris cases, which are less well identified in the registries. https://doi.org/10.1289/EHP10745
Supplemental Material is available online (https://doi.org/10.1289/EHP10745).
The authors declare no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Exposure to traffic noise is increasing because of ongoing urbanization, densification of urban settlements, and growth of the transport sector. In 2017, 113 million Europeans were estimated to be exposed to road traffic noise levels of at least 55 dB Lden, which is the health-based indicator level used for noise mapping by the European Environment Agency.1 The World Health Organization (WHO) has estimated that more than 1 million healthy years of life are lost annually due to traffic-related noise in Western Europe, primarily caused by sleep disturbance and annoyance, but cardiovascular disease also contributes.2 The burden of disease from transportation noise was ranked the second highest in Europe, after air pollution, among environmental causes, and in 2018 the WHO proposed stricter environmental noise guidelines for the European Region.
A systematic review on environmental noise exposure and cardiovascular diseases for the WHO guidelines concluded that the evidence linking exposure to road traffic noise and incidence of ischemic heart disease (IHD) was of high quality.3 Longitudinal epidemiological studies on traffic noise and incidence of IHD also including data on lifestyle factors provide the most compelling information, and such studies published after those included in the WHO review provide a mixed picture,4–11 however, mostly reporting that transportation noise exposure was associated with IHD and/or myocardial infarction. Most of the evidence relates to road traffic noise, and only few of the studies assessed exposure to railway and/or aircraft noise.5,7,9 Moreover, it is not clear whether risks associated with noise exposure differ between major subtypes of IHD, primarily myocardial infarction and angina pectoris, or between nonfatal and fatal IHD. Detailed evaluations of the shape of the exposure–response relationships were generally not performed; however, two recent studies indicated that there may be thresholds in the association between road traffic noise and incidence of IHD8 or myocardial infarction,11 which would be crucial for health impact assessments.
Assessment of interactions between traffic noise exposure and other risk factors for IHD may be important in prioritization of preventive action, e.g., for identification of vulnerable groups, as well as for the understanding of etiological mechanisms. For example, long-term exposure to air pollution such as fine particulate matter (PM2.5 <2.5μm) increases the risk of cardiovascular disease, including IHD.12 A growing number of studies on cardiovascular disease have estimated exposure to both road traffic noise and air pollution,4,6–11,13–17 but the evidence on interactions is not consistent. Other risk factors of interest include smoking, physical activity, body mass index (BMI), and socioeconomic status. However, no clear picture on interactions with traffic noise in relation to IHD has emerged, partly because of lack of data.18
The aim of this study was to assess exposure–response relations for long-term exposure to noise from road traffic, railways and aircraft, and incidence of IHD based on combined analyses of nine Scandinavian cohorts. In particular, we studied associations for common subtypes of IHD and whether the noise-related risks were modified by other risk factors, including air pollution. This constitutes a major extension of earlier publications based on the cohorts, including longer follow-up and substantially increased number of cases, which together with pooled analyses, enables detailed evaluation of the shape of exposure–response functions and interactions and for subtypes of IHD.
Methods
Study Population
The study is based on the “Nordic studies on occupational and traffic noise in relation to disease” (NordSOUND) project (www.nordsound.dk), and uses pooled data from nine Scandinavian cohorts.17 Two cohorts were included from Denmark: the nationwide Danish Nurse Cohort (DNC) and the Diet, Cancer and Health cohort (DCH) from Copenhagen/Aarhus. The seven Swedish cohorts originated from Malmö, with the Malmö Diet and Cancer Study (MDC); Gothenburg, including the Swedish Primary Prevention Cohort (PPS) and the GOT-MONICA cohort; and Stockholm County: the Swedish National Study of Aging and Care in Kungsholmen (SNAC-K), the Stockholm Screening Across the Lifespan Twin Study (SALT), the Stockholm 60 Years Old study (Sixty), and the Stockholm Diabetes Prevention Program (SDPP). The four cohorts from Stockholm used identical methodology for environmental exposure assessment and harmonized covariate information.19 The nine study cohorts are described in detail in Table S1, including key references. All cohorts had registry-based residential address history for participants, with estimated transportation noise exposure for each address. If needed, delayed entry was used for the study participants, implying that follow-up started when transportation noise data for all relevant sources were available during the 5 preceding years. This is referred to as the study baseline.
The study was conducted in accordance with the Helsinki Declaration and approved by relevant ethics review boards for the included cohorts. Informed consent was obtained from all cohort participants.
Outcome Assessment
Data on individual IHD events were obtained via linkage to national patient and mortality registers. Each event was defined based on the International Classification of Diseases (ICD) versions 9 or 10 as hospitalization or death with principal diagnosis of IHD (ICD9: 410–414; ICD10: I20–I25), and the two subgroups IHD excluding angina pectoris (ICD9: 410, 411, 412, 414; ICD10: I21–I25) and myocardial infarction (ICD9: 410; ICD10: I21–I23). The selection of the subgroup IHD excluding angina pectoris was made to achieve a better comparability between the Danish and Swedish data because angina pectoris constituted a considerably larger part of IHD in the Danish cohorts (cf. Table 1). Myocardial infarction was included because it was the outcome analyzed in several earlier studies on traffic noise and this diagnosis generally has a high quality. Supplementary analyses were also performed focusing on the subgroups angina pectoris (ICD9: 413; ICD10: I20) and IHD excluding angina pectoris and myocardial infarction (ICD9: 411, 412, 414; ICD10: I25). Subjects with an IHD diagnosis before the study baseline were excluded from the analyses because they generally receive medical treatment and may change their lifestyle, which could affect the susceptibility to noise. All first events (hospitalization or death) of IHD during follow-up were classified as incident. IHD cases were coded as fatal if they originated from a mortality registry or if deaths from any cause occurred within 28 d after hospitalization for IHD.
Table 1 Administrative and baseline characteristics of nine included cohorts from Scandinavia (n=132,801) at study baseline.
Copenhagen or Aarhus, DK Denmark Malmö, SE Gothenburg, SE Stockholm, SE
Characteristicsa DCH DNC MDC PPS GOT-MONICA SDPP SIXTY SNAC-K SALT Total
Follow-up period (y) 1994–2016 1993–2016 1992–2017 1972–2011 1985–2011 1992–2011 1997–2017 2001–2016 1998–2016 1975–2017
Median length of follow-upa (y) 20.0 (4.0–20.5) 15.5 (4.2–20.5) 19.5 (2.7–24.7) 21.5 (5.3–36.5) 12.5 (2.5–17.5) 20.1 (12.0–24.7) 19.5 (6.0–20.0) 14.5 (3.0–6.5) 16.7 (6.7–19.0) 19.7 (3.7–24.0)
Analytical sampleb (n) 52,948 26,126 26,573 5,234 2,310 7,534 3,770 2,252 6,054 132,801
Person-years (n) 891,183 427,793 441,991 115,968 29,891 159,941 68,277 30,462 97,784 2,263,290
IHD (n) 10,086 3,271 5,024 1,849 137 368 584 396 744 22,459
IHD excluding angina pectoris (n) 5,027 1,479 3,054 1,441 83 221 348 273 473 12,399
Myocardial infarction (n) 2,606 668 2,265 1,222 70 170 227 160 294 7,682
Fatal IHD (n) 590 286 882 554 22 48 80 118 160 2,740
Sex [n (%)]
Men 24,699 (46.7) 0 (0.0) 10,112 (38.1) 5,234 (100.0) 1,089 (47.1) 2,949 (39.1) 1,756 (46.6) 876 (38.9) 2,675 (44.2) 49,390 (37.2)
Women 28,249 (53.3) 26,126 (100.0) 16,461 (61.9) 0 (0.0) 1,221 (52.9) 4,585 (60.9) 2,014 (53.4) 1,376 (61.1) 3,379 (55.8) 83,411 (62.8)
Age at study baselinea (y) 56.1 (50.8–64.2) 51.0 (45.0–71.5) 57.7 (47.2–71.4) 55.0 (49.8–59.9) 50.2 (31.8–67.1) 47.9 (37.9–53.8) 60.3 (60.2–60.6) 72.2 (60.3–90.3) 56.1 (44.3–78.8) 55.4 (45.7–69.7)
Educational levelc (%)
Low 15,856 (29.9) 0 (0.0) 18,000 (67.7) 3,570 (68.2) 470 (20.3) 2,374 (31.5) 1,480 (39.3) 510 (22.6) 1,585 (26.2) 43,845 (33.0)
Medium 24,552 (46.4) 26,126 (100.0) 4,719 (17.8) 1,071 (20.5) 1,164 (50.4) 2,905 (38.6) 1,226 (32.5) 898 (39.9) 2,216 (36.6) 64,877 (48.9)
High 12,540 (23.7) 0 (0.0) 3,854 (14.5) 593 (11.3) 676 (29.3) 2,255 (29.9) 1,064 (28.2) 844 (37.5) 2,253 (37.2) 24,079 (18.1)
Marital statusd (%)
Single 12,343 (23.3) 7,681 (29.4) 9,261 (34.9) 742 (14.2) 721 (31.2) 1,246 (16.5) 976 (25.9) 1,138 (50.5) 1,945 (32.1) 36,053 (27.1)
Married 40,605 (76.7) 18,445 (70.6) 17,312 (65.1) 4,492 (85.8) 1,589 (68.8) 6,288 (83.5) 2,794 (74.1) 1,114 (49.5) 4,109 (67.9) 96,748 (72.9)
Area-level incomee (%)
1st quartile 17,557 (33.2) 8,593 (32.9) 6,164 (23.2) 1,352 (25.8) 505 (21.9) 263 (3.5) 165 (4.4) 65 (2.9) 429 (7.1) 35,093 (26.4)
2nd quartile 11,904 (22.5) 6,942 (26.6) 5,580 (21.0) 1,168 (22.3) 351 (15.2) 421 (5.6) 324 (8.6) 0 (0.0) 623 (10.3) 27,313 (20.6)
3rd quartile 8,956 (16.9) 6,462 (24.7) 7,068 (26.6) 1,283 (24.5) 500 (21.7) 1,589 (21.1) 908 (24.1) 6 (0.3) 1,107 (18.3) 27,879 (21.0)
4th quartile 14,531 (27.4) 4,129 (15.8) 7,761 (29.2) 1,431 (27.4) 954 (41.2) 5,261 (69.8) 2,373 (62.9) 2,181 (96.8) 3,895 (64.3) 42,516 (32.0)
Smoking status (%)
Current 19,127 (36.1) 9,193 (35.2) 7,571 (28.5) 2,089 (39.9) 650 (28.1) 1,983 (26.3) 788 (20.9) 349 (15.5) 1,237 (20.4) 42,987 (32.4)
Former 14,723 (27.8) 7,973 (30.5) 8,766 (33.0) 1,741 (33.3) 543 (23.5) 2,747 (36.5) 1,449 (38.4) 903 (40.1) 2,157 (35.6) 41,002 (30.9)
Never 19,098 (36.1) 8,960 (34.3) 10,236 (38.5) 1,404 (26.8) 1,117 (48.4) 2,804 (37.2) 1,533 (40.7) 1,000 (44.4) 2,660 (44.0) 48,812 (36.7)
Physical activityf (%)
Low 27,194 (51.4) 1,771 (6.8) 13,570 (51.1) 1,325 (25.3) 402 (17.4) 4,955 (65.8) 2,602 (69.0) 1,662 (73.8) 3,280 (54.2) 56,761 (42.7)
Medium 10,422 (19.7) 17,350 (66.4) 5,516 (20.8) 3,090 (59.0) 1,452 (62.9) 1,998 (26.5) 882 (23.4) 433 (19.2) 2,217 (36.6) 43,360 (32.7)
High 15,332 (28.9) 7,005 (26.8) 7,487 (28.1) 819 (15.7) 456 (19.7) 581 (7.7) 286 (7.6) 157 (7.0) 557 (9.2) 32,680 (24.6)
BMIa, kg/m2 25.5 (20.4–33.3) 23.1 (19.2–30.1) 25.2 (20.2–32.9) 25.1 (20.7–30.5) 24.5 (19.7–32.3) 25.1 (20.4–33.3) 26.1 (20.9–34.3) 25.3 (19.9–32.8) 24.1 (19.6–30.5) 24.9 (20.0–32.6)
Smoking intensitya,g [g/d (among current smokers)] 15.0 (5.0–32.1) 15.0 (2.0–26.0) 14.0 (1.0–30.0) —g 15.0 (4.0–25.0) 15.0 (2.0–25.0) 13.0 (2.0–25.0) 10.0 (0.0–30.0) 11.0 (2.0–25.0) 15.0 (3.0–30.0)
Alcohol intakeh (%)
Daily 10,418 (19.7) 3,267 (12.8) 4,338 (16.6) —h 23 (1.0) 301 (4.0) 210 (5.6) 244 (10.8) 549 (9.1) 19,350 (15.3)
Weekly 31,378 (59.3) 15,310 (60.1) 8,971 (34.3) — 814 (35.5) 4,897 (65.2) 1,508 (40.0) 1,083 (48.1) 3,867 (63.9) 67,828 (53.7)
Seldom 9,641 (18.2) 2,917 (11.5) 8,539 (32.7) — 1,292 (56.4) 2,059 (27.4) 1,689 (44.8) 768 (34.1) 1,469 (24.3) 28,374 (22.4)
Never 1,479 (2.8) 3,976 (15.6) 4,307 (16.4) — 164 (7.1) 255 (3.4) 361 (9.6) 157 (7.0) 167 (2.7) 10,866 (8.6)
Note: —, no data; BMI, body mass index; DCH, the Danish Diet Cancer and Health cohort; DK, Denmark; DNC, the Danish Nurse Cohort; GOT-MONICA, the Gothenburg cohort of “Multinational Monitoring of Trends and Determinants in Cardiovascular Diseases” (MONICA) project, Sweden; IHD, ischemic heart disease; MDC, the Malmö Diet and Cancer cohort, Sweden; PPS, the Primary Prevention Study cohort, Gothenburg, Sweden; SDPP, the Stockholm Diabetes Preventive Program, Sweden; SE, Sweden; SIXTY, the 60-Years cohort, Stockholm, Sweden; SNAC-K, the Swedish National Study of Aging and Care in Kungsholmen, Stockholm, Sweden.
a Median and 5th–95th percentiles, unless otherwise stated.
b Exclusion of prevalent IHD cases and subjects with missing information on covariates in Model 2 (age, cohort, sex, calendar year, educational level, marital status, area-income, road traffic noise, or railway noise).
c Educational level is defined as “low” for primary school or less, “medium” for up to secondary school or equivalent, and “high” for university degree and more.
d Marital status “single” category includes widowed or never married; “married” also includes those living with a partner.
e Registry-based data from small socioeconomically homogeneous areas with ∼1,000–2,000 inhabitants, categorized in country-specific quartiles.
f Physical activity during leisure-time is defined as “low” in those active once a month or <1h per week; “medium” in those active about once a week or ∼approximately 1h/wk; “high” in those active 3 times a week or more or >2 h/wk.
g No data on smoking intensity were available in the PPS cohort. In other cohorts such information was missing for up to 0.7%.
h No data on alcohol intake were available in the PPS cohort. In other cohorts such information was missing for up to 2.5%.
Noise Exposure Assessment
Noise levels for each address during the study period were calculated as the equivalent continuous A-weighted sound pressure level (LAeq) at the most exposed façade for day (07:00–19:00 h), evening (19:00–22:00 h), and night (22:00–07:00 h), and expressed as Lden, following penalties of 5 dB and 10 dB for noise occurring during the evening and night, respectively. For the Gothenburg cohorts, road and railway noise were estimated yearly, whereas for the Danish and Stockholm cohorts every fifth year, and for the Malmö cohort every 10th year. Noise levels for the years between those with estimates were calculated based on linear interpolation or other approximation methods (see Table S2).
All cohorts modeled road traffic and railway noise using the Nordic Prediction Method or an update of this method, Nord2000.20 For road traffic noise the input variables included geocodes, screening by terrain (except MDC) and buildings, and information on annual average daily traffic, distribution of light/heavy traffic, travel speed, and road type for all major road links. Furthermore, all cohorts but the Stockholm cohorts also included traffic information from minor roads (<1,000 vehicles per day), and the cohorts from Denmark and Gothenburg additionally included information on noise barriers. Ground absorption was considered in all estimations. Table S2 provides further details on the exposure assessment for road traffic noise.
Railway noise was calculated for all addresses within a 1,000m buffer around all railway tracks, and the methods used for the different cohorts are described in Table S2. Input variables included geocodes, screening by terrain (except MDC) and buildings, and average number of trains per period (day/evening/night), train types, and travel speed. In addition, cities with trams (Gothenburg and Stockholm) and/or metro (Stockholm and Copenhagen) included these in the calculations. Cohorts from Denmark and Gothenburg also used information on noise barriers. Ground absorption was considered in all estimations.
Aircraft noise was estimated in the Danish and Stockholm cohorts using noise maps obtained from local authorities and Swedavia, respectively (Table S2). For Malmö and Gothenburg, aircraft noise was not estimated due to very low numbers of exposed. In Denmark, modeled noise exposure from airports and airfields was obtained in 5 dB categories using the Danish Airport Noise Simulation Model and the Integrated Noise Model. For Stockholm, noise was estimated in 1 dB categories using the Integrated Noise Model 7.0.
Covariates
Selection of covariates was done a priori, based on existing literature and availability of harmonizable variables across cohorts. Cohort participants filled in questionnaires at recruitment with dietary and lifestyle variables, including smoking status (current, former, never), smoking intensity (among current smokers, grams per day; not available for the PPS cohort), alcohol consumption (daily, weekly, seldom, never; not available for the PPS cohort) and leisure-time physical activity (“low” as once a month or <1h per week, “medium” as about once a week or approximately 1 h per week, “high” as 3 times a week or more or >2 h per week), as well as weight and height. Information on educational level (“low” as primary school or less, “medium” as up to secondary school or equivalent, or “high” as university degree and more) and marital status (“single” as widowed or never married, or “married,” which also included those living with partner) was obtained from national registers or questionnaires, and area-level (small socioeconomically homogeneous areas with ∼1,000–2,000 inhabitants) mean income from registries, categorized in country-specific quartiles. Harmonization of covariate information between cohorts was usually achieved by relying on broad categories, such as current, former, and never for smoking. In other instances, the information was not comparable, e.g., with regard to area-level income, where country-specific distributions were used. Air pollution levels were estimated at all residential addresses during the study period using high-resolution dispersion models (see Table S3 for details). Air pollution exposure was represented by PM with a diameter <2.5μm (PM2.5), which is influenced by both long-range transport and local emissions and by nitrogen dioxide (NO2), primarily reflecting local emissions, such as from road traffic.21
Statistical Methods
Data from the nine cohorts were harmonized, checked and analyzed according to a common protocol. In pooled analyses we used Cox proportional hazards models, with age as underlying timescale, to calculate IHD hazard ratios (HRs) per 10 dB Lden higher levels of road and railway noise separately for IHD, IHD excluding angina pectoris, and myocardial infarction. Aircraft noise was estimated in 5-dB intervals in the Danish cohorts and analyzed only as a categorical variable. All road, railway, and aircraft noise values below 40 dB were set to 40 dB, due to imprecision of low-level noise estimates. Each cohort member was followed from the study baseline until the IHD outcome of interest or any other incident IHD outcome, death, emigration, loss to follow-up, or end of follow-up for the different cohorts (31 December 2011–31 December 2017; see Table S1), whichever occurred first. Exposure to noise was modeled as time-weighted means (energy-weighted) over 1- and 5-y periods preceding the IHD event, taking all addresses during these periods into account. The time periods for the estimation of air pollution exposure corresponded to those for transportation noise.
The proportional hazards assumption was tested by a correlation test between scaled Schoenfeld residuals and the rank order of event time. Deviation from the assumption was detected for sex, marital status, educational level, smoking, and physical activity, which were therefore included as strata (Table S4). All models were stratified by cohort, thereby allowing different baseline hazards across cohorts.
The association between transportation noise and IHD or IHD subtypes was analyzed in four predefined models with increasing adjustment: Model 1, adjusted for age (by design), cohort, sex, and calendar year (5 y categories); Model 2 with additional adjustment for educational status (low, medium, high), marital status (married/cohabiting, single); area-level income (percentage in national quartiles), and other transportation noise sources (binary indicator for each noise source with a cutoff at 45 dB Lden). In Model 3 we further included lifestyle factors: smoking status (never, former, current) and physical activity (low, medium, high), and in Model 4, we added time-weighted exposure to PM2.5 to Model 2. A priori, Model 2 was selected as the main model, and subjects lacking information on any of the covariates in this model were excluded in all analyses. Lifestyle factors such as smoking and physical activity were not included in the main model because they may be modified by transportation noise exposure and thus included in some causal pathways.22 Air pollution was not included in the main model, which was used in complete case analyses, primarily because we lacked information on PM2.5 and/or NO2 for a sizable fraction of the study subjects in two cohorts.
Exposure–response relationships for IHD and subtypes in relation to road traffic and railway noise exposure were explored using cubic splines with four degrees of freedom. The overall assumption of linearity was evaluated by comparing models with linear terms of noise exposure and with smoothed splines using a chi-square test and the Akaike information criterion (AIC) to assess the best fit. In case of departure from linearity, we evaluated the location of potential thresholds in the exposure–response function by comparing models with binary exposure indicators at different levels as an interaction term with exposure.
Effect modification of the association between road traffic noise and IHD was investigated in relation to the risk factors included in the adjustment models, i.e., age, sex, marital status, BMI, physical activity, smoking status, educational level, calendar year, PM2.5, and NO2, by incorporating an interaction term between the potential effect modifier and the 5 y mean noise exposure. The Wald test was used to calculate p-values of interaction. Sensitivity analyses comprised additional adjustment for BMI, alcohol consumption, and smoking intensity, inclusion of NO2 instead of PM2.5, fatal IHD as outcome, and exclusion of each of the three larger cohorts separately. Spearman’s rank correlation coefficients were used to evaluate relations between individual exposure to road, railway, and aircraft noise as well as to PM2.5 and NO2 during 5 y preceding the study baseline.
Analyses were performed in SAS (version 9.4; SAS Institute Inc.), Stata 14.2 (Stata Corp) and R (version 3.5.1; R Development Core Team).
Results
Overall, 149,894 persons were included in the original cohorts, but following exclusion of 5,186 with IHD before the study baseline, 981 with missing administrative or transportation noise exposure data, and 10,926 with missing covariate data included in the main model (Model 2) or Model 3, 132,801 persons remained for the pooled analysis, who were followed for 19.7 y on average (Table 1; Figure S1). A total of 22,459 incident cases of IHD were diagnosed during follow-up, including 7,682 cases of myocardial infarction. Angina pectoris accounted for more than 50% of the IHD cases in the Danish cohorts (DCH and DNS), whereas it was 40% or less in the Swedish cohorts. On the other hand, the fraction of fatal IHD cases was lower in the Danish cohorts. One of the cohorts was restricted to women (DNC) and another to men (PPS), whereas the other cohorts included both sexes. The median age at study baseline varied from 47.9 to 72.2 y, and the distribution of some risk factors also differed between the cohorts, such as for smoking, physical activity, educational level, and area-level income.
The median road traffic noise exposure at baseline ranged from 40.2 to 58.3 dB Lden in the different cohorts, with upper 95th percentiles from 54.2 to 72.4 dB Lden (Table 2). The distribution of road traffic noise exposure in the cohorts is further illustrated in Figure S2, showing that some cohorts contributed only marginally to exposures above 60 dB Lden, such as most cohorts from Stockholm. Overall, railway and aircraft noise exposure tended to be lower than road traffic noise exposure, with fewer exposed to 40 dB Lden and higher (Table 2). The PM2.5 levels showed a downward gradient from south to north with highest estimated concentrations in the Danish cohorts (the order of the cohorts in Table 2 generally is from south to north). NO2 levels tended to be higher in the cohorts based in urban areas (DCH, MDC, PPS, GOT-MONICA, and SNAC-K, Table S1). In general, there were low to moderate correlations between the different exposures; however, for road traffic noise exposure and NO2, the correlation coefficient was 0.62 (Table S5).
Table 2 Exposure to traffic noise and air pollution during 5 y prior to study baseline in nine cohorts from Scandinavia.
Characteristicsa Copenhagen or Aarhus, DK Denmark Malmö, SE Gothenburg, SE Stockholm, SE
DCH DNC MDC PPS GOT-MONICA SDPP SIXTY SNAC-K SALT Total
NordSOUND baseline 1993–1997 1993, 1999 1991–1996 1975 1985, 1990, 1995 1992–1998 1997–1999 2001–2004 1998–2002 1975–2002
Road traffic noise [5 y at 40 dB Lden and above (%)] 52,517 (99.2) 24,902 (95.3) 25,719 (96.8) 5,145 (98.3) 2,184 (94.6) 3,845 (51.0) 2,842 (75.4) 2,182 (96.9) 4,595 (75.9) 123,931 (93.3)
Road traffic noisea [5 y (dB Lden)] 56.6 (45.0–69.4) 53.6 (40.2–66.9) 54.6 (42.4–67.5) 57.5 (45.4–72.4) 56.1 (40.0–69.2) 40.2 (40.0–54.2) 46.6 (40.0–62.7) 58.3 (44.8–68.2) 47.3 (40.0–63.3) 54.5 (40.0–68.1)
Railway noise at 40 dB Lden and above (%) 13,638 (25.8) 5,008 (19.2) 7,438 (28.0) 788 (15.1) 460 (19.9) 1,093 (14.5) 1,196 (31.7) 1,170 (52.0) 2,063 (34.1) 32,854 (24.7)
Railway noisea,b [5 y (dB Lden)] 52.3 (42.1–66.6) 52.9 (41.8–65.9) 46.8 (40.6–67.7) 44.8 (40.4–58.8) 44.6 (40.3–57.7) 50.5 (42.2–67.3) 48.9 (40.8–63.6) 49.1 (41.6–59.5) 49.2 (40.7–64.1) 50.5 (41.0–66.0)
Aircraft noise at baselinec [5 y (%)]
≤40 dB Lden 52,215 (98.6) 25,816 (98.8) —c — — 5,638 (74.8) 3,146 (83.5) 364 (16.1) 5,037 (83.2) 126,333 (95.1)
40.1–50 dB Lden 351 (0.7) 102 (0.4) — — — 720 (9.6) 510 (13.5) 1,472 (65.4) 820 (13.5) 3,975 (3.0)
>50 dB Lden 382 (0.7) 208 (0.8) — — — 1,176 (15.6) 114 (3.0) 416 (18.6) 197 (3.3) 2,493 (1.9)
PM2.5a,d,e [5 y (μg/m3)] 19.2 (18.5–24.0) 20.7d (15.3–26.2) 11.0 (9.7–12.4) 9.7e (6.4–12.2) 9.9 (8.0–11.9) 7.6 (6.7–8.4) 8.0 (6.8–9.5) 8.3 (7.7–10.3) 7.7 (6.4–9.3) 18.7 (7.4–24.1)
NO2a,d,e [5 y (μg/m3)] 27.5 (19.6–46.5) 10.8d (5.6–28.0) 24.2 (13.7–34.9) 30.7e (22.9–43.5) 27.5 (18.4–42.4) 8.6 (4.9–13.9) 13.0 (5.3–26.2) 20.5 (15.0–32.2) 13.1 (5.5–25.6) 23.9 (7.0–40.7)
Note: —, no data; dB, decibel; DK, Denmark; DCH, the Danish Diet Cancer and Health cohort; DNC, the Danish Nurse Cohort; GOT-MONICA, the Gothenburg cohort of “Multinational Monitoring of Trends and Determinants in Cardiovascular Diseases” (MONICA) project, Sweden; Lden, day-evening-night noise level based on energy equivalent noise level over a whole day with a penalty of 10 dB for nighttime noise (23.00–7.00) and a penalty of 5 dB for evening noise (i.e., 19.00–23.00); MDC, the Malmö Diet and Cancer cohort, Sweden; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5 μm (fine particulate matter); PPS, the Primary Prevention Study cohort, Gothenburg, Sweden; NO2, nitrogen dioxide; SDPP, the Stockholm Diabetes Preventive Program, Sweden; SE, Sweden; SIXTY, the 60-Years Cohort, Stockholm, Sweden; SNAC-K, the Swedish National Study of Aging and Care in Kungsholmen, Stockholm, Sweden.
a Median and 5th–95th percentiles, unless otherwise stated.
b Calculated among those with railway noise exposure at 40 dB Lden and above.
c Only among cohorts with aircraft noise exposure (thus excluding MDC, PPS, and GOT-MONICA).
d Data on exposure to PM2.5 and NO2 missing for 9.2% in DNC.
e Data on exposure to PM2.5 and NO2 missing for 99.1% and 2.7%, respectively, in the PPS cohort. The high fraction of missing values for PM2.5 is explained by recruitment in 1970–1973 and start of PM2.5 modeling in 1990 (see Tables S1 and S3).
Table 3 shows HRs for IHD and subtypes in relation to estimated exposure 5 y prior to the event to noise from road traffic and railways, analyzed as continuous variables. Based on the main Model 2, a HR of 1.03 (95% CI: 1.00, 1.05) per 10 dB Lden was observed for both road traffic and railway noise exposure. For IHD excluding angina pectoris, the corresponding HRs were 1.06 (95% CI: 1.03, 1.08) and 1.05 (95% CI: 1.01, 1.08), respectively. Excess risks persisted after further adjustment for lifestyle factors (Model 3) or PM2.5 (Model 4), however, with some suggestion of positive confounding. Results for myocardial infarction generally pointed in the same direction but with weaker associations. Supplementary analyses focusing on angina pectoris showed HRs per 10 dB Lden based on Model 2 of 0.99 (95% CI: 0.96, 1.02) for road traffic noise and 1.01 (95% CI: 0.97, 1.04) for railway noise. Corresponding results for the category IHD excluding angina pectoris and myocardial infarction, primarily comprising “chronic IHD” (ICD9: 414 and ICD10: I25), were 1.12 (95% CI: 1.08, 1.17) and 1.05 (95% CI: 1.02, 1.07) for road and railway noise, respectively. For aircraft noise, increased HRs were noted for the IHD subgroups in Table 3 among those exposed at 40 dB Lden and above but without clear exposure–response relations. Results were quite similar for the different transportation noise sources and IHD groups when a 1-y exposure window was used instead of 5 y prior to the event (Table S6). Cohort-specific data focusing on IHD excluding angina pectoris suggested weaker associations for road traffic noise in the Stockholm cohorts, but a less coherent picture occurred for railway noise (Figure S3). Corresponding data for aircraft noise suggested some heterogeneity in exposure category specific risk estimates between cohorts, with the clearest exposure–response trend in the SDPP cohort (Figure S4).
Table 3 HRs for IHD, IHD excluding angina pectoris, and myocardial infarction in relation to transportation noise exposure during 5 y prior to the event in pooled data of nine Scandinavian cohorts (n=132,801).
Model 1a Model 2b Model 3c Model 4d
Exposure/outcome Cases Person-years HR (95% CI) HR (95% CI) HR (95% CI) HR (95% CI)
Road traffic noise, per 10 dB Lden
IHD 22,459 2,263,290 1.04 (1.02, 1.06) 1.03 (1.00, 1.05) 1.02 (0.99, 1.04) 1.02 (1.00, 1.04)
IHD excluding angina pectoris 12,399 2,263,290 1.09 (1.06, 1.11) 1.06 (1.03, 1.08) 1.05 (1.02, 1.07) 1.04 (1.01, 1.07)
Myocardial infarction 7,682 2,263,290 1.04 (1.00, 1.07) 1.02 (0.99, 1.05) 1.01 (0.97, 1.04) 1.00 (0.96, 1.04)
Railway noise, per 10 dB Lden
IHD 22,459 2,263,290 1.05 (1.02, 1.07) 1.03 (1.00, 1.05) 1.02 (1.00, 1.05) 1.03 (1.00, 1.05)
IHD excluding angina pectoris 12,399 2,263,290 1.07 (1.04, 1.10) 1.05 (1.01, 1.08) 1.04 (1.00, 1.07) 1.05 (1.01, 1.08)
Myocardial infarction 7,682 2,263,290 1.06 (1.01, 1.10) 1.04 (0.99, 1.08) 1.02 (0.98, 1.06) 1.04 (0.99, 1.08)
Aircraft noise (Lden)e
IHD
≤40 dB Lden 14,714 1,614,380 1 ref 1 ref 1 ref 1 ref
40.1–50 dB Lden 501 60,863 1.04 (0.94, 1.16) 1.06 (0.96, 1.18) 1.07 (0.96-1.19) 1.07 (0.96, 1.19)
>50 dB Lden 234 32,856 0.92 (0.80, 1.05) 0.95 (0.83, 1.09) 0.94 (0.82-1.08) 0.95 (0.83, 1.09)
IHD without angina pectoris
≤40 dB Lden 7,344 1,614,380 1 ref 1 ref 1 ref 1 ref
40.1–50 dB Lden 326 60,863 1.14 (1.00, 1.30) 1.17 (1.02, 1.34) 1.18 (1.03-1.35) 1.18 (1.02, 1.35)
>50 dB Lden 151 32,856 1.03 (0.87, 1.23) 1.09 (0.91, 1.29) 1.08 (0.91-1.28) 1.10 (0.92, 1.31)
Myocardial infarction
≤40 dB Lden 3,843 1,614,380 1 ref 1 ref 1 ref 1 ref
40.1–50 dB Lden 192 60,863 1.08 (0.90, 1.28) 1.11 (0.93, 1.32) 1.12 (0.94-1.33) 1.11 (0.93, 1.33)
>50 dB Lden 90 32,856 1.02 (0.82, 1.28) 1.06 (0.85, 1.33) 1.06 (0.84-1.32) 1.07 (0.86, 1.34)
Note: The HRs are estimated using Cox proportional hazards models with age as underlying timescale. BMI, body mass index; CI, confidence interval; HR, hazard ratio; IHD, ischemic heart disease; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5 μm (fine particulate matter); Lden, day-evening-night noise level based on energy equivalent noise level over a whole day with a penalty of 10 dB(A) for nighttime noise (23.00–7.00) and a penalty of 5 dB(A) for evening noise (i.e., 19.00–23.00); ref, reference.
aAdjusted for age (by design), cohort (strata), sex (men/women), and calendar year (in 5 y periods).
bModel 1 plus adjustment for educational level (low/medium/high), marital status (single/married), area-income (quartiles), and other noise sources (yes/no: road, railway, and aircraft noise indicators; for the three cohorts without aircraft noise information, all cohort members were assigned as no exposure).
cModel 2 plus adjustment for smoking status (current/former/never) and physical activity (low/medium/high).
dModel 2 plus adjustment for time-weighted PM2.5 exposure; 20,825 IHD, 11,245 IHD excluding angina pectoris, and 6,846 myocardial infarction cases during 2,120,816 person-years. Because the first year of PM2.5 modeling was 1990 but was 1975 for noise, relevant air pollution data for the model including both noise and PM2.5 were missing at baseline for most of the PPS cohort.
eOnly among cohorts with aircraft noise exposure (thus excluding MDC, PPS, and GOT-MONICA) with remaining 15,499 IHD cases, 7,821 IHD without angina pectoris cases, and 4,125 myocardial infarction cases.
Analyses based on noise as a continuous variable using cubic splines indicated particularly strong exposure–response trends for both road traffic and railway noise in the subgroup IHD excluding angina pectoris (Figure 1). Departure from linearity was indicated for the exposure–response function regarding road traffic noise and IHD by a better fit of the cubic spline than the linear model (p=0.02 and lower AIC) but not for the two subgroups. A threshold was suggested at 54 dB Lden with a HR and 95% CI based on the main model of 1.05 (95% CI: 1.02, 1.08) per 10 dB Lden for IHD at exposure above this level. Exposure–response relations were more uncertain for railway noise because fewer were exposed, particularly at higher levels. However, no apparent departure from linearity was suggested for any of the three outcome categories.
Figure 1. HR and 95% CI for IHD, IHD excluding angina pectoris, and myocardial infarction in relation to road traffic and railway noise exposure during 5 y prior to the event in restricted cubic spline analyses of nine cohorts from Denmark and Sweden. All results are adjusted for age (by design), cohort (strata), sex (men/women), calendar year (in 5 y periods), educational level (low/medium/high), marital status (single/married), area-income (quartiles), and other noise sources indicator (yes/no: road, railway, and aircraft noise; for the three cohorts without aircraft noise information, all cohort members were assigned as no exposure). Note: Corresponding numeric data are available in the Supplementary Excel file “Numeric data for Figures EHP10745.” CI, confidence interval; HR, hazard ratio; IHD, ischemic heart disease; Lden, day-evening night noise level based on energy equivalent noise level over a whole day with a penalty of 10 dB(A) for nighttime noise (23.00–7.00) and a penalty of 5 dB(A) for evening noise (i.e., 19.00–23.00).
Figure 1 is a set of six ribbon line graphs. The first two graphs, plotting hazard ratio for ischemic heart disease, ranging from 0.8 to 1.4 in increments of 0.1 (y-axis) across road traffic noise, day-evening-night noise level (decibel), ranging from 40 to 70 in increments of 10 and Railway noise, day-evening-night noise level (decibel), ranging from 40 to 70 in increments of 10 (x-axis). In the middle, the two graphs, plotting hazard ratio for ischemic heart disease excluding angina pectoris, ranging from 0.8 to 1.4 in increments of 0.1 (y-axis) across road traffic noise, day-evening-night noise level (decibel), ranging from 40 to 70 in increments of 10 and Railway noise, day-evening-night noise level (decibel), ranging from 40 to 70 in increments of 10 (x-axis). At the end, the two graphs, plotting hazard ratio for Myocardial infraction, ranging from 0.8 to 1.4 in increments of 0.1 (y-axis) across road traffic noise, day-evening-night noise level (decibel), ranging from 40 to 70 in increments of 10 and Railway noise, day-evening-night noise level (decibel), ranging from 40 to 70 in increments of 10 (x-axis).
Analyses of interactions focused on exposure to road traffic and railway noise in relation to HR of IHD excluding angina pectoris (Figure 2). For road traffic noise the strongest interaction was observed in relation to BMI, where those below 25 kg/m2 had a higher HR than those with BMI above this level (p-value for interaction <0.001). A BMI of 25 kg/m2 is commonly used to define overweight and is close to the overall median in our study population (cf. Table 1). There was also an interaction with BMI for railway noise. No consistent interactions were noted between other covariates and road traffic or railway noise. In particular, those with PM2.5 exposure above or equal to the median had a higher HR for IHD excluding angina pectoris related to road traffic noise exposure, whereas those with exposure below this level had a higher HR associated with exposure to railway noise. Interactions for all IHD largely confirmed the results for IHD excluding angina pectoris (Figure S5).
Figure 2. HR and 95% CI for ischemic heart disease excluding angina pectoris in relation to exposure to noise from road traffic (left) and railways (right) per 10 dB Lden during 5 y prior to the event according to covariates and air pollution exposure. p-Values are Wald pInteraction terms. Results are presented according to strata of potential effect modifiers based on separate models with interaction terms between transportation noise and each potential modifier, adjusted for age (by design), cohort (strata), sex (men/women), calendar year (in 5 y periods), educational level (low/medium/high), marital status (single/married), area-income (quartiles), and other noise sources indicator (yes/no: road, railway, and aircraft noise; for the three cohorts without aircraft noise information, all cohort members were assigned as no exposure). Note: Corresponding numeric data is available in the Supplementary Excel file “Numeric data for Figures EHP10745.” BMI, body mass index; CI, confidence interval; HR, hazard ratio; IHD, ischemic heart disease; Lden, day-evening-night noise level based on energy equivalent noise level over a whole day with a penalty of 10 dB(A) for nighttime noise (23.00–7.00) and a penalty of 5 dB(A) for evening noise (i.e. 19.00–23.00); NO2, nitrogen dioxide; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5μm (fine particulate matter).
Figure 2 is a set of two forest plots titled Road traffic noise and Railway noise, plotting subgroups with number of cases (bottom to top), air pollution nitrogen dioxide, including greater than or equal to median 19.8 milligrams per meter cubed with 5985 cases and less than median with 5628 cases; Air pollution particulate matter begin subscript 2.5 end subscript, including greater than or equal to median 13.2 milligrams per meter cubed with 5985 cases and less than median with 4814 cases and less than median with 6431 cases; Calendar year, including greater than or equal to 2005 with 7024 cases and less than 2005 with 5375 cases; Education level, including high with 1665 cases, medium with 5226 cases, and low with 5508 cases; Smoking, including current with 3541 cases, Former with 3749 cases, and never with 5109 cases; Physical activity during leisure time, including high with 2761 cases, medium with 3579 cases, and low with 6059 cases; Body mass index, including greater than or equal to 25 with 7369 cases and less than 25 with 5030 cases; Age, years, including greater than or equal to 9415 cases and less than 65 with 2984 cases; marital status, including single with 8774 and married with 3625 cases; and sex, including men with 7063 cases and women with 5336 cases (y-axis) across transportation noise and each potential modifier, ranging from 0.95 to 1.15 in increments of 0.05 and 0.95 to 1.2 in increments of 0.05 (x-axis) for hazard ratio (95 percent confidence intervals) and uppercase p of interaction, respectively.
Sensitivity analyses focusing on IHD excluding angina pectoris are presented in Figure S6. Further adjustment for covariates had limited effects on HRs for road traffic and railway noise, with the exception that adjustment for NO2 instead of PM2.5 tended to weaken the association for road traffic noise. Exposure to both road traffic and railway noise was related to an increased risk of fatal IHD excluding angina pectoris. We also calculated HRs for fatal IHD, which were 1.06 (95% CI: 1.01, 1.12) and 1.08 (95% CI: 1.02, 1.15) per 10 dB Lden for road traffic and railway noise, respectively, as well as for fatal myocardial infarction with corresponding risk estimates of 1.02 (95% CI: 0.95, 1.09) and 1.08 (95% CI: 0.99, 1.17). In analyses sequentially leaving out one of the three larger cohorts, exclusion of the Danish DCH cohort tended to reduce the HR for road traffic noise, and exclusion of the Swedish MDC cohort had the same effect for railway noise (Figure S6). The influence was only marginal following exclusions of other cohorts.
Discussion
In this large study based on Danish and Swedish cohorts we observed increased risks of IHD associated with long-term exposure to road traffic and railway noise. Higher risks were indicated following exclusion of angina pectoris cases. Associations were also observed for aircraft noise but without clear exposure–response relations. Particularly strong associations appeared for both road traffic and railway noise among those with BMI below 25 kg/m2.
We found a HR for IHD of 1.03 (95% CI: 1.00, 1.05) per 10 dB Lden of exposure to road traffic noise. A recent meta-analysis calculated an overall relative risk for IHD of 1.02 (95% CI: 1.00, 1.04) per 10 dB Lden,23 which is consistent with our estimate. The meta-analysis was based on 16 studies, including six of the cohorts in our study, but mostly with shorter follow-up and a mixture of outcomes (IHD/myocardial infarction). Our results suggest that exposure to road traffic noise below around 55 dB Lden is not associated with an increased risk of IHD. This finding is somewhat different from results regarding stroke in the same study population, where no such “threshold” was observed.17 Our risk estimate of 1.05 (95% CI: 1.02, 1.08) for IHD per 10 dB Lden at exposures above 54 dB Lden appears comparable to the estimate of 1.08 (95% CI: 1.02, 1.15) based on a meta-analysis in the systematic review for the WHO Environmental Noise Guidelines, where the weighted lower exposure level across studies was 53 dB Lden.3 One cohort (DCH) was included in both our analysis and the WHO meta-analysis; however, it contributed only 1,600 cases of myocardial infarction to the meta-analysis based on a shorter follow-up.24 Furthermore, two recent studies suggested thresholds in the exposure–response relation for road traffic noise and incidence of IHD8 or myocardial infarction.11 A threshold in the exposure–response relation could contribute to explaining the absence of association between road traffic noise and IHD in the cohorts from Stockholm in our study, where three out of four had comparatively low exposures, with few exposed over 60 dB Lden. Assessment of exposure–response relationships is crucial for health impact assessments. In particular, the shape of the exposure–response function at low exposure levels is influential, where most of the population is exposed. There is a need for further studies assessing exposure–response relations at low but common levels of traffic noise exposure.
We observed an increased risk of IHD related to exposure to railway noise, particularly following exclusion of angina pectoris cases. Besides two studies on cohorts included in our analysis,7,9 only a pair of longitudinal studies strictly based on registry data analyzed risks of myocardial infarction25 or mortality of IHD and myocardial infarction6 in relation to railway noise exposure. The risk estimates in these studies were 1.023 (95% CI: 1.005, 1.042), 1.012 (95% CI: 1.005, 1.020) and 1.020 (95% CI: 1.007, 1.033), respectively, per 10 dB Lden of exposure to railway noise. These estimates appear lower than the corresponding estimates in our study of 1.04 (95% CI: 0.99, 1.05) for incidence of myocardial infarction and 1.08 (95% CI: 1.02, 1.15) for fatal IHD and 1.08 (95% CI: 0.99, 1.17) for fatal myocardial infarction; however, the wide confidence intervals complicate a detailed quantitative comparison. Railway noise is a relatively rare exposure, which means that population-based epidemiological studies must be quite large to achieve a reasonable study power. Alternatively, studies may be conducted in areas where railway noise exposure is more common. For example, the cohort with the fewest participants in our study (SNAC-K), which had a considerably higher fraction of people exposed to railway noise than any of the other cohorts, showed an increased risk of IHD excluding angina pectoris related to railway noise exposure (Figure S3).
Our results indicated an association between aircraft noise and IHD, particularly when angina pectoris cases were excluded, but without an exposure–response relation. The clearest exposure–response trend was suggested in the SDPP cohort, with many subjects living near the major international airport in Stockholm, which also has nighttime traffic (Figure S4).26 There are few other studies on aircraft noise and IHD or myocardial infarction,6,23 including some of ecological design or strictly based on registry data. Although some of the studies showed increased risks related to aircraft noise exposure, no consistent picture emerged. A recent study suggests that short-term aircraft noise exposure may be of importance for the cardiovascular risks.27 Population studies on aircraft noise exposure are complicated by the fact that houses are more often noise insulated in the most heavily exposed areas near the airports. Such insulation may have affected the exposure–response relation in our study, but we lacked information on noise insulation of individual dwellings.
Knowing whether subtypes of IHD are differentially related to transportation noise exposure is important for the risk assessment and understanding of etiological mechanisms. We found consistent associations between road traffic noise exposure and IHD for different adjustment models in the pooled analyses, particularly when those with the diagnosis based on angina pectoris were excluded. It is compelling that the stronger association for IHD without angina pectoris was confirmed also for railway and aircraft noise, although exposure to the three noise sources showed low correlation. Furthermore, the risks of fatal IHD, which is generally not based on angina pectoris, appeared higher than those for incident IHD, where angina pectoris is more prominent. Angina pectoris constituted a larger part of IHD in the Danish studies, partly because outpatient data were included already from 1995 in the Danish National Patient Register but only from 2001 in the Swedish National Patient Register. We found no association between road or railway noise exposure and angina pectoris, which is less well captured in the national registries than other IHD diagnoses,28 but a particularly strong association between road traffic noise and chronic IHD. Traffic noise exposure has been linked to several cardiovascular outcomes, such as IHD/myocardial infarction, stroke, heart failure, and atrial fibrillation,18 but we are not aware of earlier studies on risks for angina pectoris or chronic IHD.
We observed an interaction in relation to IHD risk between road traffic noise and BMI, with a stronger association among study participants having a BMI below 25 kg/m2, which was similar for railway noise. No comparable interaction between road traffic noise exposure and BMI in relation to stroke was noted in the same study base.17 Several cohort studies have found an association between traffic noise exposure and obesity markers.26,29–32 This finding opens the possibility that BMI or other obesity markers may be mediators of the association between traffic noise and IHD; however, our results suggest that there are causal pathways not involving BMI. It is not clear why excess risks related to traffic noise exposure would be higher among those with lower BMI, but it may be speculated that other risk factors could be more influential for subjects with higher BMI. A similar absolute risk increase related to noise exposure in both BMI categories would also tend to generate a lower rate ratio among those with high BMI because of the higher IHD incidence in this group. We are not aware of other studies on noise exposure and IHD assessing interactions with BMI, and our results should be interpreted with caution in view of the exploratory nature of these analyses.
Our results did not point to effect modification by other covariates, which was consistent between road traffic and railway noise exposure. In particular, those with PM2.5 exposure above or equal to the median had a higher risk of IHD and of IHD excluding angina pectoris related to road traffic noise, whereas those with exposure below this level had a higher risk associated with exposure to railway noise. The reasons behind this apparent inconsistency are unclear, but one contributing factor is the low PM2.5 levels in the Stockholm cohorts, where the association between road traffic noise and IHD risk was weak. On the other hand, the association between railway noise and IHD was weak in the Danish cohorts, particularly for IHD excluding angina pectoris, where the PM2.5 levels were higher. Several studies on cardiovascular disease have estimated exposure to both road traffic noise and air pollution;4,6–11,13–17 however, no clear picture of interaction has emerged for IHD or myocardial infarction or for other cardiovascular end points.
Our study has several strengths. It is the largest study to date on transportation noise and IHD containing detailed information on socioeconomic and lifestyle risk factors, enabling a high statistical power and careful confounding control also in subgroup analyses. The study constitutes a major extension of earlier publications based on the cohorts, including longer follow-up and substantially increased number of cases, which together with pooled analyses, enables detailed evaluation of the shape of exposure–response functions and interactions, also for subtypes of IHD. We estimated transportation noise and air pollutant levels at residential addresses over time, using validated methods with very high spatial resolution. The study included the five major metropolitan areas in Denmark and Sweden, as well as less urbanized regions, contributing to a substantial contrast in traffic noise exposure. However, the exposure assessment still has uncertainties because only residential addresses were considered and no information was available on indoor noise levels, which depend on façade sound insulation of the building as well as to what extent windows are open or closed. Furthermore, the results may have been influenced by calendar year because exposure was assessed during a period of more than five decades, with differences between cohorts, and changes in noise exposure patterns occurred during that period. The generalizability of our findings may be affected by different building techniques, regulations for noise insulation, and behavioral characteristics in Scandinavian countries and in other countries.33 The health outcome information was obtained from high quality national patient and mortality registers; however, some diagnoses are less well captured in such registers, including angina pectoris.27 Residual confounding must also be considered in the interpretation, particularly because positive confounding was often indicated when adjusting for covariates, such as air pollution. Including data from several cohorts and countries in the analyses has advantages; however, interpretation of results from combined analyses may be complex if cohorts differ substantially in exposures, background risks of IHD, and distribution of covariates. Our results suggest that differences in risk estimates between cohorts to some extent were related to noise exposure distributions, but specifics in the exposure assessment methodologies may also have contributed.
In conclusion, our results showed that long-term exposure to road traffic noise and railway noise was associated with an increased incidence of IHD, especially for those with a diagnosis not based on angina pectoris. A threshold of around 55 dB Lden was suggested in the exposure–response relation for road traffic noise and IHD. Excess risks related to aircraft noise were also observed but without clear exposure–response relations. Our findings underscore the importance of transportation noise as a public health problem and highlight several aspects that can contribute to the understanding of causal pathways and influence the population attributable risks of IHD related to traffic noise exposure.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by NordForsk [Grant No. 83597]. Funding for the included cohorts is shown in Table S1.
==== Refs
References
1. European Environment Agency. Environmental Noise in Europe—2020. EEA Report No 22. 2020. https://anima-project.eu/fileadmin/user_upload/Eulalia_Peris_%E2%80%93ANIMA_Noise_In_Europe_Report.pdf [accessed 5 October 2021].
2. World Health Organization. Environmental Noise Guidelines for the European Region. 2018. Copenhagen, Denmark: WHO Regional Office for Europe. https://apps.who.int/iris/bitstream/handle/10665/279952/9789289053563-eng.pdf [accessed 5 October 2021].
3. van Kempen E, Casas M, Pershagen G, Foraster M. 2018. WHO environmental noise guidelines for the European region: a systematic review on environmental noise and cardiovascular and metabolic effects: a summary. Int J Environ Res Public Health 15 (2 ):379, PMID: , 10.3390/ijerph15020379.29470452
4. Bodin T, Björk J, Mattisson K, Bottai M, Rittner R, Gustavsson P, et al. 2016. Road traffic noise, air pollution and myocardial infarction: a prospective cohort study. Int Arch Occup Environ Health 89 (5 ):793–802, PMID: , 10.1007/s00420-016-1115-9.26867595
5. Dimakopoulou K, Koutentakis K, Papageorgiou I, Kasdagli M-I, Haralabidis AS, Sourtzi P, et al. 2017. Is aircraft noise exposure associated with cardiovascular disease and hypertension? Results from a cohort study in Athens, Greece. Occup Environ Med 74 (11 ):830–837, PMID: , 10.1136/oemed-2016-104180.28611191
6. Héritier H, Vienneau D, Foraster M, Eze IC, Schaffner E, Thiesse L, et al. 2018. Diurnal variability of transportation noise exposure and cardiovascular mortality: a nationwide cohort study from Switzerland. Int J Hyg Environ Health 221 (3 ):556–563, PMID: , 10.1016/j.ijheh.2018.02.005.29482991
7. Pyko A, Andersson N, Eriksson C, de Faire U, Lind T, Mitkovskaya N, et al. 2019. Long-term transportation noise exposure and incidence of ischaemic heart disease and stroke: a cohort study. Occup Environ Med 76 (4 ):201–207, PMID: , 10.1136/oemed-2018-105333.30804165
8. Andersson EM, Ögren M, Molnár P, Segersson D, Rosengren A, Stockfelt L. 2020. Road traffic noise, air pollution and cardiovascular events in a Swedish cohort. Environ Res 185 :109446, PMID: , 10.1016/j.envres.2020.109446.32278155
9. Roswall N, Raaschou-Nielsen O, Ketzel M, Gammelmark A, Overvad K, Olsen A, et al. 2017. Long-term residential road traffic noise and NO2 exposure in relation to risk of incident myocardial infarction–A Danish cohort study. Environ Res 156 :80–86, PMID: , 10.1016/j.envres.2017.03.019.28334645
10. Thacher JD, Hvidtfeldt UA, Poulsen AH, Raaschou-Nielsen O, Ketzel M, Brandt J, et al. 2020. Long-term residential road traffic noise and mortality in a Danish cohort. Environ Res 187 :109633, PMID: , 10.1016/j.envres.2020.109633.32442789
11. Lim Y-H, Jørgensen JT, So R, Cramer J, Amini H, Mehta A, et al. 2021. Long-term exposure to road traffic noise and incident myocardial infarction: a Danish nurse cohort study. Environ Epidemiol 5 (3 ):e148, PMID: , 10.1097/EE9.0000000000000148.33912785
12. Münzel T, Sørensen M, Gori T, Schmidt FP, Rao X, Brook J, et al. 2017. Environmental stressors and cardio-metabolic disease: part I-epidemiologic evidence supporting a role for noise and air pollution and effects of mitigation strategies. Eur Heart J 38 (8 ):550–556, PMID: , 10.1093/eurheartj/ehw269.27460892
13. Beelen R, Hoek G, Houthuijs D, van den Brandt PA, Goldbohm RA, Fischer P, et al. 2009. The joint association of air pollution and noise from road traffic with cardiovascular mortality in a cohort study. Occup Environ Med 66 (4 ):243–250, PMID: , 10.1136/oem.2008.042358.19017691
14. Selander J, Nilsson ME, Bluhm G, Rosenlund M, Lindqvist M, Nise G, et al. 2009. Long-term exposure to road traffic noise and myocardial infarction. Epidemiology 20 (2 ):272–279, PMID: , 10.1097/EDE.0b013e31819463bd.19116496
15. Sørensen M, Lühdorf P, Ketzel M, Andersen ZJ, Tjønneland A, Overvad K, et al. 2014. Combined effects of road traffic noise and ambient air pollution in relation to risk for stroke? Environ Res 133 :49–55, PMID: , 10.1016/j.envres.2014.05.011.24906068
16. Sørensen M, Wendelboe Nielsen O, Sajadieh A, Ketzel M, Tjønneland A, Overvad K, et al. 2017. Long-term exposure to road traffic noise and nitrogen dioxide and risk of heart failure: a cohort study. Environ Health Perspect 125 (9 ):097021, PMID: , 10.1289/EHP1272.28953453
17. Roswall N, Pyko A, Ögren M, Oudin A, Rosengren A, Lager A, et al. 2021. Long-term exposure to transportation noise and risk of incident stroke: a pooled study of nine Scandinavian cohorts. Environ Health Perspect 129 (10 ):107002, PMID: , 10.1289/EHP8949.34605674
18. Münzel T, Sørensen M, Daiber A. 2021. Transportation noise pollution and cardiovascular disease. Nat Rev Cardiol 18 (9 ):619–636, PMID: , 10.1038/s41569-021-00532-5.33790462
19. Korek MJ, Bellander TD, Lind T, Bottai M, Eneroth KM, Caracciolo B, et al. 2015. Traffic-related air pollution exposure and incidence of stroke in four cohorts from Stockholm. J Expo Sci Environ Epidemiol 25 (5 ):517–523, PMID: , 10.1038/jes.2015.22.25827311
20. Bendtsen H. 1999. The Nordic prediction method for road traffic noise. Sci Total Environ 235 (1–3 ):331–338, 10.1016/S0048-9697(99)00216-8.
21. de Hoogh K, Chen J, Gulliver J, Hoffmann B, Hertel O, Ketzel M, et al. 2018. Spatial PM2.5, NO2, O3 and BC models for Western Europe – evaluation of spatiotemporal stability. Environ Int 120 :81–92, PMID: , 10.1016/j.envint.2018.07.036.30075373
22. Roswall N, Christensen JS, Bidstrup PE, Raaschou-Nielsen O, Jensen SS, Tjønneland A, et al. 2018. Associations between residential traffic noise exposure and smoking habits and alcohol consumption-A population-based study. Environ Pollut 236 :983–991, PMID: , 10.1016/j.envpol.2017.10.093.29122366
23. Vienneau D, Eze IC, Probst-Hensch N, Rȍȍsli M. Association between transportation noise and cardio-metabolic diseases: an update of the WHO meta-analysis. Proceedings of the 23rd International Congress on Acoustics. Aachen, Germany. 9–13 September 2019. https://edoc.unibas.ch/70857/1/ICA_2019_manuscript%20Vienneau%20final.pdf [accessed 29 November 2021].
24. Sørensen M, Andersen ZJ, Nordsborg RB, Jensen SS, Lillelund KG, Beelen R, et al. 2012. Road traffic noise and incident myocardial infarction: a prospective cohort study. PLoS One 7 (6 ):e39283, PMID: , 10.1371/journal.pone.0039283.22745727
25. Seidler A, Wagner M, Schubert M, Dröge P, Pons-Kühnemann J, Swart E, et al. 2016. Myocardial infarction risk due to aircraft, road, and rail traffic noise. Dtsch Arztebl Int 113 (24 ):407–414, PMID: , 10.3238/arztebl.2016.0407.27380755
26. Eriksson C, Hilding A, Pyko A, Bluhm G, Pershagen G, Östenson C-G. 2014. Long-term aircraft noise exposure and body mass index, waist circumference, and type 2 diabetes: a prospective study. Environ Health Perspect 122 (7 ):687–694, PMID: , 10.1289/ehp.1307115.24800763
27. Saucy A, Schäffer B, Tangermann L, Vienneau D, Wunderli JM, Röösli M. 2021. Does night-time aircraft noise trigger mortality? A case-crossover study on 24 886 cardiovascular deaths. Eur Heart J 42 (8 ):835–843, PMID: , 10.1093/eurheartj/ehaa957.33245107
28. Ludvigsson JF, Andersson E, Ekbom A, Feychting M, Kim J-L, Reuterwall C, et al. 2011. External review and validation of the swedish national inpatient register. BMC Public Health 11 (1 ):450, PMID: , 10.1186/1471-2458-11-450.21658213
29. Christensen JS, Raaschou-Nielsen O, Tjønneland A, Nordsborg RB, Jensen SS, Sørensen TIA, et al. 2015. Long-term exposure to residential traffic noise and changes in body weight and waist circumference: a cohort study. Environ Res 143 (Part A ):154–161, PMID: , 10.1016/j.envres.2015.10.007.26492400
30. Pyko A, Eriksson C, Lind T, Mitkovskaya N, Wallas A, Ögren M, et al. 2017. Long-term exposure to transportation noise in relation to development of obesity—a cohort study. Environ Health Perspect 125 (11 ):117005, PMID: , 10.1289/EHP1910.29161230
31. Foraster M, Eze IC, Vienneau D, Schaffner E, Jeong A, Héritier H, et al. 2018. Long-term exposure to transportation noise and its association with adiposity markers and development of obesity. Environ Int 121 (Part 1 ):879–889, PMID: , 10.1016/j.envint.2018.09.057.30347370
32. Cai Y, Zijlema WL, Sørgjerd EP, Doiron D, de Hoogh K, Hodgson S, et al. 2020. Impact of road traffic noise on obesity measures: observational study of three European cohorts. Environ Res 191 :110013, PMID: , 10.1016/j.envres.2020.110013.32805247
33. Rasmussen B. Building acoustic regulations in Europe–brief history and actual situation. In: Baltic-Nordic Acoustics Meeting 2018. 15–18 April 2018. Rekjavik, Iceland. Nordic Acoustic Association. https://events.artegis.com/urlhost/artegis/customers/1571/.lwtemplates/layout/default/events_public/12612/Papers/Keynote_Rasmussen_BNAM2018.pdf [accessed 5 October 2021].
|
PMC009xxxxxx/PMC9819285.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36607287
EHP11618
10.1289/EHP11618
Focus
Paradox Lost? The Waning Health Advantage among the U.S. Hispanic Population
Nicole Wendee
6 1 2023
1 2023
131 1 01200125 5 2022
01 7 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
United States census form with Question 6 (“Is this person of Hispanic, Latino, or Spanish origin?”) highlighted.
==== Body
pmcFor decades epidemiologists have puzzled over an intriguing phenomenon: Hispanic Americans are healthier than might be expected, given their having, on average, lower socioeconomic status (SES) than non-Hispanic White populations.1,2 Low SES is almost unequivocally associated with worse population health and higher death rates around the globe.3 Yet studies have found that people of Hispanic origin living in the United States have better health outcomes than other racial and ethnic groups, including non-Hispanic White populations, for cardiovascular disease,4 various cancers,5 infant mortality,6 stroke,7 bronchitis,7,8 and emphysema.7 They also live an average of 3–4 years longer than non-Hispanic White populations.9
Often called the Hispanic paradox in the scientific literature, the reasons behind this population’s health and longevity advantages have remained the subject of intense research—and curiosity—for decades; in 2013, a group of authors observed, “[T]he current status of the Hispanic mortality paradox can best be described as one of great interest with significant logistical confusion.”10
Although often grouped under the catchall heading “Hispanic,” Latino immigrants—those from countries throughout Central America, South America, and the Caribbean—have very different cultural backgrounds. This reality is reflected on the current U.S. census form, which allows Hispanic respondents to identify as Mexican/Mexican American/Chicano; Puerto Rican; Cuban; or “other Hispanic, Latino, or Spanish origin.” Those who respond “other” are then asked to write in their country of origin. Acknowledging how complex a person’s origin and self-identity can be, the Census Bureau expanded this write-in field from 30 characters to 200 for the 2020 census.67 Image: Public domain via Environmental Health Perspectives.
United States census form with Question 6 (“Is this person of Hispanic, Latino, or Spanish origin?”) highlighted
Even as researchers are making progress on explanations for the paradox, their findings indicate that the health advantage is waning in Hispanic communities. Why is this so?
What Causes the Immigrant Advantage?
Studies have documented similar health advantages with African11 and Italian12 immigrants to the United States, Turkish immigrants to Germany,13 and immigrants from several regions to Canada.14 Most of the work on such advantages, however, has focused on U.S. immigrants from Mexico and Central and South America. Researchers have extensively studied the phenomenon in different generations of immigrants; approximately one-third of the U.S. Hispanic population is foreign-born.15
The selective migration (or healthy migrant) advantage is one of the most well-supported hypotheses behind the paradox.16–18 Beatriz Tapia, assistant dean of faculty development at the University of Texas Rio Grande Valley Medical School, explains that Hispanic immigrants typically fall into one of two groups. Immigrants with visas tend to have higher education and income levels and are more likely to return to their home country. Those without visas, on the other hand, tend to be both hardy enough to take on hard jobs and invested in the goal of creating a better life in the United States.19 “People [who] migrate are going to be stronger, healthier, and younger,” Tapia says. “Sick people either won’t come or they’ll die trying.”
First-generation immigrants—those who themselves came to the United States—are generally healthiest when they first arrive. By the second generation, their health status and mortality decline to rates that are much closer to U.S.-born individuals.19 One possible explanation is that toward the end of their lives, many older immigrants return to their home country, which reduces Hispanic mortality statistics in the United States.20 However, this explanation does not completely explain the mortality advantage of U.S. Hispanic populations.20,21
In research published in 2013,22 Andrew Fenelon, an assistant professor of public policy and sociology at Pennsylvania State University, identified a single factor that may explain a substantial amount of the Hispanic health advantage: smoking, or the lack thereof. “A lot of Latino immigrants to the United States, especially folks from Mexico, have very low rates of cigarette smoking, compared to native-born U.S. White individuals,” he explains. Those who smoke tend to smoke less, even after arriving in the United States. Fenelon posits that this could be due to differences between immigrants’ home and U.S. cultures, or even to the relative costliness of cigarettes in Mexico.22 In his 2013 analysis,22 which was based on data from the National Health Interview Survey (NHIS), Fenelon estimated that at least half, and in some cases more than 70%, of the mortality advantage among Mexican Americans could be attributed to differential smoking levels between them and non-Hispanic White people.
Another major explanation proposed for the mortality advantage involves familism, or strong social and family networks, which is a core value for many Hispanic families.23–25 According to Karina Corona, a postdoctoral scholar at the University of Southern California (USC), people who highly value familism tend to prioritize family over self. “Familism emphasizes the strong attachment that people build with each other,” she says, “how they identify not only with the nuclear family, but also with extended family.”
Research supports the protective role of familism in Hispanic immigrant health. “In a new place, people will try to recreate that family feeling even if family members are not near,” says environmental health professor Ana Navas-Acien. “You can easily see it in New York City through the public parks, how large ‘families’ meet every weekend just to spend time together.” Image: © Getty Images/Al Bello.
Women and men cooking food in a park
Corona found in one study that Hispanic Americans with high levels of familism reported better physical health, less loneliness, and less depression.26 Interestingly, the same study found Asian Americans and European Americans also saw a stress-buffering effect of familism, although familism was overall more prevalent among Asian Americans than European Americans. A separate study by Corona found that Hispanic mothers and daughters with high stress levels but also high levels of familism reported better self-esteem and subjective health compared with those who reported high stress but low familism.27
“Hispanic people are not experiencing less stress than others,” says Fernando Riosmena, director of the Institute for Health Disparities Research at the University of Texas at San Antonio. “However, they seem to be protected from that stress to some extent by living in places closer to family and friends.”28
Living in gateway cities—those with established immigrant populations—has also been thought to contribute to Hispanic populations’ health and mortality advantage,29 particularly via familism. However, new research casts doubt on this hypothesis.29 Using NHIS data on Mexican immigrants specifically, Fenelon and colleagues found the mortality advantage was stronger in destinations with smaller enclaves of Mexican immigrants or areas that had more recently begun to receive these populations. One reason, he surmises, is that smaller, newer immigrant communities probably have fewer generations that have adopted less healthful U.S. dietary and lifestyle choices, compared with more established communities in gateway cities.
It is important to note that some destinations may be new for one nationality but long-time gateways for others. In Miami, for example, the Hispanic population is more than half Cuban, with relatively few Mexicans, whereas in Los Angeles–Long Beach the vast majority is Mexican; other cities have majorities of Puerto Ricans, Salvadorans, or other nationalities.30 These distinctions emphasize the need for epidemiologists to break down the monolithic “Hispanic” population designation into finer categories, such as national origin, length of time in the United States, and age.
“It’s very important to make the distinction that this is not a single group,” says Ana Navas-Acien, a professor of environmental health at Columbia University. She says Hispanics tend to identify strongly with their country of origin, which affects culture, diet, and other aspects of their lives in the United States. Indeed, quite striking disparities in health outcomes have been documented between Hispanic immigrants with different birth countries, age, and immigration experience.31
Research on Hispanic Populations
As with most other immigrant groups,32 U.S. Hispanic populations overall have less access to health care, more economic disadvantage, and more severe environmental and occupational exposures, compared with non-Hispanic Whites.33–35 The authors of a recent review published in Environmental Health Perspectives noted that environmental health is an understudied area among immigrants, with little research beyond the documentation of exposure disparities.32
This girl, dressed as a popular character from the carnaval tradition of Baranquilla, Colombia, is participating in Feria de las Flores (Flower Festival) in New York City. This festival is celebrated each year by the Colombian community in Queens, recreating an iconic event of the same name in Medellín, Colombia. Although immigrant communities often have higher-than-average exposures to environmental pollutants, there is evidence that strong cultural identity has a protective effect on health. Image: © Kike Calvo/AP Images.
Child in a costume sitting on the back of a car at a festival
Now, several important cohort studies are investigating how environmental exposures disproportionately harm Hispanic populations. The Mexican Immigration to California: Agricultural Safety and Acculturation (MICASA) prospective study investigates health in Hispanic farmworkers ranging from respiratory health36 to injuries.37 The Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS)38 longitudinal study is following children of Mexican farmworkers and their environmental exposures. The Maternal and Developmental Risks from Environmental and Social Stressors (MADRES) longitudinal study of more than 900 lower-income Hispanic women and their children is looking at key maternal health and child health outcomes through 5 years of age.
“Despite [Hispanic populations] having lower overall mortality, there are a number of adverse health outcomes that are among the highest in this group, especially among children,”39 says Tracy Bastain, director of the MADRES Center for Environmental Health Disparities at USC. “Our center was initiated on the premise that air pollution exposures are disproportionately higher in communities with predominantly Hispanic/Latino residents and that these residents also have the highest rates of childhood obesity.”
Air pollution and other environmental exposures are also being addressed in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL), which examines the prevalence and development primarily of cardiovascular disease in Hispanic people, the role of acculturation, and risk factors that play protective or harmful roles. One HCHS/SOL ancillary project, SOLAir, involves understanding how air pollution affects diabetes risk in the populations of four locations: Miami, San Diego, Chicago, and the Bronx; another ancillary study focuses on how greenspace affects diabetes risk in San Diego and Miami.
Two other studies under HCHS/SOL are examining how greenspace and sedentary behaviors influence metabolic health. “These studies enhance the environmental assessments of the cohort,” says Carmen Isasi, coprincipal investigator at the HCHS/SOL Bronx field office and chair of the HCHS/SOL steering committee. “Together with rich assessments of social determinants of health, we will be able to address environmental influences from a broad perspective, that includes structural social factors, air pollution, and the built environment.” This comprehensive approach, she says, is critical to understand health inequities.
In addition to air pollution, Hispanic communities—along with other people of color—are also more likely to be exposed to drinking water contaminants, such as uranium, chromium, and arsenic.40–42 Both cadmium and arsenic have been linked to kidney disease,43 a health condition for which Hispanic individuals (particularly those from Puerto Rico44) have worse outcomes than non-Hispanic White people.45,46 Navas-Acien, along with lead author Anne Nigra, an assistant professor at Columbia University, recently showed that U.S. community water systems serving semi-urban areas with higher Hispanic populations were the most likely to exceed the U.S. Environmental Protection Agency maximum contaminant level for arsenic, even as other systems were actively working to reduce arsenic levels.42 In another study, their team found that systems serving semi-urban Hispanic communities were among those most likely to exceed the maximum contaminant levels for uranium, chromium, barium, and selenium.41 Their latest study, which examined metals in drinking water at the county population level, further demonstrates the exposure disparities that Hispanic communities face.40
Other harmful exposures can come with the jobs that Hispanic workers often take on. Physically demanding work with long hours, including agriculture and construction, is often accompanied by exposures to extreme heat, pesticides, and disease vectors, such as ticks and mosquitoes.47,48 Hispanic women comprise a large part of the housekeeping workforce, which exposes them to hazardous cleaning chemicals for long periods of time.47–49 These physical demands and harmful exposures may partly explain why, in the United States, Hispanic individuals experience higher levels of physical disability50,51 and likelihood of dying on the job52,53 than non-Hispanic White workers.
Employment in frontline jobs with little freedom to isolate or take time off meant that Hispanic workers bore a high burden of death and illness during the COVID-19 pandemic. Images, clockwise from top right: © iStock.com/davit85; iStock.com/Ivan Aaragon Alonso; © iStock.com/andresr; © iStock.com/gorodenkoff.
Images of four workers: a health care aide, a poultry plant worker, a food delivery driver, and a hospital kitchen worker
However, few studies have directly connected their findings with the Hispanic paradox paradigm, and the area is ripe for dedicated research. For example, if Hispanic populations had the environmental advantages of non-Hispanic White populations, what would their health status look like? By reducing harmful environmental exposures, would later generations of U.S. Hispanic residents live longer, like their immigrant parents and grandparents? Because so many of the health conditions for which Hispanic populations have an advantage over non-Hispanic White populations—certain cancers,54 cardiovascular disease,55 and infant mortality56—have strong environmental health components, where is their advantage coming from, and can it be extended instead of diminished?
Paradox Lost?
Some researchers are projecting that the Hispanic paradox may become a thing of the past. For example, work conducted by USC postdoctoral scholar Luis Maldonado57 using nationally representative data from 1988 to 1994 suggested that children born in Mexico had much lower obesity rates than Mexican-origin children born in the United States, but the pattern changed when he used data from 2005 to 2014, when foreign- and U.S.-born Mexican children had equivalent levels of obesity. This shift has real potential to alter health and mortality estimates for Hispanic populations in the future.58,59
This shift was the subject of another study, which assessed obesity rates as birth cohorts from the 1970s and 1980s aged, with modeled future trends suggesting the paradox may be lost.60 “Our big question was whether [the rise in] obesity is going to erode the paradox,” says Michelle Frisco, a Pennsylvania State University professor of sociology and demography. “As obesity is going up among Hispanic [individuals], especially men, we showed that the life expectancy advantage of U.S.-born Hispanic men over U.S.-born White men is likely to decline over time, and their life expectancy is going to look more similar.”
Frisco adds that her research was done pre-COVID, a disease that took an enormous toll on Hispanic adults61 due to the fact they often hold frontline worker jobs62 and are often less likely to seek out health care.63,64 “COVID took three years off life expectancy for Hispanic people,” she says, “and nobody’s done any comprehensive analysis of what that’s going to do to the paradox.”65
Understanding Hispanic populations’ health and mortality advantages largely relies on projections and studies using past data, which are rapidly changing. “I think it’s hazardous to apply old data to the present-day situation. A lot of these studies often go back to the 1980s or 1990s,” says Robert Kaplan, the other coprincipal investigator of the Bronx HCHS/SOL team. “There have been enormous technological developments in terms of lifestyle, conveniences, food production, occupations, and livelihoods. So in order to understand the needs of our present-day population, you need new data.”
Navas-Acien points again to the crucial need for disaggregated data. “Sometimes when you use general statistics, people’s race or ethnic status might not be well identified, and you might end up reaching conclusions which are not true,” she says. Navas-Acien gives the example of cardiovascular disease among American Indians/Alaska Natives: “For a long time, it was believed they have a lower burden of cardiovascular disease [than other racial and ethnic groups],” she says. “However, that was totally wrong; it was just based on statistics that were incorrect.” (In fact, cardiovascular disease is the leading cause of death among this population.66)
Shedding light on the unique aspects of Hispanic populations’ health can only help our nation reckon with its history of environmental racism and disproportionate exposures for people of color. Ultimately, a better understanding will give policy makers information to implement meaningful change. “Environmental health has a really important role to play in understanding not just the Hispanic paradox,” says Fenelon, “but what policy should do about health for Hispanic people in the United States.”
Wendee Nicole is an award-winning science writer based in San Diego. Her work has appeared in Nature, Scientific American, and other publications.
==== Refs
References
1. Arias E, Kochanek KD, Anderson RN. 2015. How does cause of death contribute to the Hispanic mortality advantage in the United States? NCHS Data Brief no. 221. https://www.cdc.gov/nchs/products/databriefs/db221.htm [accessed 29 December 2022].
2. Frisco ML, Van Hook J. 2020. Hispanics live longer than most Americans, but will the US obesity epidemic change things? The Conversation , Health section, 16 October 2020. https://theconversation.com/hispanics-live-longer-than-most-americans-but-will-the-us-obesity-epidemic-change-things-146006 [accessed 20 December 2022].
3. Franzini L, Ribble JC, Keddie AM. 2001. Understanding the Hispanic paradox. Ethn Dis 11 (3 ):496–518, PMID: .11572416
4. Cortes-Bergoderi M, Goel K, Murad MH, Allison T, Somers VK, Erwin PJ, et al. 2013. Cardiovascular mortality in Hispanics compared to non-Hispanic whites: a systematic review and meta-analysis of the Hispanic paradox. Eur J Intern Med 24 (8 ):791–799, PMID: , 10.1016/j.ejim.2013.09.003.24095273
5. Eschbach K, Mahnken JD, Goodwin JS. 2005. Neighborhood composition and incidence of cancer among Hispanics in the United States. Cancer 103 (5 ):1036–1044, PMID: , 10.1002/cncr.20885.15672387
6. Alexander GR, Kogan M, Bader D, Carlo W, Allen M, Mor J. 2003. US birth weight/gestational age-specific neonatal mortality: 1995–1997 rates for Whites, Hispanics, and Blacks. Pediatrics 111 (1 ):e61–e66, PMID: , 10.1542/peds.111.1.e61.12509596
7. Crimmins E, Kim JK, Vasunilashorn S. 2010. Biodemography: new approaches to understanding trends and differences in population health and mortality. Demography 47 (suppl 1 ):S41–S64, PMID: , 10.1353/dem.2010.0005.21302421
8. Sood A, Petersen H, Liu C, Myers O, Shore XW, Gore BA, et al. 2022. Racial and ethnic minorities have a lower prevalence of airflow obstruction than non-Hispanic whites. COPD 19 (1 ):61–68, PMID: , 10.1080/15412555.2022.2029384.35099333
9. Palloni A, Arias E. 2004. Paradox lost: explaining the Hispanic adult mortality advantage. Demography 41 (3 ):385–415, PMID: , 10.1353/dem.2004.0024.15461007
10. Ruiz JM, Steffen P, Smith TB. 2013. Hispanic mortality paradox: a systematic review and meta-analysis of the longitudinal literature. Am J Public Health 103 (3 ):e52–e60, PMID: , 10.2105/AJPH.2012.301103.23327278
11. Palarino JV. 2021. The immigrant health advantage: an examination of African-origin Black immigrants in the United States. Popul Res Policy Rev 40 (5 ):895–929, 10.1007/s11113-021-09647-6.
12. Egolf B, Lasker J, Wolf S, Potvin L. 1992. The Roseto effect: a 50-year comparison of mortality rates. Am J Public Health 82 (8 ):1089–1092, PMID: , 10.2105/ajph.82.8.1089.1636828
13. Kotwal AA. 2010. Physical and psychological health of first and second generation Turkish immigrants in Germany. Am J Hum Biol 22 (4 ):538–545, PMID: , 10.1002/ajhb.21044.20309991
14. Vang Z, Sigouin J, Flenon A, Gagnon A. 2015. The healthy immigrant effect in Canada: a systematic review. Population Change and Lifecourse Strategic Knowledge Cluster Discussion Paper Series/Un Réseau stratégique de connaissances Changements de population et parcours de vie document de travail 3 (1 ):Article 4. https://ir.lib.uwo.ca/pclc/vol3/iss1/4/ [accessed 29 December 2022].
15. Funk C, Lopez MH. 2022. Hispanic Americans’ Trust in and Engagement with Science. 1. A Brief Statistical Portrait of U.S. Hispanics. https://www.pewresearch.org/science/2022/06/14/a-brief-statistical-portrait-of-u-s-hispanics/ [accessed 20 December 2022].
16. Teruya SA, Bazargan-Hejazi S. 2013. The immigrant and Hispanic paradoxes: a systematic review of their predictions and effects. Hisp J Behav Sci 35 (4 ):486–509, PMID: , 10.1177/0739986313499004.26120244
17. Bostean G. 2013. Does selective migration explain the Hispanic paradox?: A comparative analysis of Mexicans in the U.S. and Mexico. J Immigr Minor Health 15 (3 ):624–635, PMID: , 10.1007/s10903-012-9646-y.22618355
18. Markides KS, Eschbach K. 2005. Aging, migration, and mortality: current status of research on the Hispanic paradox. J Gerontol B Psychol Sci Soc Sci 60 (spec no 2 ):68–75, PMID: , https://academic.oup.com/psychsocgerontology/article/60/Special_Issue_2/S68/2965177.16251594
19. Fenelon A, Chinn JJ, Anderson RN. 2017. A comprehensive analysis of the mortality experience of Hispanic subgroups in the United States: variation by age, country of origin, and nativity. SSM Popul Health 3 :245–254, PMID: , 10.1016/j.ssmph.2017.01.011.29349222
20. Turra CM, Elo IT. 2008. The impact of salmon bias on the Hispanic mortality advantage: new evidence from Social Security data. Popul Res Policy Rev 27 (5 ):515–530, PMID: , 10.1007/s11113-008-9087-4.19122882
21. Abraído-Lanza AF, Dohrenwend BP, Ng-Mak DS, Turner JB. 1999. The Latino mortality paradox: a test of the “salmon bias” and healthy migrant hypotheses. Am J Public Health 89 (10 ):1543–1548, PMID: , 10.2105/ajph.89.10.1543.10511837
22. Fenelon A. 2013. Revisiting the Hispanic paradox in the United States: the role of smoking. Soc Sci Med 82 :1–9, PMID: , 10.1016/j.socscimed.2012.12.028.23453311
23. Valdivieso-Mora E, Peet CL, Garnier-Villarreal M, Salazar-Villanea M, Johnson DK. 2016. A systematic review of the relationship between familism and mental health outcomes in Latino population. Front Psychol 7 :1632, PMID: , 10.3389/fpsyg.2016.01632.27826269
24. Gallegos ML, Segrin C. 2022. Family connections and the Latino health paradox: exploring the mediating role of loneliness in the relationships between the Latina/o cultural value of familism and health. Health Commun 37 (9 ):1204–1214, PMID: , 10.1080/10410236.2021.1909244.33853460
25. Katiria Perez G, Cruess D. 2014. The impact of familism on physical and mental health among Hispanics in the United States. Health Psychol Rev 8 (1 ):95–127, PMID: , 10.1080/17437199.2011.569936.25053010
26. Corona K, Campos B, Chen C. 2017. Familism is associated with psychological well-being and physical health: main effects and stress-buffering effects. Hisp J Behav Sci 39 (1 ):46–65, 10.1177/0739986316671297.
27. Corona K, Campos B, Rook KS, Biegler K, Sorkin DH. 2019. Do cultural values have a role in health equity? A study of Latina mothers and daughters. Cultur Divers Ethnic Minor Psychol 25 (1 ):65–72, PMID: , 10.1037/cdp0000262.30714768
28. Eschbach K, Ostir GV, Patel KV, Markides KS, Goodwin JS. 2004. Neighborhood context and mortality among older Mexican Americans: is there a barrio advantage? Am J Public Health 94 (10 ):1807–1812, PMID: , 10.2105/ajph.94.10.1807.15451754
29. Fenelon A. 2017. Rethinking the Hispanic paradox: the mortality experience of Mexican immigrants in traditional gateways and new destinations. Int Migr Rev 51 (3 ):567–599, PMID: , 10.1111/imre.12263.33110281
30. Brown A, Lopez MH. 2013. Mapping the Latino Population, By State, County and City. https://www.pewresearch.org/hispanic/2013/08/29/mapping-the-latino-population-by-state-county-and-city/ [accessed 20 December 2022].
31. Hummer RA. 2000. Adult mortality differentials among Hispanic subgroups and non-Hispanic whites. Soc Sci Q 81 (1 ):459–476, PMID: .17879490
32. Fong KC, Heo S, Lim CC, Kim H, Chan A, Lee W, et al. 2022. The intersection of immigrant and environmental health: a scoping review of observational population exposure and epidemiologic studies. Environ Health Perspect 130 (9 ):96001, PMID: , 10.1289/EHP9855.36053724
33. Velasco-Mondragon E, Jimenez A, Palladino-Davis AG, Davis D, Escamilla-Cejudo JA. 2016. Hispanic health in the USA: a scoping review of the literature. Public Health Rev 37 :31, PMID: , 10.1186/s40985-016-0043-2.29450072
34. Carter-Pokras O, Zambrana RE, Poppell CF, Logie LA, Guerrero-Preston R. 2007. The environmental health of Latino children. J Pediatr Health Care 21 (5 ):307–314, PMID: , 10.1016/j.pedhc.2006.12.005.17825728
35. Metzger R, Delgado JL, Herrell R. 1995. Environmental health and Hispanic children. Environ Health Perspect 103 (suppl 6 ):25–32, PMID: , 10.1289/ehp.95103s625.8549482
36. Rodriquez EJ, Stoecklin-Marois MT, Hennessy-Burt TE, Tancredi DJ, Schenker MB. 2014. Demographic and migration-related risk factors for low-level smoking in a farm working sample of Latinos (the MICASA study). Field Actions Sci Rep 20 :3286, PMID: .29643941
37. McCurdy SA, Xiao H, Hennessy-Burt TE, Stoecklin-Marois MT, Tancredi DJ, Bennett DH, et al. 2013. Agricultural injury in California Hispanic farm workers: MICASA follow-up survey. J Agromedicine 18 (1 ):39–49, PMID: , 10.1080/1059924X.2012.743380.23301889
38. University of California, Berkeley, School of Public Health. 2022. CHAMACOS Study. [Website.] https://cerch.berkeley.edu/research-programs/chamacos-study [accessed 20 December 2022].
39. Bastain TM, Chavez T, Habre R, Girguis MS, Grubbs B, Toledo-Corral C, et al. 2019. Study design, protocol and profile of the Maternal And Developmental Risks from Environmental and Social stressors (MADRES) pregnancy cohort: a prospective cohort study in predominantly low-income Hispanic women in urban Los Angeles. BMC Pregnancy Childbirth 19 (1 ):189, PMID: , 10.1186/s12884-019-2330-7.31146718
40. Martinez-Morata I, Bostick BC, Conroy-Ben O, Duncan DT, Jones MR, Spaur M, et al. 2022. Nationwide geospatial analysis of county racial and ethnic composition and public drinking water arsenic and uranium. Nat Commun 13 (1 ):7461, PMID: , 10.1038/s41467-022-35185-6.36460659
41. Ravalli F, Yu Y, Bostick BC, Chillrud SN, Schilling K, Basu A, et al. 2022. Sociodemographic inequalities in uranium and other metals in community water systems across the USA, 2006–11: a cross-sectional study. Lancet Planet Health 6 (4 ):e320–e330, PMID: , 10.1016/S2542-5196(22)00043-2.35397220
42. Nigra AE, Chen Q, Chillrud SN, Wang L, Harvey D, Mailloux B, et al. 2020. Inequalities in public water arsenic concentrations in counties and community water systems across the United States, 2006–2011. Environ Health Perspect 128 (12 ):127001, PMID: , 10.1289/EHP7313.33295795
43. Gonzales M, Erdei E, Hoover J, Nash J. 2018. A review of environmental epidemiology studies in southwestern and mountain west rural minority populations. Curr Epidemiol Rep 5 (2 ):101–113, PMID: , 10.1007/s40471-018-0146-z.30906685
44. Ricardo AC, Loop MS, Gonzalez F 2nd, Lora CM, Chen J, Franceschini N, et al. 2020. Incident chronic kidney disease risk among Hispanics/Latinos in the United States: the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). J Am Soc Nephrol 31 (6 ):1315–1324, PMID: , 10.1681/ASN.2019101008.32300066
45. Lora CM, Daviglus ML, Kusek JW, Porter A, Ricardo AC, Go AS, et al. 2009. Chronic kidney disease in United States Hispanics: a growing public health problem. Ethn Dis 19 (4 ):466–472, PMID: .20073150
46. Pereira RI, Cervantes L. 2021. Reducing the burden of CKD among Latinx: a community-based approach. Clin J Am Soc Nephrol 16 (5 ):812–814, PMID: , 10.2215/CJN.12890820.33441466
47. Eamranond PP, Hu H. 2008. Environmental and occupational exposures in immigrant health. Environ Health Insights 1 :45–50, PMID: , 10.4137/ehi.s847.21572847
48. Moyce SC, Schenker M. 2017. Occupational exposures and health outcomes among immigrants in the USA. Curr Environ Health Rep 4 (3 ):349–354, PMID: , 10.1007/s40572-017-0152-1.28812286
49. Harley KG, Calderon L, Nolan JES, Maddalena R, Russell M, Roman K, et al. 2021. Changes in Latina women’s exposure to cleaning chemicals associated with switching from conventional to “green” household cleaning products: the LUCIR intervention study. Environ Health Perspect 129 (9 ):97001, PMID: , 10.1289/EHP8831.34468180
50. Levchenko Y. 2021. Aging into disadvantage: disability crossover among Mexican immigrants in America. Soc Sci Med 285 :114290, PMID: , 10.1016/j.socscimed.2021.114290.34352506
51. Hayward MD, Hummer RA, Chiu CT, González-González C, Wong R. 2014. Does the Hispanic paradox in U.S. adult mortality extend to disability? Popul Res Policy Rev 33 (1 ):81–96, PMID: , 10.1007/s11113-013-9312-7.25821283
52. Richardson DB, Loomis D, Bena J, Bailer AJ. 2004. Fatal occupational injury rates in southern and non-southern states, by race and Hispanic ethnicity. Am J Public Health 94 (10 ):1756–1761, PMID: , 10.2105/ajph.94.10.1756.15451746
53. Forst L, Avila S, Anozie S, Rubin R. 2010. Traumatic occupational injuries in Hispanic and foreign born workers. Am J Ind Med 53 (4 ):344–351, PMID: , 10.1002/ajim.20748.19753594
54. Haile RW, John EM, Levine AJ, Cortessis VK, Unger JB, Gonzales M, et al. 2012. A review of cancer in U.S. Hispanic populations. Cancer Prev Res (Phila) 5 (2 ):150–163, PMID: , 10.1158/1940-6207.CAPR-11-0447.22307564
55. Bhatnagar A. 2017. Environmental determinants of cardiovascular disease. Circ Res 121 (2 ):162–180, PMID: , 10.1161/CIRCRESAHA.117.306458.28684622
56. Patel AP, Jagai JS, Messer LC, Gray CL, Rappazzo KM, Deflorio-Barker SA, et al. 2018. Associations between environmental quality and infant mortality in the United States, 2000–2005. Arch Public Health 76 :60, PMID: , 10.1186/s13690-018-0306-0.30356923
57. Maldonado LE, Albrecht SS. 2018. Does the immigrant advantage in overweight/obesity persist over time in Mexican American youth? NHANES 1988–1994 to 2005–2014. Obesity (Silver Spring) 26 (6 ):1057–1062, PMID: , 10.1002/oby.22178.29797556
58. Goldman N. 2016. Will the Latino mortality advantage endure? Res Aging 38 (3 ):263–282, PMID: , 10.1177/0164027515620242.26966251
59. Hummer RA, Hayward MD. 2015. Hispanic older adult health & longevity in the United States: current patterns & concerns for the future. Daedalus 144 (2 ):20–30, PMID: , 10.1162/DAED_a_00327.26082561
60. Van Hook J, Frisco ML, Graham CE. 2020. Signs of the end of the paradox? Cohort shifts in smoking and obesity and the Hispanic life expectancy advantage. Sociol Sci 7 :391–414, 10.15195/v7.a16.
61. Arias E, Tejada-Vera B, Ahmad F, Kochanek KD. 2021. Provisional Life Expectancy Estimates for 2020. Vital Statistics Rapid Release No. 015. https://www.cdc.gov/nchs/data/vsrr/VSRR015-508.pdf [accessed 20 December 2022].
62. Goldman N, Pebley AR, Lee K, Andrasfay T, Pratt B. 2021. Racial and ethnic differentials in COVID-19-related job exposures by occupational standing in the US. PLoS One 16 (99 ):e0256085, PMID: , 10.1371/journal.pone.0256085.34469440
63. Derose KP, Escarce JJ, Lurie N. 2007. Immigrants and health care: sources of vulnerability. Health Aff (Millwood) 26 (5 ):1258–1268, PMID: , 10.1377/hlthaff.26.5.1258.17848435
64. Hacker K, Anies ME, Folb B, Zallman L. 2015. Barriers to health care for undocumented immigrants: a literature review. Risk Manag Healthc Policy 8 :175–183, PMID: , 10.2147/RMHP.S70173.26586971
65. Garcia MA, Sáenz R. 2022. COVID-19 Has Reduced the Latino Mortality Advantage among Older Adults. Research Brief #71. https://surface.syr.edu/cgi/viewcontent.cgi?article=1180&context=lerner [accessed 20 December 2022].
66. Breathett K, Sims M, Gross M, Jackson EA, Jones EJ, Navas-Acien A, et al. 2020. Cardiovascular health in American Indians and Alaska Natives a scientific statement from the American Heart Association. Circulation 141 (25 ):e948–e959, PMID: , 10.1161/CIR.0000000000000773.32460555
67. Marks R, Rios-Vargas M. 2021. Improvements to the 2020 Census Race and Hispanic Origin Question Designs, Data Processing, and Coding Procedures. [Website.] https://www.census.gov/newsroom/blogs/random-samplings/2021/08/improvements-to-2020-census-race-hispanic-origin-question-designs.html [accessed 29 December 2022].
|
PMC009xxxxxx/PMC9833480.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12564
10.1289/EHP12564
Erratum
Erratum: “A Permutation Test-Based Approach to Strengthening Inference on the Effects of Environmental Mixtures: Comparison between Single-Index Analytic Methods”
https://orcid.org/0000-0003-2606-2050
Day Drew B.
Sathyanarayana Sheela
LeWinn Kaja Z.
Karr Catherine J.
Mason W. Alex
Szpiro Adam A.
11 1 2023
1 2023
11 1 2023
131 1 01900108 12 2022
15 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Environ Health Perspect. 130(8):087010 (2022), https://doi.org/10.1289/EHP10570
==== Body
pmcIn the Discussion, in the third paragraph of the “Comparing Simulation Results to Prior Evaluations of Mixture Exposure Model Performance” section, the authors stated, “Furthermore, the mixture component-specific coefficients estimated by quantile g-computation were far more biased than those of any of the WQSr models.” Given that the mean absolute percent error (MAPE) that the authors used in their analyses assesses accuracy rather than bias, the correct statement should be as follows: “Furthermore, the mixture component-specific coefficients estimated by quantile g-computation were far less accurate than those of any of the WQSr models.”
The authors regret the error.
|
PMC009xxxxxx/PMC9869870.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36688826
EHP11292
10.1289/EHP11292
Research
Estimated Transfer of Perfluoroalkyl Substances (PFAS) from Maternal Serum to Breast Milk in Women Highly Exposed from Contaminated Drinking Water: A Study in the Ronneby Mother–Child Cohort
https://orcid.org/0000-0003-1360-2580
Blomberg Annelise J. 1 2
Norén Erika 1
Haug Line S. 3
Lindh Christian 1
Sabaredzovic Azemira 3
Pineda Daniela 1
Jakobsson Kristina 4 5
Nielsen Christel 1 6
1 Division of Occupational and Environmental Medicine, Department of Laboratory Medicine, Lund University, Lund, Sweden
2 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
3 Department of Food Safety, Norwegian Institute of Public Health, Oslo, Norway
4 School of Public Health and Community Medicine, Institute of Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden
5 Department of Occupational and Environmental Medicine, Sahlgrenska University Hospital, Gothenburg, Sweden
6 Clinical Pharmacology, Pharmacy and Environmental Medicine, Department of Public Health, University of Southern Denmark, Odense, Denmark
Address correspondence to Annelise J. Blomberg, Scheelevägen 2, 22363 Lund, Sweden. Email: [email protected]
23 1 2023
1 2023
131 1 01700523 3 2022
21 11 2022
19 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Infancy perfluoroalkyl substances (PFAS) exposure from breastfeeding is partially determined by the transfer efficiencies (TEs) of PFAS from maternal serum into breast milk. However, to our knowledge there are no studies of such TEs in highly exposed populations.
Objectives:
We estimated the TEs of PFAS from maternal serum into colostrum and breast milk in a cohort of women with a wide range of PFAS exposures.
Methods:
The Ronneby Mother–Child Cohort was established in 2015 after PFAS contamination was discovered in the public drinking water of Ronneby, Sweden. We measured seven PFAS in matched samples of maternal serum at delivery and colostrum and breast milk. We calculated the TE (in percentage) as the ratio of PFAS in colostrum or breast milk to serum multiplied by 100 and evaluated whether TEs varied by PFAS, lactation stage, or exposure level using a series of linear mixed-effects models with a random intercept for each woman.
Results:
This study included 126 mothers. PFAS associated with firefighting foams [i.e., perfluorohexane sulfonic acid (PFHxS) and perfluorooctane sulfonic acid (PFOS)] were substantially elevated in the serum, colostrum, and breast milk samples of highly exposed women in the cohort and showed strong correlation. PFHxS and PFOS also contributed the largest fraction of total PFAS on average in colostrum and breast milk. Median TEs varied from 0.9% to 4.3% and were higher for perfluoroalkyl carboxylic acids, including perfluorooctanoic acid, than perfluoroalkane sulfonic acids, including PFHxS and PFOS. TEs varied by exposure level, but there was not a consistent pattern in this variation.
Discussion:
PFAS concentrations in the colostrum and breast milk of highly exposed women were higher than the concentrations in low-exposed women, and TEs were of a similar magnitude across exposure categories. This implies that breastfeeding may be an important route of PFAS exposure for breastfeeding infants with highly exposed mothers, although the relative contribution of breastfeeding vs. prenatal transplacental transfer remains to be clarified. https://doi.org/10.1289/EHP11292
Supplemental Material is available online (https://doi.org/10.1289/EHP11292).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Perfluoroalkyl substances (PFAS) are a diverse class of synthetic chemicals containing an aliphatic fluorinated carbon chain.1 PFAS have unique chemical properties, including chemical and thermal stability and water and oil repellency.2 They have been widely used since the 1950s in various industrial and commercial applications, such as personal care products and cosmetics, firefighting foams, textile treatments, food contact materials, medical devices, and membranes used by the chemical and energy sectors.3 However, the same properties that make PFAS desirable in so many applications also make them extremely persistent in the environment and bioaccumulative. Widespread human exposure through diet and other pathways, combined with the long biological half-lives of some PFAS, has led to detectable levels in most human blood samples.2,4,5
Although governmental regulation on the manufacture and use of certain PFAS is increasing,6,7 the extreme environmental persistence of PFAS means that existing contamination will remain a concern. This is particularly relevant for populations living near contaminated sites, where environmental PFAS concentrations and human exposures are typically highest.4 These sites include fluorochemical manufacturing facilities, manufacturing facilities where PFAS are used, airports and military bases that use aqueous film forming firefighting foams (AFFFs), and landfills.4,8 Highly exposed communities continue to be identified globally.2,9
Infants and children are particularly vulnerable to environmental toxicants, including PFAS. Children often have higher body burdens than adults owing to differences in exposure sources, body size, and body surface area.10 PFAS developmental toxicity has been demonstrated in animal models,11,12 and epidemiological studies have linked prenatal and childhood PFAS exposures to health outcomes, including dyslipidemia, changes in fetal and postnatal growth, impaired immunity, and delayed onset of puberty.10,13 Furthermore, it is increasingly understood that changes during sensitive periods of development may have potentially adverse consequences for health later in life.14
A critical step in the breastfeeding exposure pathway is the transfer of PFAS from maternal serum into colostrum and breast milk. Cumulative infancy exposure can be estimated by multiplying the transfer efficiency (TE; the fraction of PFAS transferred from maternal serum into breast milk), by the maternal PFAS concentration and then by the cumulative milk volume consumed. Although breastfeeding has beneficial effects for mother and child,15 it is also an important source of PFAS exposure for infants and young children. At low exposure levels, the duration of exclusive breastfeeding has been associated with an increase in infancy serum PFAS concentrations up to as high as 30% per month.16,17 One study also found that postnatal exposure from exclusive breastfeeding the month after delivery was higher than prenatal exposure.18 However, there is an overall lack of information on potential PFAS exposures from breastfeeding.19
Although a limited number of studies have estimated TEs using paired samples of maternal serum and breast milk, these studies had several major limitations.18,20–23 First, they were all conducted in populations exposed to low levels of PFAS. Therefore, PFAS concentrations in milk samples were often below the limit of detection (LOD) or limit of quantification (LOQ). To our knowledge, no studies have estimated PFAS TEs in highly exposed women although this is the population where infant exposures via breast milk are most concerning. In addition, because breast milk composition changes throughout lactation, it is possible that PFAS TEs may also change. However, no study has evaluated this possibility. Finally, all but one existing study had a sample size of <20 women.
To address these limitations, we measured the transfer of PFAS from maternal serum into colostrum and breast milk in a cohort of women from Ronneby, Sweden, where many residents were exposed to PFAS from drinking water that was highly contaminated by AFFFs.24 We hypothesized that PFAS would be measurable in colostrum and breast milk and that TEs would vary by PFAS compound and lactation stage. We also explored whether TEs varied by the level of maternal exposure (high vs. background exposure).
Methods
Ronneby Mother–Child Cohort
In 2013, it was discovered that one of the two municipal water supplies in Ronneby, Blekinge county, Sweden, had been delivering water that was highly contaminated by PFAS from AFFF runoff at a nearby military airport (Figure 1).24 Military records documented that AFFF usage at the airport began in the mid-1980s, although the start of widespread water contamination is unknown.24 The total PFAS level in outgoing drinking water in December 2013 in this municipal water supply was >10,000 ng/L (the Swedish action level at the time was 90 ng/L) and the waterworks was immediately closed.24 In the other Ronneby waterworks, the total PFAS was 48 ng/L.24 In contrast, the total PFAS concentration in the water supply of the nearby municipality of Karlshamn was <5 ng/L.24
Figure 1. (A) Map of Sweden. (B) Location of the Ronneby Mother–Child Cohort. The dashed yellow polygon outlines the area receiving highly contaminated drinking water in 2013. Background maps: ©Lantmäteriet.25
Figure 1A is a map of Sweden, marking the Ronneby locality with a scale of 100 kilometers. Figure 1B is a map showing the location of the Ronneby Mother–Child Cohort, depicting the area that received highly contaminated drinking water in 2013 with a scale of 8 kilometers.
We established the prospective Ronneby Mother–Child Cohort after the contamination was discovered and drinking water from the contaminated waterworks was turned off, with the purpose of investigating the transfer of PFAS from mother to child. All pregnant women in the Ronneby municipality between 2015 and 2020 were invited to participate at their maternal health care center. Women from Karlshamn were also invited to participate starting in 2018 after population biomonitoring found that most individuals in Ronneby, regardless of drinking water source, had higher serum PFAS concentrations compared with residents from Karlshamn.24 The final cohort included 263 women, with 225 from Ronneby, 35 from Karlshamn, and 3 who were missing location data.
A maternal blood sample was collected at the time of delivery from all participants. The blood was centrifuged and the serum sample was collected in a 2-mL micro tube (Sarstedt). It was then shipped from the Karlskrona hospital to the Laboratory of Occupational and Environmental Medicine in Lund on dry ice and stored at −80°C until analysis. Mothers collected a colostrum sample 3–4 d postpartum and a breast milk sample 4–12 wk postpartum in 50-mL screw cap polypropylene tubes (Sarstedt). Both samples were collected at home using either hand expression or a pump and were stored in the mothers’ home freezers at −20°C until they were returned to the maternal health care center during a maternal follow-up visit (∼8–12wk postpartum). They were stored at −20°C until they were retrieved and transferred to the Laboratory of Occupational and Environmental Medicine in Lund on ice and stored at −80°C. They were shipped on dry ice to the Norwegian Institute of Public Health, arrived frozen, and were stored at −80°C until analysis. Study participants also completed questionnaires with information on breastfeeding history, smoking habits, medical drug use, education, occupation, and residential history.
The study was approved by the regional ethical review board in Lund, Sweden (no. 2017/437, with amendments). Written informed consent was provided by all participants.
PFAS Analysis
Seven PFAS were measured in maternal serum, colostrum, and breast milk samples [perfluorooctanoic acid (PFOA), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), perfluoroundecanoic acid (PFUnDA), perfluorohexane sulfonic acid (PFHxS), perfluoroheptane sulfonic acid (PFHpS), and perfluorooctane sulfonic acid (PFOS)]. Serum samples were analyzed at the Division of Occupational and Environmental Medicine at Lund University using the method described by Norén et al.26 In brief, an aliquot of 100μL serum was added with an internal PFAS standard. Serum proteins were then precipitated with acetonitrile, and the mixture was vigorously shaken for 30 min. The samples were analyzed using liquid chromatography/tandem mass spectrometry (LC-MS/MS) (QTRAP 5500; AB Sciex). Total PFOS was measured as the total non–isomer-specific compounds. Four reference samples, created by pooling several serum samples, were used as a quality control to estimate between-run precision, and 164 serum samples were analyzed in duplicate to estimate between-batch precision (Table S1). The LOQ for PFAS measurements in serum was 0.1 ng/mL. The laboratory participates in the interlaboratory comparison investigations and external quality assurance schemes for analyses of PFAS and is approved by the European Human Biomonitoring Initiative (HBM4EU) project and the University of Erlangen-Nuremberg.
Colostrum and breast milk samples were analyzed for PFAS concentrations at the Department of Food Safety, Norwegian Institute of Public Health. The samples were prepared and analyzed using LC-MS/MS, following the method detailed by Thomsen et al.27 In brief, the samples were thawed and homogenized in an incubator at 37°C. An aliquot of 200μL was transferred to an Eppendorf tube, to which internal standards and acetonitrile were added. The sample was then mixed on a whirl mixer and further centrifuged. The supernatant was transferred to an autosampler vial and 0.1 M formic acid was added. The sample extracts were injected on a column-switching LC system coupled to a triple quadrupole mass spectrometer. High quality of measurements was assured by analyzing two in-house quality control samples (n=6 each). The mean concentrations of PFAS in the two in-house control samples were in the range of 0.01–0.16 ng/mL. The relative standard deviations for the two sets of replications were 28% and 11.1% for PFOA, 28.0 and 33.8% for PFNA, 48.2% and 28.8% for PFDA, 64.2% and 15.9% for PFHxS, and 20.2% and 8.8% for PFOS. The LOQ for all seven PFAS in colostrum and breast milk was 0.01 ng/mL. For the quantification of PFOS, the total area of the linear and branched isomers was integrated.
Statistical Analysis
We categorized participants into three exposure groups based on their maternal serum PFHxS concentrations at delivery given that PFHxS is a strong indicator of exposure to AFFF-contaminated water in Ronneby.24 Participants were included in the background exposure group if their PFHxS concentration was ≤90th percentile of PFHxS concentration in women receiving maternal health care in Karlshamn. This cutoff was selected to represent background levels of exposure while accounting for the fact that some women in Karlshamn may have spent time working or visiting in Ronneby. Participants who were not included in the background group were then categorized as in either the high exposure group (≥75th percentile in PFHxS concentration in the non-background population) or the intermediate exposure group (<75th percentile in PFHxS concentration in the non-background population). Of the 126 women with a serum sample at delivery and a paired colostrum or breast milk sample, 25 women were categorized as being in the background group, 76 women in the intermediate group, and 25 women in the highly exposed group (Table 1, Tables S5 and S6, and Figure S1).
Table 1. Baseline characteristics of the 126 mothers from the Ronneby Mother–Child Cohort who were included in this study, displayed as N (%; excludes missing) or mean±SD.
Characteristic Overall PFAS exposure categorya
Background Intermediate High
N 126 25 76 25
Maternal age at delivery 30.71±4.74 30.88±4.74 30.70±4.76 30.60±4.89
Year of delivery
2015 6 (5) 0 (0) 2 (3) 4 (16)
2016 27 (21) 0 (0) 20 (26) 7 (28)
2017 32 (25) 3 (12) 21 (28) 8 (32)
2018 31 (25) 10 (40) 19 (25) 2 (8)
2019 27 (21) 11 (44) 12 (16) 4 (16)
2020 3 (2) 1 (4) 2 (3) 0 (0)
Parity
Primiparous 48 (38) 10 (40) 27 (36) 11 (44)
Multiparous 77 (62) 15 (60) 48 (64) 14 (56)
Smoking status
Never smoker 80 (67) 18 (78) 45 (62) 17 (71)
Current smoker 7 (6) 0 (0) 6 (8) 1 (4)
Past smoker 32 (27) 5 (22) 21 (29) 6 (25)
Missing 7 2 4 1
Education status
Less than high school 3 (2) 0 (0) 3 (4) 0 (0)
High school 51 (42) 8 (35) 29 (40) 14 (58)
University (≥3 y) 63 (52) 15 (65) 40 (55) 8 (33)
Other 3 (2) 0 (0) 1 (1) 2 (8)
Missing 6 2 3 1
Location of maternal care
Karlshamn 23 (18) 20 (80) 3 (4) 0 (0)
Ronneby 103 (82) 5 (20) 73 (96) 25 (100)
Note: PFAS, perfluoroalkyl substances; PFHxS, perfluorohexane sulfonic acid; SD, standard deviation.
a Exposure levels were categorized based on maternal serum PFHxS concentrations (CPFHxS, ng/mL) at delivery. Background: CPFHxS≤0.78. Intermediate: 0.78<CPFHxS≤36. High: 36<CPFHxS.
We calculated the average relative contribution of each PFAS to the sum of the seven measured PFAS (PFOA, PFNA, PFDA, PFUnDA, PFHxS, PFHpS, and PFOS) for each exposure category and sample matrix (serum, colostrum, and breast milk). We estimated bivariate correlations between each PFAS in each matrix for the full cohort, as well as stratified by exposure group. Correlations were estimated using Spearman rank correlations given that the PFAS concentrations were not normally distributed. In these descriptive analyses we used the measured PFAS concentration when available and otherwise used the LOQ divided by the square root of 2.
We then calculated the TE (in percentage) of each PFAS from maternal serum (S) into colostrum (C) or breast milk (B) for all PFAS measured in the study, limiting our analyses to samples that measured >LOQ to increase the accuracy of the estimates and reduce noise from imputation-related error. TE was calculated as follows: (1) TEi:S=ConciConcS×100%,
where i was either colostrum or breast milk; ConcS was the PFAS concentration (in nanograms per milliliter) in the serum sample; and Conci was the PFAS concentration in either colostrum or breast milk. We summarized TEs using the range, median, and interquartile range (IQR). Sample quantiles were calculated using sample quantile definition 7 from Hyndman and Fan,28 which is the default quantile algorithm in R.
For PFAS measured with a high quantification frequency (>LOQ in at least 50% of maternal serum samples and 50% of combined colostrum and breast milk samples), we further investigated potential variation in TE by PFAS compound, lactation stage, and exposure level. In these secondary analyses we used measured PFAS values when available but otherwise used the LOQ divided by the square root of 2. First, we estimated TE stratified by exposure category. Then, to evaluate whether TE varied significantly by PFAS compound, lactation stage, or exposure level, we modeled TE using linear mixed-effects models with a random intercept for each woman to account for within-subject correlation. TE was log-transformed to improve the normality of the residuals. In our first model, we evaluated the significance of PFAS and lactation stage by modeling TE as a function of these two indicator variables, including an interaction between the two predictors. This can be written as follows: (2) Yi,p=∑14βpPI(PFAS=p)+βmI(milk=m)+∑24βiII(PFAS=p)×I(milk=m)+bi+εi,p,
where Yi,p is the natural logarithm of the TE for woman i and PFAS p (either PFOA, PFNA, PFHxS, or PFOS); βP is the vector of coefficients for an indicator variable of PFAS; βm is the main effect of lactation stage (colostrum or breast milk); βI is the vector of coefficients for the interaction term between PFAS and lactation stage; bi is a subject-specific intercept; and εi,p is the within-subject error. We used Wald F-tests to evaluate the overall significance of PFAS and lactation stage29 considering p<0.05 as statistically significant. In our second model, we similarly estimated TE as a function of PFAS and exposure level, allowing for an interaction between the two terms and stratifying the model by lactation stage. We again used Wald F-tests to evaluate the overall significance of exposure level.
All statistical analyses were conducted in R [version 4.2.0; R Development Core Team (2022-04-22)] using the Tidyverse.30 Mixed effects models were run using the package nlme (version 3.1.157),31 and the map of Ronneby in Figure 1 was constructed using the package sf (version 1.0.7).32
Results
Of the 263 mother–child pairs participating in the Ronneby Mother–Child Cohort, 211 had PFAS concentrations measured from maternal serum collected at the time of delivery. Of these, 126 also had PFAS concentrations measured in colostrum (n=85) or breast milk (n=109). A study selection flow chart is included as Figure 2. Summary characteristics of the study population are included in Table 1. Four mothers were enrolled twice in the study because they each had two children in the cohort and provided serum and breast milk samples in both instances.
Figure 2. Study selection flowchart. N refers to the number of mother–child pairs included or excluded. In the case of twins, each mother–twin pair is considered N=1.
Figure 2 is a flowchart with five steps. Step 1: There are 263 cases in the Ronneby Mother-Child Cohort. Step 2: There are 261 cases in the Ronneby Mother-Child Cohort after excluding two cases of twins. Step 3: There are 211 cases in the Ronneby Mother-Child Cohort after excluding 50 without a serum sample from delivery. Step 4: There are 126 cases in the Ronneby Mother-Child Cohort after excluding 85 with no colostrum or breast milk sample. Step 5: There are 85 cases in the Ronneby Mother-Child Cohort with paired colostrum and serum samples and 109 cases with paired breast milk and serum samples.
The population characteristics of women with intermediate and high PFAS exposures were generally similar to those with background exposures (Table 1). However, women in the high exposure group had lower educational attainment than the other two groups: Only 32% of women in the high exposure group attended university, compared with 60% of women in the background exposure group. Women in the background group also generally gave birth later with respect to calendar year than women in the intermediate and high exposure groups.
Women who provided at least one colostrum or breast milk sample differed from women who did not by several important characteristics. They were slightly older, more likely to be a never-smoker, and more likely to have attended university. In addition, women with a colostrum or breast milk sample were more likely to have received maternal care in Karlshamn than those who did not (18.3% compared with 8.2%) and, as a result, had lower median concentrations of PFOA, PFHxS, and PFOS (Tables S2 and S3).
PFAS Concentrations
We measured seven PFAS in all three matrices (serum, colostrum, and breast milk). All seven PFAS were quantifiable in at least 76% of serum samples (Table 2). PFAS concentrations in colostrum and breast milk were considerably lower than in maternal serum. The PFAS with the highest quantification frequency in colostrum and breast milk samples was PFOS (97% >LOQ), whereas PFDA had the lowest quantification frequency (2% >LOQ). PFHxS and PFOS had the highest median concentrations in all matrices (Table 2).
Table 2 PFAS concentrations (ng/mL) in maternal serum at delivery and in colostrum and breast milk samples.
PFAS Serum
(N=126) Colostrum
(N=85) Breast milk
(N=109)
PFOA
% >LOQ 100 98 94
5th–95th perc. 0.43–5.93 0.01–0.2 0.01–0.15
Median (IQR) 1.31 (0.84–2.6) 0.03 (0.02–0.08) 0.03 (0.02–0.06)
PFNA
% >LOQ 99 45 75
5th–95th perc. 0.17–0.73 0.01–0.02 0.01–0.01
Median (IQR) 0.38 (0.26–0.52) <0.01 (<0.01–0.01) 0.01 (0.01–0.01)
PFDA
% >LOQ 94 2 2
5th–95th perc. <0.1–0.45 <0.1–0.01 <0.1–0.01
Median (IQR) 0.22 (0.15–0.32) <0.01 (<0.01–<0.01) <0.01 (<0.01–<0.01)
PFUnDA
% >LOQ 90 9 3
5th–95th perc. <0.1–0.44 <0.01–0.01 <0.01–0.01
Median (IQR) 0.2 (0.15–0.32) <0.01 (<0.01–<0.01) <0.01 (<0.01–<0.01)
PFHxS
% >LOQ 100 71 71
5th–95th perc. 0.29–67.47 0.01–1.73 0.01–0.98
Median (IQR) 6.02 (0.88–26.12) 0.1 (<0.01–0.48) 0.07 (<0.01–0.32)
PFHpS
% >LOQ 76 34 36
5th–95th perc. <0.1–3.77 0.01–0.11 0.01–0.05
Median (IQR) 0.42 (0.1–1.74) <0.01 (<0.01–0.02) <0.01 (<0.01–0.02)
PFOS
% >LOQ 100 95 99
5th–95th perc. 1.89–92.55 0.01–0.96 0.01–1.08
Median (IQR) 11.29 (3.81–36.58) 0.12 (0.03–0.32) 0.13 (0.04–0.39)
Note: IQR, interquartile range; LOQ, limit of quantification; perc., percentile; PFAS, perfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHpS, perfluoroheptane sulfonic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; PFUnDA, perfluoroundecanoic acid.
PFAS concentrations were generally higher in women attending maternal health care in Ronneby (n=103) than in women receiving care in Karlshamn (n=23) (Table S4 and Figure S2). When we used serum PFHxS concentrations to group women by their exposure levels, 25 women were categorized as being in the background group (serum PFHxs ≤0.78 ng/mL), 76 women in the intermediate group (0.78 <serum PFHxs ≤36 ng/mL), and 25 women in the highly exposed group (serum PFHxS >36 ng/mL) (Table 1, Tables S5 and S6, and Figure S1). Concentrations of AFFF-associated PFAS were much higher among the intermediate and high exposed groups compared with the background group. For example, the median serum concentration of PFHxS in the high exposure group was 56 ng/mL compared with 6 ng/mL in the intermediate group and 0.4 ng/mL in the background group (Table S6). PFAS concentrations in colostrum and breast milk were similarly higher in the intermediate and highly exposed groups compared with the background group (Tables S7 and S8).
The relative contribution of each PFAS to the sum of seven PFAS (PFOA, PFNA, PFDA, PFUnDA, PFHxS, PFHpS, and PFOS) varied by both exposure group and sample matrix (serum, colostrum, and breast milk) (Figure 3 and Table S9). The average relative contribution of PFHxS to the sum of PFAS was higher in the intermediate and high exposure groups than the background group for all matrices. On average, PFOA contributed a higher percentage of total PFAS in colostrum and breast milk than in serum, whereas PFOS had a higher relative contribution in serum than in colostrum and breast milk.
Figure 3. The average relative contribution of each PFAS to the sum of seven PFAS (PFOA, PFNA, PFDA, PFUnDA, PFHxS, PFHpS, and PFOS), by matrix and exposure category. The underlying numeric data for this figure are presented in Table S9. Note: PFAS, perfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHpS, perfluoroheptane sulfonic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; PFUnDA, perfluoroundecanoic acid.
Figure 3 is a set of three stacked bar graphs, plotting percent contribution, ranging from 0 to 100 in increments of 25 (y-axis) across serum, colostrum, and breast milk, each in background, intermediate, and high exposure groups (x-axis) for pfas substances, perfluorooctanoic acid, other perfluoroalkyl substances, perfluorononanoic acid, perfluorooctane sulfonic acid, and perfluorohexane sulfonic acid.
Concentrations of PFOA, PFHxS, PFHpS, and PFOS were highly correlated with one another within each matrix. The strongest correlations were found for PFOS and PFHxS (Figure S3). For example, PFOS and PFHxS were strongly correlated with one another in serum (rho=0.97), colostrum (rho=0.95), and breast milk (rho=0.94). Each of these PFAS were also highly correlated across the three matrices. For example, PFHxS concentrations in maternal serum were strongly correlated with PFHxS concentrations in colostrum (rho=0.94) and breast milk (rho=0.96). Concentrations of other PFAS correlated somewhat across matrices (e.g., PFNA correlation between serum and breastmilk=0.43) but less with other PFAS. When we evaluated correlations separately in the background exposure group and in a combined intermediate and high exposure group, correlations across AFFF-related PFAS were much stronger in the combined intermediate- and high-exposure group (Figures S4 and S5).
Transfer Efficiency
TEs for each PFAS and lactation stage are shown in Table 3 and Figure 4. Median TEs varied from 0.9% (TEC:S for PFOS, n=81) to 4.3% (TEC:S for PFUnDA, n=8) and were higher for perfluoroalkyl carboxylic acids (PFCAs: PFOA, PFNA, PFDA, and PFUnDA) compared with perfluoroalkane sulfonic acids (PFSAs: PFHxS, PFHpS, and PFOS). For example, the median TEB:S for PFOA and PFNA was 2.2% (IQR: 1.7–3.1) and 2.5% (IQR: 1.8–3.4), whereas for PFHxS and PFOS it was 1.2% (IQR: 1.0–1.5) and 1.0% (IQR: 0.8–1.2), respectively. For all PFAS except PFOS, the median TEC:S was higher than the median TEB:S.
Table 3 Transfer efficiencies (%) of PFAS from maternal serum at delivery into colostrum (TEC:S) and breast milk (TEB:S), limited to samples >LOQ.
PFAS N Median (IQR) Range
TEC:S
PFOA 83 2.8 (2.31–3.81) 0.79–8.2
PFNA 38 3.18 (2.25–3.85) 1.2–12.5
PFDA 2 3.27 (2.83–3.72) 2.38–4.17
PFUnDA 8 4.26 (3.71–5.54) 2.33–9.09
PFHxS 60 1.65 (1.17–2.21) 0.68–4.13
PFHpS 29 2.19 (1.68–3.16) 0.68–5.31
PFOS 81 0.86 (0.65–1.19) 0.33–3.06
TEB:S
PFOA 103 2.16 (1.73–3.08) 0.79–6
PFNA 82 2.5 (1.79–3.42) 0.98–7.69
PFDA 2 2.72 (2–3.45) 1.28–4.17
PFUnDA 3 2.33 (2.06–2.95) 1.79–3.57
PFHxS 77 1.24 (1–1.5) 0.36–2.94
PFHpS 36 1.33 (0.93–1.87) 0.38–12.5
PFOS 108 1.02 (0.85–1.2) 0.37–3.26
Note: B, breast milk; C, colostrum; IQR, interquartile range; LOQ, limit of quantification; PFAS, perfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHpS, perfluoroheptane sulfonic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; PFUnDA, perfluoroundecanoic acid; S, serum; TE, transfer efficiency.
Figure 4. Distribution of PFAS TEs (%) from maternal serum at delivery into colostrum (TEC:S) and breast milk (TEB:S), limited to samples >LOQ. Tukey outliers are marked as points, whereas individual TEs when n≤5 are marked as Xs. The underlying numeric data for this figure are presented in Table 3. Note: B, breast milk; C, colostrum; LOQ, limit of quantification; PFDA, perfluorodecanoic acid; PFHpS, perfluoroheptane sulfonic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; PFUnDA, perfluoroundecanoic acid; S, serum; TE, transfer efficiency.
Figure 4 is a graph of transfer efficiency (percentage), ranging from 0–12 in increments of 4 (y-axis). The data is shown on the x-axis by perfluoroalkyl substance (perfluorooctanoic acid, perfluorononanoic acid, perfluorodecanoic acid, perfluoroundecanoic acid, perfluorohexane sulfonic acid, perfluoroheptane sulfonic acid, and perfluorooctane sulfonic acid) and by milk type (transfer efficiency into colostrum or breast milk).
Four PFAS (PFOA, PFNA, PFHxS, and PFOS) were >LOQ in >50% of combined colostrum and breast milk samples and were included in further analyses. When we stratified TEs by exposure group, there was no clear pattern in TE by exposure (Table 4). In our first mixed-effects model, both PFAS and lactation stage were significant predictors of TE (p<0.001 for PFAS; p<0.001 for lactation stage) although the directional effect of lactation stage was not consistent across PFAS (Table S10). In our models of TE by PFAS and exposure level (stratified by sample type), exposure level was also overall a significant predictor of TE in both colostrum (p=0.028) and breast milk (p<0.001). However, the direction of the exposure-level effect was not consistent across PFAS, agreeing with our stratified analysis (Tables S11 and S12).
Table 4 Transfer efficiencies (%) of PFAS from maternal serum at delivery into colostrum (TEC:S) and breast milk (TEB:S), stratified by exposure group.
Exposure categorya N Transfer efficiency [median (IQR)]
PFOA PFNA PFHxS PFOS
TEC:S
Background 13 2.94 (2–3.77) 2.58 (1.78–3.1) 1.86 (1.31–2.36) 0.66 (0.4–0.92)
Intermediate 56 2.74 (1.99–3.82) 2.60 (1.81–3.74) 1.33 (0.85–1.87) 0.76 (0.61–1.09)
High 16 2.83 (2.57–3.65) 2.63 (2.03–3.64) 1.81 (1.23–2.27) 1.13 (0.92–1.26)
TEB:S
Background 21 2.25 (1.82–3.45) 3.03 (2.27–3.57) 1.86 (1.39–2.36) 0.99 (0.78–1.2)
Intermediate 68 2.18 (1.72–2.96) 2.63 (2.09–3.98) 1.12 (0.79–1.36) 1.00 (0.83–1.2)
High 20 1.98 (1.7–2.51) 1.94 (1.57–2.89) 1.41 (1.14–1.54) 1.09 (1–1.19)
Note: Samples <LOQ were included and substituted with the LOQ divided by the square root of 2 when a measured concentration was not available. B, breast milk; C, colostrum; IQR, interquartile range; LOQ, limit of quantification; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; S, serum; TE, transfer efficiency.
a Exposure levels were categorized based on maternal serum PFHxS concentrations (CPFHxS, ng/mL) at delivery. Background: CPFHxS≤0.78. Intermediate: 0.78<CPFHxS≤36. High: 36<CPFHxS.
Discussion
In this study, we examined the transfer of PFAS from maternal serum into colostrum and breast milk in a cohort of women with a wide range of PFAS exposures. PFAS concentrations were correlated across both sample matrices and PFAS compound. In the intermediate and high exposure groups, correlations were extremely strong for AFFF-associated PFAS, including PFHxS and PFOS. This confirmed that AFFF runoff was indeed the dominant source of PFAS exposure in this population. Median TEs varied from 0.9% (TEC:S for PFOS) to 4.3% (TEC:S for PFUnDA), and were higher for PFCAs than PFSAs. TEs in colostrum generally had a wider spread than those in breast milk, possibly reflecting the larger interindividual variation of colostrum composition.33 In our mixed-effects models, we found significant differences in TE by PFAS and lactation stage. When we stratified women by exposure categories (background, intermediate, and high), there were also some differences in TE by exposure level. However, there was not a consistent pattern for these differences, and the magnitude was small compared with differences in TE by PFAS compound.
Comparison with Previous Studies
We conducted an informal literature review on PubMed in November 2021 with the search terms (“PFAS” OR perfluoro*) AND (breastmilk OR “breast milk” OR “milk, human” [MeSH Terms]) AND (“exposure” OR “transfer”) and identified five existing studies that estimated TEs in breast milk (Table 5). Compared with these previous studies, our study had a wider range of maternal AFFF-associated PFAS exposures and included more highly exposed women. For example, the median maternal serum PFHxS concentration in our study was 6.02 ng/mL (range: 0.15–189), compared with the median maternal concentration of 0.62 ng/mL (range: 0.8–31) in a cohort of 100 women in France,20 0.07 ng/mL (range: 0.012–0.36) in a cohort of 50 women in China,18 and 0.39 ng/mL (range: 0.16–4.1) in a cohort of 41 women in Norway.21 Similarly, the median maternal serum PFOS concentration in our study was 11.3 ng/mL (range: 0.92–310), compared with a median maternal concentration of 3.06 ng/mL (range: 0.32–24.5) in Cariou et al.,20 2.92 ng/mL (range: 0.83–13.2) in Liu et al.,18 and 6.7 ng/mL (range: 2.3–15) in Haug et al.21 This wide range in exposures allowed us to investigate potential variation in TEs by exposure levels, as well as estimate TEs for several PFAS compounds (PFDA, PFUnDA, and PFHpS) that were unquantifiable in previous studies.
Table 5 Literature comparison.
Country Reference Total paired
samples (N) Sampling period Transfer efficiency (N of paired samples >LOQ in breast milk) [% (N)]
Maternal serum Breast milk PFOA PFNA PFHxS PFOS
Sweden Kärrman et al.,22a 12 3 wk postpartum 3 wk postpartum 12 (1) 1 (2) 2 (12) 1 (12)
Norway Haug et al.,21a 19 After delivery After delivery 3.8 (10) — — 1.4 (19)
Korea Kim et al.,23a 17 1 d prior to delivery 3–10 d postpartum 2.5 (8) — 0.8 (15) 1.1 (17)
China Liu et al.,18b 50 1 wk postpartum 1 wk postpartum 9 (50) 4 (50) — 2 (50)
France Cariou et al.,20a,c 19 At delivery 4–5 d postpartum 3.8 (10) — 1.2 (9) 1.4 (19)
Sweden Present Studyb 85 At delivery 3–4 d postpartum 2.8 (83) 3.2 (38) 1.7 (60) 0.9 (81)
Sweden Present Studyb 109 At delivery 4–12 wk postpartum 2.2 (103) 2.5 (82) 1.2 (77) 1.0 (108)
Note: —, not applicable; LOQ, limit of quantification; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; TE, transfer efficiency.
a TE summarized as mean.
b TE summarized as median.
c TEs from the 2015 paper by Cariou et al.20 were received by direct correspondence with the authors and reflect a correction to their original publication.
Our results generally agree with previous studies on the transfer of PFOA and PFOS into breast milk (Table 5). In the five previous studies that estimated TEs for both PFOA and PFOS (all conducted at low levels of exposure), all five found that PFOA had a higher mean or median TE than PFOS.18,20–23 The estimated TEs for PFOS were between 1% and 2%, similar to our estimated median TEC:S of 0.9% and TEB:S of 1.0%. Estimated TEs for PFOA were more variable across previous studies. This also makes sense in the context of our results given that we found that the TEs of PFOA had a greater spread than those of PFOS.
Existing literature on the transfer of PFHxS and PFNA into human milk is limited and inconsistent. Two previous studies estimated TEs for PFNA in paired maternal serum and breast milk samples, although one study only included two samples.22 In the study by Liu et al. (N=50), the estimated median TE of PFNA was 4%,18 which is higher than the median TEC:S and TEB:S for PFNA found in the present study. We found that the median TEC:S and TEB:S were higher for PFNA than the corresponding median TEs for PFOS, a result which agrees with Liu et al.18 Notably, this contradicts the 2020 European Food Safety Authority (EFSA) report, “Risk to Human Health Related to the Presence of Perfluoroalkyl Substances in Food,” which stated “[t]he [transfer] ratio… for PFNA… occurs at similar levels for PFOS but at a lesser extent than for PFOA.”11
Three previous studies estimated TE for PFHxS, with estimated mean TEs varying from 0.8% to 2%. However, these three studies were limited by their small sample size and by the low concentrations of PFHxS in maternal serum, which ranged from 0.89 to 4.7 ng/mL.20,22,23 In contrast, we estimated TEs in a large population with a wide range of PFHxS exposures (mean maternal serum concentration=18.8 ng/mL), and found TE values that fell in the middle of these estimates (median TEC:S=1.7%; median TEB:S=1.2%). Because AFFF-contaminated groundwater often has a high level of PFHxS34–36 and high concentrations of PFHxS have been found in AFFF-exposed populations,37–39 this TE is of particular concern for contaminated populations.
This study is one of the first to quantify PFHpS in breast milk40 and provides the first estimates of TEs for this PFAS. PFHpS concentrations were elevated in the serum of exposed mothers in this cohort, as well in exposed individuals from a larger biomonitoring cohort in Ronneby.24 PFHpS has also been identified as a component PFAS in several AFFF formulations.41 This indicates that, as with PFHxS, PFHpS contamination may be of particular concern in AFFF-exposed communities, such as Ronneby. We found that the median TEC:S and TEB:S were the highest of the three PFSAs included in the study, although they were lower than the median TEs for PFCAs. The health effects of early life PFHpS exposures remain relatively unstudied.
Potential Biological Mechanisms
The specific biological processes regulating the transfer of PFAS from maternal serum into breast milk are not known. The mechanism(s) likely differ from those of other persistent organic pollutants given that PFAS do not accumulate in lipids and fatty tissue.42 Instead, PFAS in human blood are found primarily bound to albumin, the most abundant protein in the blood.43,44 Serum albumin is transported directly from plasma into human milk via transcytosis.45 Some nutrients, including zinc and copper, have been shown to cotransport with albumin,46 and it is possible that PFAS are also cotransferred this way. If so, PFAS-specific TEs may partially reflect differences in PFAS-albumin binding affinity. The concentration of albumin in colostrum and breast milk is relatively consistent,47 which could explain why we did not see large differences in TEs by lactation stage.
Another potential biological mechanism that could determine the transfer of PFAS into breast milk is active transport. PFAS have similar structures to those of natural fatty acids and may be transported in a similar manner. Long-chain fatty acids are transported across the basal layer of the mammary epithelium by specific transport proteins, such as CD36 [fatty acid translocase (FAT)] and members of the fatty acid transport proteins/solute carrier 27 (FATP/SLC2.7A) and acyl-CoA synthetase (ACSL) families, although the precise mechanisms regulating this active transfer are still unknown.48 Other transporters, especially those in the ATP-binding cassette (ABC−) and Solute Carrier (SLC−) superfamilies, are known to be highly active during lactation and act as mediators for the active transport of other toxic chemicals, including specific drugs and environmental pollutants,49 and may play a role. A better understanding of the biological processes governing the transfer of PFAS would help scientists anticipate the risk of transfer for PFAS that have not been directly measured in populations. In addition, it may help elucidate why some women have higher transfer ratios than others and identify infants at risk of high exposure.
Strengths and Limitations
The Ronneby Mother–Child Cohort is a unique cohort in which many of the women were exposed to highly contaminated drinking water for several decades. As a result, exposures in our study population spanned a wide range, with maternal serum concentrations varying from background to extremely high levels of AFFF-associated PFAS. This allowed us to explicitly test whether maternal exposure levels impacted PFAS TE. One limitation of the study was that, as in previous studies, concentrations of PFNA and PFHxS in colostrum and breast milk from low-exposed mothers were often <LOQ. This may have limited our power to detect differences in TEs across exposure groups. Other PFAS in our study (PFDA and PFUnDA) also had a low rate of quantification across all colostrum and breast milk samples, so the estimated TEs for these PFAS compounds may not be as accurate. A second limitation was that women from background-exposed Karlshamn were more likely to provide colostrum and breast milk samples than highly exposed women from Ronneby, again possibly limiting our power to detect differences by exposure level. Despite this, we were still able to include many women at high levels of exposure. The method of milk collection (hand expression vs. a pump) and time of sample was not standardized, which may have induced some sample heterogeneity. However, given that all women received the same collection instructions, this should not induce any systematic bias in our results. Finally, both TEC:S and TEB:S were estimated using the same maternal serum concentrations collected at delivery. Maternal serum PFAS concentrations have been shown to decrease over lactation,17 and therefore our estimates of TEB:S may also reflect a decreasing maternal body burden. However, obtaining paired serum and breast milk samples at two occasions would have required additional health care visits and would likely have come at the cost of declining participation rate and reduced study power.
This study adds much needed information to the existing literature. To the best of our knowledge, it is the largest study of breast milk TE to date and the first large study of PFHxS and PFHpS, two PFAS that are often elevated in AFFF-exposed communities.24,50 In addition, this study was one of the first to compare the transfer of multiple PFAS, and we found a strong pattern suggesting that functional group may play an important role in TE. Finally, this is the first study that we are aware of to measure TEs in both colostrum and breast milk.
Conclusions
In this large study of women with a wide range of PFAS exposures, we found that PFAS concentrations in colostrum and breast milk were much lower than in maternal serum at delivery. Median TEs of seven PFAS varied from 0.9% to 4.3%. Given that cumulative infant exposure is a function of maternal serum concentrations multiplied by the TE and by the cumulative milk consumption, our finding that TE is similar across exposure levels suggests that infants of highly exposed mothers are at risk of high exposures from breast milk. We also found a wide spread of TEs across individuals, especially for PFOA and PFNA, indicating that some infants may be exposed to higher levels of PFAS than others. The use of one summary TE value in risk assessments may mask this large variation and understate the risk for some children. Future research should evaluate the contribution of breastfeeding to cumulative prenatal and infancy PFAS exposures, with the ultimate goal of developing evidence-based breastfeeding recommendations for highly exposed populations.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We acknowledge the participating mothers and the staff at the Karlskrona hospital, who donated time and collected the samples. Thanks also to M. Lewandowski for her administrative support.
This work was funded by the Swedish Research Council for the Environment, Agricultural Sciences and Spatial Planning, Formas [grants 216-2014-1709 (to K.J.), 2017-00896 (to C.N.), and 2019-02344 (to C.N.)]. The funding source was not involved in the design or execution of the research. In-kind funding from the University of Gothenburg is also acknowledged.
==== Refs
References
1. Buck RC, Franklin J, Berger U, Conder JM, Cousins IT, de Voogt P, et al. 2011. Perfluoroalkyl and polyfluoroalkyl substances in the environment: terminology, classification, and origins. Integr Environ Assess Manag 7 (4 ):513–541, PMID: , 10.1002/ieam.258.21793199
2. Sunderland EM, Hu XC, Dassuncao C, Tokranov AK, Wagner CC, Allen JG. 2019. A review of the pathways of human exposure to poly- and perfluoroalkyl substances (PFASs) and present understanding of health effects. J Expo Sci Environ Epidemiol 29 (2 ):131–147, PMID: , 10.1038/s41370-018-0094-1.30470793
3. Cousins IT, Goldenman G, Herzke D, Lohmann R, Miller M, Ng CA, et al. 2019. The concept of essential use for determining when uses of PFASs can be phased out. Environ Sci Process Impacts 21 (11 ):1803–1815, PMID: , 10.1039/c9em00163h.31204421
4. De Silva AO, Armitage JM, Bruton TA, Dassuncao C, Heiger-Bernays W, Hu XC, et al. 2021. PFAS exposure pathways for humans and wildlife: a synthesis of current knowledge and key gaps in understanding. Environ Toxicol Chem 40 (3 ):631–657, PMID: , 10.1002/etc.4935.33201517
5. Lindstrom AB, Strynar MJ, Libelo EL. 2011. Polyfluorinated compounds: past, present, and future. Environ Sci Technol 45 (19 ):7954–7961, PMID: , 10.1021/es2011622.21866930
6. European Commission. 2020. Commission staff working document: poly- and perfluoroalkyl substances. https://ec.europa.eu/environment/pdf/chemicals/2020/10/SWD_PFAS.pdf [accessed 9 August 2021].
7. U.S. Environmental Protection Agency. 2021. PFAS Strategic Roadmap: EPA’s Commitments to Action 2021–2024. https://www.epa.gov/system/files/documents/2021-10/pfas-roadmap_final-508.pdf [accessed 11 November 2022].
8. Hu XC, Andrews DQ, Lindstrom AB, Bruton TA, Schaider LA, Grandjean P, et al. 2016. Detection of poly- and perfluoroalkyl substances (PFASs) in U.S. drinking water linked to industrial sites, military fire training areas, and wastewater treatment plants. Environ Sci Technol Lett 3 (10 ):344–350, PMID: , 10.1021/acs.estlett.6b00260.27752509
9. Goldenman G, Fernandes M, Holland M, Tugran T, Nordin A, Schoumacher C, et al. 2019. The Cost of Inaction: a Socioeconomic Analysis of Environmental and Health Impacts Linked to Exposure to PFAS. http://norden.diva-portal.org/smash/get/diva2:1295959/FULLTEXT01.pdf [accessed 13 July 2021].
10. Rappazzo KM, Coffman E, Hines EP. 2017. Exposure to perfluorinated alkyl substances and health outcomes in children: a systematic review of the epidemiologic literature. Int J Environ Res Public Health 14 (7 ):691, PMID: , 10.3390/ijerph14070691.28654008
11. EFSA Panel on Contaminants in the Food Chain, Schrenk D, Bignami M, Bodin L, Chipman JK, del Mazo J. 2020. Risk to human health related to the presence of perfluoroalkyl substances in food. EFSA J 18 (9 ):e06223, PMID: , 10.2903/j.efsa.2020.6223.32994824
12. Lau C, Butenhoff JL, Rogers JM. 2004. The developmental toxicity of perfluoroalkyl acids and their derivatives. Toxicol Appl Pharmacol 198 (2 ):231–241, PMID: , 10.1016/j.taap.2003.11.031.15236955
13. Liew Z, Goudarzi H, Oulhote Y. 2018. Developmental exposures to perfluoroalkyl substances (PFASs): an update of associated health outcomes. Curr Environ Health Rep 5 (1 ):1–19, PMID: , 10.1007/s40572-018-0173-4.29556975
14. Barouki R, Gluckman PD, Grandjean P, Hanson M, Heindel JJ. 2012. Developmental origins of non-communicable disease: implications for research and public health. Environ Health 11 (1 ):42, PMID: , 10.1186/1476-069X-11-42.22715989
15. Section on Breastfeeding: Eidelman AI, Schanler RJ, Johnston M, Landers S, Noble L. 2012. Breastfeeding and the use of human milk. Pediatrics 129 (3 ):e827–e841, PMID: , 10.1542/peds.2011-3552.22371471
16. Mogensen UB, Grandjean P, Nielsen F, Weihe P, Budtz-Jørgensen E. 2015. Breastfeeding as an exposure pathway for perfluorinated alkylates. Environ Sci Technol 49 (17 ):10466–10473, PMID: , 10.1021/acs.est.5b02237.26291735
17. Mondal D, Weldon RH, Armstrong BG, Gibson LJ, Lopez-Espinosa MJ, Shin HM, et al. 2014. Breastfeeding: a potential excretion route for mothers and implications for infant exposure to perfluoroalkyl acids. Environ Health Perspect 122 (2 ):187–192, PMID: , 10.1289/ehp.1306613.24280536
18. Liu J, Li J, Liu Y, Chan HM, Zhao Y, Cai Z, et al. 2011. Comparison on gestation and lactation exposure of perfluorinated compounds for newborns. Environ Int 37 (7 ):1206–1212, PMID: , 10.1016/j.envint.2011.05.001.21620474
19. LaKind JS, Verner MA, Rogers RD, Goeden H, Naiman DQ, Marchitti SA, et al. 2022. Current breast milk PFAS levels in the United States and Canada: after all this time, why don’t we know more? Environ Health Perspect 130 (2 ):25002, PMID: , 10.1289/EHP10359.35195447
20. Cariou R, Veyrand B, Yamada A, Berrebi A, Zalko D, Durand S, et al. 2015. Perfluoroalkyl acid (PFAA) levels and profiles in breast milk, maternal and cord serum of French women and their newborns. Environ Int 84 :71–81, PMID: , 10.1016/j.envint.2015.07.014.26232143
21. Haug LS, Huber S, Becher G, Thomsen C. 2011. Characterisation of human exposure pathways to perfluorinated compounds—comparing exposure estimates with biomarkers of exposure. Environ Int 37 (4 ):687–693, PMID: , 10.1016/j.envint.2011.01.011.21334069
22. Kärrman A, Ericson I, van Bavel B, Darnerud PO, Aune M, Glynn A, et al. 2007. Exposure of perfluorinated chemicals through lactation: levels of matched human milk and serum and a temporal trend, 1996–2004, in Sweden. Environ Health Perspect 115 (2 ):226–230, PMID: , 10.1289/ehp.9491.17384769
23. Kim SK, Lee KT, Kang CS, Tao L, Kannan K, Kim KR, et al. 2011. Distribution of perfluorochemicals between sera and milk from the same mothers and implications for prenatal and postnatal exposures. Environ Pollut 159 (1 ):169–174, PMID: , 10.1016/j.envpol.2010.09.008.20932617
24. Xu Y, Nielsen C, Li Y, Hammarstrand S, Andersson EM, Li H, et al. 2021. Serum perfluoroalkyl substances in residents following long-term drinking water contamination from firefighting foam in Ronneby, Sweden. Environ Int 147 :106333, PMID: , 10.1016/j.envint.2020.106333.33360412
25. Lantmäteriet (Swedish mapping and land registration authority). 2021. GSD-Sverigekartan 1:1 miljon [in Swedish]. https://www.lantmateriet.se/sv/Kartor-och-geografisk-information/geodataprodukter/produktlista/sverigekartor/ [accessed 19 January 2022].
26. Norén E, Lindh C, Glynn A, Rylander L, Pineda D, Nielsen C. 2021. Temporal trends, 2000–2017, of perfluoroalkyl acid (PFAA) concentrations in serum of Swedish adolescents. Environ Int 155 :106716, PMID: , 10.1016/j.envint.2021.106716.34144476
27. Thomsen C, Haug LS, Stigum H, Frøshaug M, Broadwell SL, Becher G. 2010. Changes in concentrations of perfluorinated compounds, polybrominated diphenyl ethers, and polychlorinated biphenyls in Norwegian breast-milk during twelve months of lactation. Environ Sci Technol 44 (24 ):9550–9556, PMID: , 10.1021/es1021922.21090747
28. Hyndman RJ, Fan Y. 1996. Sample quantiles in statistical packages. Am Stat 50 (4 ):361–365, 10.1080/00031305.1996.10473566.
29. Pinheiro JC, Bates DM. 2006. Mixed-Effects Models in S and S-PLUS. Dordrecht, Netherlands: Springer Science & Business Media.
30. Wickham H, Averick M, Bryan J, Chang W, D’Agostino McGowan L, François R, et al. 2019. Welcome to the Tidyverse. J Open Source Softw 4 (43 ):1686, 10.21105/joss.01686.
31. Pinheiro J, Bates D, DebRoy S, Sarkar D, Heisterkamp S, Van Willigen B, et al. 2021. nlme: linear and nonlinear mixed effects models. https://CRAN.R-project.org/package=nlme [accessed 24 January 2022].
32. Pebesma E. 2018. Simple features for R: standardized support for spatial vector data. R J 10 (1 ):439–446, 10.32614/RJ-2018-009.
33. Gidrewicz DA, Fenton TR. 2014. A systematic review and meta-analysis of the nutrient content of preterm and term breast milk. BMC Pediatr 14 (1 ):216, PMID: , 10.1186/1471-2431-14-216.25174435
34. Backe WJ, Day TC, Field JA. 2013. Zwitterionic, cationic, and anionic fluorinated chemicals in aqueous film forming foam formulations and groundwater from U.S. military bases by nonaqueous large-volume injection HPLC-MS/MS. Environ Sci Technol 47 (10 ):5226–5234, PMID: , 10.1021/es3034999.23590254
35. Dauchy X, Boiteux V, Colin A, Hémard J, Bach C, Rosin C, et al. 2019. Deep seepage of per- and polyfluoroalkyl substances through the soil of a firefighter training site and subsequent groundwater contamination. Chemosphere 214 :729–737, PMID: , 10.1016/j.chemosphere.2018.10.003.30293026
36. Hatton J, Holton C, DiGuiseppi B. 2018. Occurrence and behavior of per-and polyfluoroalkyl substances from aqueous film-forming foam in groundwater systems. Remediation (NY) 28 (2 ):89–99, 10.1002/rem.21552.
37. Barton KE, Starling AP, Higgins CP, McDonough CA, Calafat AM, Adgate JL. 2020. Sociodemographic and behavioral determinants of serum concentrations of per- and polyfluoroalkyl substances in a community highly exposed to aqueous film-forming foam contaminants in drinking water. Int J Hyg Environ Health 223 (1 ):256–266, PMID: , 10.1016/j.ijheh.2019.07.012.31444118
38. Rotander A, Toms LML, Aylward L, Kay M, Mueller JF. 2015. Elevated levels of PFOS and PFHxS in firefighters exposed to aqueous film forming foam (AFFF). Environ Int 82 :28–34, PMID: , 10.1016/j.envint.2015.05.005.26001497
39. Trowbridge J, Gerona RR, Lin T, Rudel RA, Bessonneau V, Buren H, et al. 2020. Exposure to perfluoroalkyl substances in a cohort of women firefighters and office workers in San Francisco. Environ Sci Technol 54 (6 ):3363–3374, PMID: , 10.1021/acs.est.9b05490.32100527
40. Zheng G, Schreder E, Dempsey JC, Uding N, Chu V, Andres G, et al. 2021. Per- and polyfluoroalkyl substances (PFAS) in breast milk: concerning trends for current-use PFAS. Environ Sci Technol 55 (11 ):7510–7520, PMID: , 10.1021/acs.est.0c06978.33982557
41. Nicol L, Kreißig J, Corden C, Keyte I, Whiting R, Matulina A, et al. 2020. The Use of PFAS and Fluorine-Free Alternatives in Fire-Fighting Foams. https://echa.europa.eu/documents/10162/28801697/pfas_flourine-free_alternatives_fire_fighting_en.pdf/d5b24e2a-d027-0168-cdd8-f723c675fa98 [accessed 7 July 2022].
42. MacManus-Spencer LA, Tse ML, Hebert PC, Bischel HN, Luthy RG. 2010. Binding of perfluorocarboxylates to serum albumin: a comparison of analytical methods. Anal Chem 82 (3 ):974–981, PMID: , 10.1021/ac902238u.20039637
43. Alesio JL, Slitt A, Bothun GD. 2022. Critical new insights into the binding of poly- and perfluoroalkyl substances (PFAS) to albumin protein. Chemosphere 287 (pt 1 ):131979, PMID: , 10.1016/j.chemosphere.2021.131979.34450368
44. Forsthuber M, Kaiser AM, Granitzer S, Hassl I, Hengstschläger M, Stangl H, et al. 2020. Albumin is the major carrier protein for PFOS, PFOA, PFHxS, PFNA and PFDA in human plasma. Environ Int 137 :105324, PMID: , 10.1016/j.envint.2019.105324.32109724
45. Monks J, Neville MC. 2004. Albumin transcytosis across the epithelium of the lactating mouse mammary gland. J Physiol 560 (pt 1 ):267–280, PMID: , 10.1113/jphysiol.2004.068403.15297572
46. Lee S, Kelleher SL. 2016. Molecular regulation of lactation: the complex and requisite roles for zinc. Arch Biochem Biophys 611 :86–92, PMID: , 10.1016/j.abb.2016.04.002.27059852
47. Lönnerdal B, Erdmann P, Thakkar SK, Sauser J, Destaillats F. 2017. Longitudinal evolution of true protein, amino acids and bioactive proteins in breast milk: a developmental perspective. J Nutr Biochem 41 :1–11, PMID: , 10.1016/j.jnutbio.2016.06.001.27771491
48. McManaman JL. 2020. Lipid Transport Across the Mammary Gland. In: Ion Transport Across Epithelial Tissues and Disease. Hamilton KL, Devor DC, eds. Cham, Switzerland: Springer, 241–277.
49. García-Lino AM, Álvarez-Fernández I, Blanco-Paniagua E, Merino G, Álvarez AI. 2019. Transporters in the mammary gland—contribution to presence of nutrients and drugs into milk. Nutrients 11 (10 ):2372, PMID: , 10.3390/nu11102372.31590349
50. Nilsson S, Smurthwaite K, Aylward LL, Kay M, Toms LM, King L, et al. 2022. Serum concentration trends and apparent half-lives of per- and polyfluoroalkyl substances (PFAS) in Australian firefighters. Int J Hyg Environ Health 246 :114040, PMID: , 10.1016/j.ijheh.2022.114040.36162311
|
PMC009xxxxxx/PMC9875724.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36695591
EHP11155
10.1289/EHP11155
Research
Prenatal Exposure to PM2.5 Oxidative Potential and Lung Function in Infants and Preschool- Age Children: A Prospective Study
https://orcid.org/0000-0003-0898-1467
Marsal Anouk 1 7
Slama Rémy 2
Lyon-Caen Sarah 2
Borlaza Lucille Joanna S. 1
Jaffrezo Jean-Luc 1
Boudier Anne 2 3
Darfeuil Sophie 1
Elazzouzi Rhabira 1
Gioria Yoann 2
Lepeule Johanna 2
Chartier Ryan 6
Pin Isabelle 2 3
Quentin Joane 2 4
Bayat Sam 4 5
Uzu Gaëlle 1 *
Siroux Valérie 2 *
the SEPAGES cohort study group
1 Université Grenoble Alpes, Centre national de la recherche scientifique (CNRS), INRAE, IRD, INP-G, IGE (UMR 5001), Grenoble, France
2 Team of Environmental Epidemiology Applied to the Development and Respiratory Health, Institute for Advanced Biosciences, Inserm U 1209, CNRS UMR 5309, Université Grenoble Alpes, 38000 Grenoble, France
3 Pediatric Department, CHU Grenoble Alpes, Grenoble, France
4 Department of Pulmonology and Physiology, CHU Grenoble Alpes, Grenoble, France
5 Université Grenoble Alpes, Inserm UA07 STOBE Laboratory, Grenoble, France
6 RTI International, Research Triangle Park, North Carolina, USA
7 Agence de l’environnement et de la Maîtrise de l’Energie, Angers, France
Address correspondence to Anouk Marsal, Institut des Géosciences de l’Environnement (IGE – UMR 5001) - 460 Rue de la piscine 38058 Grenoble Cedex 9 - France). Email: [email protected]; Gaëlle Uzu ([email protected]) and Valérie Siroux ([email protected]) Institut pour l’Avancée des Biosciences, Centre de Recherche UGA/Inserm U 1209/CNRS UMR 5309, Site Santé - Allée des Alpes, 38700 La Tronche, France.
25 1 2023
1 2023
131 1 01700423 2 2022
29 11 2022
20 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Fine particulate matter (PM2.5) has been found to be detrimental to respiratory health of children, but few studies have examined the effects of prenatal PM2.5 oxidative potential (OP) on lung function in infants and preschool children.
Objectives:
We estimated the associations of personal exposure to PM2.5 and OP during pregnancy on offspring objective lung function parameters and compared the strengths of associations between both exposure metrics.
Methods:
We used data from 356 mother–child pairs from the SEPAGES cohort. PM filters collected twice during a week were analyzed for OP, using the dithiothreitol (DTT) and the ascorbic acid (AA) assays, quantifying the exposure of each pregnant woman. Lung function was assessed with tidal breathing analysis (TBFVL) and nitrogen multiple-breath washout (N2MBW) test, performed at 6 wk, and airwave oscillometry (AOS) performed at 3 y. Associations of prenatal PM2.5 mass and OP with lung function parameters were estimated using multiple linear regressions.
Results:
In neonates, an interquartile (IQR) increase in OPvDTT (0.89 nmol/min/m3) was associated with a decrease in functional residual capacity (FRC) measured by N2MBW [β=−2.26mL; 95% confidence interval (CI): −4.68, 0.15]. Associations with PM2.5 showed similar patterns in comparison with OPvDTT but of smaller magnitude. Lung clearance index (LCI) and TBFVL parameters did not show any clear association with the exposures considered. At 3 y, increased frequency-dependent resistance of the lungs (Rrs7–19) from AOS tended to be associated with higher OPvDTT (β=0.09 hPa×s/L; 95% CI: −0.06, 0.24) and OPvAA (IQR=1.14 nmol/min/m3; β=0.12 hPa×s/L; 95% CI: −0.04, 0.27) but not with PM2.5 (IQR=6.9 μg/m3; β=0.02 hPa×s/L; 95% CI: −0.13, 0.16). Results for FRC and Rrs7–19 remained similar in OP models adjusted on PM2.5.
Discussion:
Prenatal exposure to OPvDTT was associated with several offspring lung function parameters over time, all related to lung volumes. https://doi.org/10.1289/EHP11155
Supplemental Material is available online (https://doi.org/10.1289/EHP11155).
* These authors contributed equally to this work.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Exposure to ambient particulate matter (PM) increases risk of chronic respiratory diseases and triggers asthma and chronic obstructive pulmonary diseases.1–3 Early life, including pregnancy, is a vulnerable time window for the health effects of air pollution.4,5 Exposure to PM during pregnancy is reported to influence fetal and infant lung development and respiratory health.6–10
Measures of children’s respiratory health, including spirometry outcomes,11–13 asthma incidence,14 or fraction of nitric oxide in exhaled air (FeNO),15 have been widely investigated in association with outdoor air pollution. Although studying lung function of children in early childhood is of great interest for the evaluation of their susceptibility to respiratory diseases later in life, most previous studies11–15 were limited to children older than 5 y of age, when spirometry becomes feasible. Very few studies8,10,16 have used noninvasive techniques that allow for the measurement of lung function in very young children, such as tidal breathing flow-volume loops analysis (TBFVL), nitrogen multiple-breath washout (N2MBW), or airwave oscillometry (AOS). Yet, these techniques rely on tidal breathing, making them particularly suitable and feasible in population-based cohorts. Muttoo et al.16 and Latzin et al.10 found decreases in functional residual capacity (FRC) and tidal volume (VT), respectively, estimated by MBW and TBFVL, in children who had higher prenatal exposure to nitrogen oxides (NOx) or particulate matter (PM) with diameter ≤10μm (PM10). Dutta et al.8 found higher airway reactance (Xrs5) measured by AOS in children with higher postnatal exposures to particles with <2.5μm diameter (PM2.5).
Most epidemiological studies examining the health effects of PM used the mass concentration metric in association with health parameters.17,18 Although the biological pathways are not fully understood yet, evidence suggest that oxidative stress caused by PM is a key factor in understanding PM-associated health effects.19–22 The ability of PM to generate reactive oxygen species (ROS) and thereby induce oxidative stress is measured by the oxidative potential (OP), an integrative metric of several physical and chemical properties of PM and its health effects.23 Several recent studies have presented OP as a better predictor than concentration for assessing association with some cardiorespiratory diseases.24,25 The studies addressing the effects of OP exposure on children’s lung function, although few in number, converged to a stronger detrimental effect of OP as compared to PM mass.26–28 These latter studies used average urban ambient OP measurements or OP estimated by land-use regression (LUR) models, which could lead to measurement errors, given that most people in Western countries spend more than 80% of their time indoors.29 Thereby, personal sampling has been proposed to increase the accuracy in exposure assessment; however, to the best of our knowledge, no study has estimated personal prenatal exposure to OP in relation to respiratory function in the first years of life.
The aim of this study was to assess whether maternal personal exposure to PM2.5 mass concentration and to the OP of PM2.5 is associated with lung function in newborns and in preschool children. The effects of OP and PM were also compared, and the independency of OP effects from PM2.5 were tested.
Methods
Study Population
This study is based on the data from the French mother–child SEPAGES cohort that has been set up to describe maternal and child personal exposure to environmental pollutants and their effects on health. The study design and protocol have been previously described by Lyon-Caen et al.30 Briefly, pregnant women were recruited from July 2014 to July 2017 in eight obstetrical ultrasonography practices located in the Grenoble area in the French Alps. The included women had to be pregnant by <19 gestational weeks, be older than 18 y old, to have a singleton pregnancy, to be planning to give birth in one of the four maternities clinics from Grenoble area, and to live in the study area (i.e., living 1 h driving from Grenoble city center). The volunteers were then followed during pregnancy, and their children were recruited at birth and then followed up. The mother–child pairs selected for this study had at least one period of PM2.5 sampling during pregnancy (n=405), with validated and positive OP analysis (n=387) and the children had performed at least one lung function test at either 6 wk or 3 y (n=356) (Figure 1).
Figure 1. Flow chart for the selection of the study population. Note: *PM2.5 net weight <4μg. AOS, airwave oscillometry; LF, lung function; N2MBW, nitrogen multiple-breath washout; OP, oxidative potential; PM, particulate matter; PM2.5, PM with aerodynamic diameter <2.5μm; TBFVL, tidal breathing flow-volume loops.
Figure 1 is a flowchart with 6 steps. Step 1: There are 484 cases of women in SEPAGES. Step 2: There are 405 cases of children with their mother's exposure to particulate matter, after excluding 79 cases in which no personal particulate matter was sampled. Step 3: There are 387 children with oxidative potential from their mother, excluding 17 cases of no sample valid for oxidative potential analysis and 1 case of oxidative potential less than 0. Step 4: There are 325 cases of children with lung function measurements at 6 weeks, of which 284 cases are with nitrogen multiple-breath washout and 309 cases are with tidal breathing flow-volume loops. 31 cases were excluded with no clinical visit at both 6 weeks and 3 years. Out of 325, 108 cases were excluded for no clinical visit at 3 years or for airwave oscillometry not being valid, and 31 cases were excluded for no clinical visit at 6 weeks. Step 5: There are 248 cases of children with airwave oscillometry at 3 years; of those, 197 cases are with nitrogen multiple-breath washout and 205 cases are with tidal breathing flow-volume loops. Step 6: There are 356 children with their mother's oxidative potential exposure and at least one lung function measurement.
Parents signed an informed consent for themselves and their child, and the study protocol was approved by the Comité de Protection des Personnes Sud-Est V (CPP) and the French data privacy institution (Commission Nationale de l’Informatique et des Libertés, CNIL).
Maternal Exposure
Active personal air samplers (MicroPEM™; RTI International) were used to sample PM2.5 onto Teflon filters. The participants were asked to carry the devices or keep them at close proximity during the entire sampling period (consecutive 7–8 d). The measurements took place at different periods of the pregnancy. The sample filters on which OP was measured consisted of 286 collected at a median gestational age (GA) of 18 wk (min: 12, Q1: 32, Q3: 19, max: 28) and 294 at 34 wk (min: 28, Q1: 32, Q3: 35, max: 38). Therefore, the median [interquartile range (IQR)] of time between the first and second measurement was 16 (14, 18) wk, with a minimum of 4 wk and a maximum of 23 wk, mainly due to the availability of the samplers or the volunteers. For each participant, personal exposure was estimated from one (132 out of 356, 37%) or 2 wk (224 out of 356, 63%) of sampling. An average exposure was calculated when two periods of measurements were available.
The net mass (micrograms) of PM2.5 collected was determined by gravimetric analysis (Mettler Toledo UMX2 ultra-microbalance) before and after sampling at the same hygrometric conditions (21°C, 25% relative humidity). Following gravimetric analysis, the samples filters were cold-stored (−20°C) until OP analysis, for an average of 26 wk. OP analysis followed the protocol established by Calas et al.31,32 Briefly, a simulated lung fluid [SLF, mixture of Gamble and dipalmitoylphosphatidylcholine (DPPC)] was used to extract PM2.5 from the filters for a final concentration of 10μg/mL, maintaining a constant amount of extracted PM2.5 for intercomparison. The extracts were then subjected to vortex mixing at 37°C for 1.25 h. The OP was measured using the dithiothreitol (DTT) and ascorbic acid (AA) assays.
For the DTT assay, PM2.5 extracts were mixed with a DTT solution using a 96-well plate (CELLSTAR, Greiner-Bio). Every 10 min, the remaining DTT was titrated by dithionitrobenzoic acid (DTNB) and the formation of 2-nitro-5-thiobenzoic acid (TNB) was measured by absorbance at 412 nm (TECAN spectrophotometer Infinite M200 Pro), for a total reaction time of 30 min (e.g., 3 titrations in total). For the AA assay, a modified version of the synthetic respiratory tract lining fluid (RTLF) was used.33 AA was mixed with the PM2.5 extract in a 96-well plate, and the AA consumption was evaluated measuring the change in absorbance at 265 nm over time. Absorbance measurements were collected at 4-min intervals for a total reaction time of 30 min. For both assays, the consumption rate (nanomoles per minute) was then normalized by the corresponding filtrated air sample volume (cubic meters) to represent human exposure through inhalation. OPvDTT corresponds to the consumption of DTT (nanomoles per minute per cubic meter), and OPvAA corresponds to the consumption of AA (nanomoles per minute per cubic meter). All samples were subjected to triplicate analysis, and each sample result is reported as the mean of the repeated measurements. The coefficient of variation (CV) is between 0 and 10% for each assay.
To ensure accuracy of each OP measurement, positive control tests were performed for every experiment. A 1,4-naphthoquinone (1,4-NQ) solution was used for both the DTT and AA assays. Particularly, a 40μL of 24.7μM stock solution was used for the DTT assay and an 80μL of 24.7μM 1,4-NQ solution for AA assay.31,32 The measurement quality, estimated by the CV of the positive control tests, were at <3.2% for both OP assays.
Lung Function at 6 Weeks
Lung function tests were performed on infants age 6–12 wk, using an infant face mask during natural sleep, in supine position and with the head midline, following guidelines of the European Respiratory Society (ERS) and American Thoracic Society (ATS).34 After stabilization of the breathing pattern (20–30 breaths rejected), 10 min of tidal breathing flow-volume loops (TBFVL) were recorded, and three measurements of nitrogen multiple-breath washout (N2MBW) were performed.
For TBFVL measurements, the first 30 to 50 regular breaths were used. The sighs and 10 breaths preceding and following a sigh were excluded. The following TBFVL parameters were retained in the present analysis: tidal volume (VT) and the ratio of time to peak tidal expiratory flow (tPTEF) to expiratory time (tE). Out of the 484 mother–child pairs, 325 children performed the TBFVL test.
The N2MBW technique measures lung volumes and ventilation heterogeneity. For this test, infants inhaled pure oxygen (O2) and the concentration of exhaled N2 was monitored employing the Exhalyzer© and Spiroware© equipment (Ecomedics). The main outcomes were a) functional residual capacity (FRC) and b) lung clearance index (LCI), defined as the number of respirations required to reduce the concentration of N2 below 2.5%. Up to three valid measurements were obtained, guided by the following criteria: regular breathing during quiet sleep, tidal volume within target, no swallowing or sighs in the first five breaths, no sign of leak, and N2 concentration below 2.5% for at least three consecutive breaths to end the test. A transient decrease in tidal volume may be induced by using pure oxygen during the test, which has been shown to affect FRC and LCI measures.35 Hence, the degree of hypoventilation was calculated for each N2MBW test, comparing the maximum drop of tidal volume during the first 15 breaths after O2 inhalation and the mean tidal volume before inhalation. Then, FRC and LCI values were corrected for the degree of hypoventilation using a 2-step standardization method based on regression residuals.36 First, the influence of hypoventilation was characterized using adjusted linear mixed regression models (accounting for the repeated data), and, in a second step, the model estimate was used to remove the variability in FRC (or LCI) due to hypoventilation. A total of 865 valid N2MBW tests were retained, with a median (Q1; Q3) of 3 (2; 3) tests per child. Out of the 484 mother–child pairs, 350 children performed the N2MBW test. For each child, both LCI and FRC corrected values were averaged.
Lung Function at 3 Years
At the age of 3 y (median: 3.1 y), the impedance of the respiratory system was assessed based on airwave oscillometry (AOS) using commercial device (TremoFlo; Thorasys Systems) complying with current European standards.37 The device was calibrated daily, using a reference resistance.
For this technique, pressure waves with frequencies varying from 7 to 41 Hz are applied during tidal breaths and lung impedance is calculated from the changes in flow and pressure. To ensure the quality and reproducibility of the measurements, they were performed at least 15 d after any respiratory infection (self-reported by the mother via a questionnaire administrated by a clinical research assistant at the clinical visit), with the child sitting, the head slightly extended, and wearing a nose clip. Children were asked to firmly close their lips around the mouthpiece while their cheeks and chins were maintained by the technician to avoid any signal damping by the mouth walls. After getting used to the device during approximately 30 s, three to five acceptable measurements were obtained and averaged. A rest interval of 1 min was respected between each 16-s-long measurement. We excluded measurements with the following artefacts: leakage, swallowing, glottis closure, vocalization, or obstruction of the mouthpiece by the tongue.
The key components of impedance are the resistance and the reactance of the respiratory system. The resistance is representative of friction forces mainly in the airways and the reactance depends on the inertive and elastic behaviors of the respiratory system.38 The parameters included in this study are raw values of resistance and reactance at a frequency of 7 Hz (Rrs7 and Xrs7), the area under the reactance curve (AX), and the frequency dependence of the resistance, defined by the resistance difference between 7 and 19 Hz (Rrs7–19). Rrs7 is a parameter that reflects large airway resistance, whereas AX and Rrs7–19 better characterize the peripheral airways. Rrs7–19 also evaluates the heterogeneous obstruction of the distal bronchi.39 Increased Rrs, Rrs7–19, and AX and decreased Xrs are associated with a reduced lung function.
Among the 320 children to the 3-y follow-up who performed AOS (66% follow-up rate), measurements for 306 children (96% success rate for AOS test) were retained, complying with validity and reproducibility criteria (at least two measurements with CV <15% for Rrs7). The mean value of the valid measurements was calculated for each parameter and used for the analyses. Out of them, 248 had personal prenatal exposure to OP, resulting in a total attrition rate of 51% for the exposure to personal prenatal OP–AOS parameters association study.
Statistical Methods
Both univariate and multiple linear regressions were used to study the associations between maternal personal exposure to PM2.5 and OP with each lung function parameter. The three exposure metrics used in this study (PM2.5, OPvDTT, OPvAA) were continuous and scaled by their IQR, allowing to compare their respective effects on the outcomes. The Spearman correlation coefficient (rs) was used to calculate correlations between the exposures. Linear regressions were used after confirming linearity by a likelihood ratio test between the adjusted model, modeling the exposure with a natural spline with 5 degrees of freedom and the adjusted main model (Figures S1 to S6 in the Supplement, all p≤0.05). All analyses were performed using R software (version 4.1; R Development Core Team).
Potential confounders were selected a priori, based on previous studies,10,13 a) parental characteristics: educational level (defined as the maximum number of studying years after high school degree between the parents and expressed in two classes: above or <5y; self-reported through an self-administrated questionnaire), parental history of rhinitis (binary, self-reported by a questionnaire administrated by a clinical research assistant), mother’s age (calculated with the date of birth self-reported by a questionnaire administrated by a clinical research assistant) and body mass index (BMI) before pregnancy (continuous; calculated based on self-reported weight before pregnancy and height measured by a clinical research assistant during a SEPAGES clinical visit); b) infant characteristics: child sex (male/female), age (continuous, calculated with the date of birth collected in the child health booklet), height and weight (continuous, measured by a clinical research assistant at the clinical visit), passive smoking (yes/no, in utero, including maternal passive smoking or until the clinical visit; assessed by several self-administrated questionnaires during and after the pregnancy), breastfeeding (still some breastfeeding at 6 wk, yes/no, self-reported by a questionnaire administrated by a clinical research assistant); c) exposure characteristics: season of sampling [3-class variable: cold (all filters sampled between October and March), warm (all filters sampled between April and September), and cold+warm (one filter sampled in the cold season and one filter sampled in the warm season)], mean temperature during pregnancy (continuous, assessed at home address by Hough’s model).40 The effects of the confounders were analyzed by looking at the effect of each confounder separately on the regression model adjusted for sex, height, and weight (Figures S9, S10). Missing data regarding covariates in the main model were imputed by multiple chained equation, using the R package mice,41 assuming that the data was missing completely at random (MCAR), which was checked by Little’s test42 (p-values of the test >0.05). Descriptive statistics of the covariates can be found in Table S1. Ten imputed data sets were created, and results from each data set were combined using Rubin’s rule.43 We did not correct for multiple tests, but results were interpreted by looking at the consistency of association of PM and OP exposures across the different lung function parameters.
Several sensitivity analyses were conducted to address the robustness of the results from the main model by assessing the impacts of: a) data imputation, by conducting a complete case analysis; b) extreme exposure and health outcome values, by excluding the lowest (below first percentile) and highest values (above the 99th percentile) of the outcomes and exposures, resulting in the exclusion of 4%–5% of the population of each analysis; c) the number of PM and OP measurement weeks, by excluding participants with only one measurement week (n=132); d) the independency of OP effects to PM, by adjusting OP models on PM2.5; e) LCI and FRC measurement error due to the degree of hypoventilation, by adding an analysis excluding one-fourth of the children who had the highest hypoventilation degree during the N2MBW test (n=72); f) leverage and influencing points, by excluding points that had a Cook’s distance44 higher than 4/n, where n is the number of observations in the main model (exclusion of 4%–7% of the observations); g) the independency of OP and PM effects to personal NO2 concentrations during the same weeks of sampling [passive sampler (Passam AG), worn simultaneously to the active PM sampler]. Multicollinearity was assessed using the variance inflation factor in the two-pollutant models (VIFs<2).
Results
Description of the Population
The present study was conducted with children that had at least one prenatal measurement of OP and one lung function parameter assessed, leading to 356 mother–child couples (73% of SEPAGES cohort) (Figure 1). The included children had parents with a higher educational level, had less parental history of rhinitis, had higher exposure to PM2.5, higher Rrs7, and a lower Xrs7 in comparison with the children not included in the study (Table 1). No difference between the included and excluded population was observed for both OP and lung function at 6 wk. In the study population, 52% (n=185) of the children were boys, and the majority of children were born on term (96%, n=341) by vaginal delivery (85%, n=302) from mothers who were mainly nulliparous or primiparous (45%, n=160 and 46%, n=162, respectively). In infancy, most children were still breastfed at 6 wk (86%, n=306) and <27% (n=95) were exposed to tobacco smoke in utero (including maternal passive smoking) and after birth (<6 wk). The parental level of education is high because 72% (n=256) of the parents had studied 5 y or more after receiving their French high school diploma (i.e., having at least a MSc diploma). Only 15 children were born before the 37th week, with a minimum of 34 gestational weeks. Regarding lung function tests (Figure 1), 325 children performed a valid test of the lung function at 6 wk (284 had a valid N2MBW analysis and 309 had valid TBFVL measurements), and 248 children had valid AOS measurements. Out of these 248 children, 197 had available N2MBW results and 205 had valid TBFVL test results.
Table 1 Characteristics of the included (n=356) and excluded (n=128) population from the cohort SEPAGES in this study. Included population corresponds to children who had at least both one prenatal oxidative potential assessment and one test of lung function.
Characteristics Included populationa (n=356) Excluded populationa (n=128) p-Valueb
Sex of child — — 0.2
Male 185 (52%) 73 (59%) —
Female 171 (48%) 51 (41 %) —
Missing 0 4 —
Birth weight (g) — — 0.13
Median (IQR) 3,295 (3,048, 3,580) 3,220 (2,995, 3,507) —
Missing 0 5 —
Preterm birth (<37 wk) — — 0.2
0 (No) 341 (96%) 115 (93%) —
1 (Yes) 15 (4%) 9 (7%) —
Missing 0 4 —
Parental educational level >5 y — — 0.048
0 (No) 100 (28%) 48 (38%) —
1 (Yes) 256 (72%) 80 (62%) —
Delivery mode — — 0.084
Vaginal 302 (85%) 96 (78%) —
C-section 54 (15%) 27 (22%) —
Missing 0 5 —
Still breastfed at 6 wk — — 0.11
0 (No) 49 (14%) 20 (20%) —
1 (Yes) 306 (86%) 78 (80%) —
Missing 1 30 —
Parental history of rhinitis — — 0.003
0 (No) 132 (40%) 26 (24%) —
1 (Yes) 202 (60%) 83 (76%) —
Missing 22 19 —
Parity — — 0.6
0 (nulliparous) 160 (45%) 62 (48%)
1 (primiparous) 162 (46%) 52 (41%)
2 or more (multiparous) 34 (9.6%) 14 (11%)
ETS in utero and <6 wk — — 0.7
0 (No) 259 (73%) 84 (75%) —
1 (Yes) 95 (27%) 28 (25%) —
Missing 2 16 —
ETS <3 y — — 0.7
0 (No) 270 (79%) 70 (77%) —
1 (Yes) 73 (21%) 21 (23%) —
Missing 13 37 —
Exposure to particulate air pollutionc
PM2.5 (μg/m3) 13.3 (10.6, 17.5) 12.2 (8.2, 16.6) 0.033
Missing 0 79 —
OPvDTT (nmol/min/m3) 1.49 (1.11, 2.00) 1.53 (1.05, 1.91) 0.8
Missing 0 97 —
OPvAA (nmol/min/m3) 1.56 (1.07, 2.21) 1.66 (0.93, 2.30) >0.9
Missing 0 97
Mean temperature during pregnancy (°C)
Median (IQR) 13.0 (10.6, 14.6) 11.6 (10.1, 13.6) 0.001
Missing 0 4
N2MBW parametersc (6 wk)
FRC (mL) 105 (95, 115) 108 (95, 115) 0.6
Missing 72 62 —
LCI 7.58 (6.75, 8.47) 7.49 (6.99, 8.12) 0.9
Missing 72 62 —
TBFVL parametersc (6 wk)
VT (mL) 34 (29, 39) 33 (29, 36) 0.4
Missing 47 112 —
tPTEF/tE (%) 35 (29, 42) 36 (26, 45) 0.8
Missing 47 112 —
AOS parametersc (3 y)
Rrs7 (hPa×s/L) 11.53 (10.05, 13.04) 12.67 (10.87, 14.17) 0.021
Missing 108 70 —
Rrs7–19 (hPa×s/L) 1.02 (0.56, 1.61) 1.18 (0.63, 1.98) 0.2
Missing 108 70 —
Xrs7 (hPa×s/L) −3.88 (−4.56, −3.28) −4.25 (−5.59, −3.51) 0.037
Missing 108 70 —
AX (hPa/L) 68 (45, 92) 70 (53, 105) 0.3
Missing 108 70 —
Note: —, no data; AA, ascorbic acid; AOS, airwave oscillometry; DTT, dithiothreitol; AX, area under the reactance curve; ETS, environmental tobacco smoke; FRC, functional residual capacity; LCI, lung clearance index; N2MBW, nitrogen multiple-breath washout; OP, oxidative potential; OPvAA, volume-normalized oxidative potential measured by the AA assay; OPvDTT, volume-normalized oxidative potential measured by the DTT assay; PM, particulate matter; PM2.5, PM with an aerodynamic diameter <2.5μm; Rrs7, resistance at a frequency of 7 Hz; Rrs7–19, difference between the resistance at 7 Hz and at 19 Hz; TBFVL, tidal breathing flow-volume loops; tPTEF/tE ratio of time to peak tidal expiratory flow to expiratory time; VT, tidal volume, Xrs7, reactance at a frequency of 7 Hz.
a Expressed in n (%) or Median (IQR).
b p-Value from Wilcoxon rank sum test and Pearson’s chi-squared test comparing included and excluded population.
c Variables used for population selection (selected children had prenatal exposure to PM and OP and either N2MBW or TBFVL or AOS measures).
Exposure to PM2.5 and Its OP
The median (Q1, Q3) of average prenatal personal exposures to PM2.5, OPvDTT, and OPvAA were 13.3 (10.6, 17.5) μg/m3, 1.49 (1.11, 2.00) nmol/min/m3 and 1.56 (1.07, 2.21) nmol/min/m3. Personal PM2.5 and OP (particularly OPvAA) presented a seasonal trend, with higher levels reached during the cold season (Figure 2; Table S2). OPvDTT was highly correlated with both PM2.5 concentration and OPvAA (rs=0.64 and rs=0.72, respectively; n=356, p<2.2⋅10−16 for both), whereas the correlation between PM2.5 concentration and OPvAA was moderate (rs=0.51, p<2.2⋅10−16) (Figure S7). For participants with two periods of sampling, there were no differences in PM2.5, OPvAA and OPvDTT levels at early vs. late pregnancy (Figure S8; Table S3).
Figure 2. Monthly distribution of personal measurements of PM2.5 (left), OPvDTT (center), and OPvAA (right). See Table S2 for corresponding numeric data. Note: Boxes represent 25th–75th percentiles; the middle horizontal line represents the median; whiskers extend to the most extreme point within 1.5 IQRs of the box and the dots outside boxes indicate outliers. Note: AA, ascorbic acid; DTT, dithiothreitol; IQR, interquartile range; OPvAA, volume-normalized oxidative potential measured by the AA assay (nmol/min/m3); OPvDTT, volume-normalized oxidative potential measured by the DTT assay (nmol/min/m3); PM, particulate matter; PM2.5, PM with an aerodynamic diameter <2.5μm (μg/m3).
Figure 2 is a set of three box and whiskers graph, plotting particulate matter with a diameter smaller than 2.5 microns (in microgram per cubic meter), ranging from 0 to 80 in increments of 20; volume-normalized oxidative potential measured by the dithiothreitol assay (in nanomole per minute per cubic meter), ranging from 0 to 6 in increments of 2; volume-normalized oxidative potential measured by the ascorbic acid assay (in nanomole per minute per cubic meter), ranging from 0.0 to12.5 in increments of 2.5 (y-axis) across months from January to December (x-axis).
Association between Exposures to Prenatal PM2.5 and OP and Lung Function
Lung function at 6 wk.
In the univariate analysis, increased personal prenatal exposure to PM2.5 and OPvDTT were associated with a lower FRC at 6 wk (−2.16mL; 95% CI: −4.41, 0.09 for each 6.9 μg/m3 increase of PM2.5, and −2.69mL; 95% CI: −5.28, −0.11 for each 0.89 nmol/min/m3 increase of OPvDTT). After adjusting for potential confounders (Table 2; Table S4; Figure 3), in both main and complete-case analysis, the magnitude of association between OPvDTT and FRC slightly decreased, and associations were borderline significant (β: −2.26mL; 95% CI: −4.68, 0.15 for the main model and β: −2.65mL; 95% CI: −5.16, −0.14 for the complete-case analysis). The confounders mainly driving the differences between the univariate and the main analysis were the season of sampling and the parental history of rhinitis (Figure S9). LCI and tPTEF/tE did not show any clear association trend for all exposures considered. In general, for air pollution–lung function associations showing marginal association, the sensitivity analyses showed patterns of association similar to the ones in the main model, except for the negative OPvDTT-VT association that disappeared when excluding extreme values. The analyses excluding leverage and influencing points (estimated by Cook’s distance) overall led to similar results and resulted in statistically significant association for FRC and exposure to both PM2.5 and OPvDTT (Table S4; Figure S11). The analyses further adjusted on personal NO2 sampled simultaneously with PM2.5 showed that NO2 did not modify the estimates and 95% CI for any of the studied associations. The magnitude of the associations of OPvDTT for lung volumes, estimated by FRC, remained similar in models further adjusted for PM2.5. The change in FRC in the two-pollutant model with OPvDTT and PM2.5 showed a stronger effect of OPvDTT than PM2.5 [−1.82mL (95% CI: −5.03, 1.40) for OPvDTT vs. −0.59mL (95% CI: −3.37, 2.19) for PM2.5], although this association became nonsignificant (Figure 3; Table S6).
Table 2 Associations between prenatal exposure to air pollution and lung function at 6 wk and 3 y. Regression coefficients are estimated from univariate and multiple linear models.
Age Pollutants PM2.5 (μg/m3) OPvDTT (nmol/min/m3) OPvAA (nmol/min/m3)
Regression model Unadjusted coefficients (95% CI) Adjusteda coefficients (95% CI) Unadjusted coefficients (95% CI) Adjusteda coefficients (95% CI) Unadjusted coefficients (95% CI) Adjusteda coefficients (95% CI)
6 wk FRC (mL)b −2.16 (−4.41, 0.09) −1.58 (−3.67, 0.5) −2.69 (−5.28, −0.11) −2.26 (−4.68, 0.15) −1.05 (−3.31, 1.22) −0.59 (−2.85, 1.68)
LCIb −0.02 (−0.17, 0.13) −0.01 (−0.14, 0.13) −0.05 (−0.22, 0.12) −0.06 (−0.22, 0.09) −0.02 (−0.17, 0.12) −0.05 (−0.19, 0.1)
VT (mL)c −0.52 (−1.4, 0.37) −0.54 (−1.35, 0.28) −0.65 (−1.67, 0.38) −0.58 (−1.54, 0.38) −0.11 (−1.01, 0.78) 0.13 (−0.76, 1.02)
tPTEF/tE (%)c 0.34 (−0.86, 1.54) 0.25 (−1.02, 1.51) 0.8 (−0.59, 2.19) 0.69 (−0.79, 2.17) 0.42 (−0.78, 1.63) 0.14 (−1.23, 1.51)
3 y Rrs7 (hPa×s/L)d −0.01 (−0.33, 0.32) −0.02 (−0.33, 0.3) 0.12 (−0.21, 0.46) 0.05 (−0.28, 0.37) −0.04 (−0.37, 0.28) −0.08 (−0.41, 0.25)
Rrs7–19 (hPa×s/L)d −0.01 (−0.15, 0.14) 0.02 (−0.13, 0.16) 0.08 (−0.07, 0.23) 0.09 (−0.06, 0.24) 0.1 (−0.05, 0.24) 0.12 (−0.04, 0.27)
Xrs7 (hPa×s/L)d −0.01 (−0.17, 0.16) 0.01 (−0.15, 0.17) −0.09 (−0.26, 0.08) −0.05 (−0.22, 0.11) −0.07 (−0.23, 0.1) −0.07 (−0.23, 0.1)
AX (hPa/L)d 0.65 (−4.45, 5.74) 0.22 (−4.81, 5.25) 2.34 (−2.95, 7.64) 1.07 (−4.08, 6.22) −1 (−6.07, 4.06) −2.21 (−7.48, 3.07)
Note: Coefficients are calculated for an increase of one IQR for PM2.5, OPvDTT, and OPvAA, corresponding to 6.9 μg/m3, 0.89 nmol/min/m3, and 1.14 nmol/min/m3, respectively. AA, ascorbic acid; AX, area under the reactance curve; BMI, body mass index; CI, confidence interval; FRC, functional residual capacity; IQR, interquartile range; LCI, lung clearance index; OPvAA, volume-normalized oxidative potential measured by the AA assay; OPvDTT, volume-normalized oxidative potential measured by the DTT assay; PM, particulate matter; PM2.5, PM with an aerodynamic diameter <2.5μm; Rrs7, resistance at a frequency of 7 Hz; Rrs7–19, difference between the resistance at 7 Hz and at 19 Hz; tPTEF/tE ratio of time to peak tidal expiratory flow to expiratory time; VT, tidal volume; Xrs7, reactance at a frequency of 7 Hz.
a Model adjusted for child’s height, weight, sex, age, season of sampling, breastfeeding, environmental tobacco smoke, maternal age and BMI before pregnancy, parental level of education, parental history of rhinitis, and mean temperature during pregnancy.
b Number of observations is 284 for FRC and LCI.
c Number of observations is 309 for VT and tPTEF/tE.
d Number of observations is 248 for Rrs7, Rrs7–19, Xrs7, and AX.
Figure 3. Association between personal exposure to PM2.5, OPvDTT, and OPvAA during pregnancy and lung function parameters measured at 6 wk in the univariate and multiple linear models and in the sensitivity analyses. Outcomes and exposures were scaled by their IQR. See Tables S4 and S6 for corresponding numeric data. Whiskers represent the 95% confidence interval around the estimate. The main model was adjusted on child’s height, weight, sex, age, season of sampling, breastfeeding, environmental tobacco smoke, maternal age and BMI before pregnancy, parental level of education, parental history of rhinitis, and mean temperature during pregnancy. In addition, “2 sampling periods” are the analyses reduced to the children that had 2 wk of prenatal measurements of air pollution (63%–66% of the population); “Excluding extreme values” are the analyses excluding the exposures and outcomes below the first percentile and above the 99th (exclusion of approximately 5% of the population); “Adjusted on PM” corresponds to adding personal exposure to PM2.5 in the set of confounders, “Adjusted on NO2” corresponds to adding personal exposure to NO2 in the set of confounders, and the last analyses were performed excluding children that had the highest hypoventilation degree during the nitrogen multiple breath washout test (excluding 25% of the population). Note: AA, ascorbic acid; BMI, body mass index; DTT, dithiothreitol; FRC, functional residual capacity; IQR, interquartile range; LCI, lung clearance index; OPvAA, volume-normalized oxidative potential measured by the AA assay (nmol/min/m3); OPvDTT, volume-normalized oxidative potential measured by the DTT assay (nmol/min/m3); PM, particulate matter; PM2.5, PM with an aerodynamic diameter <2.5μm (μg/m3); tPTEF/tE, ratio of time to peak tidal expiratory flow to expiratory time; VT, tidal volume.
Figure 3 is a set of four forest plots, plotting (bottom to top), volume-normalized oxidative potential measured by the ascorbic acid assay, volume-normalized oxidative potential measured by the dithiothreitol assay, particulate matter with a diameter smaller than 2.5 microns (left y-axis) and tidal volume (milliliter), ratio of time to peak tidal expiratory flow to expiratory time (percentage), lung clearance index, functional residual capacity (milliliter) (right y-axis) across beta, ranging from negative 0.2 to 0.2 in increments of 0.1 (x-axis) for Analysis, including Univariate, Main model, Complete Cases, 2 sampling periods, Excluding extreme values, Adjusted on Particulate matter, Adjusted on Nitrogen dioxide, and Excluding high degrees of H V.
Lung function at 3 y.
Increased personal prenatal exposures to OPvDTT and OPvAA were associated with an increase of 0.09 (95% CI: −0.06, 0.24) and 0.12 (95% CI: −0.04, 0.27) hPa×s/L in Rrs7–19 respectively, whereas no trend for association was found with exposure to PM2.5 (β: 0.02 hPa×s/L; 95% CI: −0.13, 0.16) (Table 2; Figure 4). The confounders mainly driving the differences between the univariate and the adjusted model were the season of sampling, parental history of rhinitis, and maternal age before pregnancy (Figure S10). The sensitivity analyses confirmed these trends of association. In particular, the analysis excluding extreme values resulted in a statistically significant positive association, with an IQR increase in OP being associated with an increase of 0.20 (95% CI: 0.04, 0.36) hPa×s/L in Rrs7–19 for OPvDTT. Likewise, the model excluding leverage and influencing points led to statistically significant results with Rrs7–19 and exposure to both OPv, whereas the results for other outcomes were not modified, with their 95% CI largely overlapping with that of the main model (Table S5; Figure S12). The analyses further adjusted on personal NO2 sampled simultaneously to PM2.5 showed that NO2 did not modify the estimates and 95% CI for any of the studied association. The two-pollutant models for Rrs7–19 showed that the effects of both OPv were stronger than the effects of PM2.5 [0.14 (−0.06, 0.34) and −0.07 (−0.27, 0.13) hPa×s/L for OPvDTT and PM2.5; 0.15 (−0.03, 0.33) and −0.05 (−0.22, 0.12) hPa×s/L for OPvAA and PM2.5], and other associations were not modified in this model (Figure 4; Table S7). No clear trends were observed for the other AOS parameters in the main model, and this was confirmed by the sensitivity analyses.
Figure 4. Association between personal exposure to PM2.5, OPvDTT, and OPvAA during pregnancy and lung function parameters measured at 3 y in the univariate and multiple linear models and in the sensitivity analyses. Outcomes and exposures were scaled by their IQR. See Tables S5 and S7 for corresponding numeric data. Whiskers represent the 95% confidence interval around the estimate. The main model was adjusted on child’s height, weight, sex, age, season of sampling, breastfeeding, environmental tobacco smoke, maternal age and BMI before pregnancy, parental level of education, parental history of rhinitis and mean temperature during pregnancy. In addition, “2 sampling periods” are the analyses reduced to the children that had 2 wk of prenatal measurements of air pollution (61% of the population); “Excluding extreme values” are the analyses excluding the exposures and outcomes below the first percentile and above the 99th (exclusion of approx. 5% of the population); “Adjusted on PM” corresponds to adding personal exposure to PM2.5 in the set of confounders; “Adjusted on NO2” corresponds to adding personal exposure to NO2 in the set of confounders. Note: AA, ascorbic acid; AX, area under the reactance curve; BMI, body mass index; DTT, dithiothreitol; IQR, interquartile range; OPvAA, volume-normalized oxidative potential measured by the AA assay (nmol/min/m3); OPvDTT, volume-normalized oxidative potential measured by the DTT assay (nmol/min/m3); PM, particulate matter; PM2.5, PM with an aerodynamic diameter <2.5μm (μg/m3); Rrs7, resistance at a frequency of 7 Hz; Rrs7–19, difference between the resistance at 7 Hz and at 19 Hz; Xrs7, reactance at a frequency of 7 Hz.
Figure 4 is a set of four forest plots, plotting (bottom to top), volume-normalized oxidative potential measured by the ascorbic acid assay, volume-normalized oxidative potential measured by the dithiothreitol assay, particulate matter with a diameter smaller than 2.5 microns (left y-axis) and area under the reactance curve (hPa per Liter), reactance at a frequency of 7 hertz (hPa seconds per liter), difference between the resistance at 7 hertz and at 19 hertz (hPa seconds per liter), and resistance at a frequency of 7 hertz (hPa seconds per liter) (right y-axis) across beta, ranging negative 0.25 to 0.50 in increments of 0.25 (x-axis) for analysis, including Univariate, Main model, Complete Cases, 2 sampling periods, Excluding extreme values, Adjusted on Particulate matter, and Adjusted on nitrogen dioxide.
Discussion
To the best of our knowledge, this study is the first one to address the associations between maternal personal exposures to PM2.5 and OP and children’s objective lung function parameters measured as early as 6 wk of age and at 3 y. Regarding OPvDTT, our findings showed consistency across some lung function parameters with higher prenatal exposure being associated with a lowered indicator of lung volumes (FRC) at 6 wk and with a trend toward reduced Rrs7–19 at 3 y, an indicator influenced by both lung volumes and ventilation heterogeneity. An interesting finding is that the effects of OPvDTT exposure on FRC were stronger than those of PM2.5 mass in the two-pollutant model.
PM and OP Exposures and Lung Function
Our results are in agreement with existing studies reporting a higher prevalence of reduced lung function in participants who are exposed to higher levels of PM2.5.12,45–48 Regarding TBFVL and N2MBW tests, our findings are in line with the results of the South African birth cohort, MACE,16 that investigated the effects of NOx from LUR models and lung function of children at 1.5, 6, 12, and 24 months of age, and with the results of a Swiss birth cohort10 that examined the association between PM10 and NO2 from an ambient monitoring station and lung function measured in neonates (median age of 34 d). Both studies showed decreases in FRC and VT in infants prenatally exposed to higher concentrations of NOx or NO2 and PM10, whereas no effects were found on LCI. Our results extend their findings by confirming the pattern of decreased FRC with exposure to PM2.5 and OP, further supporting the importance of considering the oxidative stress caused by PM during pregnancy to predict lung growth restriction of children. In our study, none of the exposures considered were associated with LCI or with tPTEF/tE, two parameters still poorly studied in association with air pollution and with conflicting results regarding LCI.10,16 The decrease in VT with OPvDTT and PM2.5 is not confirmed by all sensitivity analyses, indicating limited robustness of this association. Rrs7–19 is usually used to detect the obstruction of the distal bronchi and can be modified by both lung volumes and heterogeneity of ventilation.39 The trend for an increase of this parameter in children prenatally exposed to higher OP is in accordance with the results found at 6 wk, because lower lung volume could lead to an increased resistance of the small airways. This partially confirms the results from previous studies indicating a detrimental effect of air pollution on respiratory mechanical parameters. In the BAMSE birth cohort, Schultz et al.49 investigated the effects of early-life exposure to PM10 on lung mechanic components measured by impulse oscillometry in 2,415 adolescents and found increased frequency dependent resistance (Rrs5–20) and AX0.5 with higher PM10 exposure, although the associations were not statistically significant. Shao et al.50 found increased AX in 84 children exposed to PM2.5 from a 6-wk episode of fire during infancy. In addition, regarding acute respiratory effect of OP, He et al.27 found that an increase in OP measured 2 d prior to visit was significantly associated with increased Rrs5–20 and Rrs5 in 43 asthmatic children age 5–13 y. Although AOS parameters have been found associated with air pollution in previous studies, the parameter varies between studies.8,27,49,50 In our case, we confirmed results with Rrs7–19, a parameter specific of the small airways.
Comparison of the Exposure Metrics
Our study, which identified associations between OP and PM with FRC, an indicator of lung volume, and with Rrs7–19, an indicator also accounting for lung volumes, indicates specific effects on lung growth. These observations are supported by studies showing that prenatal exposure to environmental pollutants impacts in utero growth, including organ growth,51,52 and that oxidative stress may cause placental tissue damage, which could in turn affect lung growth in utero.53,54
Only a few cohort studies tackled the associations of PM and OP exposure with lung function.26–28 The associations found with reduced lung function seemed generally clearer with OP than with PM2.5 mass concentration, which agrees with the existing literature.24 For example, the PIAMA birth cohort study28 found associations between OPvDTT at home address and increased asthma and rhinitis prevalence and decreased lung function in 12-y-old children but no association with PM2.5 mass. The effect magnitudes of OP models adjusted on other pollutants were similar, although more sensitive to NO2 adjustment, which was not the case in our model. In children with asthma diagnosis at age 9–18 y, Delfino et al.26 found significant positive associations between ambient OPvDTT and OP measured by the in vitro ROS-macrophage assay and airway inflammation, whereas no association was found for PM2.5. Conclusions were not modified in their two-pollutant model. He et al.27 also used the ROS-macrophage assay and found associations with Rrs5, Rrs5–20, and Rrs20 for OP, whereas associations for PM were only found with Rrs5. Overall our results add to the existing evidence indicating that the OP of PM has a stronger effect on various respiratory outcomes than PM mass and is thereby a relevant complementary health metric for air pollution.26–28,55–58 The different health effects found for PM2.5 and OP could be partially explained by the difference in sources contributing to OP and PM2.5 concentration in the SEPAGES study area (Grenoble). In fact, previous studies showed that biomass burning and regional transport of secondary inorganic pollutants (nitrates and sulfates) were the main sources contributing to the ambient PM2.5 mass concentration, whereas vehicular emissions and biomass burning were the main drivers of OP levels over the area.59,60 We acknowledge that by using active personal samplers, exposure measurements incorporate both indoor- and outdoor-generated pollution, which can have different compositions and thus different health effects.61
Our study extends the findings of others by comparing OP measured by the AA and the DTT assays. In their reviews, Bates et al.24 and Rao et al.62 showed that OPDTT was a better predictor than OPAA for most health outcomes. Here, we found that OPvAA had an effect comparable to that of OPvDTT on lung function as measured by Rrs7–19 at 3 y. However, results at 6 wk were more contrasted. The effects of OPvAA on FRC seemed to be influenced by PM2.5 mass concentration, because the OPvAA coefficients in the model adjusted on PM were pulled toward zero. Overall, although both OP assays (i.e., DTT and AA) were developed to account for the toxicity of PM components, their health impact may differ, which could be explained by their different sensitivities to chemical components (traffic-related metals, organic carbon, and inorganic species for OPDTT and metals only for OPAA)63–65 and their different reactivities to specific ROS.66
Strengths and Limitations
One of the main strengths of this study is the assessment of maternal exposure by personal measurements, which was proven to be more representative of real exposure29,61,67 than assessments in studies using ambient measurement from monitoring stations or exposure models. It is also expected to be more accurate as compared to approaches modeling the personal exposure,27 combining a) self-reported time-activity patterns in different microenvironments (at home, at work, in a car, in public transport, outdoors) and b) indoor–outdoor ratios estimation for each identified microenvironment, both being at risk for errors. Additionally, the use of OP in this study is a way to consider the potential oxidative stress caused by PM, which is thought to be a better predictor of PM damages than its concentration. An interesting finding is that the similitude of the seasonality observed in personal levels of PM2.5 and OP in the present study with the results of a previous study that showed higher ambient PM2.5 and OP during winter in the Grenoble area,60 supports the external validity of our exposure data.
We acknowledge that a mixed influence of pre- and postnatal exposure cannot be totally ruled out, but such influence cannot be assessed because OP of PM2.5 was not measured in early childhood in SEPAGES. Nevertheless, other studies6,68 that considered both pre- and postnatal exposure to PM found an effect of prenatal exposure on reduced lung function in children. Although the design of the study enables evaluation of the effects of air pollution on child’s lung function at different stages of the pregnancy, we a priori decided not to perform this analysis in our study to avoid lowering the number of participants included (224 with two measurement weeks) and increasing the number of statistical tests.
One limitation of active personal samplers is that it cannot be used by the participants during their entire pregnancy. The compromise in this study was to perform sampling for two 1-wk periods during the pregnancy and to use the average of the two measures in the association studies. This approach tended to avoid the influence of seasonality and extreme pollution events during the sampling weeks, especially for OPvDTT (Figure 2; Figure S8). However, this influence could not be avoided for individuals with only 1 wk of measurement (n=132). To account for this limitation, models were adjusted for the season of sampling, and sensitivity analysis excluding participants with only one measurement week were conducted. It is interesting to note that, in general, the associations between OP or PM2.5 and FRC and Rrs7–19 were stronger in this latter sensitivity analysis consisting of a restricted population with a more accurate exposure assessment, which supports our a priori hypothesis.
The novelty of this study also lies in the repeated assessment of lung function in early life, whereas most of the other studies considered children older than 5 y old, when spirometry starts to be feasible. Assessing lung function at the youngest age allows researchers to better investigate the effects related to pregnancy and early infancy time windows, which are believed to predict long-term respiratory morbidity. However, the use of pure oxygen during the N2MBW test (SF6 being forbidden in France) induced a transient decrease in tidal volume, which could affect the measurement of FRC and LCI. Although parameters were a posteriori corrected for the degree of hypoventilation, and sensitivity analysis excluding children with the highest hypoventilation degree showed similar patterns of association, residual errors in lung function assessment that would lead to underestimated effect estimates cannot be totally excluded. Because two different techniques of lung function measurement were used at 6 wk and 3 y of age, the effect of prenatal exposure of air pollution on lung function growth could not be assessed.
The amount of data collected during the follow-up of the cohort allowed us to adjust for a number of confounding factors. However, the residual confounding due to the observational design of this study remains a limitation. An interesting finding is that the analysis excluding leverage and influencing points showed that these points tended to drive some of the regression estimates toward the null hypothesis, which indicates that influencing points might be partly related to measurement errors. Although the aim of our study was based on an a priori hypothesis derived from previous association studies and from the biological specificities of OP of PM2.5, the number of associations tested was still relatively high (n=24) and we did not apply any formal correction for multiple comparisons. Thus, we acknowledge part of the associations observed may result from chance findings and thus should be interpreted cautiously. The attrition rate of 51% for the associations between the personal prenatal OP and lung function at 3 y could not be a priori defined as low, but given the demanding protocol and the originality of the longitudinal data collected, both for exposures (personal prenatal exposure to OP) and health outcomes (with objective lung function measures in preschool children, which is rare in population-based cohorts), this can be considered acceptable. However, a selection bias cannot be totally ruled out; in particular, the associations for PM2.5 may have been underestimated because included participants tended to have both higher exposure to PM2.5 and better lung function on two AOS parameters (lower Rrs7, higher Xrs7) at 3 y in comparison with the excluded participants. Nevertheless, with no differences in OP between included and excluded children, the associations reported with OP are probably not driven by selection bias. Although a bigger sample would lead to more statistical power and therefore clearer conclusions, the use of objective and validated respiratory health parameters in early life and novel personal prenatal air pollution exposure metrics offers important and relevant information on PM exposure and its health effects.
In summary, our study shows consistency in the associations between personal prenatal OPvDTT and several early-life lung function parameters related to lung growth restriction and therefore supports findings of the detrimental health effects of PM2.5 exposure on health through oxidative stress and the relevance of OP of PM2.5 as a useful health-based metric. These findings, together with identifying sources of OP of PM, could help target emission sources that are critical in decreasing health effects of atmospheric pollution.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The SEPAGES cohort study group: E. Eyriey, A. Licinia, A. Vellement (Groupe Hospitalier Mutualiste, Grenoble), I. Pin, P. Hoffmann, E. Hullo, C. Llerena (Grenoble Alpes University Hospital, La Tronche), X. Morin (Clinique des Cèdres, Echirolles), A. Morlot (Clinique Belledonne, Saint-Martin d’Hères), J. Lepeule, S. Lyon-Caen, C. Philippat, I. Pin, J. Quentin, V. Siroux, R. Slama (Grenoble Alpes University, Inserm, CNRS, IAB). The authors express their sincere thanks to all participants in the SEPAGES study.
The authors thank all the numerous people (who could not be listed exhaustively here) from the different laboratories (IGE and Air-O-Sol analytical platform) and from the Grenoble University Hospital who performed mother recruitment and child follow-ups. The authors would like to kindly thank I. Hough, I. Kloog and A. Guilbert for their help with the assessment of temperature exposure.
The authors thank A. Benlakhryfa, L. Borges, Y. Gioria, clinical research assistants; J. Giraud, M. Marceau, M-P. Martin, nurses; E. Charvet, A. Putod, midwives; M. Graca, K. Gridel, C. Pelini, M. Barbagallo, fieldworkers; A. Bossant, K. Guichardet, J.T. Iltis, A. Levanic, C. Martel, E. Quinteiro, S. Raffin, neuropsychologists; the staff from Grenoble Center for Clinical Investigation (CIC): J.-L. Cracowski, C. Cracowski, E. Hodaj, D. Abry, N. Gonnet, and A. Tournier. The authors extend a warm thank you also to M. Althuser, S. Althuser, F. Camus-Chauvet, P. Dusonchet, S. Dusonchet, L. Emery, P. Fabbrizio, P. Hoffmann, D. Marchal André, X. Morin, E. Opoix, L. Pacteau, P. Rivoire, A. Royannais, C. Tomasella, T. Tomasella, D. Tournadre, P. Viossat, E. Volpi, S. Rey, E. Warembourg, and clinicians from Grenoble University Hospital for their support in the recruitment of the study volunteers. The authors also thank A. Buchet, S.F. Caraby, J.-N. Canonica, J. Dujourdil, E. Eyriey, P. Hoffmann, M. Jeannin, A. Licina, X. Morin, A. Nicolas, and all midwives from the four maternity wards of Grenoble urban areas.
SEPAGES data are stored thanks to Inserm RE-CO-NAI platform funded by Commissariat Général à l’Investissement, with the implication of Sophie de Visme (Inserm DSI). Many thanks to M.A. Charles, RE-CO-NAI coordinator, for her support.
This work was supported in part by the Agence de la Transition Écologique (ADEME) and by the Agence Nationale de Sécurité Sanitaire de l’Alimentation, de l’Environnement et du Travail (ANSES), both supporting A.M.’s PhD grant. OP measurements were funded by a University Grenoble Alpes grant CDP IDEX UGA MOBILAIR (ANR-15-IDEX-02) and by the French Research Agency - ANR (GetOPstandOP ANR-19-CE34-0002). The postdoctoral position of L.J.S.B. is funded by the Predict’air project (grant Fondation UGA-UGA 2022-16 and grant PR-PRE-2021 FUGAFondation Air Liquide). The SEPAGES cohort was supported by the European Research Council (No. 311765-E-DOHaD), the European Community’s Seventh Framework Programme (FP7/2007-206 - No. 308333-892 HELIX), the European Union’s Horizon 2020 research and innovation program (No. 874583 ATHLETE Project, No. 825712 OBERON Project), the French Research Agency - ANR (PAPER project ANR-12-PDOC-0029-01, SHALCOH project ANR-14-CE21-0007, ANR-15-IDEX-02 and ANR-15-IDEX5, GUMME project ANR-18-CE36-005, ETAPE project ANR - EDeN project ANR-19-CE36-0003-01), the French Agency for Food, Environmental and Occupational Health & Safety - ANSES (CNAP project EST-2016-121, PENDORE project EST-2016-121, HyPAxE project EST-2019/1/039), the Plan Cancer (Canc’Air project), the French Cancer Research Foundation Association de Recherche sur le Cancer – ARC, the French Endowment Fund AGIR for chronic diseases – APMC (projects PRENAPAR and LCI-FOT), the French Endowment Fund for Respiratory Health, the French Fund – Fondation de France (CLIMATHES – 00081169, SEPAGES 5 - 00099903).
==== Refs
References
1. Janssen NAH, Brunekreef B, van Vliet P, Aarts F, Meliefste K, Harssema H, et al. 2003. The relationship between air pollution from heavy traffic and allergic sensitization, bronchial hyperresponsiveness, and respiratory symptoms in Dutch schoolchildren. Environ Health Perspect 111 (12 ):1512–1518, PMID: , 10.1289/ehp.6243.12948892
2. Murray CJL, Aravkin AY, Zheng P, Abbafati C, Abbas KM, Abbasi-Kangevari M, et al. 2020. Global burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the global burden of disease study 2019. Lancet 396 (10258 ):1223–1249, 10.1016/S0140-6736(20)30752-2.33069327
3. WHO (World Health Organization). 2016. Ambient Air Pollution: A Global Assessment of Exposure and Burden of Disease. https://www.who.int/phe/publications/air-pollution-global-assessment/en/ [accessed 5 May 2021].
4. Capello F, Pili G. 2018. Air Pollution in Infancy, Childhood and Young Adults. In: Clinical Handbook of Air Pollution-Related Diseases. Capello F, Gaddi AV, eds. Cham, Switzerland: Springer International Publishing, 141–186, 10.1007/978-3-319-62731-1_10.
5. Sly PD, Flack F. 2008. Susceptibility of children to environmental pollutants. Ann NY Acad Sci 1140 (1 ):163–183, PMID: , 10.1196/annals.1454.017.18991915
6. Cai Y, Hansell AL, Granell R, Blangiardo M, Zottoli M, Fecht D, et al. 2020. Prenatal, Early-Life, and childhood exposure to air pollution and lung function: the ALSPAC cohort. Am J Respir Crit Care Med 202 (1 ):112–123, PMID: , 10.1164/rccm.201902-0286OC.32142356
7. Carraro S, Scheltema N, Bont L, Baraldi E. 2014. Early-life origins of chronic respiratory diseases: understanding and promoting healthy ageing. Eur Respir J 44 (6 ):1682–1696, PMID: , 10.1183/09031936.00084114.25323240
8. Dutta A, Alaka M, Ibigbami T, Adepoju D, Adekunle S, Olamijulo J, et al. 2021. Impact of prenatal and postnatal household air pollution exposure on lung function of 2-year-old Nigerian children by oscillometry. Sci Total Environ 755 (pt 2 ):143419, PMID: , 10.1016/j.scitotenv.2020.143419.33187696
9. Korten I, Ramsey K, Latzin P. 2017. Air pollution during pregnancy and lung development in the child. Paediatr Respir Rev 21 :38–46, PMID: , 10.1016/j.prrv.2016.08.008.27665510
10. Latzin P, Röösli M, Huss A, Kuehni CE, Frey U. 2009. Air pollution during pregnancy and lung function in newborns: a birth cohort study. Eur Respir J 33 (3 ):594–603, PMID: , 10.1183/09031936.00084008.19010988
11. Bergstra AD, Brunekreef B, Burdorf A. 2018. The effect of industry-related air pollution on lung function and respiratory symptoms in school children. Environ Health 17 (1 ):30, PMID: , 10.1186/s12940-018-0373-2.29587756
12. Gehring U, Gruzieva O, Agius RM, Beelen R, Custovic A, Cyrys J, et al. 2013. Air pollution exposure and lung function in children: the ESCAPE project. Environ Health Perspect 121 (11–12 ):1357–1364, PMID: , 10.1289/ehp.1306770.24076757
13. Schultz ES, Hallberg J, Bellander T, Bergström A, Bottai M, Chiesa F, et al. 2016. Early-life exposure to traffic-related air pollution and lung function in adolescence. Am J Respir Crit Care Med 193 (2 ):171–177, PMID: , 10.1164/rccm.201505-0928OC.26397124
14. He B, Huang JV, Kwok MK, Au Yeung SL, Hui LL, Li AM, et al. 2019. The association of early-life exposure to air pollution with lung function at ∼17.5 years in the “children of 1997” Hong Kong Chinese birth cohort. Environ Int 123 :444–450, PMID: , 10.1016/j.envint.2018.11.073.30622069
15. He L, Cui X, Li Z, Teng Y, Barkjohn KK, Norris C, et al. 2020. Malondialdehyde in nasal fluid: a biomarker for monitoring asthma control in relation to air pollution exposure. Environ Sci Technol 54 (18 ):11405–11413, PMID: , 10.1021/acs.est.0c02558.32822160
16. Muttoo N, Jeena Asharam K. 2019. Air pollution exposure and infant lung function in the MACE cohort, South Africa. Environ Epidemiol 3 :280, 10.1097/01.EE9.0000609000.48206.68.
17. Pope CAI. 2000. Review: epidemiological basis for particulate air pollution health standards. Aerosol Sci Technol 32 (1 ):4–14, 10.1080/027868200303885.
18. Gao D, Ripley S, Weichenthal S, Godri Pollitt KJ. 2020. Ambient particulate matter oxidative potential: chemical determinants, associated health effects, and strategies for risk management. Free Radic Biol Med 151 :7–25, PMID: , 10.1016/j.freeradbiomed.2020.04.028.32430137
19. Crobeddu B, Aragao-Santiago L, Bui LC, Boland S, Baeza Squiban A. 2017. Oxidative potential of particulate matter 2.5 as predictive indicator of cellular stress. Environ Pollut 230 :125–133, PMID: , 10.1016/j.envpol.2017.06.051.28649040
20. Hatzis C, Godleski JJ, González-Flecha B, Wolfson JM, Koutrakis P. 2006. Ambient particulate matter exhibits direct inhibitory effects on oxidative stress enzymes. Environ Sci Technol 40 (8 ):2805–2811, PMID: , 10.1021/es0518732.16683627
21. Leni Z, Cassagnes LE, Daellenbach KR, El Haddad I, Vlachou A, Uzu G, et al. 2020. Oxidative stress-induced inflammation in susceptible airways by anthropogenic aerosol. PLoS One 15 (11 ):e0233425, PMID: , 10.1371/journal.pone.0233425.33206642
22. Riva DR, Magalhães CB, Lopes AA, Lanças T, Mauad T, Malm O, et al. 2011. Low dose of fine particulate matter (PM2.5) can induce acute oxidative stress, inflammation and pulmonary impairment in healthy mice. Inhal Toxicol 23 (5 ):257–267, PMID: , 10.3109/08958378.2011.566290.21506876
23. Hellack B, Yang A, Cassee FR, Janssen NAH, Schins RPF, Kuhlbusch TAJ. 2014. Intrinsic hydroxyl radical generation measurements directly from sampled filters as a metric for the oxidative potential of ambient particulate matter. J Aerosol Sci 72 :47–55, 10.1016/j.jaerosci.2014.02.003.
24. Bates JT, Fang T, Verma V, Zeng L, Weber RJ, Tolbert PE, et al. 2019. Review of acellular assays of ambient particulate matter oxidative potential: methods and relationships with composition, sources, and health effects. Environ Sci Technol 53 (8 ):4003–4019, PMID: , 10.1021/acs.est.8b03430.30830764
25. Weichenthal S, Lavigne E, Traub A, Umbrio D, You H, Pollitt K, et al. 2021. Association of sulfur, transition metals, and the oxidative potential of outdoor PM2.5 with acute cardiovascular events: a case-crossover study of Canadian adults. Environ Health Perspect 129 (10 ):107005, PMID: , 10.1289/EHP9449.34644144
26. Delfino RJ, Staimer N, Tjoa T, Gillen DL, Schauer JJ, Shafer MM. 2013. Airway inflammation and oxidative potential of air pollutant particles in a pediatric asthma panel. J Expo Sci Environ Epidemiol 23 (5 ):466–473, PMID: , 10.1038/jes.2013.25.23673461
27. He L, Norris C, Cui X, Li Z, Barkjohn KK, Brehmer C, et al. 2021. Personal exposure to PM2.5 oxidative potential in association with pulmonary pathophysiologic outcomes in children with asthma. Environ Sci Technol 55 (5 ):3101–3111, PMID: , 10.1021/acs.est.0c06114.33555874
28. Yang A, Janssen NAH, Brunekreef B, Cassee FR, Hoek G, Gehring U. 2016. Children’s respiratory health and oxidative potential of PM2.5: the PIAMA birth cohort study. Occup Environ Med 73 (3 ):154–160, PMID: , 10.1136/oemed-2015-103175.26755634
29. Avery CL, Mills KT, Williams R, McGraw KA, Poole C, Smith RL, et al. 2010. Estimating error in using residential outdoor PM2.5 concentrations as proxies for personal exposures: a meta-analysis. Environ Health Perspect 118 (5 ):673–678, PMID: , 10.1289/ehp.0901158.20075021
30. Lyon-Caen S, Siroux V, Lepeule J, Lorimier P, Hainaut P, Mossuz P, et al. 2019. Deciphering the impact of Early-Life exposures to highly variable environmental factors on foetal and child health: design of SEPAGES Couple–Child cohort. Int J Environ Res Public Health 16 (20 ):3888, PMID: , 10.3390/ijerph16203888.31615055
31. Calas A, Uzu G, Martins JMF, Voisin D, Spadini L, Lacroix T, et al. 2017. The importance of simulated lung fluid (SLF) extractions for a more relevant evaluation of the oxidative potential of particulate matter. Sci Rep 7 (1 ):11617, PMID: , 10.1038/s41598-017-11979-3.28912590
32. Calas A, Uzu G, Kelly FJ, Houdier S, Martins JMF, Thomas F, et al. 2018. Comparison between five acellular oxidative potential measurement assays performed with detailed chemistry on PM10 samples from the city of Chamonix (France). Atmos Chem Phys 18 (11 ):7863–7875, 10.5194/acp-18-7863-2018.
33. Kelly FJ, Mudway IS. 2003. Protein oxidation at the air-lung interface. Amino Acids 25 (3–4 ):375–396, PMID: , 10.1007/s00726-003-0024-x.14661098
34. Bates JH, Schmalisch G, Filbrun D, Stocks J. 2000. Tidal breath analysis for infant pulmonary function testing. ERS/ATS task force on standards for infant respiratory function testing. European Respiratory Society/American Thoracic Society. Eur Respir J 16 (6 ):1180–1192, PMID: , 10.1034/j.1399-3003.2000.16f26.x.11292125
35. Gustafsson PM, Bengtsson L, Lindblad A, Robinson PD. 2017. The effect of inert gas choice on multiple breath washout in healthy infants: differences in lung function outcomes and breathing pattern. J Appl Physiol (1985) 123 (6 ):1545–1554, PMID: , 10.1152/japplphysiol.00524.2017.28860172
36. Mortamais M, Chevrier C, Philippat C, Petit C, Calafat AM, Ye X, et al. 2012. Correcting for the influence of sampling conditions on biomarkers of exposure to phenols and phthalates: a 2-step standardization method based on regression residuals. Environ Health 11 (1 ):29, PMID: , 10.1186/1476-069X-11-29.22537080
37. Beydon N, Davis SD, Lombardi E, Allen JL, Arets HGM, Aurora P, et al. 2007. An official American Thoracic Society/European Respiratory Society statement: pulmonary function testing in preschool children. Am J Respir Crit Care Med 175 (12 ):1304–1345, PMID: , 10.1164/rccm.200605-642ST.17545458
38. Gosselink R, Stam H. 2005. Lung Function Testing: European Respiratory Monograph. Lausanne, Switzerland: European Respiratory Society.
39. Lundblad LKA, Siddiqui S, Bossé Y, Dandurand RJ. 2021. Applications of oscillometry in clinical research and practice. Can J Respir Crit Care Sleep Med 5 (1 ):54–68, 10.1080/24745332.2019.1649607.
40. Hough I, Just AC, Zhou B, Dorman M, Lepeule J, Kloog I. 2020. A multi-resolution air temperature model for France from MODIS and Landsat thermal data. Environ Res 183 :109244, PMID: , 10.1016/j.envres.2020.109244.32097815
41. van Buuren S, Groothuis-Oudshoorn K. 2011. MICE: Multivariate imputation by chained equations in R. J Stat Softw 45 (3 ):1–67, 10.18637/jss.v045.i03.
42. Little RJA. 1988. A test of missing completely at random for multivariate data with missing values. J Am Stat Assoc 83 (404 ):1198–1202, 10.1080/01621459.1988.10478722.
43. Rubin DB. 1987. Multiple Imputation for Nonresponse in Surveys. New York: John Wiley & Sons, i–xxix, 10.1002/9780470316696.fmatter.
44. Cook RD. 1977. Detection of influential observation in linear regression. Technometrics 19 (1 ):15, 10.2307/1268249.
45. Gauderman WJ, Avol E, Gilliland F, Vora H, Thomas D, Berhane K, et al. 2004. The effect of air pollution on lung development from 10 to 18 years of age. N Engl J Med 351 (11 ):1057–1067, PMID: , 10.1056/NEJMoa040610.15356303
46. Guo C, Hoek G, Chang L-Y, Bo Y, Lin C, Huang B, et al. 2019. Long-term exposure to ambient fine particulate matter (PM2.5) and lung function in children, adolescents, and young adults: a longitudinal cohort study. Environ Health Perspect 127 (12 ):127008, PMID: , 10.1289/EHP5220.31873044
47. Hwang BF, Chen YH, Lin YT, Wu XT, Leo Lee Y. 2015. Relationship between exposure to fine particulates and ozone and reduced lung function in children. Environ Res 137 :382–390, PMID: , 10.1016/j.envres.2015.01.009.25614339
48. Rice MB, Rifas-Shiman SL, Litonjua AA, Oken E, Gillman MW, Kloog I, et al. 2016. Lifetime exposure to ambient pollution and lung function in children. Am J Respir Crit Care Med 193 (8 ):881–888, PMID: , 10.1164/rccm.201506-1058OC.26575800
49. Schultz ES, Hallberg J, Gustafsson PM, Bottai M, Bellander T, Bergström A, et al. 2016. Early life exposure to traffic-related air pollution and lung function in adolescence assessed with impulse oscillometry. J Allergy Clin Immunol 138 (3 ):930–932.e5, PMID: , 10.1016/j.jaci.2016.04.014.27297996
50. Shao J, Zosky GR, Hall GL, Wheeler AJ, Dharmage S, Melody S, et al. 2020. Early life exposure to coal mine fire smoke emissions and altered lung function in young children. Respirology 25 (2 ):198–205, PMID: , 10.1111/resp.13617.31231911
51. Lavigne É, Burnett RT, Stieb DM, Evans GJ, Godri Pollitt KJ, Chen H, et al. 2018. Fine particulate air pollution and adverse birth outcomes: effect modification by regional nonvolatile oxidative potential. Environ Health Perspect 126 (7 ):077012, PMID: , 10.1289/EHP2535.30073952
52. Saadeh R, Klaunig J. 2014. Child’s development and respiratory system toxicity. J Environ Anal Toxicol 4 (5 ), 10.4172/2161-0525.1000233.
53. Øvrevik J. 2019. Oxidative potential versus biological effects: a review on the relevance of cell-free/abiotic assays as predictors of toxicity from airborne particulate matter. Int J Mol Sci 20 (19 ):4772, PMID: , 10.3390/ijms20194772.31561428
54. Veras MM, de Oliveira Alves N, Fajersztajn L, Saldiva P. 2017. Before the first breath: prenatal exposures to air pollution and lung development. Cell Tissue Res 367 (3 ):445–455, PMID: , 10.1007/s00441-016-2509-4.27726025
55. Abrams JY, Weber RJ, Klein M, Sarnat SE, Chang HH, Strickland MJ, et al. 2017. Associations between ambient fine particulate oxidative potential and cardiorespiratory emergency department visits. Environ Health Perspect 125 (10 ):107008, PMID: , 10.1289/EHP1545.29084634
56. Fang T, Verma V, Bates JT, Abrams J, Klein M, Strickland MJ, et al. 2016. Oxidative potential of ambient water-soluble PM2.5 in the southeastern United States: contrasts in sources and health associations between ascorbic acid (AA) and dithiothreitol (DTT) assays. Atmos Chem Phys 16 (6 ):3865–3879, 10.5194/acp-16-3865-2016.
57. Janssen NAH, Strak M, Yang A, Hellack B, Kelly FJ, Kuhlbusch TAJ, et al. 2015. Associations between three specific a-cellular measures of the oxidative potential of particulate matter and markers of acute airway and nasal inflammation in healthy volunteers. Occup Environ Med 72 (1 ):49–56, PMID: , 10.1136/oemed-2014-102303.25104428
58. Weichenthal S, Crouse DL, Pinault L, Godri-Pollitt K, Lavigne E, Evans G, et al. 2016. Oxidative burden of fine particulate air pollution and risk of cause-specific mortality in the Canadian census health and environment cohort (CanCHEC). Environ Res 146 :92–99, PMID: , 10.1016/j.envres.2015.12.013.26745732
59. Borlaza LJS, Weber S, Uzu G, Jacob V, Cañete T, Micallef S, et al. 2021. Disparities in particulate matter (PM10) origins and oxidative potential at a city scale (Grenoble, France) – Part I: Source apportionment at three neighbouring sites. Atmos Chem Phys 21 (7 ):5415–5437, 10.5194/acp-21-5415-2021.
60. Borlaza LJS, Weber S, Jaffrezo JL, Houdier S, Slama R, Rieux C, et al. Disparities in particulate matter (PM10) origins and oxidative potential at a city-scale (Grenoble, France) – Part II: sources of PM10 oxidative potential using multiple linear regression analysis and the predictive applicability of multilayer perceptron neural network analysis. Atmos Chem Phys 21 (12 ):9719–9739, 10.5194/acp-2021-57.
61. Evangelopoulos D, Katsouyanni K, Keogh RH, Samoli E, Schwartz J, Barratt B, et al. 2020. PM2.5 and NO2 exposure errors using proxy measures, including derived personal exposure from outdoor sources: a systematic review and meta-analysis. Environ Int 137 :105500, PMID: , 10.1016/j.envint.2020.105500.32018132
62. Rao L, Zhang L, Wang X, Xie T, Zhou S, Lu S, et al. 2020. Oxidative potential induced by ambient particulate matters with acellular assays: a review. Processes 8 (11 ):1410, 10.3390/pr8111410.
63. Fang T, Zeng L, Gao D, Verma V, Stefaniak AB, Weber RJ. 2017. Ambient size distributions and lung deposition of aerosol dithiothreitol-measured oxidative potential: contrast between soluble and insoluble particles. Environ Sci Technol 51 (12 ):6802–6811, PMID: , 10.1021/acs.est.7b01536.28548846
64. Janssen NAH, Yang A, Strak M, Steenhof M, Hellack B, Gerlofs-Nijland ME, et al. 2014. Oxidative potential of particulate matter collected at sites with different source characteristics. Sci Total Environ 472 :572–581, PMID: , 10.1016/j.scitotenv.2013.11.099.24317165
65. Visentin M, Pagnoni A, Sarti E, Pietrogrande MC. 2016. Urban PM2.5 oxidative potential: importance of chemical species and comparison of two spectrophotometric cell-free assays. Environ Pollut 219 :72–79, PMID: , 10.1016/j.envpol.2016.09.047.27661730
66. Xiong Q, Yu H, Wang R, Wei J, Verma V. 2017. Rethinking dithiothreitol-based particulate matter oxidative potential: measuring dithiothreitol consumption versus reactive oxygen species generation. Environ Sci Technol 51 (11 ):6507–6514, PMID: , 10.1021/acs.est.7b01272.28489384
67. Ouidir M, Giorgis-Allemand L, Lyon-Caen S, Morelli X, Cracowski C, Pontet S, et al. 2015. Estimation of exposure to atmospheric pollutants during pregnancy integrating space–time activity and indoor air levels: does it make a difference? Environ Int 84 :161–173, PMID: , 10.1016/j.envint.2015.07.021.26300245
68. Stapleton A, Casas M, García J, García R, Sunyer J, Guerra S, et al. 2022. Associations between pre- and postnatal exposure to air pollution and lung health in children and assessment of CC16 as a potential mediator. Environ Res 204 (pt A ):111900, PMID: , 10.1016/j.envres.2021.111900.34419474
|
PMC009xxxxxx/PMC9875838.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36696105
EHP12514
10.1289/EHP12514
Science Selection
From Drinking Water to Individual Body Burden: Modeling Toxicokinetics of Four PFAS
Seltenrich Nate
25 1 2023
1 2023
131 1 01400130 11 2022
22 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
A person holding a glass of water
==== Body
pmcA steady stream of research and media coverage has raised public awareness of the prevalence and potential harms of per- and polyfluoroalkyl substances (PFAS).1–3 “One of the things that we have heard in countless community meetings is that a lot of people want to know what’s in their body,” says Rachel Rogers, a senior health scientist with the U.S. Agency for Toxic Substances and Disease Registry (ATSDR), which performs outreach in communities affected by PFAS contamination. “There’s a real interest in understanding more about PFAS blood levels. People want that information.” Now a new web-based tool, released in November 2022,4 is available to help people estimate their PFAS exposure themselves. The tool is based on research by Rogers and colleagues that was recently published in Environmental Health Perspectives.5
PFAS are present in a wide range of common consumer products, including food packaging, stain- and water-resistant fabrics, nonstick cookware, and insecticides.6 They are also commonly included in firefighting foams.7 As a result of this widespread use and the chemicals’ persistence in the environment, individuals may be exposed via dermal contact,8 inhalation, and ingestion—including through drinking water.6 Despite many people’s interest in understanding their own PFAS exposures, blood tests are not widely available and can be difficult to access.9 So ATSDR developed a privacy-protected web tool that individuals can use to predict their own blood levels, based upon local drinking water concentrations of four PFAS: perfluorooctanoic acid (PFOA), perfluorooctane sulfonate (PFOS), perfluorohexane sulfonate (PFHxS), and perfluorononanoic acid (PFNA).10
Users enter their tap water levels for one or more of the four PFAS. This information may be available from public utilities, state agencies, or testing results—as in the case of drinking water from private wells. Alternatively, users can follow a link to look up tap water testing data from the U.S. Environmental Protection Agency (EPA). Users also input their age, sex, weight, and relative usage of bottled and filtered water.
The tool shows users how their blood levels compare with average levels across the U.S. population. Although the tool does not provide guidance regarding health effects, its core function of predicting blood levels based upon drinking water exposures represents a significant advance, according to the authors.
The tool’s predictions are made possible through extensive analyses and computations, described in the paper, that followed careful review and synthesis of data from 21 published studies reporting PFAS concentrations in human blood and corresponding drinking water measurements. The authors calculated new estimates for half-lives of the four chemicals in blood serum. They also estimated the chemicals’ volume of distribution (Vd), which refers to a substance’s tendency to either remain in the blood or redistribute into body tissues.5
“Accurate estimates of Vd in addition to half-life are important because together they determine the relationship between PFAS exposure levels and blood concentrations,” says first author Weihsueh Chiu, a professor at Texas A&M University and expert in pharmacokinetic modeling and risk assessment. “These are the drivers of any health risks.”
To date, PFAS half-life estimates have been subject to some disagreement.11–13 Yet their precise values have significant implications not only for using the new tool but, more broadly, for regulatory policy, says Tony Fletcher, an associate professor and PFAS researcher with the London School of Hygiene & Tropical Medicine, who was not involved with the study. A shorter half-life often indicates that a chemical is excreted more rapidly and will remain in the body for a shorter period—and thus may be deemed safe or permissible at higher concentrations, Fletcher says. “So a good, reliable estimation of half-life is a hot topic that is important to get right, because of the implications for regulatory standard setting.”
Study coauthor Meghan Lynch, a principal associate with Abt Associates, the consulting firm that partnered with ATSDR to develop the PFAS tool, acknowledges there are several aspects of the model that warrant further investigation. For example, it is based on data that came mostly from adults and does not account for the potentially thousands of other PFAS that may be present in drinking water.
Yet the study’s findings are clear, says John Wambaugh, a research physical scientist with the U.S. EPA who was not affiliated with the study. “The overall understanding that some of these chemicals stay in us a very long time was supported by this paper,” he says. “The existing regulations were developed using that sort of assumption, and this paper provides further evidence that that’s the right way to go. The numbers might adjust slightly, but the important thing is that we’re talking years.”
Authors of the new paper synthesized data from 21 published studies that reported drinking water measurements and corresponding concentrations of four PFAS in human blood. Their results include new estimates of half-lives in blood serum for those chemicals. Image: © iStock.com/Beyhes Evren.
A person holding a glass of water
Nate Seltenrich covers science and the environment from the San Francisco Bay Area. His work on subjects including energy, ecology, and environmental health has appeared in a wide variety of regional, national, and international publications.
==== Refs
References
1. Perkins T. 2022. Toxic ‘forever chemicals’ detected in commonly used insecticides in US, study finds. The Guardian . 7 October 2022. https://www.theguardian.com/environment/2022/oct/07/forever-chemicals-found-insecticides-study [accessed 17 January 2023].
2. Sterling M. 2022. Where did the PFAS in your blood come from? These computer models offer clues. Environmental Health News . 28 November 2022. https://www.ehn.org/pfas-testing-2658727343.html [accessed 17 January 2023].
3. Perkins T. 2022. PFAS left dangerous blood compounds in nearly all US study participants. The Guardian . 29 October 2022. https://www.theguardian.com/environment/2022/oct/29/pfas-left-dangerous-blood-compounds-in-nearly-all-us-study-participants [accessed 17 January 2023].
4. U.S. Centers for Disease Control and Prevention. 2022. Introducing CDC/ATSDR’s PFAS blood level estimation tool. https://blogs.cdc.gov/yourhealthyourenvironment/2022/11/15/introducing-cdc-atsdrs-pfas-blood-level-estimation-tool/ [accessed 17 January 2023].
5. Chiu WA, Lynch MT, Lay CR, Antezana A, Malek P, Sokolinski S, et al. 2022. Bayesian estimation of human population toxicokinetics of PFOA, PFOS, PFHxS, and PFNA from studies of contaminated drinking water. Environ Health Perspect 130 (12 ):127001, PMID: , 10.1289/EHP10103.36454223
6. U.S. EPA (U.S. Environmental Protection Agency). 2022. Our current understanding of the human health and environmental risks of PFAS. https://www.epa.gov/pfas/our-current-understanding-human-health-and-environmental-risks-pfas [accessed 17 January 2023].
7. California Water Boards. 2023. PFAS: per- and polyfluoroalkyl substances. https://www.waterboards.ca.gov/drinking_water/certlic/drinkingwater/pfas.html [accessed 17 January 2023].
8. Shane HL, Baur R, Lukomska E, Weatherly L, Anderson SE. 2020. Immunotoxicity and allergenic potential induced by topical application of perfluorooctanoic acid (PFOA) in a murine model. Food Chem Toxicol 136 :111114, PMID: , 10.1016/j.fct.2020.111114.31904477
9. ATSDR (Agency for Toxic Substances and Disease Registry). 2022. PFAS blood testing. https://www.atsdr.cdc.gov/pfas/health-effects/blood-testing.html [accessed 17 January 2023].
10. ATSDR. 2022. Estimating levels of PFAS in your blood. https://www.atsdr.cdc.gov/pfas/resources/estimating-pfas-blood.html [accessed 17 January 2023].
11. Bartell SM, Calafat AM, Lyu C, Kato K, Ryan PB, Steenland K, et al. 2010. Rate of decline in serum PFOA concentrations after granular activated carbon filtration at two public water systems in Ohio and West Virginia. Environ Health Perspect 118 (2 ):222–228, PMID: , 10.1289/ehp.0901252.20123620
12. Xu Y, Fletcher T, Pineda D, Lindh CH, Nilsson C, Glynn A, et al. 2020. Serum half-lives for short- and long-chain perfluoroalkyl acids after ceasing exposure from drinking water contaminated by firefighting foam. Environ Health Perspect 128 (7 ):077004, PMID: , 10.1289/EHP6785.32648786
13. Dourson M, Gadagbui B. 2021. The dilemma of perfluorooctanoate (PFOA) human half-life. Regul Toxicol Pharmacol 126 :105025, PMID: , 10.1016/j.yrtph.2021.105025.34400261
|
PMC009xxxxxx/PMC9875842.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36696102
EHP11164
10.1289/EHP11164
Research
Racial, Ethnic, and Socioeconomic Disparities in Multiple Measures of Blue and Green Spaces in the United States
Klompmaker Jochem O. 1 2
Hart Jaime E. 1 2
https://orcid.org/0000-0002-2652-596X
Bailey Christopher R. 3
Browning Matthew H.E.M. 4
Casey Joan A. 5
https://orcid.org/0000-0003-2963-2637
Hanley Jared R. 3
Minson Christopher T. 3 6
Ogletree S. Scott 7
Rigolon Alessandro 8
Laden Francine 1 2 9
James Peter 1 10
1 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
2 Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s Hospital, Boston, Massachusetts, USA
3 NatureQuant, Eugene, Oregon, USA
4 Department of Parks, Recreation and Tourism Management, Clemson University, Clemson, South Carolina, USA
5 Department of Environmental Health Sciences, Columbia University Mailman School of Public Health, New York, New York, USA
6 Department of Human Physiology, University of Oregon, Eugene, Oregon, USA
7 OPENspace Research Centre, School of Architecture and Landscape Architecture, University of Edinburgh, Edinburgh, UK
8 Department of City and Metropolitan Planning, University of Utah, Salt Lake City, Utah, USA
9 Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
10 Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, Massachusetts, USA
Address correspondence to Jochem O. Klompmaker, Department of Environmental Health, Harvard T.H. Chan School of Public Health, Landmark Center, 401 Park Dr., Boston, Massachusetts 02215 USA. Email: [email protected]
25 1 2023
1 2023
131 1 01700724 2 2022
03 11 2022
22 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Several studies have evaluated whether the distribution of natural environments differs between marginalized and privileged neighborhoods. However, most studies restricted their analyses to a single or handful of cities and used different natural environment measures.
Objectives:
We evaluated whether natural environments are inequitably distributed based on socioeconomic status (SES) and race/ethnicity in the contiguous United States.
Methods:
We obtained SES and race/ethnicity data (2015–2019) for all U.S. Census tracts. For each tract, we calculated the Normalized Different Vegetation Index (NDVI) for 2020, NatureScore (a proprietary measure of the quantity and quality of natural elements) for 2019, park cover for 2020, and blue space for 1984–2018. We used generalized additive models with adjustment for potential confounders and spatial autocorrelation to evaluate associations of SES and race/ethnicity with NDVI, NatureScore, park cover, and odds of containing blue space in all tracts (n=71,532) and in urban tracts (n=45,338). To compare effect estimates, we standardized NDVI, NatureScore, and park cover so that beta coefficients presented a percentage increase or decrease of the standard deviation (SD).
Results:
Tracts with higher SES had higher NDVI, NatureScore, park cover, and odds of containing blue space. For example, urban tracts in the highest median household income quintile had higher NDVI [44.8% of the SD (95% CI: 42.8, 46.8)] and park cover [16.2% of the SD (95% CI: 13.5, 19.0)] compared with urban tracts in the lowest median household income quintile. Across all tracts, a lower percentage of non-Hispanic White individuals and a higher percentage of Hispanic individuals were associated with lower NDVI and NatureScore. In urban tracts, we observed weak positive associations between percentage non-Hispanic Black and NDVI, NatureScore, and park cover; we did not find any clear associations for percentage Hispanics.
Discussion:
Multiple facets of the natural environment are inequitably distributed in the contiguous United States. https://doi.org/10.1289/EHP11164
Supplemental Material is available online (https://doi.org/10.1289/EHP11164).
C.R.B. (co-founder), J.R.H. (founder), and C.T.M. (co-founder) are affiliated with NatureQuant. NatureQuant had no role in the study design in the collection, analysis, and interpretation of data; or in the decision to submit the article for publication. All other authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Of all high- and middle-income countries, the United States has among the highest income-related disparities in self-reported health and health care measures.1 Health disparities have been attributed to several factors, such as health behaviors, housing conditions, and access to health care.2,3 Recently, increasing attention has been paid to the role of environmental exposures.4,5 Research suggests that exposure to natural environments (e.g., green space, parks, and blue space) may protect against several adverse health outcomes, including depression,6–9 cardiovascular disease,6,9,10 and mortality.6–10 Protective associations of green space are generally stronger for low-socioeconomic status (SES) individuals than for individuals in more affluent groups.4 Therefore, an inequitable distribution of natural environments could partially explain the observed health disparities.
Several studies have evaluated whether the distribution of green and blue spaces differs between marginalized and privileged neighborhoods.5,11–13 A meta-analysis reported that higher income households or neighborhoods have more urban forest cover than lower income households or neighborhoods.12 A review showed that marginalized neighborhoods have access to fewer acres of parks and have parks with lower quality than more privileged neighborhoods but found mixed results for park proximity.11 Results for differences in natural environment measures between race/ethnic groups are less clear. Another review reported significant race-based inequity in urban forest cover, but this inequity disappeared when only studies that adjusted for income were included.13
Most studies included in natural environment–inequity reviews restricted their analyses to a single or handful of cities, used a wide range of different constructs to quantify SES or race/ethnicity, and differed in their control for potential confounders.5,11–13 This may have led to differences in associations between studies and limits the generalizability of the results. Moreover, two studies showed that patterns of natural environment inequity varied by measure of the natural environment considered.14,15 Different measures of the natural environment (e.g., greenness, parks, or tree cover) capture different aspects of the natural environment that may result in differing associations between studies. For example, the Normalized Difference Vegetation Index (NDVI) captures private greenery (backyards) and could therefore be more strongly related to SES measures than parks. To the best of our knowledge, there is no study that covers the entire contiguous United States and compares diverse natural environment measures to assess whether natural environments vary by SES and race/ethnicity.
Our aim was to evaluate whether natural environments were inequitably distributed by U.S. Census tract SES and race/ethnicity in the contiguous United States. We analyzed whether median household income, percentage of households below the U.S. poverty level, percentage of the population with less than a high school education (%<high school education), %non-Hispanic White and Black, and %Hispanic across all census tracts were associated with several measures capturing different aspects of the natural environment. To compare findings between all and urban tracts, we additionally analyzed associations in all urban tracts (≥1,000 persons/mi2) in the contiguous United States.
Methods
SES and Race/Ethnicity
We downloaded data on several SES indicators and race/ethnicity for each census tract in the contiguous United States from the National Historical Geographic Information System (https://www.nhgis.org).16 Census tracts are small, relatively permanent statistical subdivisions of the United States with an average population size of between 1,200 and 8,000 people.17 The spatial size of the tracts varies widely depending on the density of the settlement. For each tract, we obtained SES indicators and racial/ethnic composition from the 2015–2019 American Community Survey (ACS), which is a nationwide survey and has an annual sample size of about 3.5 million addresses.18 The U.S. Census Bureau combines 5 consecutive years of ACS data to produce more reliable and precise estimates, especially for small geographic areas and small population subgroups.
Given that SES has multiple components and that associations with measures of the natural environment may differ between SES components, we examined three indicators in our analyses. Specifically, we examined a) median household income in the past 12 months (in 2019 inflation-adjusted U.S. dollars), b) percentage of households with an income in the past 12 months below the poverty level, and c) percentage of the population ≥25 years of age with <high school education. Further, we examined the percentages of non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, and non-Hispanic people of other races (American Indian and Alaska Native, Native Hawaiian and Other Pacific Islander, some other race, or two or more races) and Hispanic individuals in each tract. Using R (Version 1.2.5042, R Development Core Team), we plotted the spatial variation of the SES and race/ethnicity indicators across the contiguous United States (Figure S1).
Natural Environment Measures
We considered four natural environment measures: a) NDVI (an indicator of greenness), b) NatureScore (a proprietary measure of the quantity and quality of natural elements), c) park cover, and d) blue space. We selected these measures because they (or similar measures) have been studied in previous natural environment–inequity studies,4,5,12,13 capture different aspects/pathways of the natural environment that may be relevant to health, or may protect against adverse health outcomes.7,8,19,20 These measures vary across the contiguous United States and within cities. Using Google Earth Engine,22 we created detailed maps of the metropolitan areas of Boston, Massachusetts, Washington, DC, and San Francisco, California (Figure 1). Using R, we plotted the spatial variation of tract-level NDVI, NatureScore, park cover, and blue space (tract +100-m buffer) in the contiguous United States (Figure S2).
Figure 1. The spatial variation of NDVI (2020 data), NatureScore (2019 data), park cover (2020 data), and blue space (1984–2018 data) in the metropolitan areas of Boston, Massachusetts, Washington, DC, and San Francisco, California. NDVI is based on Landsat 8 images, NatureScore is a proprietary measure created by NatureQuant, Park cover was based on the USGS Protected Areas Database of the U.S. (PAD-US), and blue space was based on the European Commision’s Joint Research Centre’s Global Surface Water data set. Note: NDVI, Normalized Difference Vegetation Index; USGS, U.S. Geological Survey.
Figure 1 is a set of twelve aerial views of maps. On the left, four maps of Boston depict the Normalized Difference Vegetation Index, NatureScore, park cover, and blue space. Four maps of Washington, D.C., in the center, depict the normalized difference vegetation index, NatureScore, park cover, and blue space. On the right, four maps of San Francisco depict the Normalized Difference Vegetation Index, NatureScore, park cover, and blue space. The scale of the Normalized Difference Vegetation Index ranges from 0.0 to 1.0 and NatureScore ranges from 0 to 100. The values of park cover ranges as 0 and 1 and the values of blue space ranges as 0 and 1.
Normalized Different Vegetation Index.
We estimated the NDVI, an indicator of greenness, using satellite imagery. NDVI is the ratio between the red and near infrared values, and values range from –1 to 1.21 Values close to 1 correspond to areas with complete coverage by live vegetation, values close to zero correspond to areas without much live vegetation (e.g., rocks, sand), and negative values correspond to water/ice/snow. We used Landsat 8 images (Collection 1 Tier 1 DN values, representing scaled values, calibrated at sensor radiance) from June 1 through 31 August 2020, to maximize variability in NDVI values. Landsat 8 images are generated every 16 d at a 30-m2 spatial resolution. Using Google Earth Engine,22 we created cloud-free Landsat composites for the United States. We calculated the mean summer NDVI for each tract in the contiguous United States by averaging all pixel values in the tract polygons, after setting negative NDVI values to zero. In addition, we used Landsat 8 images from January 1 through 31 December 2020, to calculate the annual average NDVI for each tract in the contiguous United States.
NatureScore.
NatureScore is a proprietary measure of the quantity and quality of natural elements and was created by NatureQuant.23,24 NatureScore is a blend of park space, open water, park features, tree canopy, computer vision (aerial and street view analysis), noise, air pollution, light pollution, human modifications (road densities and impervious surfaces), geographic information system and land classification databases, and satellite infrared vegetation measurements.23,24 These elements are weighted to create the highest correlation with observed health measures of given natural elements using a proprietary machine learning algorithm.23,24 The NatureScore values range from 0 (poor NatureScore, lacking beneficial natural elements) to 100 (high NatureScore, abundant beneficial natural elements). The data used in this study were based on calendar year 2019 averages. Each tract’s NatureScore was based on a combination of raster (predominantly 10-m2 spatial resolution) and vector data that fell within the tract boundary.
Park cover.
Park cover was based on the U.S. Geological Survey (USGS) Protected Areas Database of the United States (PAD-US). The PAD-US compiles the “best available” data provided by land managing agencies and organizations, and strives to be a complete inventory of public land and other protected areas in the United States.25 PAD-US differentiates between multiple types of public lands. Therefore, we retrieved polygon data from PAD-US (version 2.1; 2020) and selected land types likely to be known and used by the general public for outdoor recreation to create a park cover data set. This included open and restricted access areas but not closed access areas, therein providing a recreational and accessible version of the PAD-US (i.e., PAD-US-RA). An overview of the included land types can be found in the Supplemental Material in the section “Park cover.” To assess park cover, we converted the park data set to a raster image with a spatial resolution of <2 m2 and calculated park cover (area park/area census tract) for each tract using Google Earth Engine.22
Blue space.
We estimated blue space using satellite imagery based on the European Commision’s Joint Research Centre’s Global Surface Water data set.26 This data set contains maps of the location and temporal distribution of surface water from 1984 to 2018 based on imagery from Landsat 5, 7, and 8 satellites at a 30-m2 spatial resolution. Surface water data was aggregated over the entire time period and not available for each year. Using Google Earth Engine,22 we selected the Occurrence band (the frequency with which water was present). If water was present in a pixel for ≥50% of the time, we classified the pixel as blue space. If water was present in a pixel for <50% of the time, we classified the pixel as no blue space. Because adjacent water bodies, such as lakes, rivers, and oceans, are not always included within tract boundaries, we calculated the mean blue space within tracts and a 100-m buffer around each tract.
Potential Confounders
We downloaded data about median age and population size for each tract based on 5-y ACS estimates (2015–2019) via https://www.nhgis.org.16 We calculated population density by dividing population size by tract land area. We defined urban tracts as tracts with ≥1,000 persons/mi2.27 For each tract, we estimated the annual average maximum temperature and daily total precipitation for the year 2020 using data from the Gridded Surface Meteorological data set at an ∼4-km2 spatial resolution.28
Statistical Analysis
We calculated the Spearman correlation between all four natural environment measures, SES, and race/ethnicity indicators. We used linear generalized additive models (GAMs) to evaluate associations between SES indicators and race/ethnicity with NDVI, NatureScore, and park cover. After checking modeling assumptions, we decided to use quintiles of SES indicators and race/ethnicity measures. Given that >40% of all tracts (and >60% of urban tracts) contained no blue space, we used a binary indicator for this measure (0=absent, 1=present), and modeled associations with logistic GAMs. In all models, SES and race/ethnicity indicators were the independent variables and natural environment measures were the dependent variables. We analyzed associations in all tracts and in urban tracts (≥1,000 persons/mi2) For the analyses in urban tracts, we recalculated SES and race/ethnicity quintiles.
To evaluate potential confounding, we specified models with increasing levels of adjustment. Model 1 included the independent variable(s) %<high school education; median household income; % below poverty; %non-Hispanic White+%non-HispanicBlack+%non-Hispanic Asian+%non-Hispanic other+%Hispanic, as well as median age and splines (a full tensor product smooth)29 for the combination of latitude and longitude of the centroid of the tract to account for spatial autocorrelation between tracts. We additionally adjusted for population density in model 2. In model 3, we additionally adjusted for annual average temperature and precipitation to account for climatic factors. For models including an SES indicator, we added all race/ethnicity measures to model 4. For models including race/ethnicity, we added median household income to model 4. We did not include all SES measures simultaneously in any single model because they were strongly correlated with each other (Spearman rho ≥0.65).
For sensitivity analyses, we included a random effect by state to model 4 to account for differences between states. Further, we performed analyses by U.S. Census divisions, to evaluate whether associations of median household income, %non-Hispanic Black and %Hispanic differed between geographic areas. We maintained the quintiles used in the main analyses for all stratified analyses.
To compare effect estimates for NDVI, NatureScore, and park cover, we standardized these outcomes so that beta coefficients presented a percentage increase or decrease of the standard deviation (SD) of these indicators (based on all tracts). Given that the correlation between summer and annual mean NDVI was very strong (Pearson r=0.96), we only used summer NDVI (referred to as NDVI) in our analyses. Analyses were performed in RStudio (version 1.4.1717; RStudio) and used the following packages (sp, raster, dplyr, sf, ggplot2, grid, data.table, and mgcv).
Results
Descriptive Statistics
We excluded 1.4% of all tracts in the contiguous United States because of missing data, resulting in 71,532 included tracts. Approximately 63% of the tracts were urban (≥1,000 persons/mi2). NDVI levels were generally higher in the eastern United States, whereas park cover was higher in the western United States (Figure S2). Mean NDVI and NatureScore and percentage of tracts that contain blue space were substantially lower in urban tracts than across all tracts, mean park cover was weakly lower in urban tracts (Table 1, Table S1). Means±SDs of median household income, %<poverty level, %<high school education, %non-Hispanic Black, and %Hispanic were higher in urban tracts than across all tracts.
Table 1 Descriptive statistics of all census tracts (n=71,532) and urban census tracts (n=45,338) in the contiguous United States after excluding census tracts with missing data.
Variable All census tracts Urban census tracts
Mean±SD or n (%) Mean±SD or n (%)
Land area (km2) 108.7 ± 557.6 3.6 ± 4.0
Natural environment measures
NDVI 0.48 ± 0.18 0.41 ± 0.15
NatureScore 64.4 ± 33.4 48.9 ± 31.5
Park cover (proportion) 0.08 ± 0.13 0.07 ± 0.10
Blue space +100m (continuous, proportion) 0.03 ± 0.09 0.02 ± 0.08
Contains blue space +100m 41,632 (58.2) 17,304 (38.2)
Race/ethnicity
% non-Hispanic White 61.5 ± 29.9 51.7 ± 29.6
% non-Hispanic Black 13.5 ± 21.5 16.8 ± 23.9
% non-Hispanic Asian 4.8 ± 8.9 6.7 ± 10.3
% non-Hispanic other 3.4 ± 5.2 3.4 ± 3.0
% Hispanic 16.7 ± 21.5 21.4 ± 23.5
SES indicators
%<high school education 12.7 ± 10.3 13.2 ± 11.3
% below the U.S. poverty level 14.1 ± 10.7 14.9 ± 11.7
Median household income (USD) 66,976 ± 33,471 68,495 ± 35,892
Potential confounders
Population density (persons/miles2) 5,277 ± 11,669 8,175 ± 13,852
Median age (y) 39.6 ± 7.8 37.8 ± 7.5
Annual average daily maximum temperature (°C) 20.4 ± 4.8 20.9 ± 4.8
Annual average daily total precipitation (mm) 2.9 ± 1.5 2.8 ± 1.6
Note: %, percentage; NDVI, Normalized Difference Vegetation Index; SD, standard deviation; SES, socioeconomic status; USD, U.S. dollars.
The correlation between NDVI and NatureScore was very strong across all tracts (Spearman rho=0.87; Figure S3), whereas correlations between other natural environment measures were weak to moderate (Spearman rho ≤0.40). NDVI and park cover were not correlated (Spearman rho=0.00) across all tracts and weakly positively correlated across urban tracts (Spearman rho=0.11). Natural environment measures were generally negatively correlated with %<poverty level, %<high school education, and %Hispanic, but positively correlated with median household income and %non-Hispanic White across all and urban tracts. We observed a clear trend between both NDVI and NatureScore and %non-Hispanic White and %Hispanic; the higher the NDVI or NatureScore, the higher the %non-Hispanic White and the lower the %Hispanic (Figure 2, Tables S2–S9).
Figure 2. Average 2015–2019 race/ethnicity composition by levels of NDVI (2019 data), NatureScore (2020 data), park cover (2020 data), and blue space (1984–2018 data) in all census tracts (n=71,532) and in urban census tracts (n=45,338) in the contiguous United States after excluding census tracts with missing data. See Tables S2–S9 for corresponding numeric data. The x-axis of each plot was truncated by the 2.5 and 97.5 percentile. Note: NDVI, Normalized Difference Vegetation Index.
Figure 2 is a set of eight area graphs. On the left, a set of four area graphs titled All, plotting Neighborhood Racial Composition, ranging from 0 to 100 percent in increments of 25 (y-axis) across Normalized Difference Vegetation Index, ranging from 0.125 to 0.625 in increments of 0.125, NatureScore, ranging from 2.5 to 77.5 in increments of 25, Park cover (proportion), ranging from 0.00 to 0.50 in increments of 0.125, and Blue space (proportion), ranging from 0.00 to 0.30 in increments of 0.10 (x-axis) for Non-Hispanic White, Non-Hispanic Black, Non-Hispanic Asian, Non-Hispanic Other, and Hispanic. On the right, a set of four area graphs titled Urban, plotting Neighborhood Racial Composition, ranging from 0 to 100 percent in increments of 25 (y-axis) across Normalized Difference Vegetation Index, ranging from 0.125 to 0.625 in increments of 0.125, NatureScore, ranging from 2.5 to 77.5 in increments of 25, Park cover (proportion), ranging from 0.00 to 0.375 in increments of 0.125, and Blue space (proportion), ranging from 0.00 to 0.30 in increments of 0.10 (x-axis) for Non-Hispanic White, Non-Hispanic Black, Non-Hispanic Asian, Non-Hispanic Other, and Hispanic.
Relations of SES and Race/Ethnicity with Natural Environment Measures
Urban tracts with higher median household incomes, lower %<poverty level, and lower %<high school education had higher levels of NDVI, NatureScore, park cover, and odds of containing blue space (Figure 3, Table S10). Urban tracts in the highest median household income quintile had higher NDVI [44.8% of the SD; 95% confidence interval (CI): 42.8, 46.8], corresponding to a 0.08 higher NDVI; NatureScore [54.9% of the SD (95% CI: 52.6, 57.3)], corresponding to a 18.3 higher NatureScore; and park cover [16.2% of the SD (95% CI: 13.5, 19.0)], corresponding to a 2.1% higher park cover, compared with urban tracts in the lowest median household income quintile. Associations with SES indicators were generally strongest for NatureScore and weakest for park cover. Across all tracts, lower median household income and higher %<poverty level were associated with lower NDVI and NatureScore; we found nonlinear associations for park cover and blue space. Associations of SES indicators with NDVI and NatureScore were generally stronger in urban tracts than across all tracts.
Figure 3. Associations of 2015–2019 median household income, %<poverty level, and %<high school education with NDVI (2020 data), NatureScore (2019 data), park cover (2020 data), and blue space (1984–2018 data) in all census tracts (n=71,532, ●) and in urban census tracts (n=45,338, ) in the contiguous United States after excluding census tracts with missing data. See Table S10 for corresponding numeric data. NDVI, NatureScore, and park cover were standardized and multiplied by 100, so that the beta represents the percentage increase/decrease of the SD (NDVI: 0.18, NatureScore: 33.4, Park cover: 0.13). The error bars correspond to 95% CIs. Models included median household income / %<poverty level/%<high school education and were adjusted for median age, population density, temperature, precipitation, %non-Hispanic White, %non-Hispanic Black, %non-Hispanic Asian, %non-Hispanic other, %Hispanic, and latitude and longitude of the centroid. For all census tracts, the following percentiles (20, 40, 60, 80) were used to create median household income (in U.S. dollars) quintiles: 41,135, 53,214, 66,468, 88,640; %<poverty level quintiles: 5.5, 9.1, 13.7, 21.3; %<high school education quintiles: 4.4, 7.8, 12.1, 19.5. For urban census tracts, the following percentiles (20, 40, 60, 80) were used to create median household income (in U.S. dollars) quintiles: 39,435, 53,283, 69,168, 93,216; %<poverty level quintiles: 5.4, 9.2, 14.5, 23.4; %<high school education quintiles: 4.0, 7.5, 12.4, 21.2. Note: %, percentage; CI, confidence interval; NDVI, Normalized Difference Vegetation Index; OR, odds ratio; q, quartile; SD, standard deviation.
Figure 3 is a set of twelve line graphs. The first set on the left, three line graphs, plotting Normalized Difference Vegetation Index beta (95 percent confidence interval), ranging from 0 to 60 in increments of 20; negative 50 to 10 in increments of 20; and negative 50 to 10 in increments of 20 (y-axis) across Median household income, ranging from quarter 1 to 5 in unit increments; percentage less than poverty level, ranging from quarter 1 to 5 in unit increments; and percentage less than high school education, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban. The second set, three line graphs, plotting NatureScore, ranging from 0 to 60 in increments of 20; negative 50 to 10 in increments of 20; and negative 50 to 10 in increments of 20 (y-axis) across Median household income, ranging from quarter 1 to 5 in unit increments; percentage less than poverty level, ranging from quarter 1 to 5 in unit increments; and percentage less than high school education, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban. The third set, three line graphs, plotting Park cover, ranging from 0 to 60 in increments of 20; negative 50 to 10 in increments of 20; and negative 50 to 10 in increments of 20 (y-axis) across Median household income, ranging from quarter 1 to 5 in unit increments; percentage less than poverty level, ranging from quarter 1 to 5 in unit increments; and percentage less than high school education, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban. The fourth set, three line graphs, plotting Blue space odds ratio (95 percent confidence interval), ranging from 1.0 to 1.6 in increments of 0.2; 0.7 to 1.3 in increments of 0.2; and 0.6 to 1.4 in increments of 0.2 (y-axis) across Median household income, ranging from quarter 1 to 5 in unit increments; percentage less than poverty level, ranging from quarter 1 to 5 in unit increments; and percentage less than high school education, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban.
Urban tracts with lower %non-Hispanic White had lower NDVI and NatureScore but not park cover (Figure 4, Table S11). For %non-Hispanic Black, we observed weak positive associations with NDVI, NatureScore, and park cover in urban tracts. For %Hispanic, we did not find any clear associations with NDVI, NatureScore, and park cover. Across all tracts, higher %Hispanic and lower %non-Hispanic White had lower NDVI and NatureScore but not lower park cover. We did not observe any clear patterns between %non-Hispanic Black and NDVI, NatureScore, and park cover. Higher %non-Hispanic White and %Hispanic and lower %non-Hispanic Black were generally associated with higher odds of the tracts containing blue space.
Figure 4. Associations of 2015–2019% non-Hispanic White, %non-Hispanic Black, and %Hispanic with NDVI (2020 data), NatureScore (2019 data), park cover (2020 data), and blue space (1984–2018 data) in all census tracts (n=71,532, ●) and in urban census tracts (n=45,338, ) in the contiguous United States after excluding census tracts with missing data. See Table S11 for corresponding numeric data. NDVI, NatureScore, and park cover were standardized and multiplied by 100, so that the beta represents the percentage increase/decrease of the SD (NDVI: 0.18, NatureScore: 33.4, Park cover: 0.13). The error bars correspond to 95% CIs. Models included %non-Hispanic White, %non-Hispanic Black, %non-Hispanic Asian, %non-Hispanic Other, %Hispanic and were adjusted for median age, population density, temperature, precipitation, median household income, and latitude and longitude of the centroid. For all census tracts, the following percentiles (20, 40, 60, 80) were used to create %non-Hispanic White quintiles: 30.2, 60.4, 78.0, 89.9; %non-Hispanic Black quintiles: 0.6, 2.4, 6.8, 20.3; %Hispanic quintiles: 2.1, 5.2, 11.1, 27.0. For urban census tracts, the following percentiles (20, 40, 60, 80) were used to create %non-Hispanic White: 18.0, 45.8, 66.3, 81.1; %non-Hispanic Black quintiles: 1.3, 4.0, 9.7, 26.0; %Hispanic quintiles: 3.7, 8.3, 16.6, 36.6. Note: %, percentage; CI, confidence interval; NDVI, Normalized Difference Vegetation Index; OR, odds ratio; q, quartile; SD, standard deviation.
Figure 4 is a set of twelve line graphs. The first set on the left, three line graphs, plotting Normalized Difference Vegetation Index beta (95 percent confidence interval), ranging from 0 to 60 in increments of 20; negative 20 to 10 in increments of 10; and negative 20 to 10 in increments of 10 (y-axis) across percentage of non-Hispanic White, ranging from quarter 1 to 5 in unit increments; percentage of non-Hispanic Black, ranging from quarter 1 to 5 in unit increments; and percentage of Hispanic, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban. The second set, three line graphs, plotting NatureScore, ranging from 0 to 60 in increments of 20; negative 20 to 10 in increments of 10; and negative 20 to 10 in increments of 10 (y-axis) across percentage of non-Hispanic White, ranging from quarter 1 to 5 in unit increments; percentage of non-Hispanic Black, ranging from quarter 1 to 5 in unit increments; and percentage of Hispanic, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban. The third set, three line graphs, plotting Park cover, ranging from 0 to 60 in increments of 20; negative 20 to 10 in increments of 10; and negative 20 to 10 in increments of 10 (y-axis) across percentage of non-Hispanic White, ranging from quarter 1 to 5 in unit increments; percentage of non-Hispanic Black, ranging from quarter 1 to 5 in unit increments; and percentage of Hispanic, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban. The fourth set, three line graphs, plotting Blue space odds ratio (95 percent confidence interval), ranging from 1.0 to 4.0 in increments of 1.0; 0.5 to 1.1 in increments of 0.1; and 0.9 to 1.5 in increments of 0.1 (y-axis) across percentage of non-Hispanic White, ranging from quarter 1 to 5 in unit increments; percentage of non-Hispanic Black, ranging from quarter 1 to 5 in unit increments; and percentage of Hispanic, ranging from quarter 1 to 5 in unit increments (x-axis) for all and urban.
Across all tracts and in urban tracts, associations of SES and %non-Hispanic White with NDVI, NatureScore, and blue space were generally strongest in minimally adjusted models (model 1) and mildly attenuated after adjustments for potential confounders (Figures S4–S7, Tables S12–S17). Associations of SES and %non-Hispanic White with park cover barely changed with increasing levels of adjustment. Associations of %non-Hispanic Black with NDVI, NatureScore, and park cover in urban tracts became slightly stronger after adjustment for potential confounders, especially median household income. %Hispanic was associated with lower NDVI and NatureScore in urban tracts in minimally adjusted models, but we did not notice any pattern after adjustment for potential confounders. %non-Hispanic Asian was negatively associated with NDVI and NatureScore, but positively associated with blue space (only in urban census tracts) and park cover (Table S18). Sensitivity analyses including a random effect by state showed similar results as the fully adjusted models (Tables S19–S20).
In urban tracts, positive associations of median household income with NDVI and NatureScore were generally consistent across U.S. Census divisions, whereas associations with park cover showed no clear pattern in any U.S. Census division (Table S21). Median household income was generally negatively associated with the odds of containing blue space in the Northeastern (New England, Middle Atlantic) and Western (Mountain, Pacific) divisions and positively associated with the odds of containing blue space in the Midwestern (East North Central, West North Central) and Southern (South Atlantic, East South Central, West South Central) divisions. For most divisions, associations of %non-Hispanic Black and %Hispanic with NDVI, NatureScore, park cover, and blue space in urban tracts were not consistent (Tables S22–S23); for some divisions we observed positive associations and for others we observed negative associations. For example, we observed positive associations of %non-Hispanic Black with NDVI and NatureScore in the Middle and South Atlantic divisions, but we found negative associations in the East South Central division.
Discussion
In urban areas across the contiguous United States, census tracts with lower SES had less greenness (i.e., NDVI), park cover, and presence of blue space and lower NatureScores. Urban tracts with lower percentages of non-Hispanic White individuals had less greenness and lower NatureScores but not less park cover; we did not find any clear patterns for percentages of Hispanic individuals. Urban tracts with higher percentages of non-Hispanic Black individuals had more NDVI, NatureScore, and park cover. Across all tracts in the contiguous United States, associations between SES and natural environment measures were mixed and differed by the specific natural environment measure. Associations with race/ethnicity were more definitive across all tracts; tracts with larger proportions of Hispanic individuals and smaller proportions of non-Hispanic White individuals had less greenness and lower NatureScores but not less park cover. The inequitable distribution of natural environments could partially explain the health disparities between SES and race/ethnicity groups in the United States given that multiple reviews have documented protective associations of natural environment with adverse health outcomes.6–10
Associations of SES with natural environment measures in this study are generally consistent with recent studies.5,11,12,30,31 These associations may be due to the fact that green and blue spaces are highly valued, especially in urban areas, and proximity to natural environments and private greenery may increase house prices.32–34 Another possible explanation could be that there is less green infrastructure investment in low SES and minority race areas than in other areas.35 We note that associations of SES measures with NDVI and NatureScore in urban tracts were weakest for %<high school education. However, associations with park cover and blue space in urban tracts showed more consistent patterns for %<high school education than for the other measures. We have no clear explanation for this, but note that although education and income are strongly related to each other, education is generally considered an early life SES measure36 and only indirectly (via employment and income) does it affect material resources, such as housing.
In general, we observed that SES was more strongly associated with NDVI compared with park cover, which supports findings by Nesbitt et al.15 This is probably because NDVI captures street and private greenery, which may contain a large portion of the total greenness in urban areas, and high SES neighborhoods generally have larger residential properties. We also observed a very weak correlation between NDVI and park cover in urban tracts (Spearman rho=0.11). Not all parks contain dense vegetation, especially in dry areas of the western United States. Further, urban parks could include paved paths, playgrounds, and basketball courts that are not captured by NDVI. Moreover, there might be fewer parks in (suburban) areas with large backyards and tree-lined streets. We also note that NDVI was very strongly correlated with NatureScore and that associations with both measures and SES and race/ethnicity were similar. NatureScore is a blend of several natural elements, including park space, noise, air pollution, and satellite infrared vegetation measurements, and proximity to greenness is one of the most heavily weighted elements in the NatureScore. Further, associations of SES measures with blue space showed nonlinear patterns across all tracts. This could be due to harbors and air pollution sources (e.g., ships) in/close to blue spaces that may not be highly valued or provide beneficial health effects.
We found weak positive associations between %non-Hispanic Black and NDVI, NatureScore, and park cover in urban tracts. Associations became more pronounced after adjustment for median household income. This is likely due to the moderate negative correlation (Spearman rho=−0.41) between %non-Hispanic Black and median household income. A review by Watkins and Gerrish reported no urban forest inequity for Black populations.13 Casey et al. reported a positive association of %non-Hispanic White and a weak negative association of %non-Hispanic Black with a change in NDVI 2001–2011 in urban census tracts.37 A review by Rigolon reported that White individuals had more urban park acreage than Black or Hispanic individuals,11 whereas we observed no clear associations of %non-Hispanic White or %Hispanic with park cover in urban tracts. Differences in associations between these studies might be due to differences in adjustment for potential confounders and spatial autocorrelation, exposure assessment, and study years. Moreover, we observed that associations of race/ethnicity with natural environments differed between U.S. Census divisions, indicating that associations could differ between study regions.
Associations of %non-Hispanic White with NDVI and NatureScore were generally weaker in urban tracts than across all tracts, in contrast to patterns with SES indicators. For %non-Hispanic Black and Hispanics, we observed negative associations with NDVI and NatureScore in all tracts, but not in urban tracts. This may partially be due to urban–rural patterns that emerge in models with all the tracts. In densely populated areas, NDVI, NatureScore, and percentage of tracts containing blue space were lower than in more rural areas. Given that population density is more strongly correlated with %non-Hispanic White, %non-Hispanic Black, and %Hispanic across all tracts compared with urban tracts, this may have resulted in stronger associations with race/ethnicity in models that included all tracts than in models with only urban tracts, despite adjustments for population density. Because SES measures were weakly correlated with population density, this pattern may not have affected associations of SES indicators. Further, we note that associations of %non-Hispanic Black and %Hispanic with NDVI and NatureScore differed between divisions, whereas associations of median household income were generally consistent across all divisions.
We observed no clear pattern between %Hispanic and NDVI, NatureScore, and park cover in urban tracts. A weak negative association of %Hispanic with change in NDVI 2001–2011 was observed by Casey et al.37 Choi et al. observed negative associations between %Latino with street greenery, but they observed positive associations with green space accessibility in 12 U.S. cities.38 The review by Watkins and Gerrish also reported urban forest inequities for Hispanic populations.13 However, when studies that did not control for income were removed, no inequities were found.13 We observed a similar trend; patterns of associations with NDVI, NatureScore, or park cover disappeared after adjustment for income. In all tracts, we observed that a higher %Hispanics was associated with lower NDVI and NatureScore. We note that the percentage of Hispanic residents was highest in the southwest United States, which is a region with generally low NDVI (Figures S1 and S2). Hence, the negative associations with NDVI and NatureScore might be due to the large-scale spatial variation of these variables.
Our results are comparable to two studies that showed redlined neighborhoods (i.e., those ineligible for federal mortgage programs in the 1930s) had less green space in urban areas.39,40 Today, redlined neighborhoods are still generally composed of lower SES, Black, and Hispanic populations.41 Both studies that focused on redlined neighborhoods used green space measures that included private greenery,39,40; they did not evaluate associations with park cover or blue space.
Limitations
This study has a few limitations. We used cross-sectional data, so we did not evaluate how associations of SES or race/ethnicity with natural environment change over time. NDVI and NatureScore do not differentiate between private and public greenery; hence, we do not know whether vegetation on private or public land is primarily responsible for the observed inequities. We acknowledge that the park data set is based on “best available” data provided by land management agencies and organizations; therefore, it may not be completely accurate or comprehensive. We used satellite imagery from 1984–2018 at a 30-m2 spatial resolution with a 50% threshold to classify blue spaces and may have missed some water bodies that are more ephemeral. Although blue space data was aggregated over multiple years in this data set, we do not think that this would affect our results because the spatial distribution of most water bodies is relatively stable over time.42 We used a binary blue space indicator because of the limited variability and did not evaluate whether tracts with low SES or high %non-Hispanic Black or %Hispanic had lower blue space levels. Given that blue space is based on satellite images, we did not differentiate between types of blue spaces (e.g., oceans, lakes, rivers). We used four diverse natural environment metrics, but we note that other studies used metrics based on land use/cover databases.6,7 Recently, studies have also classified greenness based on street view images.43,44 Associations with these metrics may differ from associations with the metrics used in this study. We also used census tracts as our unit of observation, but residents may travel across tract boundaries to access nature. Further, we note that this is an ecological study and that ecological studies should not be used to make inferences about individuals.
Strengths
This study also has several strengths. To the best of our knowledge, this is the first study to cover all census tracts in the contiguous United States and to use diverse natural environments metrics to assess whether natural environments vary by SES and race/ethnicity. We evaluated associations across all tracts and in urban census tracts. We adjusted for several potential confounders and corrected for spatial autocorrelation. In addition, we used four metrics of natural environments that capture different aspects of nature that might have independent influences on health. SES and race/ethnicity measures were based on 2015–2019 estimates, NatureScore was based on data from 2019, and NDVI and park cover were based on data from 2020.
Conclusion
In short, we observed an inequitable distribution of natural environments by SES and race/ethnicity in the contiguous United States. The strength of the associations differed between the natural environment measures used. Associations with SES were stronger in urban tracts than across all census tracts, whereas associations with %non-Hispanic White were stronger across all census tracts than in urban tracts. Assuming that exposure to natural environments is protective against several adverse health outcomes, the inequitable distribution of natural environments may partly explain health disparities observed in the United States. Increasing green and blue spaces in urban areas can be challenging and may result in green gentrification.45 Therefore, urban planners should target green and blue space interventions in an equitable way that could promote healthier environments in marginalized communities.46
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This study was supported by National Institute of Environmental Health Sciences (R01 ES028033, P30 ES000002), and the National Heart, Lung, and Blood Institute (R01 HL150119). The funders had no role in the study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.
==== Refs
References
1. Hero JO, Zaslavsky AM, Blendon RJ. 2017. The United States leads other nations in differences by income in perceptions of health and health care. Health Aff (Millwood) 36 (6 ):1032–1040, PMID: , 10.1377/hlthaff.2017.0006.28583961
2. OECD (Organisation for Economic Co-operation, Development). 2019. Health for Everyone?: Social Inequalities in Health and Health Systems, 10.1787/3c8385d0-en [accessed 20 May 2021].
3. Braveman P, Arkin E, Orleans T, Proctor D, Acker J, Plough A. 2018. What is health equity? Behav Sci Policy 4 (1 ):1–14, 10.1353/bsp.2018.0000.
4. Rigolon A, Browning MHEM, McAnirlin O, Yoon HV. 2021. Green space and health equity: a systematic review on the potential of green space to reduce health disparities. Int J Environ Res Public Health 18 (5 ):2563, PMID: , 10.3390/ijerph18052563.33806546
5. Schüle SA, Hilz LK, Dreger S, Bolte G. 2019. Social inequalities in environmental resources of green and blue spaces: a review of evidence in the WHO European Region. Int J Environ Res Public Health 16 (7 ):1216, PMID: , 10.3390/ijerph16071216.30987381
6. James P, Banay RF, Hart JE, Laden F. 2015. A review of the health benefits of greenness. Curr Epidemiol Rep 2 (2 ):131–142, PMID: , 10.1007/s40471-015-0043-7.26185745
7. Fong KC, Hart JE, James P. 2018. A review of epidemiologic studies on greenness and health: updated literature through 2017. Curr Environ Health Rep 5 (1 ):77–87, PMID: , 10.1007/s40572-018-0179-y.29392643
8. Gascon M, Zijlema W, Vert C, White MP, Nieuwenhuijsen MJ. 2017. Outdoor blue spaces, human health and well-being: a systematic review of quantitative studies. Int J Hyg Environ Health 220 (8 ):1207–1221, PMID: , 10.1016/j.ijheh.2017.08.004.28843736
9. Yang BY, Zhao T, Hu LX, Browning MHEM, Heinrich J, Dharmage SC, et al. 2021. Greenspace and human health: an umbrella review. Innovation (Camb) 2 (4 ):100164, PMID: , 10.1016/j.xinn.2021.100164.34622241
10. Twohig-Bennett C, Jones A. 2018. The health benefits of the great outdoors: a systematic review and meta-analysis of greenspace exposure and health outcomes. Environ Res 166 :628–637, PMID: , 10.1016/j.envres.2018.06.030.29982151
11. Rigolon A. 2016. A complex landscape of inequity in access to urban parks: a literature review. Landsc Urban Plan 153 :160–169, 10.1016/j.landurbplan.2016.05.017.
12. Gerrish E, Watkins SL. 2018. The relationship between urban forests and income: a meta-analysis. Landsc Urban Plan 170 :293–308, PMID: , 10.1016/j.landurbplan.2017.09.005.29249844
13. Watkins SL, Gerrish E. 2018. The relationship between urban forests and race: a meta-analysis. J Environ Manage 209 :152–168, PMID: , 10.1016/j.jenvman.2017.12.021.29289843
14. Park Y, Guldmann JM. 2020. Understanding disparities in community green accessibility under alternative green measures: a metropolitan-wide analysis of Columbus, Ohio, and Atlanta, Georgia. Landsc Urban Plan 200 :103806, 10.1016/j.landurbplan.2020.103806.
15. Nesbitt L, Meitner MJ, Girling C, Sheppard SRJ, Lu Y. 2019. Who has access to urban vegetation? A spatial analysis of distributional green equity in 10 US cities. Landsc Urban Plan 181 :51–79, 10.1016/j.landurbplan.2018.08.007.
16. Manson S, Schroeder J, Van Riper D, Kugler T, Ruggles S. 2020. IPUMS National Historical Geographic Information System: version 15.0 [Database]. 10.18128/D050.V15.0 [accessed 1 April 2021].
17. U.S. Census Bureau. 2021. Glossary. https://www.census.gov/programs-surveys/geography/about/glossary.html [accessed 5 May 2021].
18. U.S. Census Bureau. 2021. American Community Survey (ACS). https://www.census.gov/programs-surveys/acs/ [accessed 31 March 2021].
19. Nieuwenhuijsen MJ, Khreis H, Triguero-Mas M, Gascon M, Dadvand P. 2017. Fifty shades of green: pathway to healthy urban living. Epidemiology 28 (1 ):63–71, PMID: , 10.1097/EDE.0000000000000549.27525811
20. White MP, Elliott LR, Gascon M, Roberts B, Fleming LE. 2020. Blue space, health and well-being: a narrative overview and synthesis of potential benefits. Environ Res 191 :110169, PMID: , 10.1016/j.envres.2020.110169.32971082
21. NASA (National Aeronautics and Space Administration). 2000. Measuring Vegetation (NDVI & EVI). https://earthobservatory.nasa.gov/features/MeasuringVegetation/measuring_vegetation_2.php [accessed 16 June 2020].
22. Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. 2017. Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sens Environ 202 :18–27, 10.1016/j.rse.2017.06.031.
23. NatureQuant. 2021. Nature Access and Environmental Justice: Addressing Nature Access Disparities in the United States. https://www.naturequant.com/NatureQuant-Nature Access_and_Environmental_Justice.pdf [accessed 23 July 2021].
24. NatureQuant. 2021. NatureScore Priority Index: A Tool to Identify Areas of Nature Deficiency and High SES Deprivation. https://www.naturequant.com/blog/NatureScore-Priority-Index/ [accessed 4 August 2021].
25. USGS GAP (U.S. Geological Survey Gap Analysis Project). 2020. Protected Areas Database of the United States (PAD-US) 2.1, 10.5066/P92QM3NT [accessed 1 March 2021].
26. Pekel JF, Cottam A, Gorelick N, Belward AS. 2016. High-resolution mapping of global surface water and its long-term changes. Nature 540 (7633 ):418–422, PMID: , 10.1038/nature20584.27926733
27. USDA ERS (U.S. Department of Agriculture Economic Research Service). 2019. What is Rural? https://www.ers.usda.gov/topics/rural-economy-population/rural-classifications/what-is-rural/ [accessed 17 September 2021].
28. Abatzoglou JT. 2013. Development of gridded surface meteorological data for ecological applications and modelling. Int J Climatol 33 (1 ):121–131, 10.1002/joc.3413.
29. Wood SN. 2017. Generalized Additive Models: An Introduction with R. 2nd ed. New York, NY: Chapman and Hall/CRC.
30. Viinikka A, Paloniemi R, Assmuth T. 2018. Mapping the distributive environmental justice of urban waters. Fennia 196 (1 ):9–23, 10.11143/fennia.64137.
31. Wüstemann H, Kalisch D, Kolbe J. 2017. Accessibility of urban blue in German major cities. Ecol Indic 78 :125–130, 10.1016/j.ecolind.2017.02.035.
32. Staats H, Swain R. 2020. Cars, trees, and house prices: evaluation of the residential environment as a function of numbers of cars and trees in the street. Urban For Urban Green 47 :126554, 10.1016/j.ufug.2019.126554.
33. Evangelio R, Hone S, Lee M, Prentice D. 2019. What makes a locality attractive? Estimates of the amenity value of parks for Victoria. Econ Pap 38 (3 ):182–192, 10.1111/1759-3441.12259.
34. Chen WY, Li X, Hua J. 2019. Environmental amenities of urban rivers and residential property values: a global meta-analysis. Sci Total Environ 693 :133628, PMID: , 10.1016/j.scitotenv.2019.133628.31377374
35. Kephart L. 2022. How racial residential segregation structures access and exposure to greenness and green space: a review. Environ Justice 15 (4 ):204–213, 10.1089/env.2021.0039.
36. Galobardes B, Shaw M, Lawlor DA, Lynch JW, Smith GD. 2006. Indicators of socioeconomic position (part 1). J Epidemiol Community Health 60 (1 ):7–12, PMID: , 10.1136/jech.2004.023531.16361448
37. Casey JA, James P, Cushing L, Jesdale BM, Morello-Frosch R. 2017. Race, ethnicity, income concentration and 10-year change in urban greenness in the United States. Int J Environ Res Public Health 14 (12 ):1546, PMID: , 10.3390/ijerph14121546.29232867
38. Choi DA, Park K, Rigolon A. 2020. From XS to XL urban nature: examining access to different types of green space using a “just sustainabilities” framework. Sustainability 12 (17 ):6998, 10.3390/su12176998.
39. Nardone A, Rudolph KE, Morello-Frosch R, Casey JA. 2021. Redlines and greenspace: the relationship between historical redlining and 2010 greenspace across the United States. Environ Health Perspect 129 (1 ):017006, PMID: , 10.1289/EHP7495.33502254
40. Locke DH, Hall B, Grove JM, Pickett STA, Ogden LA, Aoki C, et al. 2021. Residential housing segregation and urban tree canopy in 37 US cities. npj Urban Sustain 1 (1 ):15, 10.1038/s42949-021-00022-0.
41. Mitchell B, Franco J. 2018. HOLC “redlining” maps: the persistent structure of segregation and economic inequality. https://ncrc.org/holc [accessed 19 July 2021].
42. Homer C, Dewitz J, Jin S, Xian G, Costello C, Danielson P, et al. 2020. Conterminous United States land cover change patterns 2001–2016 from the 2016 National Land Cover Database. ISPRS J Photogramm Remote Sens 162 :184–199, PMID: , 10.1016/j.isprsjprs.2020.02.019.35746921
43. Larkin A, Hystad P. 2019. Evaluating street view exposure measures of visible green space for health research. J Expo Sci Environ Epidemiol 29 (4 ):447–456, PMID: , 10.1038/s41370-018-0017-1.29352209
44. Jimenez MP, Suel E, Rifas-Shiman SL, Hystad P, Larkin A, Hankey S, et al. 2022. Street-view greenspace exposure and objective sleep characteristics among children. Environ Res 214 (pt 1 ):113744, PMID: , 10.1016/j.envres.2022.113744.35760115
45. Wolch JR, Byrne J, Newell JP. 2014. Urban green space, public health, and environmental justice: the challenge of making cities ‘just green enough.’ Landsc Urban Plan 125 :234–244, 10.1016/j.landurbplan.2014.01.017.
46. Anguelovski I, Connolly JJT, Pearsall H, Shokry G, Checker M, Maantay J, et al. 2019. Why green “climate gentrification” threatens poor and vulnerable populations. Proc Natl Acad Sci USA 116 (52 ):26139–26143, 10.1073/pnas.1920490117.
|
PMC009xxxxxx/PMC9875843.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36696106
EHP10967
10.1289/EHP10967
Research
Exposure to Air Pollution during Pre-Hypertension and Subsequent Hypertension, Cardiovascular Disease, and Death: A Trajectory Analysis of the UK Biobank Cohort
Zhang Shiyu 1
Qian Zhengmin Min 2
https://orcid.org/0000-0002-8775-4301
Chen Lan 1
Zhao Xing 3
Cai Miao 1
Wang Chongjian 4
Zou Hongtao 1
Wu Yinglin 1
Zhang Zilong 1
Li Haitao 5
https://orcid.org/0000-0002-3643-9408
Lin Hualiang 1
1 Department of Epidemiology, School of Public Health, Sun Yat-sen University, Guangzhou, China
2 Department of Epidemiology and Biostatistics, College for Public Health and Social Justice, Saint Louis University, Saint Louis, Missouri, USA
3 West China School of Public Health and West China Fourth Hospital, Sichuan University, Chengdu, Sichuan, China
4 Department of Epidemiology and Biostatistics, College of Public Health, Zhengzhou University, Zhengzhou, China
5 Shenzhen University General Hospital, Shenzhen University Clinical Medical Academy, Shenzhen, China
Address correspondence to Hualiang Lin, Department of Epidemiology, School of Public Health, Sun Yat-sen University, No. 74 Zhongshan Rd. 2, Yuexiu District, Guangzhou 510080, China. Telephone: 86-20-87332455. Email: [email protected]
25 1 2023
1 2023
131 1 01700818 1 2022
27 11 2022
15 12 2022
27 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
The associations between air pollution exposure and morbidity and mortality of cardiovascular diseases (CVDs) have been widely reported; however, evidence on such associations across different dynamic disease trajectories remain unknown.
Objective:
We examined whether ambient air pollution during the prehypertension (pre-HTN) stage could aggravate the progression from hypertension (HTN) to CVD, and consequent death.
Methods:
A total of 168,010 adults with pre-HTN (120–139 mmHg systolic blood pressure or 80–89 mmHg diastolic blood pressure) from the UK Biobank were included in this analysis. We used a multistate model to explore the associations between five air pollutants (PM2.5, PM2.5 absorbance, PM10, NO2, and NOx) and the risk of six disease transitions (from pre-HTN to HTN, from pre-HTN to CVD, from pre-HTN to death, from HTN to CVD, from HTN to death, and from CVD to death). Mediation analyses were further conducted to explore the role of intermediate diseases in the dynamic progression of CVDs.
Results:
During a median follow-up of 12 y, 13,743 (8.18%) of participants with pre-HTN developed HTN, whereas 12,825 (7.63%) and 4,467 (2.66%) directly developed CVD or died, respectively. Air pollution was positively associated with the dynamic disease progression. For example, a per-interquartile range increase of PM2.5 was significantly associated with the hazard ratios (HRs) of 1.105 [95% confidence intervals (CI): 1.083, 1.127], 1.045 (95% CI: 1.022, 1.068), and 1.086 (95% CI: 1.047, 1.126) in the transition from pre-HTN to HTN, CVD, and death, respectively. Higher levels of air pollution were associated with increased transition probability of disease progression. Mediation analyses indicated that intermediate diseases subsequently significantly mediated air pollutant-associated risk to develop more serious disease.
Conclusions:
This study provides evidence that air pollution might play a role in the early stages of CVD progression. Controlling air pollution might be an effective measure to prevent CVD progression and reduce the disease burden of CVD. https://doi.org/10.1289/EHP10967
Supplemental Material is available online (https://doi.org/10.1289/EHP10967).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Cardiovascular diseases (CVDs), including ischemic heart disease (IHD), atrial fibrillation (AF), stroke, and heart failure (HF) and a number of other heart and blood vessel disorders, comprise the leading cause of global mortality and disability, accounting for >18 million deaths worldwide in 2019.1 Hypertension (HTN) is the most important driving factor for CVD.2,3 Previous studies have shown that prevention of high blood pressure could substantially reduce the risk of CVD.4,5 Prehypertension (pre-HTN) is a precursor of clinical HTN and subsequent CVD, but it is often inadequately addressed in the prevention of CVD.6 Although pharmacological treatment is advisable,7 improving the living environment, changes in lifestyle, and other nonpharmacological approaches during pre-HTN might be more effective in preventing disease progression.8
Ambient air pollution has been widely regarded as one important environmental risk factor of HTN and CVD.9,10 Among various air pollutants, particulate matter (PM) pollution is widely reported to be associated with CVD, whereas the evidence for gaseous air pollutants is relatively scarce.11,12 For example, one study of seven northeast cities of China (the China Seven Cities Study) reported that long-term exposure to airborne particulates with an aerodynamic diameter of≤2.5μm (PM2.5) was positively associated with systolic blood pressure (SBP) and HTN.13 A cohort study in the United States reported that exposure to high levels of PM2.5 was associated with cardiovascular mortality, with a hazard ratio (HR) of 1.19 [95% confidence interval (CI): 1.10, 1.28].14 In addition, previous studies also found significant associations of nitrogen dioxide (NO2) and nitrogen oxides (NOx) with cardiovascular morbidity and mortality.15,16 Numerous studies support the hypothesis that long-term air pollution exposure is associated with increased risk of CVD and death.17,18
Most previous studies have focused on only a single trajectory of the disease progression and such segmented analyses have resulted in limited understanding of the comprehensive impacts of ambient air pollution on the dynamic trajectories of CVD progression. For example, most previous studies have focused only on studying a single trajectory, such as the transition risk from healthy status to incidence of CVD, or to death. The analysis considering the role of air pollution in the dynamic progression of CVDs would strengthen our understanding on the etiology of CVD and pave the way to proper prevention and management of HTN and CVD at different disease stages. We hypothesized that exposure to ambient air pollution during the pre-HTN stage could lead to blood vessel damage and thus induce HTN and CVD and, consequently, death. Furthermore, the intermediate diseases might play an important role in mediating the associations between air pollution and CVD progression.
We thus conducted this study with the aims of investigating the associations of ambient air pollution with HTN and subsequent CVD and death among the UK Biobank participants who had pre-HTN. We applied multistate regression models to assess the associations of air pollution with the risk of transitions from pre-HTN to HTN, subsequently to CVD, and further to death. Furthermore, a mediation analysis was applied to explore whether intermediate diseases could mediate the association between air pollution and cardiovascular progression.
Methods
Study Population
This analysis was based on the UK Biobank, a large-scale prospective cohort of more than half a million participants 40–70 years of age. The baseline survey was conducted during 2006–2010, and the participants were recruited from the general population of the United Kingdom. At baseline, a comprehensive set of individual-level data was provided by the participants using touchscreen questionnaires while biological and physical measurements were also collected. Blood pressure was measured twice at enrollment by trained nurses with a digital or manual sphygmomanometer. Two sets of measurements were taken at a 1-min interval. The mean value of the two measurements was used to minimize measurement error. In the present study, pre-HTN was defined as a participant having a baseline SBP in the range of 120–139 mmHg or diastolic blood pressure (DBP) in the range of 80–89 mmHg and who not using blood pressure medications.19
To study the association between air pollution exposure and new-onset CVDs, we used a relatively strict exclusion criterion. We excluded participants who had a self-reported history of CVD or had hospital medical records of CVD before the study baseline [Fields 130000–132605 and Field 6150, International Classification of Diseases, Tenth Revision20 (ICD-10) codes I00–I99] (n=174,203). Participants with a history of blood pressure medication use were further excluded (Field 6153, n=21,900) according to the response to the touchscreen question “Do you regularly take any of the following medications?” Participants with missing data on blood pressure measurement (n=771) or who had withdrawn from the UK Biobank (n=37) were also excluded. According to the definition of pre-HTN, 182,848 participants were identified as having pre-HTN. We additionally excluded participants without available data on air pollution exposure (Fields 24003–24007 and 24016–24019, n=14,838). The remaining 168,010 participants with pre-HTN were included in the present study (Figure 1).
Figure 1. Flow chart of the selection of the participants with pre-HTN from the UK Biobank study. Note: CVD, cardiovascular disease; DBP, diastolic blood pressure; pre-HTN, prehypertension; SBP, systolic blood pressure.
Figure 1 is a flowchart with two steps. Step 1: There are 502,461 baseline cases, excluding 174,203 cases with previous history of C V D, 21,900 cases with blood pressure medication use, 37 cases who withdrew from the study, 771 cases with missing data on blood pressure, and 122,702 cases who did not meet the definition of pre-hypertension leading to pre-H T N, including systolic blood pressure of 120 to 139 millimeters of mercury and or diastolic blood pressure of 80 to 89 millimeters of mercury with 182,848 cases. Step 2: From 182,848 cases, 14,838 were excluded due to missing data on air pollution, leading to 168,010 cases in the study population.
All participants provided written informed consent. UK Biobank was approved by the North West Multi-Centre Research Ethical Committee (REF: 11/NW/03820). This study complied with the Declaration of Helsinki.
Assessment of Air Pollution
Air pollution exposure was estimated for each participant, including PM2.5, PM2.5 absorbance (a measurement that reflects the concentration of elemental or black carbon of PM2.5), PM10 (PM with an aerodynamic diameter of <10μm), NO2, and NOx. Annual concentration levels of NO2 were estimated in 4 consecutive years (2005, 2006, 2007, and 2010), whereas PM10 concentration levels were estimated in 2 years (2007 and 2010) and the other air pollutants were estimated in the year 2010. The annual average air pollution in 2010 was estimated using a land use regression model developed as a part of the European Study of Cohorts for Air Pollution Effects (ESCAPE),21,22 and linked to participants’ geocoded residential addresses at the baseline visit. Air pollution estimates during 2005–2007 were derived from EU-wide air pollution maps (resolution 100×100m), which covered the residential locations of the participants. More detailed information about the model and model performance has been described elsewhere.21,22 In brief, leave-one-out cross-validation showed a good model performance. We averaged the values in different years to represent the long-term air pollution exposures.
Covariates
We developed a directed acyclic graph to identify the potential confounders. A minimally sufficient adjustment set was selected to account for potential confounding in the multivariate models, including age, sex, ethnicity, physical activity, income, and educational attainment (Figure S1). Specifically, age was derived from the birth date and the date of attending an initial assessment center and was included as a time-varying variable. Self-reported ethnicity categories included White, Asian, Black, Mixed, and Other ethnic group. Thereafter, the Asian, Black, Mixed, and Other ethnic group categories were combined into the non-White category because of the small proportions. Physical activity was assessed using the International Physical Activity Questionnaire (IPAQ) and categorized into three levels: low, moderate, and high. Average yearly household income was categorized as <18,000 £, 18,000–30,999 £, 31,000–51,999 £, 52,000–100,000 £, and >100,000 £. Educational attainment was categorized as high (college or university degree, nursing, teaching, and others), intermediate [A (advanced)/AS levels or equivalent, O (ordinary) levels/General Certificate of Secondary Education (GCSE) or equivalent, Certificate of Secondary Education (CSE) or equivalent], low [National Vocational Qualification (NVQ) or Higher National Diploma (HND) or equivalent)], and other (none of the above). Body mass index was calculated as weight divided by height squared (in kilograms per meter squared) and categorized as nonobese (<30.0 kg/m2) or obese (≥30.0 kg/m2). Diabetes at baseline was defined based on HbA1c levels (≥48 mmol/mol), the use of insulin, and diagnosis of diabetes by a doctor.23 Smoking status was categorized as never, previous, or current smoker.
Follow-up and Outcome Ascertainment
Incidence of health outcomes was obtained by the following sources, including self-reports, inpatient, and primary care records. Date and diagnosis of hospital admissions/visits were obtained through linkage to the Hospital Episode Statistics for England (HES), Scottish Morbidity Record (SMR), and Patient Episode Database for Wales (PEDW). Participants’ date of death was obtained from death certificates provided by the National Health Service (NHS) Information Center (England and Wales) and the NHS Central Register Scotland (Scotland). The mortality data were available up to 28 February 2021. The health data were available up to 31 March 2021, 31 March 2021, and 28 February 2018 for England, Scotland, and Wales, respectively. Therefore, follow-up was censored at the date of death, loss to follow-up, or the last date with available health data, whichever occurred first.
In the present analysis, disease progression of pre-HTN could go through the following states: HTN (ICD-10 codes I10–I13, I15), IHD (ICD-10 codes I20–I25), stroke (ICD-10 codes I63, I61, I60), AF (ICD-10 code I48), HF (ICD-10 code I50), and all-cause death. An intermediate disease was the state between two disease states in the dynamic disease transition; for example, CVD is an intermediate state in the trajectory from pre-HTN to death and from HTN to death, and HTN is an intermediate state in the trajectory from pre-HTN to CVD. An intermediate disease may act as an intermediary state leading to a more severe disease.
Statistical Analysis
We reported the characteristics of the included participants with mean±standard deviation for continuous variables and frequency and proportions for categorical variables. Spearman’s correlation coefficients were computed to present the pairwise associations among the air pollutants.
We first used the Cox proportional hazards model to estimate the associations between air pollution and the transitions between different states. Further analysis was performed based on multistate regression models using a clock-forward approach as the timescale (Figure 2). The multistate model consisted of three states (HTN, CVD, and all-cause death), and there were six transitions between these states: a) the transition from pre-HTN to HTN, b) the transition from pre-HTN to CVD, c) the transition from pre-HTN to all-cause death, d) the transition from HTN to CVD, e) the transition from HTN to all-cause death, and f) the transition from CVD to all-cause death. For the participants whose death date was the same as their CVD/HTN diagnosis date, we calculated the diagnosis date as the death date minus 0.5 d. Participants who were diagnosed with HTN and CVD on the same date were directly included in the status CVD. Restricted cubic spline models were used to evaluate potential nonlinear relationships between air pollutants and the progression of CVDs, with 3 knots, at the 10th, 50th, and 90th percentiles. We calculated the transition probability at an extreme exposure level of air pollution for a hypothetical subject with mean values of all covariates. High exposure levels of the air pollutants were set as 4.6×(10–5/m) for PM2.5 absorbance, 20.19 μg/m3 for PM2.5, 30.52 μg/m3 for PM10, 109.388 μg/m3 for NO2, and 265.94 μg/m3 for NOx. The low exposure levels were set as 0.83×(10–5/m) for PM2.5 absorbance, 8.17 μg/m3 for PM2.5, 11.84 μg/m3 for PM10, 8.863 μg/m3 for NO2, and 19.74 μg/m3 for NOx. Sex-stratified analyses were further performed to investigate the difference of disease transition probability between males and females. Paired sample t-tests were used to compare the transition probability at the two different exposure levels.
Figure 2. Trajectories of disease progression from pre-HTN to HTN, CVD, and death in the participants with pre-HTN (N=168,010). Note: CVD, cardiovascular disease; HTN, hypertension; pre-HTN, prehypertension.
Figure 2 is a flowchart with three steps. Step 1: There are 168,010 pre-H T N cases; the disease progression of 13,743 cases (8.18 percent) leads to 13,743 H T N cases; the disease progression of 12,825 cases (7.63 percent) leads to 14,262 C V D cases; and the disease progression of 4,467 cases (2.66 percent) leads to 7,643 death cases. Step 2: From 13,743 H T N cases, disease progression of 1,437 (10.46 percent) leads to 14,262 C V D cases, and disease progression of 804 cases (5.85 percent) leads to 7,643 deaths. Step 3: 14,262 C V D cases with a disease progression of 2,372 cases (16.63 percent) leads to 7,643 deaths.
Considering that intermediate diseases might mediate the association between air pollution and disease progression (Figure S2), we further conducted a mediation analysis to examine whether HTN could mediate the association between air pollution and CVD incidence or death, and whether CVD could mediate the association between air pollution and death.24,25 The direct effect (DE) represents the effect of the exposure on the outcomes that is not explained by the mediator, and the indirect effect (IE) is the effect that is explained by the mediator. It is important to note that the mediation analyses required four important potential confounding assumptions: a) no unmeasured exposure–outcome confounding, b) no unmeasured exposure–mediator confounding, c) no unmeasured mediator–outcome confounding, and d) no mediator–outcome confounder is affected by the exposure.26 This study included major confounders in mediation analysis and we therefore considered these assumptions as reasonable.
Sensitivity Analysis
Several sensitivity analyses were conducted to examine the robustness of our results. a) We treated death as a competing risk and conducted an additional sensitivity analysis using Fine–Gray subdistribution hazards regression models.27 b) We calculated the entering date of the prior status for the participants who experienced death and CVD/HTN event on the same day using different time intervals (0.5, 1, 3, 5 y and the mean interval from CVD/HTN event to all-cause death) and checked the consistency of the findings. c) The associations between air pollution and each specific CVD in the dynamic cardiovascular trajectory were also estimated separately in the multistate model. d) To mitigate multiple testing issues for various air pollutants, a Bonferroni correction was used to correct for multiple comparisons. e) Because residential mobility might lead to exposure misclassification, we further conducted multistate analyses among the participants who did not change their residential addresses during the follow-up.28 f) To identify potentially susceptible populations, we conducted stratified analyses by smoking status, diabetes, and obesity to examine their potential effect modifications. Wald chi-square tests were calculated for the interaction terms to determine whether the results between subgroups were statistically different. g) Because our analysis might be subject to potential collider bias because of the participant selection, we further performed a multistate analysis in participants without existing CVD (Figure S3) and compared the results with those of participants with pre-HTN. h) Considering that participants with existing HTN and CVD at baseline can also contribute to the HTN→CVD, HTN→death, and CVD→death transitions, we further included these participants in the Cox regression models.
Multiple imputation was performed to impute missing covariate values. Analyses were estimated on each of the imputed data sets, and the results were then combined using the Rubin rule.29,30 The results were presented as HRs and 95% CIs per interquartile range (IQR) increase in air pollutant. All statistical tests were two-sided, and p<0.05 were considered as statistically significant. The multistate models were performed using the mstate package of R (version 4.1.3; R Development Core Team).
Results
Descriptive Results
The characteristics of the study participants are presented in Table 1. During a median follow-up of 12 y, 13,743 (8.18%) of participants with pre-HTN developed HTN, and 12,825 (7.63%) and 4,467 (2.66%) directly developed CVD and died, respectively. In addition, low economic and educational attainment and low physical activity levels were significantly associated with disease progression in the three baseline transitions. The characteristics of included and excluded participants with pre-HTN were similar (Table S1).
Table 1 Baseline characteristics of the 168,101 participants from the UK Biobank study, grouped by pre-HTN without disease progression and pre-HTN to HTN, CVD, and death [mean±SD or n (%)].
Characteristics Pre-HTN without disease progression
(n=136,975) Pre-HTN to HTN
(n=13,743) Pre-HTN to CVD
(n=12,825) Pre-HTN to death
(n=4,467) p-Value
Age at recruitment (y) 54.20±7.96 58.26±7.72 59.15±7.19 59.54±7.14 <0.001
Sex <0.001
Male 60,734 (44.34) 6,117 (44.51) 7,805 (60.86) 2,193 (49.09)
Female 76,241 (55.66) 7,626 (55.49) 5,020 (39.14) 2,274 (50.91)
Ethnicity <0.001
White 128,973 (94.16) 12,650 (92.05) 12,251 (95.52) 4,322 (96.75)
Asian 3,055 (2.23) 441 (3.21) 284 (2.21) 50 (1.12)
Black 2,084 (1.52) 291 (2.12) 80 (0.62) 30 (0.67)
Mixed 897 (0.65) 87 (0.63) 63 (0.49) 22 (0.49)
Others 1,304 (0.95) 163 (1.19) 83 (0.65) 17 (0.38)
Prefer not to answer 662 (0.48) 111 (0.81) 64 (0.50) 26 (0.58)
Physical activity <0.001
Low 19,703 (14.38) 2,139 (15.56) 1,889 (14.73) 695 (15.56)
Moderate 45,099 (32.92) 4,327 (31.49) 3,954 (30.83) 1,438 (32.19)
High 47,321 (34.55) 4,125 (30.02) 4,461 (34.78) 1,362 (30.49)
Missing 24,852 (18.14) 3,152 (22.94) 2,521 (19.66) 972 (21.76)
Educational attainment <0.001
High 69,206 (50.52) 5,688 (41.39) 5,701 (44.45) 1,919 (42.96)
Intermediate 42,610 (31.11) 3,864 (28.12) 3,393 (26.46) 1,210 (27.09)
Low 7,084 (5.17) 880 (6.40) 883 (6.88) 299 (6.69)
Other 15,795 (11.53) 2,876 (20.93) 2,522 (19.66) 936 (20.95)
Missing 2,280 (1.66) 435 (3.17) 326 (2.54) 103 (2.31)
Yearly income <0.001
>100,000 £ 7,978 (5.82) 413 (3.01) 466 (3.63) 139 (3.11)
52,000–100,000 £ 29,368 (21.44) 1,796 (13.07) 1,929 (15.04) 524 (11.73)
31,000–51,999 £ 33,685 (24.59) 2,688 (19.56) 2,617 (20.41) 863 (19.32)
18,000–30,999 £ 27,592 (20.14) 3,192 (23.23) 2,947 (22.98) 1,066 (23.86)
<18,000 £ 19,580 (14.29) 3,149 (22.91) 2,735 (21.33) 1,109 (24.83)
Missing 18,772 (13.70) 2505 (18.23) 2,131 (16.62) 766 (17.15)
Smoking status <0.001
Past or current smoker 55,206 (40.30) 6,782 (49.35) 6,627 (51.67) 2,418 (54.13)
Never-smoked 81,192 (59.28) 6,824 (49.65) 6,132 (47.81) 2,023 (45.29)
Missing 577 (0.42) 137 (1.00) 66 (0.51) 26 (0.58)
BMI <0.001
Obese 9,545 (6.97) 1,899 (13.82) 1,178 (9.19) 348 (7.79)
Nonobese 127,017 (92.73) 11,771 (85.65) 11,575 (90.25) 4,092 (91.61)
Missing 413 (0.30) 73 (0.53) 72 (0.56) 27 (0.60)
Diabetes <0.001
Yes 2,786 (2.03) 1,148 (8.35) 611 (4.76) 157 (3.51)
No 134,189 (97.97) 12,595 (91.65) 12,214 (95.24) 4,310 (96.49)
Note: The chi-square test was used for categorical variables and ANOVA for continuous variables. ANOVA, analysis of variance; BMI, body mass index; CVD, cardiovascular disease; HTN, hypertension; pre-HTN, prehypertension.
Participants who developed HTN had significantly higher exposure to air pollution compared with those without disease progression (Table 2). Baseline concentrations of air pollutants were below the current air quality limits in Europe but substantially higher than the latest World Health Organization recommendations. High correlations were observed among various air pollutants (Table S2). For example, PM2.5 levels were significantly and positively correlated with NOx levels (r=0.88).
Table 2 Summary statistics of air pollutant concentrations of the participants grouped by pre-HTN without disease progression and from pre-HTN to HTN, CVD, and death.
Statistics Pre-HTN without disease progression
(n=136,975) Pre-HTN to HTN
(n=13,743) Pre-HTN to CVD
(n=12,825) Pre-HTN to death
(n=4,467) p-Value
PM2.5 absorbance (10–5/m) <0.001
Mean±SD 1.18±0.27 1.20±0.28 1.18±0.27 1.18±0.26
Max 4.60 3.46 3.65 3.09
P75 1.30 1.32 1.29 1.30
P25 0.99 1.01 0.99 1.00
Min 0.83 0.83 0.83 0.83
PM2.5 (μg/m3) <0.001
Mean±SD 9.95±1.05 10.06±1.07 9.97±1.05 10.01±1.08
Max 19.69 18.61 19.56 20.19
P75 10.53 10.63 10.55 10.59
P25 9.25 9.35 9.27 9.29
Min 8.17 8.17 8.17 8.17
PM10 (μg/m3) <0.001
Mean±SD 19.28±1.97 19.41±1.95 19.23±1.89 19.25±1.88
Max 30.52 28.69 29.25 26.86
P75 20.38 20.54 20.28 20.32
P25 18.02 18.15 18.04 18.09
Min 11.84 13.34 13.00 13.46
NO2 (μg/m3) <0.001
Mean±SD 29.06±9.29 29.93±9.40 29.00±8.92 29.17±9.09
Max 109.39 101.76 107.81 81.26
P75 33.77 34.59 33.62 33.77
P25 22.65 23.52 22.88 22.94
Min 8.863 9.185 11.29 10.97
NOx (μg/m3) <0.001
Mean±SD 43.54±15.46 44.93±16.32 43.68±15.68 44.21±15.66
Max 263.96 265.94 255.33 174.82
P75 50.38 51.59 50.39 51.12
P25 33.70 34.91 33.96 34.24
Min 19.74 19.74 19.74 19.74
Note: Analysis was performed using ANOVA for continuous variables. ANOVA, analysis of variance; CVD, cardiovascular disease; HTN, hypertension; Max, maximum; Min, minimum; NO2, nitrogen dioxide; NOx, nitrogen oxides; P25, the 25th percentile; P75, the 75th percentile; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM10, particulate matter with aerodynamic diameter ≤10μm; pre-HTN, prehypertension; SD, standard deviation.
Associations between Air Pollution and Dynamic CVD Progression Using Cox Regressions
Associations between air pollution and disease transition were almost linear in multivariate Cox regression models (Figure S4). Exposure to higher levels of air pollution were significantly associated with six transitions of CVD progression (Table 3). Higher levels of exposure to PM2.5 absorbance, PM2.5, PM10, NO2, and NOx levels were positively associated with the risk of transitioning from pre-HTN to HTN, CVD, and death, whereas PM10 exposure was significantly associated with transitioning from pre-HTN to HTN and death. For example, a per-IQR increase of PM2.5 absorbance was associated with the largest increase in incident HTN (HR=1.075; 95% CI: 1.056, 1.094). PM2.5 exposure was positively associated with transitioning from pre-HTN to HTN (HR=1.100; 95% CI: 1.079, 1.120), to CVD (HR=1.044; 95% CI: 1.023, 1.066), and to death (HR=1.082; 95% CI: 1.053, 1.113). PM2.5 absorbance, PM2.5, PM10, NO2 and NOx exposures were significantly associated with an increased risk of death among participants who had CVD; the HRs were 1.051 (95% CI: 1.003, 1.101), 1.059 (95% CI: 1.008, 1.113), 1.080 (95% CI: 1.026, 1.136), 1.093 (95% CI: 1.038, 1.150), 1.077 (95% CI: 1.034, 1.122), respectively.
Table 3 Hazard ratio (95% CI) for per-IQR increase of air pollution associated with HTN and subsequent CVD and death by using the Cox model (N=168,010).
Model PM2.5 absorbance PM2.5 PM10 NO2 NOx
Model 1a
Pre-HTN→HTN 1.108 (1.090, 1.127) 1.157 (1.136, 1.178) 1.123 (1.102, 1.143) 1.144 (1.124, 1.164) 1.120 (1.104, 1.137)
Pre-HTN→CVDs 1.030 (1.011, 1.049) 1.073 (1.052, 1.094) 1.026 (1.006, 1.047) 1.041 (1.021, 1.061) 1.055 (1.037, 1.073)
Pre-HTN→Death 1.070 (1.044, 1.097) 1.128 (1.099, 1.158) 1.075 (1.046, 1.105) 1.102 (1.074, 1.132) 1.108 (1.085, 1.133)
HTN→CVDs 1.010 (0.952, 1.071) 1.011 (0.950, 1.075) 0.996 (0.935, 1.062) 1.006 (0.945, 1.071) 1.022 (0.969, 1.077)
HTN→Death 1.075 (1.008, 1.146) 1.017 (0.948, 1.092) 1.066 (0.993, 1.145) 1.087 (1.015, 1.164) 1.044 (0.985, 1.107)
CVDs→Death 1.070 (1.023, 1.120) 1.106 (1.054, 1.161) 1.095 (1.042, 1.151) 1.115 (1.062, 1.170) 1.110 (1.068, 1.154)
Model 2b
Pre-HTN→HTN 1.075 (1.056, 1.094) 1.100 (1.079, 1.120) 1.089 (1.068, 1.110) 1.105 (1.085, 1.126) 1.077 (1.060, 1.094)
Pre-HTN→CVDs 1.020 (1.000, 1.040) 1.044 (1.023, 1.066) 1.019 (0.999, 1.041) 1.030 (1.009, 1.052) 1.034 (1.016, 1.053)
Pre-HTN→Death 1.060 (1.032, 1.088) 1.082 (1.053, 1.113) 1.072 (1.042, 1.103) 1.094 (1.064, 1.126) 1.079 (1.054, 1.104)
HTN→CVDs 1.006 (0.946, 1.069) 0.979 (0.918, 1.044) 0.996 (0.932, 1.065) 0.999 (0.935, 1.067) 1.000 (0.946, 1.057)
HTN→Death 1.072 (1.004, 1.145) 0.990 (0.921, 1.065) 1.071 (0.994, 1.153) 1.087 (1.012, 1.169) 1.027 (0.966, 1.091)
CVDs→Death 1.051 (1.003, 1.101) 1.059 (1.008, 1.113) 1.080 (1.026, 1.136) 1.093 (1.038, 1.150) 1.077 (1.034, 1.122)
Note: IQR increments are 0.31 (10–5/m) for PM2.5 absorbance, 1.28 μg/m3 for PM2.5, 2.35 μg/m3 for PM10, 11.09 μg/m3 for NO2, and 16.69 μg/m3 for NOx. CI, confidence interval; CVD, cardiovascular disease; HTN, hypertension; IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM10, particulate matter with aerodynamic diameter ≤10μm; pre-HTN, prehypertension.
a Model 1: analysis adjusted for age, sex.
b Model 2: further adjusted for ethnicity, physical activity, income, and educational attainment.
Associations between Air Pollution and Dynamic CVD Progression Using Multistate Models
Multistate regression models showed that air pollutants were associated with an increased risk of disease progression (Figure 3). Estimates of the association between air pollution and each disease transition are illustrated in Table 4. All the air pollutants were associated with increased risk of HTN in the participants with pre-HTN, and the associations remained significant after adjusting for sex, age, ethnicity, physical activity, income, and educational attainment. The associations of NO2 and NOx with disease progression were stronger than those of PM (PM2.5, PM2.5 absorbance, and PM10). Specifically, per-IQR increases of PM2.5, PM2.5 absorbance, and PM10 exposure were significantly associated with the risk of transitioning from pre-HTN to HTN; the HRs were 1.105 (95% CI: 1.083, 1.127), 1.084 (95% CI: 1.064, 1.104), and 1.097 (95% CI: 1.075, 1.120), respectively. The HRs of transitioning from pre-HTN to HTN for NO2 and NOx exposure were 1.114 (95% CI: 1.091, 1.137) and 1.079 (95% CI: 1.061, 1.098), respectively. PM2.5, NO2, and NOx exposure were associated with the risk of incident CVD among participants with pre-HTN; the HRs were 1.045 (95% CI: 1.022, 1.068), 1.028 (95% CI: 1.005, 1.050), and 1.033 (95% CI: 1.014, 1.053), respectively. In addition, PM2.5 absorbance, PM2.5, and PM10 exposure were also associated with the risk of all-cause death from pre-HTN, with HRs of 1.043 (95% CI: 1.008, 1.079), 1.086 (95% CI: 1.047, 1.126), and 1.049 (95% CI: 1.011, 1.088), respectively. PM2.5 absorbance, PM2.5, PM10, NO2 and NOx exposure were associated with increased risk of transitioning from CVD to death; the HRs were 1.071 (95% CI: 1.022, 1.121), 1.082 (95% CI: 1.029, 1.137), 1.100 (95% CI: 1.045, 1.158), 1.119 (95% CI: 1.063, 1.177), and 1.099 (95% CI: 1.055, 1.144), respectively.
Figure 3. Associations between air pollution and different statuses in six transitions in the participants with pre-HTN by using a multistate model (N=168,010). The hazard ratio is represented by a bold line; the 95% confidence interval is represented by the shaded area. The corresponding numerical data are listed in Table S3. Note: CVD, cardiovascular disease; HTN, hypertension; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM10, particulate matter with aerodynamic diameter ≤10μm; pre-HTN, prehypertension.
Figure 3 is a set of thirty ribbon and line graphs. The first set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance (10 begin superscript negative 5 end superscript per meter), particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed), particulate matter begin subscript 10 end subscript (micrograms per meter cubed), nitrogen dioxide (micrograms per meter cubed), and nitrogen oxides (micrograms per meter cubed), plotting hazard ratio, ranging from 0.9 to 1.2 in increments of 0.1 (left y-axis) and pre-H T N to H T N (right y-axis) across air pollution, ranging from 1.00 to 1.75 in increments of 0.25, 9 to 11 in unit increments, 16 to 22 in increments of 2; 20 to 40 in increments of 10, and 30 to 70 in increments of 10 (x-axis), respectively. The second set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance (10 begin superscript negative 5 end superscript per meter), particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed), particulate matter begin subscript 10 end subscript (micrograms per meter cubed), nitrogen dioxide (micrograms per meter cubed), and nitrogen oxides (micrograms per meter cubed), plotting hazard ratio, ranging from 0.90 to 1.10 in increments of 0.05 (left y-axis) and pre-H T N to C V Ds (right y-axis) across air pollution, ranging from 1.00 to 1.75 in increments of 0.25, 9 to 11 in unit increments, 16 to 22 in increments of 2; 20 to 40 in increments of 10, and 30 to 70 in increments of 10 (x-axis), respectively. The third set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance (10 begin superscript negative 5 end superscript per meter), particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed), particulate matter begin subscript 10 end subscript (micrograms per meter cubed), nitrogen dioxide (micrograms per meter cubed), and nitrogen oxides (micrograms per meter cubed), plotting hazard ratio, ranging from 0.9 to 1.2 in increments of 0.1 (left y-axis) and pre-H T N to death (right y-axis) across air pollution, ranging from 1.00 to 1.75 in increments of 0.25, 9 to 11 in unit increments, 16 to 22 in increments of 2; 20 to 40 in increments of 10, and 30 to 70 in increments of 10 (x-axis), respectively. The fourth set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance (10 begin superscript negative 5 end superscript per meter), particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed), particulate matter begin subscript 10 end subscript (micrograms per meter cubed), nitrogen dioxide (micrograms per meter cubed), and nitrogen oxides (micrograms per meter cubed), plotting hazard ratio, ranging from 0.9 to 1.2 in increments of 0.1 (left y-axis) and H T N to C V D (right y-axis) across air pollution, ranging from 1.00 to 1.75 in increments of 0.25, 9 to 11 in unit increments, 16 to 22 in increments of 2; 20 to 40 in increments of 10, and 30 to 70 in increments of 10 (x-axis), respectively. The fifth set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance (10 begin superscript negative 5 end superscript per meter), particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed), particulate matter begin subscript 10 end subscript (micrograms per meter cubed), nitrogen dioxide (micrograms per meter cubed), and nitrogen oxides (micrograms per meter cubed), plotting hazard ratio, ranging from 0.8 to 1.2 in increments of 0.2 (left y-axis) and H T N to death (right y-axis) across air pollution, ranging from 1.00 to 1.75 in increments of 0.25, 9 to 11 in unit increments, 16 to 22 in increments of 2; 20 to 40 in increments of 10, and 30 to 70 in increments of 10 (x-axis), respectively. The sixth set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance (10 begin superscript negative 5 end superscript per meter), particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed), particulate matter begin subscript 10 end subscript (micrograms per meter cubed), nitrogen dioxide (micrograms per meter cubed), and nitrogen oxides (micrograms per meter cubed), plotting hazard ratio, ranging from 0.8 to 1.3 in increments of 0.1 (left y-axis) and C V D to death (right y-axis) across air pollution, ranging from 1.00 to 1.75 in increments of 0.25, 9 to 11 in unit increments, 16 to 22 in increments of 2; 20 to 40 in increments of 10, and 30 to 70 in increments of 10 (x-axis), respectively.
Table 4 Hazard ratio (95% CI) for per-IQR increase of air pollution associated with HTN and subsequent CVD and death by using a multistate model (N=168,010).
Model PM2.5 absorbance PM2.5 PM10 NO2 NOx
Model 1a
Pre-HTN→HTN 1.120 (1.100, 1.140) 1.163 (1.141, 1.186) 1.135 (1.112, 1.158) 1.156 (1.134, 1.178) 1.124 (1.107, 1.142)
Pre-HTN→CVDs 1.025 (1.005, 1.046) 1.069 (1.047, 1.092) 1.022 (1.001, 1.043) 1.036 (1.014, 1.057) 1.051 (1.032, 1.070)
Pre-HTN→Death 1.046 (1.011, 1.081) 1.120 (1.082, 1.160) 1.044 (1.008, 1.082) 1.070 (1.033, 1.108) 1.091 (1.060, 1.124)
HTN→CVDs 1.009 (0.951, 1.070) 1.014 (0.953, 1.079) 0.992 (0.931, 1.057) 1.006 (0.945, 1.070) 1.024 (0.971, 1.079)
HTN→Death 1.067 (0.989, 1.151) 1.001 (0.922, 1.088) 1.056 (0.971, 1.149) 1.084 (1.000, 1.175) 1.025 (0.956, 1.100)
CVDs→Death 1.089 (1.041, 1.139) 1.123 (1.070, 1.178) 1.117 (1.063, 1.174) 1.141 (1.086, 1.198) 1.128 (1.085, 1.173)
Model 2b
Pre-HTN→HTN 1.084 (1.064, 1.104) 1.105 (1.083, 1.127) 1.097 (1.075, 1.120) 1.114 (1.091, 1.137) 1.079 (1.061, 1.098)
Pre-HTN→CVDs 1.017 (0.996, 1.038) 1.045 (1.022, 1.068) 1.017 (0.995, 1.039) 1.028 (1.005, 1.050) 1.033 (1.014, 1.053)
Pre-HTN→Death 1.043 (1.008, 1.079) 1.086 (1.047, 1.126) 1.049 (1.011, 1.088) 1.071 (1.033, 1.111) 1.070 (1.038, 1.104)
HTN→CVDs 1.007 (0.947, 1.070) 0.984 (0.923, 1.050) 0.994 (0.930, 1.062) 1.001 (0.937, 1.069) 1.004 (0.950, 1.061)
HTN→Death 1.072 (0.992, 1.158) 0.993 (0.912, 1.081) 1.068 (0.979, 1.165) 1.095 (1.006, 1.191) 1.021 (0.950, 1.098)
CVDs→Death 1.071 (1.022, 1.121) 1.082 (1.029, 1.137) 1.100 (1.045, 1.158) 1.119 (1.063, 1.177) 1.099 (1.055, 1.144)
Note: IQR increments are 0.31 (10–5/m) for PM2.5 absorbance, 1.28 μg/m3 for PM2.5, 2.35 μg/m3 for PM10, 11.09 μg/m3 for NO2, and 16.69 μg/m3 for NOx. CI, confidence interval; CVD, cardiovascular disease; HTN, hypertension; IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM10, particulate matter with aerodynamic diameter ≤10μm; pre-HTN, prehypertension.
a Model 1: analysis adjusted for age, sex.
b Model 2: further adjusted for ethnicity, physical activity, income, and educational attainment.
Transition Probability in Different Trajectories Using Multistate Models
The estimated probabilities of transitioning from one state to another at high or low levels of air pollution exposure are shown in Figure 4. Exposure to higher levels of air pollution were associated with a higher transition probability to a more serious state, except for the transition from HTN to CVD. Specific transition probabilities in year 1, 5, and 15 are shown in Table S4. For example, the probability of transitioning from pre-HTN to HTN at high levels of PM2.5 was 3.03% in the fifth year, whereas it was 1.26% for low levels of PM2.5. In addition, the transition probability from CVD to death was the highest in the first 5 y, but it leveled off thereafter.
Figure 4. Transition probabilities over time from pre-HTN to HTN, CVD, and death, from HTN to CVD and death, and from CVD to death in a high or low level of air pollution by using a multistate model (N=168,010). The corresponding numerical data are listed in Table S4. High exposure was 4.6×10–5/m for PM2.5 absorbance, 20.19 μg/m3 for PM2.5, 30.52 μg/m3 for PM10, 109.388 μg/m3 for NO2, and 265.94 μg/m3 for NOx. Low exposure was 0.83×10–5/m for PM2.5 absorbance, 8.17 μg/m3 for PM2.5, 11.84 μg/m3 for PM10, 8.863 μg/m3 for NO2, and 19.74 μg/m3 for NOx. Note: CVD, cardiovascular disease; HTN, hypertension; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM10, particulate matter with aerodynamic diameter ≤10μm; pre-HTN, prehypertension.
Figure 4 is set of thirty clustered bar graphs. The first set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen dioxide, and nitrogen oxides, plotting transition probability in percentages, ranging from 0 to 15 in increments of 5 (left y-axis) and pre-H T N to H T N (right y-axis) across years since recruitment, ranging from 1 to 15 in increments of 2 (x-axis) for group, including low exposure and high exposure, respectively. The second set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen dioxide, and nitrogen oxides, plotting transition probability in percentages, ranging from 0.0 to 10.0 in increments of 2.5 (left y-axis) and pre-H T N to C V Ds (right y-axis) across years since recruitment, ranging from 1 to 15 in increments of 2 (x-axis) for group, including low exposure and high exposure, respectively. The third set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen dioxide, and nitrogen oxides, plotting transition probability in percentages, ranging from 0 to 6 in increments of 2 (left y-axis) and pre-H T N to death (right y-axis) across years since recruitment, ranging from 1 to 15 in increments of 2 (x-axis) for group, including low exposure and high exposure, respectively. The fourth set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen dioxide, and nitrogen oxides, plotting transition probability in percentages, ranging from 0 to 15 in increments of 5 (left y-axis) and H T N to C V D (right y-axis) across years since recruitment, ranging from 1 to 15 in increments of 2 (x-axis) for group, including low exposure and high exposure, respectively. The fifth set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen dioxide, and nitrogen oxides, plotting transition probability in percentages, ranging from 0 to 20 in increments of 5 (left y-axis) and H T N to death (right y-axis) across years since recruitment, ranging from 1 to 15 in increments of 2 (x-axis) for group, including low exposure and high exposure, respectively. The sixth set of five graphs titled particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen dioxide, and nitrogen oxides, plotting transition probability in percentages, ranging from 0 to 60 in increments of 20 (left y-axis) and C V Ds to death (right y-axis) across years since recruitment, ranging from 1 to 15 in increments of 2 (x-axis) for group, including low exposure and high exposure, respectively.
In the early stages of CVD progression, a higher transition probability was observed in males than in females (Figure S5 and Table S5). In the early three trajectories starting from pre-HTN, the transition probabilities consistently continued to rise with time, whereas in the subsequent trajectories (from HTN to CVD, from HTN to death, and from CVD to death), the transition probability increased significantly and approached being flat. As CVDs progress, a higher transition probability was observed in females than in males in the transition from CVD to death. For example, the transition probability from CVD to death in males and females at exposures to high levels of PM2.5 was 25.05% vs. 36.45% in the fifth year.
Mediation Analyses of Intermediate Diseases
Potential mediation effects of intermediate diseases on the association between air pollution and a more serious disease status were evaluated by a mediation analysis (Table 5). We observed that the association of PM2.5 on incident CVD was significantly mediated by HTN, and the IE effect was 1.017 (95% CI: 1.010, 1.024), with a proportion of IE of 31.23% (95% CI: 19.02%, 44.81%). Similarly, HTN significantly mediated the association between PM10 and CVD, with IE=1.014 (95% CI: 1.008, 1.020), and the corresponding proportion of IE=45.16% (95% CI: 26.85%, 66.18%). HTN and CVD significantly mediated 17.81%–46.26% of the associations between all the five air pollutants and death. For example, the IE for CVD and HTN was 1.065 (95% CI: 1.038, 1.094), 1.038 (95% CI: 1.028, 1.048) for PM2.5 and the proportions of IE were 46.26% (95% CI: 26.93%, 65.68%) and 32.41% (95% CI: 24.58%, 41.00%).
Table 5 Mediation effect [hazard ratio (95% CI)] for per-IQR increase of air pollution associated with an intermediate disease status on disease progression (N=168,010).
Air pollutant Air pollution→HTN→CVD Air pollution→HTN→death Air pollution→CVD→death
DE IE TE Proportion Mediated (%) DE IE TE Proportion Mediated (%) DE IE TE Proportion Mediated (%)
PM2.5 absorbance 1.021 (0.999, 1.042) 1.012 (1.007, 1.018) 1.033 (1.011, 1.056) 37.76 (22.36, 55.50) 1.058 (1.032, 1.086) 1.025 (1.018, 1.034) 1.085 (1.057, 1.115) 30.53 (21.45, 40.50) 1.058 (1.031, 1.086) 1.020 (0.995, 1.046) 1.079 (1.041, 1.119) 25.95 (−7.11, 59.09)
PM2.5 1.038 (1.015, 1.061) 1.017 (1.010, 1.024) 1.055 (1.031, 1.080) 31.23 (19.02, 44.81) 1.080 (1.051, 1.110) 1.038 (1.028, 1.048) 1.120 (1.089, 1.153) 32.41 (24.58, 41.00) 1.076 (1.047, 1.106) 1.065 (1.038, 1.094) 1.147 (1.104, 1.191) 46.26 (26.93, 65.68)
PM10 1.017 (0.994, 1.040) 1.014 (1.008, 1.020) 1.031 (1.007, 1.055) 45.16 (26.85, 66.18) 1.071 (1.041, 1.101) 1.028 (1.020, 1.038) 1.101 (1.069, 1.134) 29.14 (20.71, 38.41) 1.072 (1.042, 1.102) 1.015 (0.989, 1.042) 1.088 (1.047, 1.131) 17.81 (−13.56, 49.24)
NO2 1.027 (1.004, 1.050) 1.017 (1.010, 1.025) 1.044 (1.020, 1.069) 39.14 (23.91, 56.10) 1.092 (1.062, 1.123) 1.036 (1.027, 1.046) 1.131 (1.099, 1.164) 28.56 (21.48, 36.34) 1.092 (1.062, 1.123) 1.035 (1.008, 1.063) 1.131 (1.088, 1.175) 28.13 (6.50, 49.84)
NOx 1.029 (1.009, 1.049) 1.013 (1.008, 1.018) 1.042 (1.021, 1.063) 30.79 (18.53, 44.68) 1.078 (1.053, 1.103) 1.028 (1.021, 1.036) 1.108 (1.082, 1.135) 27.06 (20.06, 34.75) 1.076 (1.051, 1.101) 1.046 (1.023, 1.071) 1.125 (1.090, 1.163) 38.37 (18.95, 57.87)
Note: Models adjusted for age, sex, ethnicity, physical activity, income, and educational attainment. IQR increments are 0.31 (10–5/m) for PM2.5 absorbance, 1.28 μg/m3 for PM2.5, 2.35 μg/m3 for PM10, 11.09 μg/m3 for NO2, and 16.69 μg/m3 for NOx. CI, confidence interval; CVD, cardiovascular disease; DE, direct effect; HTN, hypertension; HR, hazard ratio; IE, indirect effect; IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM10, particulate matter with aerodynamic diameter ≤10μm; pre-HTN, prehypertension; TE, total effect.
Sensitivity Analyses
The associations between air pollution and CVD progression in participants with pre-HTN remained generally robust in multiple sensitivity analyses. Specifically, a) the associations remained stable in the transition from pre-HTN to HTN, from pre-HTN to CVD, and from HTN to CVD when treating all-cause death as a competing risk event (Table S6). b) We chose different time intervals for participants who experienced death and a CVD/HTN event on the same day. The results remained stable at different time intervals (Table S7). c) The associations between air pollution and four specific CVD (AF, stroke, IHD, and HF) during the disease progression are shown in Table S8. Significant positive associations were consistently observed for the five air pollutants from pre-HTN to incidence of specific CVD. The associations between air pollution and transition from pre-HTN to AF were slightly stronger than those to other diseases. d) Using a Bonferroni correction, the associations of air pollution with progression of CVDs remained significant (Table S9). e) In participants who did not change residences during follow-up, we found that the direction and magnitude of estimates from multistate models were consistent (Table S10). f) In the subgroup analyses, interaction terms between air pollution and each of the variables were generally not significant (Tables S11–S13). In previous or current smokers, those who experienced obesity, and those who had diabetes, the associations between air pollution and CVD progression were stronger in all disease transitions except for from HTN to CVD. g) In addition, the associations between air pollution and the dynamic disease progression were generally consistent in the participants with pre-HTN and the general participants (Figure S6). For example, a per-IQR increase of PM2.5 was significantly associated with the risk of transitioning from pre-HTN to HTN (HR=1.105; 95% CI: 1.083, 1.127) in participants with pre-HTN, and the HR was 1.117 (95% CI: 1.088, 1.147) in general participants. h) When participants with preexisting CVD were included in the Cox regression analyses, we found stronger associations in these disease transitions compared with those in participants without preexisting CVD (Table S14).
Discussion
Based on the baseline assessment of ambient air pollution exposure, we recruited participants with pre-HTN as the starting point, and followed up their disease progression to evaluate the association between air pollution and disease transition using multistate models. We found that five major air pollutants were positively associated with subsequent HTN, CVD, and death among the participants with pre-HTN. We also observed stronger associations of air pollution with the transition from pre-HTN to HTN, CVD, and death, whereas the associations of air pollution with the disease progression from hypertensive status (i.e., HTN) to CVD and death were not significant. Furthermore, our analysis indicated that HTN might mediate the associations of air pollution with the disease transition from pre-HTN to CVD and death.
Numerous studies have assessed the role of air pollution on morbidity and mortality of HTN and CVD.14,31 Less attention, however, has been paid to pre-HTN, the subclinical status of HTN that places great public health significance on the prevention of CVD. Pre-HTN is an early stage of HTN that can be reversed by effective early intervention. Thus, estimating the association between air pollution and pathophysiological changes during the subclinical period has significant meaning to the prevention and management of CVD. Several studies have reported air pollution might first influence cardiac conduction systems and, subsequently, increase the risk of CVD. According to the analysis of our study, exposure to air pollution is associated with the risk of disease transition from pre-HTN to HTN. This might suggest that the strong positive association between air pollution and CVD might be attributable to the sensitive window of pre-HTN to a great extent. What is more, individuals with pre-HTN might be a vulnerable population and deserve more attention.32
Compared with previous studies, this study applied a multistate model, which provided a more extensive view of the association between air pollution exposure and CVD progression. The multistate model could adequately describe the transitions from one state to a more serious state in disease progression.33 A major advantage of the multistate model is that all the time-to-event outcomes during the dynamic disease progression can be simultaneously considered in one model rather than using separate models. Separate and segmented analyses of a disease course can reduce the power of the estimated associations or result in false-negative findings.34 Moreover, the occurrence of intermediate events can be taken into consideration in multistate models, thus providing a useful insight into their relationship with a subsequent end point, usually death.35 Previous studies reported that Cox models might underestimate HRs compared with those in multistate models.36 We also observed this trend in some transitions of our study. For example, the association between air pollution and transition from pre-HTN to HTN were consistently higher using multistate models compared with those using Cox models. In general, the results of separate Cox models showed much consistency with those of multistate models. However, investigators tend to use the multistate model in interval-censored illness-death-type data analyses given that it can provide a more comprehensive view of the exposure impact on the progression of dynamic diseases.37,38
Our study provided strong evidence that ambient air pollution, even below the current European standard, was associated with an increased risk of CVD progression. The observed associations between air pollutants and CVD in our study were consistent with those reported from other studies using UK Biobank data but with a different magnitude of associations. For example, for the incident CVD outcomes, we estimated an HR of 1.044 (95% CI: 1.023, 1.066) an IQR increase (1.3 μg/m3) in PM2.5, whereas Cai et al. reported an HR of 1.068 (95% CI: 1.026, 1.112),39 which was larger than the estimate in this study. Wang et al. reported that ambient PM2.5 was associated with increased risk of all-cause mortality (HR=1.032; 95% CI: 1.006, 1.059) for an IQR increase in PM2.5,40 whereas the HR in our study was 1.082 (95% CI: 1.053, 1.113). The difference might be attributable to the strict inclusion/exclusion criterion and different lengths of follow-up. The present study focused on studying the new-onset cardiovascular events in our trajectory; thus, we used a stricter exclusion criterion that accounted for CVD history, blood pressure medication use, and blood pressure measurements. In addition, varying follow-up lengths might be another reason for the heterogeneity in the effect estimates.
The present study showed that the associations between air pollution and CVD could act directly on each disease trajectory and promote the disease progression through intermediate diseases statuses. The findings were in line with the current understanding that individuals with existing CVDs were more vulnerable to the types of CVD induced by air pollution.41 Furthermore, the intermediate disease status caused by air pollution could further mediate CVD progression.11 Although the underlying biological mechanisms are not completely understood, different potential mechanisms have been proposed that inhaled air pollution can directly affect the cardiovascular system, blood, and lung receptors42 but also indirectly lead to pulmonary oxidative stress and inflammatory responses.43 These changes would further play an important pathophysiological role in the transition from HTN to CVD and, further, to death.44 Although a prehypertensive population with early hemodynamic changes, accompanied by both slightly elevated cardiac output and systemic resistance, might be more sensitive to air pollution exposure.45
Our study has several strengths. Comprehensive studies on the role of air pollution in the dynamic progression of CVDs are still limited. Our multistate regression results and mediation analyses first provided a further understanding about the association between air pollution and dynamic progression of CVDs. In addition, as chronic conditions, CVD develop through the influence of multiple risk factors over a long period. This study had a long follow-up time that enabled us to observe significant associations between air pollution and CVD progression.
Our study has several limitations. First, exposure measurement error may have biased our estimates given that the UK Biobank did not provide air pollution concentrations for each year. However, according to the UK official statistics (https://www.gov.uk/government/statistics) from government departments, agencies, and commercial organizations, the trend of yearly average concentration levels for major air pollutants remained stable during the observation period. Furthermore, the residential addresses of >71.7% of the participants remained unchanged during the follow-up period, indicating a relatively low residential mobility. Thus, it is reasonable to assume that the baseline exposure estimates of air pollutants could represent the long-term exposure during the disease progression. In addition, outdoor air pollution was estimated based on the participants’ home address and therefore cannot account for all variation in indoor concentrations. Thus, the association between long-term exposure to air pollution and CVD progression in this study should be interpreted with caution. Moreover, information of the potential time-varying confounders was not available because of the low response rate during the follow-up survey; therefore, residual confounding due to potential unmeasured time-varying confounders (e.g., income, educational attainment, physical activity) was another concern that could not be completely ruled out. Because of the lack of precise information of blood pressure medication history during the follow-up, we could not define HTN incidence according to the use of blood pressure medicine, possibly resulting in underestimating the associations between air pollution exposure levels and CVD progression. Furthermore, the exclusion of participants with existing HTN and CVD in the multistate study might have underestimated the associations in the subsequent disease transitions to some extent. However, under these circumstances, we still observed significant associations in these disease transitions. It is also possible that the restriction of the study population to those with pre-HTN could lead to potential collider bias. However, the results of the sensitivity analysis showed that the potential collider bias would not essentially affect the conclusion of this study.
Conclusions
In conclusion, we investigated the association between air pollution exposure and dynamic disease trajectories from pre-HTN to subsequent HTN, CVD, and death using a multistate analysis and a mediation analysis based on the UK Biobank. We found statistically significant associations between five air pollutants and the progression of CVDs. Moreover, CVD progression induced by air pollution exposure might be partly mediated by the individual’s intermediate disease status. This might suggest that advanced prevention on the status of pre-HTN could significantly reduce the risk of CVD progression. More research is required to further explore the driving factors of air pollution in the developing trajectories of CVDs to provide better scientific evidence for formulating health care policy and interventions.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
All the authors contributed to data collection, and to the design, analysis, interpretation, and redrafting of this report. H.T.L. and H.L.L. had full access to the combined data. S.Y.Z., M.C., L.C., and H.T.Z. did the statistical analysis. S.Y.Z., H.L.L., Z.M.Q., X.Z., and C.J.W. drafted the work or revised it critically for important intellectual content. S.Y.Z. and H.L.L. drafted the manuscript and had responsibility for submission of the manuscript for publication.
We are grateful to the UK Biobank participants. The UK Biobank was established by the Wellcome Trust Medical Charity, Medical Research Council, Department of Health, Scottish Government, and the Northwest Regional Development Agency. It has also had funding from the Welsh Assembly Government, British Heart Foundation, and Diabetes United Kingdom.
The work was supported by the National Natural Science Foundation of China (grant 82041021 to H.L.L.) and by the Bill & Melinda Gates Foundation (grant INV-016826 to H.L.L.).
We are thankful to the participants in the study and the members of the survey teams, as well as to the project development and management teams. This research has been conducted using the UK Biobank Resource under the project number of 69550.
==== Refs
References
1. Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM, et al. 2020. Global Burden of Cardiovascular Diseases and Risk Factors, 1990–2019: update from the GBD 2019 study. J Am Coll Cardiol 76 (25 ):2982–3021, PMID: , 10.1016/j.jacc.2020.11.010.33309175
2. Oparil S, Acelajado MC, Bakris GL, Berlowitz DR, Cífková R, Dominiczak AF, et al. 2018. Hypertension. Nat Rev Dis Primers 4 :18014, PMID: , 10.1038/nrdp.2018.14.29565029
3. Wang C, Yuan Y, Zheng M, Pan A, Wang M, Zhao M, et al. 2020. Association of age of onset of hypertension with cardiovascular diseases and mortality. J Am Coll Cardiol 75 (23 ):2921–2930, PMID: , 10.1016/j.jacc.2020.04.038.32527401
4. Fuchs FD, Whelton PK. 2020. High blood pressure and cardiovascular disease. Hypertension 75 (2 ):285–292, PMID: , 10.1161/HYPERTENSIONAHA.119.14240.31865786
5. Yano Y, Reis JP, Colangelo LA, Shimbo D, Viera AJ, Allen NB, et al. 2018. Association of blood pressure classification in young adults using the 2017 American College of Cardiology/American Heart Association blood pressure guideline with cardiovascular events later in life. JAMA 320 (17 ):1774–1782, PMID: , 10.1001/jama.2018.13551.30398601
6. Svetkey LP. 2005. Management of prehypertension. Hypertension 45 (6 ):1056–1061, PMID: , 10.1161/01.HYP.0000167152.98618.4b.15897368
7. Egan BM, Laken MA. 2013. Pre-hypertension: rationale for pharmacotherapy. Curr Hypertens Rep 15 (6 ):669–675, PMID: , 10.1007/s11906-013-0387-7.24142744
8. Kanegae H, Oikawa T, Kario K. 2017. Should pre-hypertension be treated? Curr Hypertens Rep 19 (11 ):91, PMID: , 10.1007/s11906-017-0789-z.29046988
9. Yusuf S, Joseph P, Rangarajan S, Islam S, Mente A, Hystad P, et al. 2020. Modifiable risk factors, cardiovascular disease, and mortality in 155 722 individuals from 21 high-income, middle-income, and low-income countries (PURE): a prospective cohort study. Lancet 395 (10226 ):795–808, PMID: , 10.1016/S0140-6736(19)32008-2.31492503
10. Cai Y, Zhang B, Ke W, Feng B, Lin H, Xiao J, et al. 2016. Associations of short-term and long-term exposure to ambient air pollutants with hypertension: a systematic review and meta-analysis. Hypertension 68 (1 ):62–70, PMID: , 10.1161/HYPERTENSIONAHA.116.07218.27245182
11. Grande G, Ljungman PLS, Eneroth K, Bellander T, Rizzuto D. 2020. Association between cardiovascular disease and long-term exposure to air pollution with the risk of dementia. JAMA Neurol 77 (7 ):801–809, PMID: , 10.1001/jamaneurol.2019.4914.32227140
12. Miller MR. 2020. Oxidative stress and the cardiovascular effects of air pollution. Free Radic Biol Med 151 :69–87, PMID: , 10.1016/j.freeradbiomed.2020.01.004.31923583
13. Wu QZ, Li S, Yang BY, Bloom M, Shi Z, Knibbs L, et al. 2020. Ambient airborne particulates of diameter ≤1 μm, a leading contributor to the association between ambient airborne particulates of diameter ≤2.5 μm mum and children’s blood pressure. Hypertension 75 (2 ):347–355, PMID: , 10.1161/HYPERTENSIONAHA.119.13504.31838909
14. Chen Q, Wang Q, Xu B, Xu Y, Ding Z, Sun H. 2021. Air pollution and cardiovascular mortality in Nanjing, China: evidence highlighting the roles of cumulative exposure and mortality displacement. Chemosphere 265 :129035, PMID: , 10.1016/j.chemosphere.2020.129035.33246705
15. Su TC, Chen SY, Chan CC. 2011. Progress of ambient air pollution and cardiovascular disease research in Asia. Prog Cardiovasc Dis 53 (5 ):369–378, PMID: , 10.1016/j.pcad.2010.12.007.21414472
16. Gariazzo C, Carlino G, Silibello C, Tinarelli G, Renzi M, Finardi S, et al. 2021. Impact of different exposure models and spatial resolution on the long-term effects of air pollution. Environ Res 192 :110351, PMID: , 10.1016/j.envres.2020.110351.33130163
17. Hart JE, Hohensee C, Laden F, Holland I, Whitsel EA, Wellenius GA, et al. 2021. Long-term exposures to air pollution and the risk of atrial fibrillation in the Women’s Health Initiative cohort. Environ Health Perspect 129 (9 ):97007, PMID: , 10.1289/EHP7683.34523977
18. Pope CA III, Lefler JS, Ezzati M, Higbee JD, Marshall JD, Kim SY, et al. 2019. Mortality risk and fine particulate air pollution in a large, representative cohort of U.S. adults. Environ Health Perspect 127 (7 ):77007, PMID: , 10.1289/EHP4438.31339350
19. de Menezes ST, Giatti L, Brant LCC, Griep RH, Schmidt MI, Duncan BB, et al. 2021. Hypertension, prehypertension, and hypertension control: association with decline in cognitive performance in the ELSA-Brasil cohort. Hypertension 77 (2 ):672–681, PMID: , 10.1161/HYPERTENSIONAHA.120.16080.33307849
20. WHO (World Health Organization). 2016. International Statistical Classification of Diseases and Related Health Problems, 10th Revision. http://apps.who.int/classifications/icd10/browse/2016/en [accessed 1 January 2023].
21. Beelen R, Hoek G, Vienneau D, Eeftens M, Dimakopoulou K, Pedeli X, et al. 2013. Development of NO2 and NOx land use regression models for estimating air pollution exposure in 36 study areas in Europe—the ESCAPE project. Atmos Environ 72 :10–23, 10.1016/j.atmosenv.2013.02.037.
22. Eeftens M, Beelen R, de Hoogh K, Bellander T, Cesaroni G, Cirach M, et al. 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PMcoarse in 20 European study areas; results of the ESCAPE project. Environ Sci Technol 46 (20 ):11195–11205, PMID: , 10.1021/es301948k.22963366
23. Cai M, Xie Y, Bowe B, Gibson AK, Zayed MA, Li T, et al. 2021. Temporal trends in incidence rates of lower extremity amputation and associated risk factors among patients using Veterans Health Administration services from 2008 to 2018. JAMA Netw Open 4 (1 ):e2033953, PMID: , 10.1001/jamanetworkopen.2020.33953.33481033
24. Huang YT, Yang HI. 2017. Causal mediation analysis of survival outcome with multiple mediators. Epidemiology 28 (3 ):370–378, PMID: , 10.1097/EDE.0000000000000651.28296661
25. Richiardi L, Bellocco R, Zugna D. 2013. Mediation analysis in epidemiology: methods, interpretation and bias. Int J Epidemiol 42 (5 ):1511–1519, PMID: , 10.1093/ije/dyt127.24019424
26. VanderWeele TJ. 2016. Mediation analysis: a practitioner’s guide. Annu Rev Public Health 37 :17–32, PMID: , 10.1146/annurev-publhealth-032315-021402.26653405
27. Donoghoe MW, Gebski V. 2017. The importance of censoring in competing risks analysis of the subdistribution hazard. BMC Med Res Methodol 17 (1 ):52, PMID: , 10.1186/s12874-017-0327-3.28376736
28. Cai M, Li H, Wu Y, Zhang S, Wang X, Zhang Z, et al. 2022. Ambient air pollution associated with body fat percentages at different body compartments: a cohort study of UK Biobank participants. Environ Health Perspect 130 (6 ):67702, PMID: , 10.1289/EHP10920.35767013
29. Toutenburg H. 1990. Rubin, D.B: multiple imputation for nonresponse in surveys. Stat Pap (Berl) 31 (1 ):180, 10.1007/BF02924688.
30. van Buuren S, Groothuis-Oudshoorn K. 2011. mice: multivariate imputation by chained equations in R. J Stat Softw 45 (3 ):1–67, 10.18637/jss.v045.i03.
31. Kim OJ, Lee SH, Kang SH, Kim SY. 2020. Incident cardiovascular disease and particulate matter air pollution in South Korea using a population-based and nationwide cohort of 0.2 million adults. Environ Health 19 (1 ):113, PMID: , 10.1186/s12940-020-00671-1.33167999
32. Fukuhara M, Arima H, Ninomiya T, Hata J, Yonemoto K, Doi Y, et al. 2012. Impact of lower range of prehypertension on cardiovascular events in a general population: the Hisayama Study. J Hypertens 30 (5 ):893–900, PMID: , 10.1097/HJH.0b013e328351d380.22388232
33. Andersen PK, Keiding N. 2002. Multi-state models for event history analysis. Stat Methods Med Res 11 (2 ):91–115, PMID: , 10.1191/0962280202SM276ra.12040698
34. de Bock GH, Putter H, Bonnema J, van der Hage JA, Bartelink H, van de Velde CJ. 2009. The impact of loco-regional recurrences on metastatic progression in early-stage breast cancer: a multistate model. Breast Cancer Res Treat 117 (2 ):401–408, PMID: , 10.1007/s10549-008-0300-2.19148746
35. Putter H, Fiocco M, Geskus RB. 2007. Tutorial in biostatistics: competing risks and multi-state models. Stat Med 26 (11 ):2389–2430, PMID: , 10.1002/sim.2712.17031868
36. Binder N, Balmford J, Schumacher M. 2019. A multi-state model based reanalysis of the Framingham Heart Study: is dementia incidence really declining? Eur J Epidemiol 34 (11 ):1075–1083, PMID: , 10.1007/s10654-019-00567-6.31612352
37. Cannon J, Roberts K, Milne C, Carapetis JR. 2017. Rheumatic heart disease severity, progression and outcomes: a multi-state model. JAHA 6 (3 ):e003498, PMID: , 10.1161/JAHA.116.003498.28255075
38. Shad R, Fong R, Quach N, Bowles C, Kasinpila P, Li M, et al. 2021. Long-term survival in patients with post-LVAD right ventricular failure: multi-state modelling with competing outcomes of heart transplant. J Heart Lung Transplant 40 (8 ):778–785, PMID: , 10.1016/j.healun.2021.05.002.34167863
39. Cai Y, Hodgson S, Blangiardo M, Gulliver J, Morley D, Fecht D, et al. 2018. Road traffic noise, air pollution and incident cardiovascular disease: a joint analysis of the HUNT, EPIC-Oxford and UK Biobank cohorts. Environ Int 114 :191–201, PMID: , 10.1016/j.envint.2018.02.048.29518662
40. Wang M, Zhou T, Song Q, Ma H, Hu Y, Heianza Y, et al. 2022. Ambient air pollution, healthy diet and vegetable intakes, and mortality: a prospective UK Biobank study. Int J Epidemiol 51 (4 ):1243–1253, PMID: , 10.1093/ije/dyac022.35179602
41. Pope CA III, Turner MC, Burnett RT, Jerrett M, Gapstur SM, Diver WR, et al. 2015. Relationships between fine particulate air pollution, cardiometabolic disorders, and cardiovascular mortality. Circ Res 116 (1 ):108–115, PMID: , 10.1161/CIRCRESAHA.116.305060.25348167
42. Robertson S, Miller MR. 2018. Ambient air pollution and thrombosis. Part Fibre Toxicol 15 (1 ):1, PMID: , 10.1186/s12989-017-0237-x.29298690
43. Fiordelisi A, Piscitelli P, Trimarco B, Coscioni E, Iaccarino G, Sorriento D. 2017. The mechanisms of air pollution and particulate matter in cardiovascular diseases. Heart Fail Rev 22 (3 ):337–347, PMID: , 10.1007/s10741-017-9606-7.28303426
44. Ekström M, Hellman A, Hasselström J, Hage C, Kahan T, Ugander M, et al. 2020. The transition from hypertension to hypertensive heart disease and heart failure: the PREFERS Hypertension study. ESC Heart Fail 7 (2 ):737–746, PMID: , 10.1002/ehf2.12612.32073753
45. Izzo JL Jr. 2007. Prehypertension: demographics, pathophysiology, and treatment. Curr Hypertens Rep 9 (4 ):264–268, PMID: , 10.1007/s11906-007-0049-8.17686375
|
PMC009xxxxxx/PMC9875846.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36696103
EHP11089
10.1289/EHP11089
Research
Soy-Based Infant Formula Feeding and Uterine Fibroid Development in a Prospective Ultrasound Study of Black/African-American Women
https://orcid.org/0000-0001-8649-8863
Langton Christine R. 1
Harmon Quaker E. 1
Upson Kristen 2
Baird Donna D. 1
1 Women’s Health Group, Epidemiology Branch, National Institute of Environmental Health Sciences, Durham, North Carolina, USA
2 Department of Epidemiology and Biostatistics, College of Human Medicine, Michigan State University, East Lansing, Michigan, USA
Address correspondence to Christine R. Langton, Women’s Health Group, Epidemiology Branch, National Institute of Environmental Health Sciences, P.O. Box 12233, Research Triangle Park, NC 27709 USA. Email: [email protected]
25 1 2023
1 2023
131 1 01700610 2 2022
14 11 2022
19 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Uterine fibroids are highly prevalent, benign tumors. They are the leading indication for hysterectomy, and Black women are disproportionally burdened. Soy-based infant formula contains phytoestrogens, and exposure during sensitive developmental windows may adversely affect the developing uterus; early phytoestrogen treatment in rodent studies led to detrimental uterine effects, including increased fibroid risk in Eker rats. Limited epidemiological studies also have suggested increased fibroid development with soy formula infant feeding.
Objective:
The goal of this study was to examine the association between soy formula feeding in infancy and fibroid development in adulthood.
Methods:
We evaluated this association among 1,610 Black/African-American women age 23–35 y in the Study of Environment, Lifestyle & Fibroids (SELF). Soy formula feeding data was gathered directly from the participants’ mothers (89%). A standardized ultrasound examination was conducted during 4 clinic visits over 5 y to detect fibroids ≥0.5cm in diameter. We used Cox proportional hazards regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between soy formula feeding and incident fibroids adjusted for early-life and adult factors. Fibroid growth was calculated as change in log-volume for fibroids matched at successive visits.
Results:
Of 1,121 fibroid-free participants at baseline, 150 (13%) were ever fed soy formula as infants, and 269 (24%) developed incident fibroids. We did not observe an association between ever being fed soy formula and incident fibroid risk (HR=1.08; 95% CI: 0.75, 1.54). However, participants fed soy formula within 2 months of birth and for >6 months (n=53) had an elevated risk of fibroid incidence in comparison with those never fed soy formula (HR=1.56; 95% CI: 0.92, 2.65). Fibroid growth rates did not differ.
Discussion:
Adding support to limited human data, this prospective fibroid study found that soy-based formula feeding during infancy was associated with a suggestive increase in risk of ultrasound-identified incident fibroids in adulthood. https://doi.org/10.1289/EHP11089
Supplemental Material is available online (https://doi.org/10.1289/EHP11089).
The authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Uterine fibroids are noncancerous tumors of the myometrium that develop in over 70% of women of reproductive age.1 Symptomatic fibroids may cause heavy menstrual bleeding, pelvic pain, and urinary incontinence, and they are the leading cause of hysterectomy in the United States.2,3 African-American women experience fibroid onset an estimated 10 y earlier than U.S. White women4 and have a disproportionate health burden from fibroids.5,6
Phytoestrogens are compounds produced by plants that can act as estrogens by binding to estrogen receptors.7 Isoflavones, a subgroup of phytoestrogens found primarily in legumes and soybeans, provide antioxidant and anti-inflammatory benefits; however, they can also act as endocrine disruptors, leading to adverse health conditions.7,8 Researchers hypothesize that exposure to these estrogen-like compounds during sensitive developmental windows can have detrimental effects on reproductive systems.9,10 Two of the most well-characterized phytoestrogens are the isoflavones daidzein and genistein that have chemical structures that resemble 17-β estradiol and are commonly found in soy-based infant formula.7,8,11 Multiple laboratory animal studies have demonstrated that early exposure to phytoestrogens adversely affects reproductive tract development, including the uterus (reviewed in Suen et al.9). Female mice postnatally exposed to genistein exhibit posteriorization of the uterus that persists into adulthood and are infertile.12–14 Eker rats treated with genistein postnatally exhibit epigenetic alterations in the myometrium and have increased fibroid incidence as adults.15
Though studies in humans are few, soy formula feeding during infancy has been linked to uterine fibroids in adulthood,16–19 as well as to other female reproductive conditions including early and late menarche,20,21 menstrual irregularities,22,23 and endometriosis.24 In the Infant Feeding and Early Development (IFED) Study that examined the postnatal development of estrogen-responsive tissues during the first 9 months of life, the uterine volume of girls fed soy formula decreased more slowly in comparison with the uterine volume of girls who were fed cow’s milk formula, and vaginal tissue of the soy-fed infants was proliferative, indicating estrogenization.25
Medical indications for soy-based infant formula include use in term infants with congenital galactosemia or hereditary lactase deficiency, in families following a strict vegan diet, and for secondary lactose intolerance from acute gastroenteritis.26,27 Despite indications that apply to a small percentage of infants,27 soy formula is consumed by 12% of U.S. infants in the first year of life, with close to 16% of infants from higher-income households consuming soy-based formulas.28 This widespread use of soy-based formula is likely due to other conditions, including cows’ milk allergy or intolerance and desires to have relief of gas, fussiness, or colic symptoms.25,29 In addition, because soy food consumption is beneficial for a variety of health outcomes,30 some parents may believe that soy formula feeding in infancy protects against development of diseases later in life.31 In the U.S., most infants are fed infant formula by 2 months of life or earlier, despite recommendations for exclusive breastfeeding for the first 6 months of an infant’s life.27,32 These patterns of infant feeding result in many infants exposed to soy formula during a sensitive developmental window.33 Therefore, we assessed the association between soy formula feeding in infancy and fibroid development in adulthood in our cohort of young Black/African-American women, the group who develop the highest fibroid burden.5,6 The Study of Environment, Lifestyle & Fibroids (SELF) followed fibroid development with standardized ultrasound examinations at 20-month intervals over 5 y, and most of our soy formula data were collected from participants’ mothers.
Methods
Study Population
SELF is a prospective cohort study designed to evaluate risk factors for incidence and growth of uterine fibroids among young women with no prior clinical diagnosis of fibroids.34 Established in 2010–2012, study recruitment was implemented in collaboration with the Henry Ford Health System (HFHS) in Detroit, Michigan. SELF enrollment was limited to women who self-identified as “Black or African American” among a list of racial and ethnic categories from which they were instructed to choose all that applied. Of 3,200 women screened, 89% met eligibility criteria, and 1,693 women ages 23–35 y attended an orientation and completed all additional enrollment activities. To assess fibroids, a transvaginal ultrasound examination was conducted at the enrollment clinic visit and during three subsequent clinic visits at approximately 20-month intervals through 2018. Self-reported medical history and health-related behaviors, such as pregnancy history, use of hormonal contraception, and smoking status were collected at each visit via computer-assisted telephone interviews, web-based questionnaires, and hard-copy questionnaires. Participants who missed a visit were invited to attend the next study visit. Ninety-five percent of enrolled participants attended at least two visits, 79% attended all four study visits, and over 90% attended the final visit. SELF was approved by the institutional review boards of the National Institute of Environmental Health Sciences and HFHS. All participants provided informed consent as part of the enrollment process.
Assessment of Soy-Based Formula Feeding during Infancy
Exposure to soy-based infant formula was assessed via an early-life questionnaire given out at time of enrollment. Two versions of the early-life questionnaire were created with the same questions. Participants who reported being able to speak with their mother were given a version designed in an interview format so the questions could be systematically asked of mothers. Remaining participants were given a version that simply listed the questions, and they were instructed to get help answering the questions from relatives and family friends who were present during their infancy and childhood. The early-life questionnaire was completed by 1,628 participants (96%), of whom 89% got answers from their mothers.
Participants were asked if they were ever fed soy formula as an infant with response options of “yes,” “no,” or “do not know.” For those who answered yes, they were asked about how many months they were fed soy formula with the following response options: “<1 month,” “1 to 3 months,” “4 to 6 months,” “>6months,” or “do not know.” Participants fed soy formula were also asked whether they were started on soy formula within the first 2 months of their life, with response options of “yes,” “no,” or “do not know.”
Using these data, we created the following four exposure variables: dichotomous exposure of soy formula feeding in infancy (ever or never), timing of soy formula initiation (never fed, within first 2 months after birth, or more than 2 months after birth), soy formula feeding duration (never fed, ≤6 months, or >6 months), and a composite variable combining timing of initiation and duration of soy formula feeding (never fed, initiated within 2 months after birth and >6 months duration, or initiated more than 2 months after birth or ≤6 months duration).
Assessment of Fibroids
The methods for assessing fibroid incidence and growth in the SELF cohort have been previously documented in detail.34,35 Briefly, transvaginal ultrasounds were conducted by experienced and trained sonographers using 2-D equipment at each clinic visit. A standardized protocol was followed to detect, measure, and document fibroids ≥0.5cm in diameter. The largest six fibroids were measured in three perpendicular planes at three separate times during the examination. Fibroid volume was calculated from each of the three fibroid measurements based on the ellipsoid formula, and these calculations were averaged to estimate the volume of each fibroid. Video and still images were archived, and an 8% sample for each sonographer per month, oversampled for fibroid cases, was reviewed by the lead sonographer for quality-control purposes.
Our overall sample of 1,610 participants who returned for one or more follow-up ultrasound visit (Figure 1) included 23% (n=364) who had fibroids detected at enrollment35 and who were excluded from the incidence analysis. Also excluded were five participants who had a hysterectomy for nonfibroid indications prior to their first follow-up visit. Last, 9 participants were excluded due to factors that impeded ultrasound visualization, resulting in a total of 1,232 participants available for analysis of incidence. Incident fibroid cases were defined as participants who were fibroid-free at the initial ultrasound but had fibroids detected at a subsequent ultrasound.
Figure 1. Flowchart of participant selection for incidence analysis, Study of Environment, Lifestyle & Fibroids (SELF), 2010–2018. Of the 1,610 participants at baseline with follow-up data available, a total of 1,121 were included in the analytical sample for incident fibroids (n=269).
Figure 1 is a flowchart with four steps. Step 1: There were 1610 participants with follow-up data. Of these, the following participants were excluded: 364 participants who had fibroids detected at enrollment, 5 participants who had a nonfibroid-related hysterectomy before their first visit, and 9 participants who had ultrasound quality issues. Step 2: There were 1232 participants who were eligible for incidence analysis. Of these, the following were excluded due to missing soy formula data: 34 participants who did not return the early-life questionnaire, 27 participants who responded "don't know" to soy formula questions, and 6 participants who skipped the soy formula question. Step 3: There were 1165 participants who were eligible with soy formula data. Of these, 44 participants had missing covariate data and were excluded for the following reasons: 1 for missing maternal education, 8 for missing mother's age at birth, 11 for missing maternal diabetes or G D M, and 15 for missing maternal HDP. Step 4: There were 1121 participants in the analytical sample.
Fibroids included in the growth analysis were matched across two consecutive clinic visits by the lead sonographer and principal investigator using archived images and fibroid location. A total of 399 participants were included in the growth analysis, of which 245 had prevalent fibroids detected at enrollment and 154 had fibroids that were detected over the course of follow-up. There were 1,259 interval growth measurements from successive visits. The median interval length was 19 months (25th–75th percentiles: 18–21).
Covariates
Characteristics of each participant’s mother during pregnancy with the participant and early-life characteristics of each participant were ascertained on the early-life questionnaire. Prepregnancy and gestational diabetes (GDM) were assessed separately by asking whether the participant’s mother had diabetes or “sugar” before or during the pregnancy of the participant. Maternal hypertensive disorders of pregnancy (HDP) were assessed by asking whether the mother developed preeclampsia, eclampsia, or toxemia during the relevant pregnancy, and a separate question asked whether the mother developed pregnancy-related high blood pressure. Mother’s age at the time of the participant’s birth and participant’s birth weight were also assessed on the early-life questionnaire. Participants were asked in a separate questionnaire to report the highest year or level of school completed by their mothers or primary caregivers when they were ∼10y old. Other factors of interest were asked of the participant at enrollment and at each follow-up visit by computer-assisted questionnaires and telephone interviews. These time-varying factors included participant age, hormonal contraception history, pregnancy history, current cigarette use, and household income. Body mass index (BMI), also a time-varying factor, was calculated using height measured at enrollment and weight measured at each clinic visit.
Statistical Analyses
Maternal pregnancy factors, early-life factors, and adult characteristics of participants were descriptively examined according to ever vs. never soy-based formula feeding during infancy. To examine the association between infant soy formula feeding and fibroid incidence, we used Cox proportional hazard regression, with age as the time scale to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). When fitting Cox models, we assigned fibroid incidence at the time when a fibroid was first seen on ultrasound among those fibroid-free at enrollment. Participants contributed follow-up time from the enrollment clinic visit until they had an incident fibroid detected at a study visit, nonfibroid-related hysterectomy, loss to follow-up, or their final study visit, whichever came first.
To identify important covariates for adjustment, we examined the literature on indications for soy formula feeding,27 risk factors for fibroids,2,36 and studies that examined the association of the two.16–19 Given the lack of consistent findings to identify causal risk factors for fibroid development, we conducted the analyses considering three models for adjustment using factors with support from prior studies, or associations in our data. We completed three models: minimal adjustment for age by using age as the time scale (Model 1), adjustment for early-life factors only (Model 2), and adjustment for both early-life factors and time-varying participant adult factors (Model 3). The early-life factors were maternal prepregnancy diabetes or GDM (no or yes), maternal HDP (no or yes), mother’s age at participant’s birth (<20, 20–29, or ≥30y), highest education level of participant’s mother at age 10 (≤high school/GED or some college/college degree), and birth weight (<2,500 or ≥2,500 grams). We used birth weight as a surrogate for preterm birth because low birth weight is a result of preterm birth or intrauterine growth restriction37 and our data on gestational age were incomplete.38 These early-life factors might have influenced choice of soy formula feeding and have been associated with fibroid prevalence, though the studies are few, and data are limited. Maternal fibroid history was assessed as a potential confounder, but it did not affect observed associations between soy formula feeding and fibroid development, so it was not included in final models. Participant adult factors were time since last injectable depot medroxyprogesterone acetate (DMPA) use (never, <2y, or ≥2y since last use), parity (0, 1–2, or ≥3 births), time since last birth (<3 or ≥3y ago including no births), current smoking (no or yes), BMI (<25.0, 25.0–<30.0, 30.0–<35.0, 35.0–<40.0, or ≥40.0kg/m2), and household income (<USD $20,000 or ≥USD $20,000 per year). All participant adult factors were included as time-varying covariates in the models. Because these factors were associated with fibroid incidence and/or growth in our sample, adjusting for them could increase precision of the association of interest and improve model fit. We ran complete case analysis on all models. After excluding participants missing data on soy formula feeding during infancy (n=67) and those missing any covariate data (n=44), our final analytical data set for incidence comprised 1,121 participants (Figure 1). We tested proportionality of hazards based on a test of interaction between our composite soy formula feeding variable and age in our fully adjusted model. Tests of proportionality of hazards did not indicate violation of model assumptions (p=0.84).
Fibroid growth was calculated as the difference in the natural logarithm of the volumes, and this volume change was scaled to a growth rate over 18 months. Factors affecting growth were analyzed using a mixed model [GLIMMIX procedure in SAS (version 9.4; SAS Institute Inc.)].35,39 The random effects portion of our mixed models accounted for correlation among fibroids from the same participant and for correlation over time for the same fibroid as well as greater variability among our volume measures for small vs. large fibroids.35 For ease of interpretation, the logarithmic growth rate scale was back transformed to estimate percent difference between exposed and unexposed in volume change per 18 months. When examining fibroid growth, all models were adjusted for fibroid volume, number of fibroids, and age,35 as well as for covariates considered in the three models for adjustment as described for the incidence analyses (minimal adjustment, additional adjustment for early-life factors, and further adjustment for adult factors).
We conducted several sensitivity analyses. First, we restricted the incidence and growth analyses to only those participants whose mothers directly provided data on maternal and early-life exposures, excluding participants who completed the questionnaire with help from others. This restriction allowed us to evaluate the sensitivity of the findings to potential misclassification based on use of reports from relatives and family friends. Second, to account for varying infant feeding patterns, we repeated the incidence analyses, adjusting for whether participants were breastfed during their infancy. Third, we tested the robustness of our incidence findings by moving the time of incidence to the midpoint of each interval instead of the end. Last, we examined the extent to which results from our growth analyses might be influenced by outliers by excluding fibroids that had residuals for growth >3 standard deviations from the mean as had been done in prior fibroid growth analyses.35,39
All statistical analyses were conducted with SAS (version 9.4; SAS Institute Inc.).
Results
Maternal, early-life, and enrollment characteristics for the 1,610 participants who had one or more follow-up visits and by exposure to soy-based infant formula for the incidence (n=1,121) and growth (n=399) analytical samples are shown in Table 1. Mothers of soy formula–exposed participants in comparison with nonexposed tended to be older at the time of the participant’s birth and more educated. Participants ever fed soy formula were more likely to have been breastfed and to have come from a pregnancy complicated by hypertension. At time of enrollment in SELF, participants ever fed soy formula as infants tended to be younger and have higher household incomes in comparison with those who were unexposed. In adulthood, parity, smoking, and use of DMPA were similar for those ever and never fed soy formula as infants.
Table 1 Maternal, early-life, and enrollment characteristics of 1,610 participants by soy-based formula feeding during infancy, Study of Environment, Lifestyle & Fibroids (SELF), 2010–2018.
Characteristic Overall cohort Incidence analysis sample Growth analysis sample
Overall (n=1,610) n (%) Never fed soy formula (n=971) n (%) Ever fed soy formula (n=150) n (%) Never fed soy formula (n=355) n (%) Ever fed soy formula (n=44) n (%)
Pregnancy and demographic factors of participant’s mother
Prepregnancy or gestational diabetesa
No 1,441 (95) 921 (95) 137 (91) 340 (96) 42 (95)
Yes 80 (5) 50 (5) 13 (9) 15 (4) 2 (5)
Don’t know response/missing 28 0 0 0 0
Hypertensive disorders of pregnancya
No 1,313 (87) 866 (89) 116 (77) 309 (87) 35 (80)
Yes 193 (13) 105 (11) 34 (23) 46 (13) 9 (20)
Don’t know response/missing 43 0 0 0 0
Smoked during pregnancya
No 1,154 (76) 729 (76) 117 (78) 271 (77) 35 (80)
Yes 366 (24) 234 (24) 33 (22) 83 (23) 9 (20)
Don’t know response/missing 29 8 0 1 0
Age at participant’s birth (y)a
<20 328 (21) 217 (22) 30 (20) 74 (21) 5 (11)
20–29 906 (59) 575 (59) 81 (54) 210 (59) 29 (66)
≥30 303 (20) 179 (19) 39 (26) 71 (20) 10 (23)
Missing 12 0 0 0 0
Highest education at age 10 y of participantb
≤High school/GED 742 (46) 463 (48) 56 (37) 165 (46) 13 (30)
Some college or associate/technical degree 672 (42) 414 (42) 69 (46) 141 (40) 22 (50)
Bachelor/master/doctoral degree 194 (12) 94 (10) 25 (17) 49 (14) 9 (20)
Missing 2 0 0 0 0
Early life factors of participant
Participant’s birth weight (g)a
<2,500 204 (13) 135 (14) 19 (13) 43 (12) 4 (9)
≥2,500 1,327 (87) 836 (86) 131 (87) 312 (88) 40 (91)
Don’t know response/missing 18 0 0 0 0
Breastfed during infancya
Never breastfed 1,039 (68) 685 (71) 72 (48) 254 (72) 24 (55)
Breastfed 488 (32) 283 (29) 78 (52) 100 (28) 20 (45)
Don’t know response/missing 22 3 0 1 0
Duration, among those breastfed:
≤6 months 320 (70) 176 (67) 58 (78) 59 (63) 19 (95)
>6 months 138 (30) 88 (33) 16 (22) 34 (37) 1 (5)
Don’t know response/missing 30 19 4 7 0
Fed soy formula during infancya
Never fed soy formula 1,313 (87) 971 (100) 0 (0) 355 (100) 0 (0)
Ever fed soy formula 196 (13) 0 (0) 150 (100) 0 (0) 44 (100)
Don’t know response/missing 40 0 0 0 0
Timing of initiation, among those ever fed soy formula:
Started within first 2 months after birth 105 (57) 0 (0) 81 (58) 0 (0) 29 (66)
Started later than 2 months after birth 79 (43) 0 (0) 59 (42) 0 (0) 15 (34)
Don’t know response/missing 12 0 10 0 0
Duration, among those ever fed soy formula:
≤6 months 89 (48) 0 (0) 65 (45) 0 (0) 19 (43)
>6 months 97 (52) 0 (0) 78 (55) 0 (0) 25 (57)
Don’t know response/missing 10 0 7 0 0
Initiation and duration, among those ever fed soy formula:
Initiated within 2 months after birth and >6-month duration 66 (36) 0 (0) 53 (38) 0 (0) 21 (48)
Initiated more than 2 months after birth or ≤6-month duration 115 (64) 0 (0) 85 (62) 0 (0) 23 (52)
Don’t know response/missing 15 0 12 0 0
Participant characteristics at enrollment
Age at ultrasound (y)
23–25 362 (22) 243 (25) 50 (33) 47 (13) 3 (7)
26–28 395 (25) 249 (26) 37 (25) 80 (23) 14 (32)
29–31 442 (27) 254 (26) 41 (27) 113 (32) 13 (29)
32–35 411 (26) 225 (23) 22 (15) 115 (32) 14 (32)
Yearly household income of participant (USD)
<$20,000 734 (46) 441 (46) 63 (43) 141 (40) 13 (30)
$20,000−$50,000 590 (37) 377 (39) 58 (39) 131 (37) 19 (43)
>$50,000 275 (17) 145 (15) 27 (18) 81 (23) 12 (27)
Don’t know response/missing 11 8 2 2 0
Parity
0 births 626 (39) 348 (36) 65 (43) 173 (49) 21 (48)
1–2 births 708 (44) 441 (45) 60 (40) 141 (40) 19 (43)
≥3 births 276 (17) 182 (19) 25 (17) 41 (11) 4 (9)
Time since last birth
Within 3 y 365 (23) 246 (25) 36 (24) 58 (16) 5 (11)
≥3y ago, or no births 1,245 (77) 725 (75) 114 (76) 297 (84) 39 (89)
Smoking status
Non/former 1,296 (80) 781 (80) 130 (87) 293 (83) 37 (84)
Current 314 (20) 190 (20) 20 (13) 62 (17) 7 (16)
Body mass index (kg/m2)
<25.0 322 (20) 197 (20) 25 (17) 66 (19) 7 (16)
25.0 to <30.0 341 (21) 203 (21) 34 (22) 72 (20) 8 (18)
30.0 to <35.0 307 (19) 182 (19) 31 (21) 72 (20) 14 (32)
35.0 to <40.0 268 (17) 168 (17) 22 (15) 68 (19) 4 (9)
≥40.0 372 (23) 221 (23) 38 (25) 77 (22) 11 (25)
DMPA use
Never used 918 (57) 502 (52) 87 (58) 238 (67) 28 (64)
<2y since last use 188 (12) 105 (11) 17 (11) 28 (8) 3 (7)
≥2y since last use 503 (31) 364 (37) 46 (31) 89 (25) 13 (29)
Missing 1 0 0 0 0
Note: DMPA, depot medroxyprogesterone acetate; GED, high school equivalency diploma.
a Frequencies and percentages for the overall cohort are based on a total of 1,549 participants because 61 participants did not complete the early-life questionnaire.
b Maternal education data were collected from all participants on the enrollment questionnaire.
During 4,841 person-years of follow-up, participants had an average 3.8 (±0.5) study visits and a median length of study participation of 4.7 y (25th–75th percentiles: 4.6–4.9). Five participants (0.4%) were censored due to hysterectomy for nonfibroid indications and 269 participants (24%) had incident fibroids detected; median volume at detection was 0.6 cm3 (25th–75th percentiles: 0.2–1.4 cm3). In this sample of young women, with no clinical diagnosis of fibroids before enrollment, most of the fibroids followed for growth were also small (median volume, 3.3 cm3, 25th–75th percentiles: 0.8–13.7 cm3; average diameter, 1.8cm).
In analyses adjusted for age (Table 2, Model 1), we did not observe an association between ever being fed soy formula as an infant and incident fibroid risk (HR=1.03; 95% CI: 0.73, 1.47). However, our data showed that participants fed soy formula within 2 months of birth in comparison with those never fed soy formula had a 24% increased risk of incident fibroids (HR=1.24; 95% CI: 0.81, 1.91). Similarly, participants exposed to soy formula feeding for more than 6 months in infancy in comparison with those never fed soy formula had increased fibroid incidence (HR=1.21; 95% CI: 0.77, 1.90). Considering both the timing and duration of soy formula feeding, soy formula feeding within 2 months of birth and for >6months′ duration (vs. never fed soy formula) was associated with a 37% increased risk of incident fibroids (HR=1.37; 95% CI: 0.82, 2.29). The magnitudes of the associations were stronger in models additionally adjusted for early-life characteristics (Model 2) and both early-life and time-varying adult characteristics (Model 3), except for models that examined duration of soy formula feeding ≤6 months, where the estimates were slightly attenuated or fluctuated. For example, considering soy formula feeding initiated within 2 months after birth and >6months in duration, after adjustment for early-life factors (Model 2), the aHR was 1.44 (95% CI: 0.86, 2.42), and after adjustment for both early-life and adult factors (Model 3), the aHR was 1.56 (95% CI: 0.92, 2.65).
Table 2 Association between soy-based formula feeding in infancy and fibroid incidence in adulthood among 1,121 participants in Study of Environment, Lifestyle & Fibroids (SELF), 2010–2018.
Exposure No. exposed Incident cases Person-years Model 1a HR (95% CI) Model 2b HR (95% CI) Model 3c HR (95% CI)
Soy formula feeding
Never fedd 971 233 4,178 Ref Ref Ref
Ever fed 150 36 663 1.03 (0.73, 1.47) 1.05 (0.74, 1.51) 1.08 (0.75, 1.54)
Timing of soy formula feeding initiatione
Within 2 months after birth 81 23 351 1.24 (0.81, 1.91) 1.32 (0.85, 2.04) 1.36 (0.88, 2.12)
More than 2 months after birth 59 12 265 0.86 (0.48, 1.53) 0.84 (0.47, 1.51) 0.81 (0.45, 1.46)
Duration of soy formula feedinge
≤6 months 65 14 290 0.88 (0.51, 1.51) 0.90 (0.52, 1.54) 0.91 (0.52, 1.56)
>6 months 78 21 341 1.21 (0.77, 1.90) 1.24 (0.79, 1.96) 1.28 (0.81, 2.03)
Initiation and duration of soy formula feedinge
Initiated within 2 months after birth and >6-month duration 53 16 227 1.37 (0.82, 2.29) 1.44 (0.86, 2.42) 1.56 (0.92, 2.65)
Initiated more than 2 months after birth or ≤6-month duration 85 19 379 0.94 (0.59, 1.49) 0.95 (0.59, 1.52) 0.92 (0.57, 1.47)
Note: CI, confidence interval; DMPA, depot medroxyprogesterone acetate; GED, high school equivalency diploma; HR, hazard ratio; Ref, referent.
a Cox model with age as the time scale and no additional covariate adjustment.
b Adjusted for early-life factors of maternal prepregnancy or gestational diabetes (no or yes), maternal hypertensive disorders of pregnancy (no or yes), mother’s age at participant’s birth (<20, 20–29, or ≥30y), birth weight (<2,500 or ≥2,500 grams), and highest education of mother at age 10 y (≤high school/GED or some college/college degree).
c Adjusted for early-life factors as well as time-varying adult factors of time since last DMPA use (never, <2y, or ≥2y since last use), parity (0, 1–2, or ≥3 births), time since last birth (<3 or ≥3y ago including no births), current smoking (no or yes), body mass index (<25.0, 25.0 to<30.0, 30.0 to<35.0, 35.0 to<40.0, or ≥40.0 kg/m2), and household income (<USD $20,000 or ≥USD $20,000 per year).
d Never fed is referent for all exposure categories.
e Numbers of exposed do not sum to 150 because of missing data: timing of soy formula initiation (n=10), duration of soy formula feeding (n=7), or combination of initiation and duration of soy formula feeding (n=12).
Fibroid growth rates did not differ based on exposure to soy formula in infancy (Table 3). In comparison with an estimated measure of 69% growth per 18 months for all fibroids in our analytical dataset, the estimated difference in growth rate over 18 months comparing participants ever and never fed soy formula as infants was small and accompanied by a wide CI (−3.1%; 95% CI: −16.4%, 12.4%). Estimated differences between participants ever and never fed soy formula as infants were similarly of small magnitude after adjustment for early-life and adult factors, and growth rates did not differ when initiation and duration of soy formula feeding were considered (Table 3, Models 2 and 3).
Table 3 Association between soy-based formula feeding in infancy and fibroid growth over 18 months in adulthood among 399 participants in Study of Environment, Lifestyle & Fibroids (SELF), 2010–2018.
Exposure Growth intervalsa Model 1b Model 2c Model 3d
Estimated percentage difference in growth (95% CI)
Soy formula feeding
Never fede 1,109 Ref Ref Ref
Ever fed 150 −3.1 (−16.4, 12.4) −2.9 (−16.4, 12.8) −1.3 (−14.0, 13.4)
Timing of soy formula feeding initiation
Within 2 months after birth 96 −0.6 (−17.0, 18.9) −0.5 (−17.1, 19.3) 1.5 (−14.3, 20.2)
More than 2 months after birth 54 −7.4 (−27.1, 17.7) −7.1 (−27.1, 18.4) −5.9 (−24.5, 17.3)
Duration of soy formula feeding
≤6 months 78 −6.2 (−23.4, 14.9) −6.3 (−23.8, 15.2) −3.1 (−19.7, 17.0)
>6 months 72 0.2 (−18.2, 22.7) 0.5 (−18.0, 23.3) 0.6 (−17.1, 22.2)
Initiation and duration of soy formula feeding
Initiated within 2 months after birth and >6-month duration 55 2.6 (−18.1, 28.5) 2.0 (−18.7, 27.9) 4.7 (−15.7, 30.1)
Initiated more than 2 months after birth or ≤6-month duration 95 −6.6 (−22.4, 12.5) −6.0 (−22.2, 13.5) −4.7 (−19.7, 13.0)
Note: CI, confidence interval; DMPA, depot medroxyprogesterone acetate; GED, high school equivalency diploma; Ref, referent.
a Growth analyses were conducted among fibroids that could be matched across successive visits. This includes fibroids from 399 participants with 1,259 interval growth measurements. Participants could contribute multiple fibroids and fibroids could be followed across multiple intervals.
b Adjusted for fibroid characteristics of volume of fibroid (<0.5 cm3, 0.5 to<4.2 cm3, 4.2 to<14.1 cm3, or ≥14.1 cm3), number of fibroids (ordinal; 1, 2, 3, or ≥4), and age (continuous).
c Adjusted for fibroid characteristics and age as well as early-life factors of maternal prepregnancy or gestational diabetes (no or yes), maternal hypertensive disorders of pregnancy (no or yes), mother’s age at participant’s birth (<20, 20–29, or ≥30y), birth weight (<2,500 or ≥2,500 grams), and highest education of mother at age 10 y (≤high school/GED or some college/college degree).
d Adjusted for fibroid characteristics, age, early-life factors, and time-varying adult factors of time since last DMPA use (never, <2y, or ≥2y since last use), parity (0, 1–2, or ≥3 births), time since last birth (<3 or ≥3 years ago including no births), current smoking (no or yes), body mass index (<25.0, 25.0 to<30.0, 30.0 to<35.0, 35.0 to<40.0, or ≥40.0 kg/m2), and household income (<USD $20,000 or ≥USD $20,000).
e Never fed is referent for all exposure categories.
In our sensitivity analyses restricting the study population to participants whose mothers completed or helped complete the early-life questionnaire, estimates for risk of incidence fibroids (Table S1) and differences in fibroid growth (Table S2) were similar to the estimates obtained in our main analyses. When we adjusted for breastfeeding during the participants’ infancies, our estimates for the association between soy formula feeding and fibroid incidence were slightly strengthened (Table S3), and assigning fibroid onset to the midpoint of the interval did not substantively alter the estimates of association (Table S4). Outlier analysis identified 16 fibroids with residuals for growth >3 standard deviations from the mean. After exclusion of these outliers estimated growth difference by soy formula exposure remained of small magnitude (Table S5).
Discussion
In this community-based sample of young Black/African-American women, soy-based formula feeding during infancy was associated with a suggestive increased risk of ultrasound-identified incident fibroids in adulthood. The strongest association was observed for participants who were fed soy-based formula soon after birth and for a duration longer than 6 months. However, fibroid growth rates did not differ based on exposure to soy-based infant formula. Our incidence findings are consistent with that observed in an animal model of Eker rats: Genistein exposure on postnatal days 10 to 12 increased uterine fibroid incidence in adulthood to 93% in genistein-exposed rats vs. a 65% spontaneous tumor incidence observed in control rats.15 We are unable to compare our growth findings to fibroid development in the treated Eker rats because data pertaining to fibroid growth was not reported.
Overall, our findings align with previous epidemiological studies that examined the association between soy formula feeding in infancy and fibroid development in adulthood. A recent meta-analysis reports that soy formula feeding in infancy increased the risk of uterine fibroids by 19% in adulthood.40 Consistent with our findings, the Sister Study reported increased risk of early onset fibroids for Black women diagnosed at ≤30y of age [relative risk (RR)=1.26; 95% CI: 0.83, 1.89] and White women diagnosed at ≤35y of age (RR=1.33; 95% CI: 1.08, 1.64) who were fed soy formula during their infancy when compared with those who were not, and these associations were further strengthened when feeding within the first 2 months of infancy was considered [relative risk (RR)=1.48 (95% CI: 0.84, 2.63) and RR=1.43 (95% CI: 1.10, 1.86) for Black and White women, respectively].18 Although sample sizes were large (n=3,201 and n=27,048 for Black and White women, respectively), the Sister Study was limited by a cross-sectional analysis that relied on retrospective self-report of fibroid diagnosis data. In a prospective analysis that examined self-reported new clinical diagnoses of uterine fibroids among 23,505 participants age 23–50 y in the Black Women’s Health Study, risk was increased for women diagnosed at <30y of age [incidence rate ratio (IRR)=1.28; 95% CI: 0.91, 1.79] but not women diagnosed at ≥30y of age (IRR=0.99; 95% CI: 0.87, 1.13; p for interaction=0.19).19 A cross-sectional assessment in the baseline SELF cohort found no association for prevalent fibroid at baseline with soy formula feeding in infancy, but among participants with fibroids detected, those exposed to soy formula had larger fibroids in comparison with unexposed participants, consistent with earlier onset.16 Despite similarity to previous findings, our results must be interpreted with caution because exposure numbers were small, leading to imprecise estimates with wide CIs, especially among those exposed early in infancy and for a duration longer than 6 months (n=53). Nonetheless, our study notably extends prior analyses by capturing soy formula feeding data directly from mothers for most participants, restricting the analytical sample to participants who were fibroid-free at study entry and by using standardized ultrasound imaging for the detection of incident fibroids over a 5-y follow-up period.
Our study has limitations, but also important strengths. Although the composition of soy infant formula has varied since the product was first introduced to the U.S. food supply over 100 y ago, all soy formulas on the market throughout the birth years of SELF participants (1975–1989) contained isolated soy protein,41,42 the component that has high concentrations of isoflavones.43 Although the SELF population is young, information pertaining to the pregnancy characteristics of the participants’ births and feeding patterns during their infancies was gathered from mothers approximately 25–35 y after the participants were born. Validation studies examining the long-term maternal recall of pregnancy characteristics have shown that mothers are able to recall their child’s birth weight with reasonable accuracy44,45; however, maternal recall of GDM46,47 and HDP48,49 is less consistent. A systematic review of 10 validation studies of maternal recall of HDP found sensitivity estimates ranged from 57% to 87% for preeclampsia and from 31% to 100% for gestational hypertension.50 Furthermore, most studies of maternal recall have been conducted among predominately White, highly educated individuals49,51; thus, future validation studies in more diverse populations are needed. To our knowledge, mothers’ recall of soy formula feeding during their children’s infancy has not been assessed, but validation studies have demonstrated short- and long-term recall of other infant feeding histories to be fairly accurate. When recalling infant formula feeding after 10 y, 94% of mothers recalled feeding formula to their babies and 65% recalled the exact brand.52 Among a cohort of 374 Norwegian mothers, breastfeeding duration recorded during infancy and recalled 20 y later was found to be strongly correlated [intraclass correlation coefficient (ICC)=0.82, p<0.001].53 Nonetheless, the potential for recall error is an important limitation of this study. Yet, 89% of participants were able to gather infant feeding patterns directly from their mothers. Moreover, prevalence of soy formula feeding among SELF participants (13%) was similar to prevalence in the most recent report of soy-based formula consumption (12%)28 and to prevalence during the birth years of our participants (11%).16,42 More accurate exposure and covariate data could be available in the future from follow-up of pregnancy and childhood studies that collected such data at or near the time of pregnancy/infancy. Bias due to unmeasured confounders is a risk inherent with all observational studies,54 but we have a rich database of covariates, and we used the available literature and prior analyses in the SELF cohort to identify potential confounders. To our knowledge, this study was the first large epidemiological study to assess incident fibroids via prospective ultrasound imaging, providing the best data on fibroid incidence available.
Experimental animal studies clearly show adverse reproductive effects from postnatal exposure to phytoestrogens at exposure levels comparable to levels that infants fed soy formula experience (reviewed in Suen et al.9). The human data are limited. Our findings, based on exposure data collected primarily from mothers and outcome data from prospectively assessed fibroids, add to the human data that have suggested possible grounds for concern. The early months after birth when there is transient activation of the hypothalamic-pituitary-gonadal axis may be a particularly susceptible window for exogenous estrogen exposure.33,55 Currently, expert panels deem soy-based infant formula safe for the growth and development of term infants,55,56 yet the associated risk of adverse health outcomes later in life are not well understood nor are they factored into current recommendations.9,31,57–59 Well-designed prospective studies are needed to accurately capture both exposure and outcome data. Future studies that build on prospective birth cohorts that recorded infant feeding patterns from birth and follow participants into adulthood with gynecological ultrasound could provide further critical data on the long-term effects of early-life estrogenic exposures.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by the Intramural Research Program of the National Institutes of Health (NIH), National Institute of Environmental Health Sciences, and funds from the American Recovery and Reinvestment Act funds designated for NIH research.
The authors thank C. Williams and W. Jefferson for their helpful review of an earlier draft of this paper. S. Patchel (Westat, Durham, North Carolina) assisted with the growth analysis and calculation of fibroid size.
==== Refs
References
1. Baird DD, Dunson DB, Hill MC, Cousins D, Schectman JM. 2003. High cumulative incidence of uterine leiomyoma in black and white women: ultrasound evidence. Am J Obstet Gynecol 188 (1 ):100–107, PMID: , 10.1067/mob.2003.99.12548202
2. Wise LA, Laughlin-Tommaso SK. 2016. Epidemiology of uterine fibroids: from menarche to menopause. Clin Obstet Gynecol 59 (1 ):2–24, PMID: , 10.1097/GRF.0000000000000164.26744813
3. Giuliani E, As-Sanie S, Marsh EE. 2020. Epidemiology and management of uterine fibroids. Int J Gynaecol Obstet 149 (1 ):3–9, PMID: , 10.1002/ijgo.13102.31960950
4. Laughlin SK, Schroeder JC, Baird DD. 2010. New directions in the epidemiology of uterine fibroids. Semin Reprod Med 28 (3 ):204–217, PMID: , 10.1055/s-0030-1251477.20414843
5. Catherino WH, Eltoukhi HM, Al-Hendy A. 2013. Racial and ethnic differences in the pathogenesis and clinical manifestations of uterine leiomyoma. Semin Reprod Med 31 (5 ):370–379, PMID: , 10.1055/s-0033-1348896.23934698
6. Eltoukhi HM, Modi MN, Weston M, Armstrong AY, Stewart EA. 2014. The health disparities of uterine fibroid tumors for African American women: a public health issue. Am J Obstet Gynecol 210 (3 ):194–199, PMID: , 10.1016/j.ajog.2013.08.008.23942040
7. Patisaul HB, Jefferson W. 2010. The pros and cons of phytoestrogens. Front Neuroendocrinol 31 (4 ):400–419, PMID: , 10.1016/j.yfrne.2010.03.003.20347861
8. Yu L, Rios E, Castro L, Liu J, Yan Y, Dixon D. 2021. Genistein: dual role in women’s health. Nutrients 13 (9 ):3048, PMID: , 10.3390/nu13093048.34578926
9. Suen AA, Kenan AC, Williams CJ. 2021. Developmental exposure to phytoestrogens found in soy: new findings and clinical implications. Biochem Pharmacol 114848, 10.1016/j.bcp.2021.114848.34801523
10. Jefferson WN, Patisaul HB, Williams CJ. 2012. Reproductive consequences of developmental phytoestrogen exposure. Reproduction 143 (3 ):247–260, PMID: , 10.1530/REP-11-0369.22223686
11. Yu J, Bi X, Yu B, Chen D. 2016. Isoflavones: anti-inflammatory benefit and possible caveats. Nutrients 8 (6 ):361, PMID: , 10.3390/nu8060361.27294954
12. Jefferson WN, Padilla-Banks E, Suen AA, Royer LJ, Zeldin SM, Arora R, et al. 2020. Uterine patterning, endometrial gland development, and implantation failure in mice exposed neonatally to genistein. Environ Health Perspect 128 (3 ):37001, PMID: , 10.1289/EHP6336.32186404
13. Jefferson WN, Padilla-Banks E, Goulding EH, Lao SP, Newbold RR, Williams CJ. 2009. Neonatal exposure to genistein disrupts ability of female mouse reproductive tract to support preimplantation embryo development and implantation. Biol Reprod 80 (3 ):425–431, PMID: , 10.1095/biolreprod.108.073171.19005167
14. Jefferson WN, Padilla-Banks E, Phelps JY, Gerrish KE, Williams CJ. 2011. Permanent oviduct posteriorization after neonatal exposure to the phytoestrogen genistein. Environ Health Perspect 119 (11 ):1575–1582, PMID: , 10.1289/ehp.1104018.21810550
15. Greathouse KL, Bredfeldt T, Everitt JI, Lin K, Berry T, Kannan K, et al. 2012. Environmental estrogens differentially engage the histone methyltransferase EZH2 to increase risk of uterine tumorigenesis. Mol Cancer Res 10 (4 ):546–557, PMID: , 10.1158/1541-7786.MCR-11-0605.22504913
16. Upson K, Harmon QE, Baird DD. 2016. Soy-based infant formula feeding and ultrasound-detected uterine fibroids among young African-American women with no prior clinical diagnosis of fibroids. Environ Health Perspect 124 (6 ):769–775, PMID: , 10.1289/ehp.1510082.26565393
17. D’Aloisio AA, Baird DD, DeRoo LA, Sandler DP. 2010. Association of intrauterine and early-life exposures with diagnosis of uterine leiomyomata by 35 years of age in the Sister Study. Environ Health Perspect 118 (3 ):375–381, PMID: , 10.1289/ehp.0901423.20194067
18. D’Aloisio AA, Baird DD, DeRoo LA, Sandler DP. 2012. Early-life exposures and early-onset uterine leiomyomata in black women in the Sister Study. Environ Health Perspect 120 (3 ):406–412, PMID: , 10.1289/ehp.1103620.22049383
19. Wise LA, Radin RG, Palmer JR, Rosenberg L. 2012. Association of intrauterine and early life factors with uterine leiomyomata in black women. Ann Epidemiol 22 (12 ):847–854, PMID: , 10.1016/j.annepidem.2012.09.006.23089164
20. Adgent MA, Daniels JL, Rogan WJ, Adair L, Edwards LJ, Westreich D, et al. 2012. Early-life soy exposure and age at menarche. Paediatr Perinat Epidemiol 26 (2 ):163–175, PMID: , 10.1111/j.1365-3016.2011.01244.x.22324503
21. D’Aloisio AA, DeRoo LA, Baird DD, Weinberg CR, Sandler DP. 2013. Prenatal and infant exposures and age at menarche. Epidemiology 24 (2 ):277–284, PMID: , 10.1097/EDE.0b013e31828062b7.23348069
22. Upson K, Adgent MA, Wegienka G, Baird DD. 2019. Soy-based infant formula feeding and menstrual pain in a cohort of women aged 23–35 years. Hum Reprod 34 (1 ):148–154, PMID: , 10.1093/humrep/dey303.30412246
23. Upson K, Harmon QE, Laughlin-Tommaso SK, Umbach DM, Baird DD. 2016. Soy-based infant formula feeding and heavy menstrual bleeding among young African American women. Epidemiology 27 (5 ):716–725, PMID: , 10.1097/EDE.0000000000000508.27196806
24. Upson K, Sathyanarayana S, Scholes D, Holt VL. 2015. Early-life factors and endometriosis risk. Fertil Steril 104 (4 ):964–971.e5, PMID: , 10.1016/j.fertnstert.2015.06.040.26211883
25. Adgent MA, Umbach DM, Zemel BS, Kelly A, Schall JI, Ford EG, et al. 2018. A longitudinal study of estrogen-responsive tissues and hormone concentrations in infants fed soy formula. J Clin Endocrinol Metab 103 (5 ):1899–1909, PMID: , 10.1210/jc.2017-02249.29506126
26. Agostoni C, Axelsson I, Goulet O, Koletzko B, Michaelsen KF, Puntis J, et al. 2006. Soy protein infant formulae and follow-on formulae: a commentary by the ESPGHAN Committee on Nutrition. J Pediatr Gastroenterol Nutr 42 (4 ):352–361, PMID: , 10.1097/01.mpg.0000189358.38427.cd.16641572
27. Bhatia J, Greer F, American Academy of Pediatrics Committee on Nutrition. 2008. Use of soy protein-based formulas in infant feeding. Pediatrics 121 (5 ):1062–1068, PMID: , 10.1542/peds.2008-0564.18450914
28. Rossen LM, Simon AE, Herrick KA. 2016. Types of infant formulas consumed in the United States. Clin Pediatr (Phila) 55 (3 ):278–285, PMID: , 10.1177/0009922815591881.26149849
29. Polack FP, Khan N, Maisels MJ. 1999. Changing partners: the dance of infant formula changes. Clin Pediatr (Phila) 38 (12 ):703–708, PMID: , 10.1177/000992289903801202.10618762
30. Li N, Wu X, Zhuang W, Xia L, Chen Y, Zhao R, et al. 2020. Soy and isoflavone consumption and multiple health outcomes: umbrella review of systematic reviews and meta-analyses of observational studies and randomized trials in humans. Mol Nutr Food Res 64 (4 ):e1900751, PMID: , 10.1002/mnfr.201900751.31584249
31. Testa I, Salvatori C, Di Cara G, Latini A, Frati F, Troiani S, et al. 2018. Soy-based infant formula: are phyto-oestrogens still in doubt? Front Nutr 5 :110, PMID: , 10.3389/fnut.2018.00110.30533415
32. U.S. Department of Agriculture and U.S. Department of Health and Human Services. Dietary Guidelines for Americans, 2020–2025. DietaryGuidelines.gov [accessed 15 November 2021].
33. Kuiri-Hanninen T, Sankilampi U, Dunkel L. 2014. Activation of the hypothalamic-pituitary-gonadal axis in infancy: minipuberty. Horm Res Paediatr 82 (2 ):73–80, PMID: , 10.1159/000362414.25012863
34. Baird DD, Harmon QE, Upson K, Moore KR, Barker-Cummings C, Baker S, et al. 2015. A prospective, ultrasound-based study to evaluate risk factors for uterine fibroid incidence and growth: methods and results of recruitment. J Womens Health (Larchmt) 24 (11 ):907–915, PMID: , 10.1089/jwh.2015.5277.26334691
35. Baird DD, Patchel SA, Saldana TM, et al. 2020. Uterine fibroid incidence and growth in an ultrasound-based, prospective study of young African Americans. Am J Obstet Gynecol 223 (3 ):402.e1–e18, PMID: , 10.1016/j.ajog.2020.02.016.32105679
36. Pavone D, Clemenza S, Sorbi F, Fambrini M, Petraglia F. 2018. Epidemiology and risk factors of uterine fibroids. Best Pract Res Clin Obstet Gynaecol 46 :3–11, PMID: , 10.1016/j.bpobgyn.2017.09.004.29054502
37. Cutland CL, Lackritz EM, Mallett-Moore T, Bardají A, Chandrasekaran R, Lahariya C, et al. 2017. Low birth weight: case definition & guidelines for data collection, analysis, and presentation of maternal immunization safety data. Vaccine 35 (48 pt A ):6492–6500, PMID: , 10.1016/j.vaccine.2017.01.049.29150054
38. Hennessy D, Torvaldsen S, Bentley JP, Bowen JR, Moore HA, Roberts CL. 2022. Alternatives to low birthweight as a population-level indicator of infant and child health. Public Health Res Pract 32 (1 ):e31122106, PMID: , 10.17061/phrp31122106.33942046
39. Peddada SD, Laughlin SK, Miner K, Guyon J-P, Haneke K, Vahdat HL, et al. 2008. Growth of uterine leiomyomata among premenopausal black and white women. Proc Natl Acad Sci USA 105 (50 ):19887–19892, PMID: , 10.1073/pnas.0808188105.19047643
40. Qin H, Lin Z, Vasquez E, Luan X, Guo F, Xu L. 2019. High soy isoflavone or soy-based food intake during infancy and in adulthood is associated with an increased risk of uterine fibroids in premenopausal women: a meta-analysis. Nutr Res 71 :30–42, PMID: , 10.1016/j.nutres.2019.06.002.31668644
41. Vandenplas Y, Castrellon PG, Rivas R, Gutiérrez CJ, Garcia LD, Jimenez JE, et al. 2014. Safety of soya-based infant formulas in children. Br J Nutr 111 (8 ):1340–1360, PMID: , 10.1017/S0007114513003942.24507712
42. Foman SJ. 1987. Reflections on infant feeding in the 1970s and 1980s. Am J Clin Nutr 46 :171–182, PMID: , 10.1093/ajcn/46.1.171.3300256
43. Setchell KDR, Zimmer-Nechemias L, Cai J, Heubi JE. 1998. Isoflavone content of infant formulas and the metabolic fate of these phytoestrogens in early life. Am J Clin Nutr 68 (suppl 6 ):1453S–1461S, PMID: , 10.1093/ajcn/68.6.1453S.9848516
44. Chin HB, Baird DD, McConnaughey DR, Weinberg CR, Wilcox AJ, Jukic AM. 2017. Long-term recall of pregnancy-related events. Epidemiology 28 (4 ):575–579, PMID: , 10.1097/EDE.0000000000000660.28346268
45. Lumey LH, Stein AD, Ravelli ACJ. 1994. Maternal recall of birthweights of adult children: validation by hospital and well baby clinic records. Int J Epidemiol 23 (5 ):1006–1012, PMID: , 10.1093/ije/23.5.1006.7860151
46. Carter EB, Stuart JJ, Farland LV, Rich-Edwards JW, Zera CA, McElrath TF, et al. 2015. Pregnancy complications as markers for subsequent maternal cardiovascular disease: validation of a maternal recall questionnaire. J Womens Health (Larchmt) 24 (9 ):702–712, PMID: , 10.1089/jwh.2014.4953.26061196
47. Yawn BP, Suman VJ, Jacobsen SJ. 1998. Maternal recall of distant pregnancy events. J Clin Epidemiol 51 (5 ):399–405, PMID: , 10.1016/s0895-4356(97)00304-1.9619967
48. Diehl CL, Brost BC, Hogan MC, Elesber AA, Offord KP, Turner ST, et al. 2008. Preeclampsia as a risk factor for cardiovascular disease later in life: validation of a preeclampsia questionnaire. Am J Obstet Gynecol 198 (5 ):e11–e13, PMID: , 10.1016/j.ajog.2007.09.038.18241822
49. Bokslag A, Fons AB, Zeverijn LJ, Teunissen PW, de Groot CJM. 2020. Maternal recall of a history of early-onset preeclampsia, late-onset preeclampsia, or gestational hypertension: a validation study. Hypertens Pregnancy 39 (4 ):444–450, PMID: , 10.1080/10641955.2020.1818090.32981372
50. Stuart JJ, Bairey Merz CN, Berga SL, Miller VM, Ouyang P, Shufelt CL, et al. 2013. Maternal recall of hypertensive disorders in pregnancy: a systematic review. J Womens Health (Larchmt) 22 (1 ):37–47, PMID: , 10.1089/jwh.2012.3740.23215903
51. Coolman M, de Groot CJ, Jaddoe VW, Hofman A, Raat H, Steegers EA. 2010. Medical record validation of maternally reported history of preeclampsia. J Clin Epidemiol 63 (8 ):932–937, PMID: , 10.1016/j.jclinepi.2009.10.010.20189760
52. van Zyl Z, Maslin K, Dean T, Blaauw R, Venter C. 2016. The accuracy of dietary recall of infant feeding and food allergen data. J Hum Nutr Diet 29 (6 ):777–785, PMID: , 10.1111/jhn.12384.27333813
53. Natland ST, A LF, Lund Nilsen TI, Forsmo S, Jacobsen GW. 2012. Maternal recall of breastfeeding duration twenty years after delivery. BMC Med Res Methodol 12 :179, PMID: , 10.1186/1471-2288-12-179.23176436
54. Lash TL, VanderWeele TJ, Haneuse S, Rothman KJ. 2021. Chapter 12, Confounding and Confounders. In: Modern Epidemiology. 4th ed. Philadelphia, PA: Wolters Kluwer, 276.
55. McCarver G, Bhatia J, Chambers C, Clarke R, Etzel R, Foster W, et al. 2011. NTP-CERHR expert panel report on the developmental toxicity of soy infant formula. Birth Defects Res B Dev Reprod Toxicol 92 (5 ):421–468, PMID: , 10.1002/bdrb.20314.21948615
56. Vandenplas Y, Hegar B, Munasir Z, Astawan M, Juffrie M, Bardosono S, et al. 2021. The role of soy plant-based formula supplemented with dietary fiber to support children’s growth and development: an expert opinion. Nutrition 90 :111278, PMID: , 10.1016/j.nut.2021.111278.34004412
57. Ho SM, Cheong A, Adgent MA, Veevers J, Suen AA, Tam NNC, et al. 2017. Environmental factors, epigenetics, and developmental origin of reproductive disorders. Reprod Toxicol 68 :85–104, 10.1016/j.reprotox.2016.07.011.27421580
58. Verduci E, Di Profio E, Cerrato L, Nuzzi G, Riva L, Vizzari G, et al. 2020. Use of soy-based formulas and cow’s milk allergy: lights and shadows. Front Pediatr 8 :591988, PMID: , 10.3389/fped.2020.591988.33313028
59. Helfer B, Leonardi-Bee J, Mundell A, Parr C, Ierodiakonou D, Garcia-Larsen V, et al. 2021. Conduct and reporting of formula milk trials: systematic review. BMJ 375 :n2202, PMID: , 10.1136/bmj.n2202.34645600
|
PMC009xxxxxx/PMC9875848.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36696107
EHP12509
10.1289/EHP12509
Invited Perspective
Invited Perspective: A Multivariate Disease Process Perspective for Environmental Epidemiology
https://orcid.org/0000-0003-0664-1194
Valeri Linda 1
1 Columbia University Mailman School of Public Health, New York, New York, USA
Address correspondence to Linda Valeri, Columbia University Mailman School of Public Health, 722 West 168th St., New York, NY 10032 USA. Email: [email protected]
25 1 2023
1 2023
131 1 01130229 11 2022
19 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Author also holds an adjunct position at Harvard T.H. Chan School of Public Health, Boston, MA, USA.
The author declares she has no conflicts of interest to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP10967
==== Body
pmcLife course epidemiological studies provide longitudinal information, with varying degrees of frequency, over a long follow-up for a rich set of environmental, social, and biological factors.1 Life course cohort studies constitute a large investment of funds and time holding the promise of uncovering the effect of complex environmental exposures and of informing prevention strategies when randomized controlled trials are unethical or unfeasible. To inform effective prevention strategies it is important to identify which environmental pollutants are the key drivers of health outcomes as well as identify which timing of exposure influences which phase of the health condition of interest and through which mechanism. Such promise can be fully realized when we acknowledge that disease and exposure are multivariate intertwined dynamic processes in continuous time and consequently apply appropriate methods to investigate their dynamic dependence. Data science methods can be shaped by causal inference principles to overcome key methodological challenges toward the accomplishment of this goal.
The work by Zhang et al. published in this issue of Environmental Health Perspectives2 provides an interesting perspective and an important direction for the study of the life course influence of environmental pollutants. The authors hypothesized that exposure to ambient air pollution during the prehypertension (pre-HTN) stage could lead to blood vessel damage and thus induce HTN and cardiovascular disease (CVD). The authors obtained estimates of associations between a mixture of air pollutants and dynamic progression of CVD using the multistate model, a joint modeling approach for time-to-event outcomes.3 The model consisted of three outcomes (HTN, CVD, and all-cause death)—also referred to as states—and six transitions between these states. The authors conducted mediation analyses for the association between pollutants and incident CVD through HTN and for the association between pollutants and survival through incident CVD and HTN jointly. They found that intermediate diseases significantly mediated the air pollutant–associated risk to develop more serious disease subsequently. Distinguishing aspects of the study are a) multiple exposures and outcomes, b) joint modeling of time-to-event outcomes, and c) mediation analysis involving time-to-event mediators and time-to-event outcomes.
There are limitations to the available methodology for analyzing multivariate dynamic processes that hamper a robust causal interpretation. First, mediation analyses require four assumptions: a) no unmeasured exposure–outcome confounding, b) no unmeasured exposure–mediator confounding, c) no unmeasured mediator–outcome confounding, and d) no effect of the exposure on any mediator–outcome confounder.4 Zhang et al.2 included baseline confounders but did not address the issue of time-varying confounding, although the latter is likely in this longitudinal setting and renders the mediation contrasts not identifiable. Furthermore, the authors did not clarify the impact of competing risk of death in the causal interpretation of the estimates for the association between air pollution and HTN and CVD, or in the mediation analysis of the association between air pollution and CVD potentially mediated by HTN.5 Individuals may die prior to developing HTN or CVD, rendering these outcomes undefined. Finally, measurement error in the time-to-event mediators and confounders is likely, given that the information is obtained through linkage to error-prone medical records.6
Emerging literature in causal inference is providing solutions for the investigation of average causal effects and mediating mechanisms when competing risks are present and the mediators are longitudinal or time-to-event. New causal contrasts, such as survivor average causal effect7 or survivor natural direct and indirect effects,8 have been proposed. Alternatively, separable causal effects have been introduced in the context of mediation with competing risks.9 These are important contributions that could greatly benefit causal interpretation of the analyses of dynamic disease processes across the life course.
However, the current literature falls short of addressing key issues peculiar to the field of environmental health. First, these methodological advancements consider single binary exposures. Therefore, when complex environmental mixtures are under study, implementation of the approaches might not be feasible owing to multicollinearity among the pollutants or incorrect specification of dose–response relationships. Machine learning models might come to the rescue, as discussed by Billionnet et al.10
Second, most cohort studies collect information at discrete time points or resort to electronic medical record linkage, methods that are potentially prone to measurement error in the evaluation of time-to-event outcomes and longitudinal processes. The adoption of wearable and mobile devices for intensive longitudinal data collection within cohort studies may offer opportunities to improve measurement.11 Although wearables may be less effective at capturing rare events, they still might be useful for monitoring subclinical and continuous surrogate measures for rare outcomes. Validation of such measures obtained via wearable devices is critical to establish clinical relevance of the subsequent analysis. When intensive longitudinal data collection is not feasible, statistical approaches to correct for measurement error and misclassification should be employed.12–14
Finally, the competing risk of death is typically handled by interpreting the causal effects in a counterfactual world where everyone survives.7,8 Such interpretation is unrealistic when long-term effects of environmental pollutants are of interest. Incorporating survival process as an additional outcome (rather than focusing on survivor causal effects) is potentially more relevant in life course studies, and novel causal contrasts could be introduced toward this goal.15
In conclusion, the well-conceived study by Zhang et al.2 makes an important contribution to the existing literature owing to its comprehensive consideration of dynamic disease processes to elucidate the role of air pollution as a risk factor for CVD. It adds considerable evidence that ambient fine particles affect CVD risk. Although the causal interpretation of this multivariate time-to-event outcome analysis is hindered by several potential biases, this should not reduce the enthusiasm for these findings. Rather, this article is exciting for its improvement in study design and application of statistical approaches to uncover the effects of environmental pollutants across the life course.
==== Refs
References
1. Kuh D, Ben-Shlomo Y, Lynch J, Hallqvist J, Power C. 2003. Life course epidemiology. J Epidemiol Community Health 57 (10 ):778–783, PMID: , 10.1136/jech.57.10.778.14573579
2. Zhang S, Qian ZM, Chen L, Xhao X, Cai M, Wang C, et al. 2023. Exposure to air pollution during pre-hypertension and subsequent hypertension, cardiovascular disease and death: a trajectory analysis of the UK Biobank cohort. Environ Health Perspect 131 (1 ):017008, 10.1289/EHP10967.36696106
3. Andersen PK, Keiding N. 2002. Multi-state models for event history analysis. Stat Methods Med Res 11 (2 ):91–115, PMID: , 10.1191/0962280202SM276ra.12040698
4. VanderWeele TJ. 2015. Explanation in Causal Inference: Methods for Mediation and Interaction. New York, NY: Oxford University Press.
5. McConnell S, Stuart EA, Devaney B. 2008. The truncation-by-death problem: what to do in an experimental evaluation when the outcome is not always defined. Eval Rev 32 (2 ):157–186, PMID: , 10.1177/0193841X07309115.18319423
6. Young JC, Conover MM, Jonsson Funk M. 2018. Measurement error and misclassification in electronic medical records: methods to mitigate bias. Curr Epidemiol Rep 5 (4 ):343–356, PMID: , 10.1007/s40471-018-0164-x.35633879
7. Tchetgen Tchetgen EJ. 2014. Identification and estimation of survivor average causal effects. Stat Med 33 (21 ):3601–3628, PMID: , 10.1002/sim.6181.24889022
8. Tai AS, Tsai CA, Lin SH. 2021. Survival mediation analysis with the death‐truncated mediator: the completeness of the survival mediation parameter. Stat Med 40 (17 ):3953–3974, PMID: , 10.1002/sim.9008.34111901
9. Stensrud MJ, Young JG, Didelez V, Robins JM, Hernán MA. 2022. Separable effects for causal inference in the presence of competing events. J Am Stat Assoc 117 (537 ):175–183, 10.1080/01621459.2020.1765783.
10. Billionnet C, Sherrill D, Annesi-Maesano I, GERIE study. 2012. Estimating the health effects of exposure to multi-pollutant mixture. Ann Epidemiol 22 (2 ):126–141, PMID: , 10.1016/j.annepidem.2011.11.004.22226033
11. Jia P, Lakerveld J, Wu J, Stein A, Root ED, Sabel CE, et al. 2019. Top 10 research priorities in spatial lifecourse epidemiology. Environ Health Perspect 127 (7 ):74501, PMID: , 10.1289/EHP4868.31271296
12. Sheppard L, Burnett RT, Szpiro AA, Kim SY, Jerrett M, Pope CA III, et al. 2012. Confounding and exposure measurement error in air pollution epidemiology. Air Qual Atmos Health 5 (2 ):203–216, PMID: , 10.1007/s11869-011-0140-9.22662023
13. Valeri L. 2021. Measurement error in causal inference. In: Handbook of Measurement Error Models. Yi GY, Delaigle A, Gustafson P, eds. Boca Raton, FL: Chapman and Hall/CRC, 453–480.
14. Oh EJ, Shepherd BE, Lumley T, Shaw PA. 2018. Considerations for analysis of time‐to‐event outcomes measured with error: bias and correction with SIMEX. Stat Med 37 (8 ):1276–1289, PMID: , 10.1002/sim.7554.29193180
15. Valeri L, Proust-Lima C, Fan W, Chen JT, Jacqmin-Gadda H. 2021. A multistate approach for mediation analysis in the presence of semi-competing risks with application in cancer survival disparities. arXiv. Preprint posted online February 26, 2021, 10.48550/arXiv.2102.13252.
|
PMC009xxxxxx/PMC9875855.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36696104
EHP12536
10.1289/EHP12536
Invited Perspective
Invited Perspective: Nature Is Unfairly Distributed in the United States—But That’s Only Part of the Global Green Equity Story
https://orcid.org/0000-0002-4018-1825
Nesbitt Lorien 1
Quinton Jessica 1
1 Department of Forest Resources Management, Faculty of Forestry, University of British Columbia, Vancouver, British Columbia, Canada
Address correspondence to Lorien Nesbitt, 2022–2424 Main Mall, Vancouver, British Columbia V6T 1Z4, Canada. Email: [email protected]
25 1 2023
1 2023
131 1 01130105 12 2022
21 12 2022
22 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have no actual or potential conflicts of interest to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11164
==== Body
pmcAs we experience increased impacts of climate change and urbanization, the many benefits to health and well-being provided by green and blue spaces are becoming more important,1,2 and research has shown these health benefits to be particularly strong for lower-income populations.3,4 Given these benefits, we have seen increased societal concern about the inequitable distribution of urban green and blue spaces. Researchers have responded with studies on this topic in multiple jurisdictions, analyzing whether low-income, less-educated, or racialized—that is, those who have been marginalized owing to the societal assignment of a specific racial identity—populations are less likely to have access to urban green5–7 and blue8 spaces. In general, we now know that populations with higher incomes and more education have better access to green and blue spaces, especially in cities,5,7,9 whereas associations between greenness and racialization are variable.10
Klompmaker et al.11 contribute to this growing body of evidence with an analysis of the distributional equity of natural environments in all census tracts in the contiguous United States. The authors found that census tracts with higher socioeconomic status had greater access to nature, as measured by the Normalized Difference Vegetation Index (NDVI), NatureScore, park cover, and presence of blue space. U.S. Census tracts with larger percentages of White residents and smaller percentages of Hispanic residents had lower NDVI and NatureScore values, whereas some urban tracts showed weak positive associations between racialization and natural elements. As one of the most extensive analyses of distributional green and blue equity in the United States to date, this research confirms that patterns observed in previous studies extend across the country.
Much of the research on this topic has been produced in the United States, which also has had a large influence on environmental justice theory in the past decades.12,13 However, as distributional green equity analyses have emerged in other jurisdictions, the patterns seen in U.S. research are not as prevalent elsewhere.14 These findings may indicate that environmental inequities are not as stark outside of the United States. Alternatively, they may highlight an important gap in environmental justice theory that also has been raised by global South scholars15; namely, that theory derived from patterns of injustice observed in the United States may not be applicable outside of that country, and researchers attempting to apply U.S.-based theory to non-U.S. locations may not even be asking the right questions.
For example, the census variables typically used as proxies for deprivation, such as income or categories of racialization, may not accurately reflect the social power dynamics and histories of urban development at play in diverse societies around the world. If researchers apply nonapplicable environmental justice theory to other areas, they may produce what is known as recognitional injustice rather than inform solutions. Recognition is defined in this context as respect for identities and cultural difference and the ways in which agents, ideas, and cultures are valued in discourse, practices, and policies.16,17 High-quality and ethical research enacts recognitional justice by attending to the nuances of place and the context-specific dynamics of injustice.
The theme of recognitional justice must also inform policy responses to findings of distributional inequity. As cities around the world attempt to rectify green inequity, for example, through tree planting or park establishment, environmental justice research has expanded to examine the phenomenon of green gentrification—the physical or psychological displacement of underresourced populations as a result of urban greening.18 Although research on this topic is still emerging19 and findings are variable across jurisdictions,20 there is growing evidence that the installation of new green amenities under capitalist development paradigms—which prioritize profit and invite financial investment alongside urban greening—risks displacing those that the greening was intended to serve.21 Recent research has found that green gentrification processes often include breakdowns in recognitional and procedural justice, applying a top-down “green is always good” approach to greening that does not consider the needs and desires of local communities or their potential vulnerability within a capitalist system.22
As cities engage in greening efforts to improve resilience to climate change and address environmental inequities, there is an urgent need for place-based research in understudied jurisdictions to inform these efforts. This should include research that examines processes—such as green gentrification—that may frustrate efforts to improve existing inequities. A business-as-usual approach is unlikely to bring urban nature to those who need it most. If we want to create equitable cities and healthy communities, we need to think outside of the systems that created harm in the first place.
==== Refs
References
1. Britton E, Kindermann G, Domegan C, Carlin C. 2020. Blue care: a systematic review of blue space interventions for health and wellbeing. Health Promot Int 35 (1 ):50–69, PMID: , 10.1093/heapro/day103.30561661
2. van den Bosch M, Ode Sang Å. 2017. Urban natural environments as nature-based solutions for improved public health—a systematic review of reviews. Environ Res 158 :373–384, PMID: , 10.1016/j.envres.2017.05.040.28686952
3. Mitchell R, Popham F. 2008. Effect of exposure to natural environment on health inequalities: an observational population study. Lancet 372 (9650 ):1655–1660, PMID: , 10.1016/S0140-6736(08)61689-X.18994663
4. White MP, Elliott LR, Gascon M, Roberts B, Fleming LE. 2020. Blue space, health and well-being: a narrative overview and synthesis of potential benefits. Environ Res 191 :110169, PMID: , 10.1016/j.envres.2020.110169.32971082
5. Nesbitt L, Meitner MJ, Girling C, Sheppard SRJ, Lu Y. 2019. Who has access to urban vegetation? A spatial analysis of distributional green equity in 10 US cities. Landsc Urban Plan 181 :51–79, 10.1016/j.landurbplan.2018.08.007.
6. Rigolon A. 2016. A complex landscape of inequity in access to urban parks: a literature review. Landsc Urban Plan 153 :160–169, 10.1016/j.landurbplan.2016.05.017.
7. Schwarz K, Fragkias M, Boone CG, Zhou W, McHale M, Grove JM, et al. 2015. Trees grow on money: urban tree canopy cover and environmental justice. PLoS One 10 (4 ):e0122051, PMID: , 10.1371/journal.pone.0122051.25830303
8. Schüle SA, Hilz LK, Dreger S, Bolte G. 2019. Social inequalities in environmental resources of green and blue spaces: a review of evidence in the WHO European Region. Int J Environ Res Public Health 16 (7 ):1216, PMID: , 10.3390/ijerph16071216.30987381
9. Landry SM, Chakraborty J. 2009. Street trees and equity: evaluating the spatial distribution of an urban amenity. Environ Plan A 41 (11 ):2651–2670, 10.1068/a41236.
10. Watkins SL, Gerrish E. 2018. The relationship between urban forests and race: a meta-analysis. J Environ Manage 209 :152–168, PMID: , 10.1016/j.jenvman.2017.12.021.29289843
11. Klompmaker JO, Hart JE, Bailey CR, Browning MHEM, Casey JA, Hanley JR, et al. 2023. Racial, ethnic, and socioeconomic disparities in multiple measures of blue and green spaces in the United States. Environ Health Perspect 131 (1 ):017007, 10.1289/EHP11164.36696102
12. Schlosberg D. 2013. Theorising environmental justice: the expanding sphere of a discourse. Env Polit 22 (1 ):37–55, 10.1080/09644016.2013.755387.
13. Taylor DE. 2000. The rise of the environmental justice paradigm: injustice framing and the social construction of environmental discourses. Am Behav Sci 43 :508–580, 10.1177/0002764200043004003.
14. Quinton J, Nesbitt L, Czekajlo A. 2022. Wealthy, educated, and… non-millennial? Variable patterns of distributional inequity in 31 Canadian cities. Landsc Urban Plan 227 :104535, 10.1016/j.landurbplan.2022.104535.
15. López-Morales E. 2015. Gentrification in the global South. City 19 (4 ):564–573, 10.1080/13604813.2015.1051746.
16. Martin A, Coolsaet B, Corbera E, Dawson NM, Fraser JA, Lehmann I, et al. 2016. Justice and conservation: the need to incorporate recognition. Biol Conserv 197 :254–261, 10.1016/j.biocon.2016.03.021.
17. Young IM. 1990. Justice and the Politics of Difference. Princeton, NJ: Princeton University Press.
18. Anguelovski I, Connolly JJT, Garcia-Lamarca M, Cole H, Pearsall H. 2019. New scholarly pathways on green gentrification: what does the urban ‘green turn’ mean and where is it going? Prog Hum Geogr 43 (6 ):1064–1086, 10.1177/0309132518803799.
19. Quinton J, Nesbitt L, Sax D. 2022. How well do we know green gentrification? A systematic review of the methods. Prog Hum Geogr 46 (4 ):960–987, PMID: , 10.1177/03091325221104478.35971517
20. Anguelovski I, Connolly JJT, Cole H, Garcia-Lamarca M, Triguero-Mas M, Baró F, et al. 2022. Green gentrification in European and North American cities. Nat Commun 13 (1 ):3816, PMID: , 10.1038/s41467-022-31572-1.35780176
21. Rigolon A, Németh J. 2020. Green gentrification or ‘just green enough’: Do park location, size and function affect whether a place gentrifies or not? Urban Stud 57 (2 ):402–420, 10.1177/0042098019849380.
22. Sax DL, Nesbitt L, Quinton J. 2022. Improvement, not displacement: a framework for urban green gentrification research and practice. Environ Sci Policy 137 :373–383, 10.1016/j.envsci.2022.09.013.
|
PMC009xxxxxx/PMC9881649.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36705937
EHP12495
10.1289/EHP12495
Science Selection
Global Beauty Hazard: Assessing Mercury in Skin-Lightening Products
Lewis Jori
27 1 2023
1 2023
131 1 01400225 11 2022
15 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Women seated at a bus stop in Thailand in front of a billboard for skin lightening products
==== Body
pmcIn 2019, a woman with blurred vision, weak muscles, and motor and speech difficulties was admitted to a hospital in Sacramento, California, where her condition declined to agitated delirium.1 After tests revealed extremely high levels of mercury in her blood and urine,1 public health officials learned that the likely source of her mercury poisoning was long-term use of skin-lightening cream that she obtained from Mexico.1 Such products are used daily by people around the world.2 A new systematic review in Environmental Health Perspectives3 characterizes global exposure to mercury from using skin-lightening products, which remain widely available despite bans in the European Union, North America, the Philippines, and several African countries.4
Mercury is a common additive in such products because it interferes with melanin production.5 But the element is also a neurotoxicant and can harm the kidneys, as well as the digestive and immune systems.6 The Minamata Convention on Mercury limits the amount of mercury to 1 μg/g in cosmetics manufactured, imported, or exported by the parties to the convention.7 The authors of the new paper reported that mercury exceeded this amount in approximately 25% of the 787 products tested across the studies they reviewed. Mercury content varied widely, with many products containing hundreds or thousands of micrograms per gram; one cream had a concentration of 314,387μg/g.
The researchers organized peer-reviewed studies into four topic groups: a) mercury levels in skin-lightening products, b) skin-lightening product usage, c) health impacts related to such products, and d) biomarkers of mercury exposure. “Societal perception of beauty continues to perpetuate the notion that lighter skin is more desirable,” the authors wrote, calling the market for skin-lightening products “one of the world’s fastest-growing beauty industries.”3
Niladri Basu, the review’s senior author and a professor of environmental health sciences at McGill University in Montreal, has been studying mercury exposures for years, but this focus on skin-lightening products is new. “I can tell you how much mining and fossil fuel combustion contribute to mercury pollution,” he says. But scientists do not know the extent to which cosmetic products contribute to the global mercury burden. “It’s going to be significant because there are tens, if not hundreds, of millions of people who use these products,” says Basu.
The review describes baseline conditions and provides data that Basu expects will be helpful to scientists, regulators, and policy makers. For example, five studies measured urinary mercury in populations in Hong Kong and the western United States. Across these studies, urinary mercury concentrations ranged between 0 and 770μg/L, with roughly two-thirds of all the study participants’ levels exceeding 20μg/L (identified as a reference value by all studies reviewed, the authors noted).3 By comparison, when Basu and colleagues assessed global mercury exposure from all sources, not just skin lighteners, urinary levels were generally below 3 μg/L.8 Mercury concentrations in tested products varied across geographic regions, the authors found, with the highest median concentrations measured in products purchased from countries in the Eastern Mediterranean, Southeast Asian, and Western Pacific regions.
The high variability of mercury concentrations in skin-lightening products caught the attention of Kyla Taylor, a health scientist in the Division of Translational Toxicology at the National Institute of Environmental Health Sciences. “These products are popular in many different countries and are widely available for purchase on the Internet,” says Taylor, who was not involved in the new work. She adds that the paper may understate the problem. Because most studies analyzed the use of skin-lighting products purchased in stores, Taylor explains, this review likely missed the growing number of products advertised on social media and sold online.7
The authors recognized other gaps, including the need for more studies from Europe, Southeast Asia, and Latin America, where people are known to use skin lighteners.2 In addition, some of the biomarker studies did not differentiate between mercury from skin lighteners and exposures from other sources, such as seafood consumption and amalgam dental fillings.
After combing through more than 2,300 peer-reviewed scientific papers, only 41 papers from 22 countries met the authors’ criteria for inclusion: that they be original articles published in or after 2000 and that their data about human use could be extracted and used in the analysis. Basu was surprised to find so few high-quality, well-designed studies. “This topic has largely not been investigated that well,” he says. “The data that are published are not the strongest, unfortunately,” he explained, citing shortcomings such as small sample sizes, not controlling for other mercury exposures, and too little information on the amount of product applied and duration of use. Basu calls for more resources to be dedicated to the work. His team’s next steps will be to fill in some of the data gaps, he says, noting that the foundational work done in this assessment is “the tip of the iceberg.”
People around the world seek products that promise to lighten skin, which sometimes contain mercury. Besides health reasons such as pigmentary disorders, such products are sought in response to media images (such as this billboard at a bus stop in Thailand) associating beauty with lighter skin, and the pressure of colorism, which refers to discrimination against those within one’s same ethnic or racial group who have darker skin.9 Image: ©iStock.com/oneclearvision.
Women seated at a bus stop in Thailand in front of a billboard for skin lightening products
Jori Lewis writes about the environment, agriculture, and international development. She is the author of Slaves for Peanuts: A Story of Conquest, Liberation, and a Crop That Changed History.
==== Refs
References
1. Mudan A, Copan L, Wang R, Pugh A, Lebin J, Barreau T, et al. 2019. Notes from the field: methylmercury toxicity from a skin lightening cream obtained from Mexico—California, 2019. MMWR Morb Mortal Wkly Rep 68 (50 ):1166–1167, PMID: , 10.15585/mmwr.mm6850a4.31856147
2. Sagoe D, Pallesen S, Dlova NC, Lartey M, Ezzedine K, Dadzie O. 2019. The global prevalence and correlates of skin bleaching: a meta-analysis and meta-regression analysis. Int J Dermatol 58 (1 ):24–44, PMID: , 10.1111/ijd.14052.29888464
3. Bastiansz A, Ewald J, Rodríguez Saldaña V, Santa-Rios A, Basu N. 2022. A systematic review of mercury exposures from skin-lightening products. Environ Health Perspect 130 (11 ):116002, PMID: , 10.1289/EHP10808.36367779
4. World Health Organization. 2019. Preventing Disease Through Healthy Environments: Mercury in Skin Lightening Products. No. WHO/CED/PHE/EPE/19.13. https://apps.who.int/iris/bitstream/handle/10665/330015/WHO-CED-PHE-EPE-19.13-eng.pdf?sequence=1&isAllowed=y [accessed 10 January 2023].
5. Ladizinski B, Mistry N, Kundu RV. 2011. Widespread use of toxic skin lightening compounds: medical and psychosocial aspects. Dermatol Clin 29 (1 ):111–123, PMID: , 10.1016/j.det.2010.08.010.21095535
6. Ha E, Basu N, Bose-O’Reilly S, Dórea JG, McSorley E, Sakamoto M, et al. 2017. Current progress on understanding the impact of mercury on human health. Environ Res 152 :419–433, PMID: , 10.1016/j.envres.2016.06.042.27444821
7. Zero Mercury Working Group. 2022. Skin Lighteners Still Online Despite Mercury Findings. https://eeb.org/wp-content/uploads/2022/03/ZMWG-Skin-2022-Report-Final.pdf [accessed 10 January 2023].
8. Basu N, Horvat M, Evers DC, Zastenskaya I, Weihe P, Tempowski J. 2018. A state-of-the-science review of mercury biomarkers in human populations worldwide between 2000 and 2018. Environ Health Perspect 126 (10 ):106001, PMID: , 10.1289/EHP3904.30407086
9. Pollock S, Taylor S, Oyerinde O, Nurmohamed S, Dlova N, Sarkar R, et al. 2021. The dark side of skin lightening: an international collaboration and review of a public health issue affecting dermatology. Int J Womens Dermatol 7 (2 ):158–164, PMID: , 10.1016/j.ijwd.2020.09.006.33937483
|
PMC009xxxxxx/PMC9885850.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36715545
EHP11906
10.1289/EHP11906
Invited Perspective
Invited Perspective: Environmental Health Interventions Are Only as Good as Their Adoption
https://orcid.org/0000-0002-0968-9401
Levy Karen 1
1 Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, USA
Address correspondence to Karen Levy, Department of Environmental and Occupational Health Sciences, University of Washington, Box 351618, 2980 15th Ave. NE, Seattle, WA 98195 USA. Telephone: (206) 543-4341. Email: [email protected]
30 1 2023
1 2023
131 1 01130324 7 2022
14 9 2022
23 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The author declares she has nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP10839
==== Body
pmc“I have not failed. I’ve just found 10,000 ways that will not work.” –Thomas Edison
Scientists and engineers often come up with brilliant ideas, but in the end technology is only as good as its adoption and proper use. This is true for regardless of how complicated or simple the technology.
Household water treatment, also commonly referred to as point-of-use (POU) water treatment, is a strategy to improve access to safe water that was introduced within the water, sanitation, and hygiene (WASH) field1 because many communities in lower- and middle-income country (LMIC) settings lack access to a full-scale water treatment plant. With POU treatment, residents can conduct at least some elements of treatment (coagulation and sedimentation, filtration, ultraviolet disinfection, or chemical inactivation) in their own homes, offering them the capacity to disinfect their own water. Treating water in the home also has the benefit of minimizing recontamination after collection, a well-documented phenomenon in locations where people fetch their water from locations outside the home.2,3
Chlorination has been seen as a particularly promising POU approach because it leaves a residual disinfectant in the water to address microbial intrusion during storage.4 In LMICs, many studies have been carried out on the effectiveness of household drinking water chlorination for preventing diarrhea. A recent systematic review found that, compared with untreated drinking water from an unimproved source, risk of diarrhea was reduced by 44% with POU chlorination of water [n=25 studies; 0.66 (0.56–0.77)].5
However, although understanding the potential health impact of a technology is important, understanding whether and why people use, or do not use, a proposed technology is equally important. In their review in this issue of Environmental Health Perspectives, Crider et al. do just this, addressing barriers to the adoption of POU chlorination for household drinking water treatment by 46 target populations.6
Considering users’ needs and interests is essential to adoption of a technology. For example, the authors found that bad taste, smell, or appearance of treated water was cited by a large percentage of households, as was lack of time to spend on disinfection (a time burden usually placed on women). Most of the intervention groups received chlorination products for free; households in the groups that did not cited price or availability of products as a barrier to repurchase and continued use.
In addition to the identification of specific barriers to adoption two other aspects stood out in the review. First, the authors identified a sheer lack of information on barriers to adoption. “Much of the time, the reasons for low adoption are poorly understood simply because the relevant data are not systematically collected,” they stated. The authors excluded 27 of 63 otherwise-eligible studies because quantitative measures of adoption were not reported. Among those that did report a measure of adoption, there was no consensus definition of adoption, and several studies emphasized reasons for use rather than nonuse.
Second, lack of attention in the field to user adoption as a signal of intervention success is also belied by the language that researchers use to describe it. Different words used in the literature to describe what Crider et al. appropriately refer to as “adoption” of water chlorination practices range from “adherence” or “compliance” to “use/usage” or “uptake.”6 Public health inherited the language of “compliance” and “adherence” from medicine, where it describes how often patients follow through with a medication regimen.7,8 Although subtle, this language is important because the medical words put the burden of failed adoption on the user, whereas “adoption,” “use/usage,” and “uptake” put this burden on the implementer.
Time and again we see environmental health–based interventions fail because we are not focused enough on the actual uptake of a technology. This is also true for other areas of environmental health, such as household air pollution (HAP).9,10 The WASH and HAP fields are increasingly recognizing the importance of applying approaches from systems science and implementation science to increase the chances of success for environmental health interventions.10,11
This work is critical. Our practices, and our language, must center on users’ needs and interests if we hope for adoption of new technologies to improve population health related to environmental conditions.
==== Refs
References
1. Sobsey M. 2002. Managing Water in the Home: Accelerated Health Gains from Improved Water Supply. https://apps.who.int/iris/bitstream/handle/10665/67319/WHO_SDE_WSH_02.07.pdf?sequence=1&isAllowed=y [accessed 24 July 2022].
2. Bain R, Johnston R, Khan S, Hancioglu A, Slaymaker T. 2021. Monitoring drinking water quality in nationally representative household surveys in low- and middle-income countries: cross-sectional analysis of 27 multiple indicator cluster surveys 2014–2020. Environ Health Perspect 129 (9 ):97010, PMID: , 10.1289/EHP8459.34546076
3. Wright J, Gundry S, Conroy R. 2004. Household drinking water in developing countries: a systematic review of microbiological contamination between source and point-of-use. Trop Med Int Health 9 (1 ):106–117, PMID: , 10.1046/j.1365-3156.2003.01160.x.14728614
4. Mintz ED, Reiff FM, Tauxe RV. 1995. Safe water treatment and storage in the home: a practical new strategy to prevent waterborne disease. JAMA 273 (12 ):948–953, PMID: , 10.1001/jama.1995.03520360062040.7884954
5. Wolf J, Hubbard S, Brauer M, Ambelu A, Arnold BF, Bain R, et al. 2022. Effectiveness of interventions to improve drinking water, sanitation, and handwashing with soap on risk of diarrhoeal disease in children in low-income and middle-income settings: a systematic review and meta-analysis. Lancet 400 (10345 ):48–59, PMID: , 10.1016/S0140-6736(22)00937-0.35780792
6. Crider Y, Tsuchiya M, Mukundwa M, Ray I, Pickering A. 2023. Adoption of point-of-use chlorination for household drinking water treatment: a systematic review. Environ Health Perspect 131 (1 ):016001, 10.1289/EHP10839.36715546
7. Chakrabarti S. 2014. What’s in a name? Compliance, adherence and concordance in chronic psychiatric disorders. World J Psychiatry 4 (2 ):30–36, PMID: , 10.5498/wjp.v4.i2.30.25019054
8. De las Cuevas C. 2011. Towards a clarification of terminology in medicine taking behavior: compliance, adherence and concordance are related although different terms with different uses. Curr Clin Pharmacol 6 (2 ):74–77, PMID: , 10.2174/157488411796151110.21592067
9. Puzzolo E, Pope D, Stanistreet D, Rehfuess EA, Bruce NG. 2016. Clean fuels for resource-poor settings: a systematic review of barriers and enablers to adoption and sustained use. Environ Res 146 :218–234, PMID: , 10.1016/j.envres.2016.01.002.26775003
10. Rosenthal J, Arku RE, Baumgartner J, Brown J, Clasen T, Eisenberg JNS, et al. 2020. Systems science approaches for global environmental health research: enhancing intervention design and implementation for household air pollution (HAP) and water, sanitation, and hygiene (WASH) programs. Environ Health Perspect 128 (10 ):105001, PMID: , 10.1289/EHP7010.33035121
11. Haque SS, Freeman MC. 2021. The applications of implementation science in water, sanitation, and hygiene (WASH) research and practice. Environ Health Perspect 129 (6 ):65002, PMID: , 10.1289/EHP7762.34132602
|
PMC009xxxxxx/PMC9885856.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36715546
EHP10839
10.1289/EHP10839
Review
Adoption of Point-of-Use Chlorination for Household Drinking Water Treatment: A Systematic Review
https://orcid.org/0000-0003-3812-6517
Crider Yoshika S. 1 2 3
Tsuchiya Miki 4
Mukundwa Magnifique 5
Ray Isha 1
Pickering Amy J. 6
1 Energy and Resources Group, University of California, Berkeley (UC Berkeley), Berkeley, California, USA
2 Division of Epidemiology and Biostatistics, UC Berkeley, Berkeley, California, USA
3 King Center on Global Development, Stanford University, Stanford, California, USA
4 Master of Development Practice Program, UC Berkeley, Berkeley, California, USA
5 Department of Civil and Environmental Engineering, Tufts University, Medford, Massachusetts, USA
6 Department of Civil and Environmental Engineering, UC Berkeley, Berkeley, California, USA
Address correspondence to Yoshika S. Crider, John A. and Cynthia Fry Gunn Bldg., 366 Galvez St., Stanford, CA 94305 USA. Email: [email protected]
30 1 2023
1 2023
131 1 01600121 12 2021
20 12 2022
23 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Centralized chlorination of urban piped water supplies has historically contributed to major reductions in waterborne illness. In locations without effective centralized water treatment, point-of-use (POU) chlorination for households is widely promoted to improve drinking water quality and health. Realizing these health benefits requires correct, consistent, and sustained product use, but real-world evaluations have often observed low levels of use. To our knowledge, no prior reviews exist on adoption of chlorine POU products.
Objectives:
Our objectives were to identify which indicators of adoption are most often used in chlorine POU studies, summarize levels of adoption observed, understand how adoption changes over time, and determine how adoption is affected by frequency of contact between participants and study staff.
Methods:
We conducted a systematic review of household POU chlorination interventions or programs from 1990 through 2021 that reported a quantitative measure of adoption, were conducted in low- and middle-income countries, included data collection at households, and reported the intervention start date.
Results:
We identified 36 studies of household drinking water chlorination products that met prespecified eligibility criteria and extracted data from 46 chlorine intervention groups with a variety of chlorine POU products and locations. There was no consensus definition of adoption of household water treatment; the most common indicator was the proportion of household stored water samples with free chlorine residual >0.1 or 0.2mg/L. Among studies that reported either free or total chlorine–confirmed adoption of chlorine POU products, use was highly variable (across all chlorine intervention groups at the last time point measured in each study; range: 1.5%–100%; sample size-weighted median=47%; unweighted median=58%). The median follow-up duration among intervention groups was 3 months. On average, adoption declined over time and was positively associated with frequency of contact between respondents and study staff.
Discussion:
Although prior research has shown that POU chlorine products improve health when correctly and consistently used, a reliance on individual adoption for effective treatment is unlikely to lead to the widespread public health benefits historically associated with pressurized, centralized treatment of piped water supplies. https://doi.org/10.1289/EHP10839
Supplemental Material is available online (https://doi.org/10.1289/EHP10839).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Chlorination of urban piped water supplies contributed to substantial declines in waterborne disease in major cities in the early 1900s.1 Today the addition of chlorine-based disinfectants is a standard step in effective municipal water treatment processes.2 For the >2 billion people globally lacking access to safe and effectively treated water,3 point-of-use (POU) chlorination at the household level has been promoted as an alternative and interim strategy to realize the benefits of safe water in the absence of large-scale infrastructure.4 However, there has been debate about whether evidence supports widespread investment in household water treatment,4–6 and recent randomized controlled trials that included chlorine POU products found little to no impact on child health outcomes that have previously been linked to safe water consumption.7–9
A key difference between the systems that have historically delivered enormous public health benefits and household-level chlorination strategies is that the latter relies on individuals to implement treatment. Modeling studies have concluded that high levels of correct, consistent, and sustained use of household water treatment products are required to realize the health benefits of such treatment,10,11 and meta-analyses have confirmed that greater health benefits are associated with higher levels of adoption.12,13 Thus, an important question is whether these POU products can achieve high levels of correct and consistent use: what is often referred to as adherence to treatment or product adoption.
Substantial research efforts have been made to identify ways to increase adoption.14 Adoption of POU treatment is not uniformly reported in the literature, despite being a critical determinant of the benefits for water quality or health.10,11 It is difficult to measure and lacks a standard definition.15–17 Use is commonly based on self-report or on an indirect measure, such as observed product presence in the home.16 Chlorine POU products offer a meaningful advantage for measuring use as compared with non-chlorine POU products: the ability to measure chlorine in stored drinking water as an objective measure of current product use. In a recent large trial where adoption was measured through both self-report and residual chlorine measurement, self-reported use was higher than objectively measured use.18 A second component of adoption is exclusive consumption of treated water. However, because this is less reported in the literature, and much harder to objectively verify, we decided to focus here on product use (a lower bar).
In this systematic review, we aimed to summarize the evidence on adoption of chlorine POU products and factors associated with high levels of adoption. Our objectives were to a) identify which indicators have been used to assess adoption in chlorine POU studies, b) describe the levels of adoption and barriers to use observed across studies, c) determine trends in adoption over time, and d) assess the relationship between adoption and frequency of contact between study staff and participants.
Methods
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines19 to develop a review protocol prior to beginning our search. The full protocol is available at https://osf.io/ptc3m/. Our search strategy was developed to first identify studies that included a chlorine POU product as a component of an intervention or program, recognizing that adoption is typically not considered a main outcome in household water treatment studies and therefore unlikely to be included in keywords, titles, or abstracts. The search terms for previous systematic reviews of household water treatment studies, which summarized evidence on health or water quality impacts, were used as a starting point and further refined for our purposes.12,20
We searched for “drinking water,” “potable water,” “tap water,” “household water,” or “domestic water” in combination with terms and brands associated with chlorine POU products: “chemical disinfectant,” chlorin*, chlorate, chlorite, disinfec*, hypochlorite, “sodium hypochlorite,” “calcium hypochlorite,” “sodium dichloroisocyanurate,” NaDCC, trichlor, Aquatab, Waterguard or WaterGuard, Klorin, Pur, “water quality,” “free residual chlorine,” or “free chlorine.” We additionally included the names of all countries included in the World Bank 2019 low- and lower-middle income country categories21 and limited our search to articles published after 1 January 1990. We conducted database searches in PubMed/MEDLINE, Web of Science, Global Health [CAB Abstracts and Global Health (CABI)], and Embase. The exact search term sets are included in the Supplemental Material in the section “Full Search Terms.” We also hand searched the reference sections of four prior systematic reviews of household safe water interventions to ensure all relevant studies were included.12,13,20,22 While screening full texts, we identified additional references that we screened for inclusion.
We downloaded search results from each database search and screened titles and available abstracts in Covidence systematic review software (Veritas Health Innovation). Two authors independently reviewed each title/abstract. Inter-reviewer agreement at this stage was >96%. Subsequently, full texts of articles were collected in a Google drive folder and assessed using full eligibility criteria. Y.S.C. screened all full texts for inclusion, and ∼60% were screened by authors M.T. or M.M. Y.S.C. did data extraction for all included full texts. Other authors partially replicated data extraction for ∼70% of texts, and A.J.P. fully replicated this step for ∼10% of texts.
Selection Criteria and Data Extraction
Eligible studies included a) a clearly described drinking water intervention or program with a chlorine POU product, including combined flocculant–disinfectants; b) studies conducted in countries in World Bank low- and middle-income country categories (2019 data)21; c) studies in which data were collected at households (e.g., not solely in health facilities or schools); d) studies including a quantitative measure of adoption; and e) an intervention or program start date. Titles and abstracts were screened for criteria 1–3; criteria 4–5 were confirmed during full text review. Cross-sectional studies were eligible if the start date of the chlorination intervention could be approximated. Because 1990 has been used as a baseline for measuring progress in global safe water access,23 we included all English language studies published from 1 January 1990 through 31 December 2021; we conducted our final search of this date range on 13 April 2022. Any non-English language studies identified during the hand search of prior systematic reviews were also eligible for title and abstract screening.
We extracted data related to the intervention or program design and measures of adoption. Where both self-reported and presence of chlorine were reported, we used the latter as the more objective measure. We categorized adoption as increasing if there was a ≥10 percentage point increase between the first and last measure of adoption, decreasing if there was a ≥10 percentage point decrease, and sustained if the change was <10 percentage points. We determined 10 percentage points to be a reasonable and meaningful threshold to capture changes in adoption; previous work has modeled that a decline from 100% to 90% use negates nearly all health benefits of household water treatment.10 To visually illustrate these trends in plotted data, we fit a linear trend line fit to all adoption data points weighted by intervention group enrolled sample size. Given the expected heterogeneity in adoption measurement and reporting, we did not plan to report a pooled summary statistic. However, we chose to make two departures from our protocol after reviewing the available data and determining that additional analyses were possible. First, because we found comparable measures of adoption across studies, we calculated median values for adoption measures that used free or total chlorine. We calculated weighted medians by multiplying the number of observations of each group’s reported adoption by its enrolled sample size prior to calculation. Second, we additionally extracted data about which individuals, if specified, were specifically targeted for product usage instruction and were the primary implementer of the intervention at the household level. Our goal was to systematically extract data identifying upon whom the nonmonetary costs of household water treatment fall. Global water access data have established that women and girls are primarily responsible for water fetching,3 but less research has explicitly discussed the highly gendered household allocation of water treatment responsibilities.24,25
Our objective was to evaluate the adoption of chlorine POU products when provided directly to households. To separate household product use from less-than-perfect implementation fidelity, which may mean products do not reach households, we excluded program evaluations that included data from households that had not received chlorine POU products. These included, for example, large-scale programs that bundled chlorine POU promotion and distribution with antenatal care26–28 and humanitarian relief efforts that distributed chlorine POU products in the aftermath of disasters that damaged water infrastructure.29–31 We excluded studies of manual chlorine products that were installed outside of the household, that may have been associated with different barriers to and drivers of adoption (e.g., limited by availability at specific sources only, usage motivated by public peer pressure). Finally, although we did not prespecify a method of study quality assessment, we considered objective measurements of chlorine [i.e., free chlorine residual (FCR)] as higher quality data compared with self-reports. Adoption is not a primary outcome for any of the studies included in this review and only chlorine product arms (groups with different interventions within the same study) of trials are included; thus, many of the commonly used quality assessment criteria therefore do not apply (e.g., justified sample size, clearly defined outcome). Therefore, we did not exclude any studies owing to potential biases, which we acknowledge may be present. For example, response bias is a known issue with self-reported outcomes for which a respondent feels pressure to respond in a more socially acceptable32 or courteous way. These biases may be mitigated with objective outcomes, such as FCR measurements. An additional bias, unmitigated by objective outcomes, is the Hawthorne effect, which could result in overestimates of adoption because of respondents changing their behavior in reaction to being observed.33,34 Ultimately, we excluded one eligible study because data were reported in an unusable format.
To report adoption measures for crossover trials, we used the method as described by Albert et al.35 and pooled all households over the duration they experienced each product. For example, for Geremew et al.,36 we combined results from Group 1 crossover period 1 with Group 2 crossover period 2 to calculate adoption of a single product for all households over the duration of one crossover period. Depending on the order that households were assigned products, they may have experienced another chlorine product just prior. Where studies had multiple arms assigned the same chlorine POU product, Table 1 adoption calculations reflect pooled data across those arms. When possible, we used the number of units (e.g., households, children) measured at each time point so that each arm is appropriately weighted even if there was differential attrition between arms. Luby et al.37 was a follow-up study to Chiller et al.,38 and we included the follow-up study’s adoption as the last measured adoption in the sample. Our search identified a 6- to 12-month follow-up study39 to George et al.40 However, because it is unclear whether households had continued access to Aquatabs, which were provided for free in the original trial, we used only adoption data from the original trial.
Table 1 Included studies of point-of-use chlorine water treatment with reported adoption, by product type.
Product type Reference Setting Study design Enrolled households (or other specified unit) (n)a Indicator of adoption Last measured adoption (%) Time point at measurement Change in adoption (+/−/=)b
Flocculant–disinfectant Albert et al. (b)35c Kenya (rural) RCT (crossover) 400 Self-report 62 2 months NA
Flocculant–disinfectant Chiller et al.38 and Luby et al.37 Guatemala (rural) RCT 268 FCR >0.1mg/L 44 and 1.5 10 wk and 8.5 months −
Flocculant–disinfectant Colindres et al.61 Haiti (rural) Cross-sectional study 100 FCR ≥0.1mg/L 12 1 month NA
Flocculant–disinfectant Crump et al. (a)41c Kenya (rural) cRCT 201 family compounds FCR >0.1mg/L 44 Pooled (5 months) NA
Flocculant–disinfectant Doocy and Burnham55 Liberia (IDP camp) cRCT 200 FCR ≥0.1mg/L 95 Pooled (12 wk) NA
Flocculant–disinfectant Geremew et al. (b)36c Ethiopia (rural) cRCT (crossover) 400 Detectable free chlorine 25 2 months =
Flocculant–disinfectant Luoto et al. (c)42 Bangladesh (urban) RCT (crossover) 600 Detectable free chlorine 3 6 weeks NA
Flocculant–disinfectant Norton et al.64 Bangladesh (rural) Nonrandomized trial 105 women FCR >0.2mg/L 43 Pooled (12 wk) NA
Flocculant–disinfectant Rangel et al. (b)44c Guatemala (rural) RCT 60 FCR ≥0.5mg/L 83 Pooled (3 wk) NA
Flocculant–disinfectant Reller et al. (b)45c Guatemala (rural) RCT 199 FCR >0.1mg/L 30 Pooled (9 months) NA
Flocculant–disinfectant Shaheed et al. (a)46c Pakistan (rural) RCT (crossover) 247 Total chlorine ≥0.2mg/L 59 8 wk −
Flocculant–disinfectant Shaheed et al. (b)46c Zambia (urban) RCT (crossover) 214 Total chlorine ≥0.2mg/L 18 8 wk −
Liquid Albert et al. (a)35c Kenya (rural) RCT (crossover) 400 Self-report 76 2 months NA
Liquid Crump et al. (b)41c Kenya (rural) cRCT 203 family compounds FCR >0.1mg/L 61 Pooled (5 months) NA
Liquid Geremew et al. (a)36c Ethiopia (rural) cRCT (crossover) 400 Detectable free chlorine 41 2 months =
Liquid Humphrey et al.9 Zimbabwe (rural) cRCT 2,035 womend FCR >0.1mg/L 58 12 months NA
Liquid Luby et al.82 Pakistan (urban) Trial (unclear if randomized) 50 FCR >0.1mg/L 71 Pooled (10 wk) NA
Liquid Luoto et al. (a)42c Bangladesh (urban) RCT (crossover) 600 Detectable free chlorine 11 6 wk NA
Liquid Macy and Quick83 Nicaragua (rural) Nonrandomized trial 100 FCR >0.1mg/L 52 3 months NA
Liquid Mellor et al.53 Guatemala (rural) RCT 34 FCR ≥0.2mg/L 65 7 months −
Liquid Mengistie et al.84 Ethiopia (rural) cRCT 286 FCR ≥0.2mg/L 77 12 wk =
Liquid Murray et al.57 Haiti (peri-urban) Nonrandomized trial 60 FCR >0.1mg/L 13 13 months −
Liquid Null et al.8 Kenya (rural) cRCT 2,737 Detectable free chlorine 21 2 y −
Liquid Opryszko et al.48 Afghanistan (rural) cRCT 607 Self-report 80 1 y NA
Liquid Potgieter et al. (a)43c South Africa (rural) RCT 20 Detectable free chlorine 88 3 months =
Liquid Potgieter et al. (b)43c South Africa (rural) RCT 20 Detectable free chlorine 98 3 months +
Liquid Quick et al.49 Bolivia (urban) RCT 15 FCR ≥0.1 100 9 wk =
Liquid Quick et al.51 Bolivia (peri-urban) RCT 64 Detectable total chlorine 95 6 months +
Liquid Quick et al.50 Zambia (peri-urban) Nonrandomized trial 166 Detectable total chlorine 85 13 wk +
Liquid Rangel et al. (a)44c Guatemala (rural) RCT 20 FCR ≥0.5mg/L 83 Pooled (3 wk) NA
Liquid Reller et al. (a)45c Guatemala (rural) RCT 197 FCR >0.1mg/L 40 Pooled (9 months) NA
Liquid Sobsey et al. (a)47c Bangladesh (urban) RCT ∼138 Detectable free chlorine 89 Pooled (8 months) NA
Liquid Sobsey et al. (b)47c Bolivia (peri-urban) RCT ∼70 Detectable free chlorine 77 Pooled (6 months) NA
Liquid Solomon et al.85 Ethiopia (rural) cRCT 203 FCR ≥0.2mg/L 81 Pooled (4 months) NA
Liquid Sugar et al.52 Kenya (urban) Program evaluation 392 children Smell of chlorine 97 Pooled (12 months) NA
Tablet Altmann et al.56 Chade cRCT 850 children FCR ≥0.1mg/L 98 2 months =
Tablet Boisson et al.62 India (urban/rural) RCT 1,080 FCR ≥0.1mg/L 47 12 months +
Tablet Clasen et al.54 Bangladesh (urban) RCT 50 FCR ≥0.1mg/L 100 4 months =
Tablet Ercumen et al.86 Bangladesh (rural) RCT 600 FCR ≥0.2mg/L 79 1 y −
Tablet George et al. (a)39 Bangladesh (urban) cRCT 84 FCR ≥0.2mg/L 94 Pooled (9 d) NA
Tablet Jain et al.60 Ghana (peri-urban) RCT 120 FCR ≥0.2mg/L 83 12 wk =
Tablet Luby et al.7 Bangladesh (rural) cRCT 2,086 compounds FCR >0.1mg/L 84 2 y =
Tablet Luoto et al. (b)42 Bangladesh (urban) RCT (crossover) 600 Detectable free chlorine 10 6 wk NA
Tablet Pickering et al.59 Bangladesh (urban) cRCT 90 Total chlorine >0.1mg/L 55 10 months −
Multiple Blanton et al.58 Kenya (rural) Program evaluation 662 FCR ≥0.1mg/L 18 13 months =
Granular Tsai et al.63 Haiti (rural) cRCT 447 FCR ≥0.5mg/L 27 180 d −
Note: Flocculant–disinfectant products included PuR, PureIt, Purifier of Water, and Bishan Gari (local brand in Ethiopia). Liquid chlorine products included sodium hypochlorite, WaterGuard, Klorin/Clorin, bleach, calcium hypochlorite solution, and electrochlorinator (for at-home sodium hypochlorite production). Tablet chlorine products included Aquatabs. Multiple products in a single study included PuR and WaterGuard. Granular chlorine products included Klorfasil. cRCT, cluster randomized controlled trial; FCR, free chlorine residual; IDP, internally displaced persons; NA, not applicable; RCT, randomized controlled trial.
a Sample size was included for chlorine arm(s) only.
b Change in adoption: +indicates ≥10 percentage point increase from first to last measure; – indicates ≥10 percentage point decrease; =indicates <10 percentage point change.
c Albert et al.,35 Crump et al.,41 Geremew et al.,36 Shaheed et al.,46 Luoto et al.,42 Potgieter et al.,43 Rangel et al.44 Reller et al.,45 and Sobsey et al.47 had multiple eligible chlorine intervention arms that are separately listed because they had different settings or chlorine products. Separate arms (referred to here as intervention groups) are indicated with letters in parentheses. We combined intervention arms with the same product, even if other intervention components differed. Studies in which measurement across multiple time points are presented as a pooled statistic are indicated as such, with the study duration in parentheses.
d FCR was measured in only 752 households; self-reported adoption across the entire sample was 87%.
e Urban, rural, or peri-urban was not specified and could not be inferred from the main text of the paper.
Results
Our search identified 8,617 unique results. After reviewing all available titles and abstracts, we obtained 127 full-text articles to assess using full eligibility criteria. This step yielded 36 eligible texts, including 28 cluster or individually randomized controlled trials, 1 cross-sectional study, 2 program evaluations, 4 nonrandomized trials, and 1 trial in which method of intervention assignment was unspecified (Figure S1). Four studies had a crossover design, and studies were conducted in 16 countries (Table 1). Nine studies had multiple intervention arms or were crossover trials with different chlorine POU products35,36,41–45 or were conducted in more than one country.46,47 We categorized each product- or site-specific group within each study as a separate intervention group, and the results of each unique intervention group are separately listed (Table 1). We pooled data into a single intervention group if a single study had multiple arms that used the same chlorine product in the same setting but had different additional components (e.g., in combination with a safe storage container, handwashing stations, or latrines).7,8,43–45,48
Studies were conducted in rural (n=20), urban (n=7), and peri-urban (n=4) settings in addition to three studies in multiple settings, one in an internally displaced persons (IDP) camp, and one with an unspecified setting description. The enrolled sample size (of chlorine POU intervention groups) ranged from 15 households49 to 2,737 households.8 Nearly all provided the POU products for free for the duration of the study. The most common chlorine POU products were WaterGuard (liquid sodium hypochlorite by Population Services International), PuR (flocculant–disinfectant with calcium hypochlorite by Procter & Gamble), and Aquatabs [sodium dichloroisocyanurate (NaDCC) tablets by Medentech].
Defining and Measuring Adoption
A variety of metrics were used to assess product adoption, and various terms were used to describe adoption, including uptake, use/usage, adherence, and compliance. Here, we chose the terms adoption and use, rather than adherence or compliance, so as not to suggest a failure of the end user but rather of the product or its implementation. The most common reported indicator of adoption was the proportion of households with stored drinking water having a FCR greater than or equal to a specified threshold, typically 0.1 or 0.2mg/L, but as high as 0.5mg/L. The choice of threshold can significantly change conclusions about adoption. Using a threshold of 0.5mg/L, Altmann et al. reported that 51% of households adhered to chlorine POU treatment, compared with 98% when using a threshold of 0.1mg/L (their instrument limit of detection).56 Six studies defined adoption as the presence of “detectable” free or total chlorine without specifying a detection limit,8,36,42,43,47,50,51 and one study used the smell of chlorine in stored water because no test instruments were available.52 Two studies measured only self-reported adoption, which was defined as use “during the previous two weeks”48 or undefined.35 All others measured free or total chlorine as either a primary or secondary measure of adoption. Self-reported adoption was higher than FCR-confirmed adoption in studies that reported both.9,42 Quick et al. defined adoption as “any detectable total chlorine residual,” but they also measured and reported FCR.50 Over four time points, the percentage of households with detectable total chlorine ranged from 72% to 95%, in contrast to 55% to 81% with FCR >0.2mg/L. Eleven studies reported a single pooled measure of adoption across their entire study duration; studies that reported time point-specific measures included between 1 and 24 adoption measurements (across all intervention groups, median=3). Of the adoption results reported in Table 1, 18 were measured at unannounced visits, 2 at announced visits, 24 were not specified as either, and 2 were from studies that had both announced and unannounced visits. Measures of variance, such as standard deviation or range, were typically not reported with adoption results.
Observed Levels of Adoption Across Studies
Final measured adoption was highly variable and was not associated with study length (Figure 1). Across all intervention groups, there were 18,480 observations (each study-level final adoption data point times the enrolled group sample size). The studies reporting FCR-confirmed (≥0.1 or ≥0.2mg/L) adoption at any time point ranged from 1.5%37 to 100%.49,53,54 Of the studies that confirmed adoption with either free or total chlorine, eight groups had >90% adoption40,43,49,51,52,54–56 and two groups had <10% adoption at the final time point measured.37,42 Sugar et al. also reported >90% with adoption defined as having the smell of chlorine in stored water.52 The rest reported adoption ranging from 10% to 90% at the final time point measured. With the exception of Luby et al.7 and Null et al.,8 which included 2 y of follow-up, study durations were ≤13 months (Figure 1).
Figure 1. Measured product adoption at the last follow-up. The point sizes are scaled to indicate relative sample size [of the group(s) receiving chlorine only]. Open circles indicate that the data point is reported as multiple adoption measures pooled over the months of follow-up up until the time point shown. Closed circles are a single time point result. Letters in parentheses indicate different intervention groups within a single study. Opryszko et al.48 and Albert et al.35 used self-reported adoption; Sugar et al.52 used the smell of chlorine in stored water as adoption; the rest used either free or total chlorine to measure adoption. The dashed line shows a linear trend line using adoption measures at last follow-up weighted by sample size. Data are from Table 1, columns “Time point at measurement” and “Last measured adoption.” Note that papers are indicated by first author only.
Figure 1 is a bubble chart, plotting adoption (percentage) at the final follow-up time point, ranging from 0 to 100 in increments of 25 (y-axis) across month of follow-up, ranging from 0 to 25 in increments of 5 (x-axis) for Quick et al. reference 51, Altmann et al. reference 56, Clasen et al. reference 54, Potgieter et al. (b) reference 43, George et al. reference 40, Doocy and Burnham reference 55, Quick et al. reference 51, Sugar et al. reference 52, Potgieter et al. (a) reference 43, Sobsey et al. (a) reference 47, Rangel et al. (b) reference 44, Quick et al. reference 50, Rangel et al. (a) reference 44, Jain et al. reference 60, Solomon et al. reference 85, Opryszko et al. reference 48, Luby et al. reference 7, Mengistie et al. reference 84, Albert et al. (a) reference 35, Sobsey et al. (b) reference 47, Ercumen et al. reference 86, Luby et al. reference 37, Albert et al. (b) reference 35, Crump et al. (b) reference 41, Mellor et al. reference 53, Shaheed et al. (a) reference 46, Humphrey et al. reference 9, Macy and Quick reference 83, Pickering et al. reference 59, Norton et al. reference 64, Crump et al. (a) reference 41, Boisson et al. reference 62, Geremew et al. (a) reference 36, Reller et al. (a) reference 45, Geremew et al. (b) reference 36, Tsai et al. reference 65, Reller et al. (b) reference 45, Shaheed et al. (b) reference 46, Colindres et al. reference 61, Blanton et al. reference 58, Null et al. reference 8, Luoto et al. (a) reference 42, Murray et al. reference 57, Luoto et al. (b) reference 42, Luoto et al. (c) reference 42, Chiller et al. reference 38, and Luby et al. reference 37. The sample size ranges from 15 to 2,737, with the bubble size scaled to reflect sample size relative to this minimum and maximum.
Changing Adoption Over Time
On average across all intervention groups, adoption declined slightly over time (Figure 2), although some groups had increasing or sustained adoption. When restricting to studies <13 months in duration, adoption remained stable on average. Among 23 groups that reported multiple time point-specific measures, adoption increased in 4 chlorine intervention groups, decreased in 9 groups, and was sustained in 10 groups. The total pooled sample size was similar for studies with increasing and sustained adoption, but there were approximately one-third as many observations across all studies with decreasing adoption. There were 52,349 observations (group average multiplied by enrolled sample size) from studies that reported one or more single time point adoption measurements and provided products entirely for free (Figure 2), which included one study with self-reported adoption.48 However, we note that data from >13 months after intervention delivery were from only two related studies (i.e., WASH Benefits Bangladesh7 and WASH Benefits Kenya8), and when these studies were excluded, the remaining data (total n=42,703) showed no change in adoption over time.
Figure 2. Reported product adoption over time after the start of intervention. The line width is scaled to indicate relative sample size [of the group(s) receiving chlorine only]. Letters in parentheses indicate different intervention groups within a single study. The studies in the graph are restricted to studies reporting one or more single time point measures of adoption and provided chlorine products entirely for free for the study duration. The latter restriction excludes Tsai et al.,63 Blanton et al.,58 Mellor et al.,53 and Luby et al.37 (which provided the final time point follow-up to Chiller et al.38). Among the studies included here, only Opryszko et al.48 had self-reported adoption. The dashed line shows a linear trend line using all adoption measures weighted by sample size. Data are available in Excel Tables S1 and S2. Note that papers are indicated by first author only.
Figure 2 is a line graph, plotting adoption (percentage), over time, ranging from 0 to 100 in increments of 25 (y-axis) across month of follow-up, ranging from 0 to 25 in increments of 5 (x-axis) for Quick et al. reference 49, Clasen et al. reference 54, Altmann et al. reference 56, Potgieter et al. (b) reference 43, Quick et al. reference 51, Potgieter et al. (a) reference 43, Quick et al. reference 50, Jain et al. reference 60, Opryszko et al. reference 48, Mengistie et al. reference 84, Luby et al. reference 7, Ercumen et al. reference 86, Shaheed et al. (a) reference 46, Humphrey et al. reference 9, Macy and Quick reference 83, Pickering et al. reference 59, Geremew et al. (a) reference 36, Boisson et al. reference 62, Geremew et al. (b) reference 36, Shaheed et al. (b) reference 46, Colindres et al. reference 61, Murray et al. reference 57, Null et al. reference 8, Luoto et al. (a) reference 42, Luoto et al. (b) reference 42, and Luoto et al. (c) reference 42.
Contact Frequency between Participants and Study Staff
Adoption was positively associated with contact frequency between study participants and study staff across studies (Figure 3, Figure S3) and within studies as well. Restricting to studies that used free or total chlorine–confirmed use, adoption ranged from a sample-size-weighted median of 84%, when households were visited one or more times per week by study staff (20 groups, total n=5,560), to 47%, when visits were once or more per month (14 groups, total n=7,810), to 11% with less frequent visits (8 groups, total n=3,311) (Figure 3). In a cluster randomized controlled trial in urban Dhaka,59 the intervention included promotional visits every 2 wk for the first half of the 10-month study, and adoption was >90% when promotions were ongoing. After these visits concluded, free delivery of Aquatabs and water quality testing continued, but adoption quickly dropped by ∼50% and remained relatively stable (42%–56% from months 5–10). In a 2-y cluster randomized controlled trial in rural Kenya, Null et al.8 observed adoption decline by ∼50% between year 1, during which households received monthly promotional visits, and year 2, when households were visited approximately every other month.
Figure 3. Weighted box plots showing the relationship between contact frequency over the study and final measured adoption, restricted to only groups that used free chlorine residual (FCR) or total chlorine to measure adoption. This restriction excludes Opryszko et al.,48 Albert et al.,35 and Sugar et al.52 Dots show each group average. Groups were weighted by sample size and 25th, 50th, and 75th percentiles of the pooled data are displayed above as the midline and box limits. The whiskers extend to data within 1.5 times the interquartile range (IQR) above and below the 75th and 25th percentiles. This figure includes studies that reported single time point and pooled adoption measures. Summary data are presented in Table S1.
Figure 3 is a box and whiskers plot, plotting the 25th percentile, median, and 75th percentile of adoption (percentage) weighted by sample size, ranging from 0 to 100 in increments of 25 (y-axis) across frequency of contact at 1 or more per week, 1 or more per month, and 1 or more per 3 months (x-axis). On the same axes, points show adoption (percentage) of individual groups not weighted by sample size.
Type of Chlorine Product and Adoption
Across all 46 intervention groups, 23 received liquid chlorine products, including branded products, such as WaterGuard and Clorin/Klorin, and generic sodium hypochlorite. One group received a locally mixed calcium hypochlorite solution.49 Twelve groups received flocculant–disinfectant products, most commonly PuR brand, but one study in Ethiopia used a local product called Bishan Gari.36 Nine groups received tablets, all Aquatabs brand. Among groups that used FCR or total chlorine to measure adoption, tablet chlorine product interventions had the highest adoption (9 groups; total n=5,560; weighted median=84%), followed by liquid products (20 groups, total n=7,418; weighted median=41%), then flocculant–disinfectants (11 groups, total n=2,594; weighted median=25%) (Figures S2 and S4).
Barriers to Use
Reasons for nonuse of products, reported by respondents, were provided for only 23 of the 46 intervention groups. This suggests that, much of the time, the reasons for low adoption are poorly understood simply because the relevant data are not systematically collected. Bad taste, smell, or appearance of treated water was identified by one or more households in 17/23 intervention groups (74%), and lack of time was identified in 10/23 groups (43%). Although most of the studies provided products for free, 4/23 groups (17%) identified price or availability as a barrier to repurchase and continued use. Often, however, each of these reasons was reported by a small proportion of households. Several studies emphasized reasons for use rather than nonuse, reporting instead that households preferred treated water41,55,59–61 or that households felt the time required to treat the water was worth it.45
Taste and Smell Concerns
The included studies suggest that taste and smell concerns can be a reason for nonuse, but they are not universal barriers. In Ethiopia, the majority (66% of n=377) of households said they disliked the chlorine taste.36 In rural South Africa, use of 3.5% sodium hypochlorite was slightly higher than use of a 1% solution, although households in the former group did mention disliking the taste of water and the sample size was small.43 However, in both studies that evaluated chlorine POU products in humanitarian crisis settings, respondents reported that they preferred the taste of the chlorinated water over the untreated water,55,61 as did respondents in some households in nonemergency situations.60 In urban Bangladesh, where FCR-confirmed use of Aquatabs, WaterGuard, and PuR was 10%, 11%, and 3%, respectively, only around half of respondents (across 1,737 household visits) said, unprompted, that taste and smell were obstacles to use.42
The success that blinded studies have had in blinding participants to treatment assignment also suggests that taste and smell are not the overwhelming problems sometimes ascribed to chlorination. Two blinded, placebo-controlled trials with Aquatabs60,62 found no difference between placebo and chlorine arm respondents in their beliefs about their group assignment. Jain et al.60 found that 16% of respondents (n=238, reporting both placebo and chlorine arms together), said the tablets made their water taste better, compared with 2% and 1% reporting bad smell and taste, respectively. However, Boisson et al.62 found higher dissatisfaction with taste and smell among the intervention group compared with the placebo group.
Price of Chlorine POU Products
All but three studies provided the chlorine POU products free to respondents for the study duration, and one study37 reported data from households 6 months following the conclusion of the original study,38 after which households could continue to purchase the product on their own. Blanton et al.58 provided rural Kenyan schoolchildren with free samples of PuR to take home to parents, after which they could repurchase the widely available products in the markets. In rural Haiti, Tsai et al.63 provided half of respondents with a free trial of Klorfasil, a granular chlorine POU product, followed by the opportunity to purchase at a subsidized price. The other half received no free trial. Although over half of total respondents repurchased the product, <30% of 236 respondents at the final follow-up had FCR in stored water. In Mexico, Mellor et al.53 provided 34 households with a free bottle of sodium hypochlorite, with the option to later purchase a 6-month supply for USD $3.14 from a local distributor; half a year later, 65% of 20 available households had FCR >0.2mg/L in stored water. In rural Guatemala, 93% (430/462) of households reported that they would be willing to pay half the market price for PuR (USD $0.14 to treat 10L of water), but only 1.5% (7/462) had FCR in stored water.37
Gender and the Time Cost of Chlorine POU Interventions
The time required to treat the water was the second-most reported reason for nonuse by respondents, and this time burden was usually placed on women. Women were the primary respondents in 36/46 groups (78%), targeted for inclusion as either the primary caretaker of children <5 years of age or the individual in charge of household water management. In one study in rural Afghanistan, intervention messaging was targeted to female caretakers, but nearly half of households allowed only males to participate as respondents.48 In the remaining studies, the gender of respondents was not addressed. In Guatemala, Luby et al. observed very low (1.5% of stored water with FCR 8.5 months after the start of intervention) sustained use of PuR for drinking water treatment, and respondents reported lack of time as one reason for nonuse.37 The authors observed: “Female heads of household already spent substantial time collecting water and on other innumerable household tasks required for family survival in a low-income setting. Using the flocculant–disinfectant required extra steps for water treatment and extra time spent washing the filter cloths.”37
Three studies provided instructions that included time estimates for treatment steps, with ∼2h per week of active time spent stirring and filtering required for median reported product use.44,55,64 This doubles when including wait time required for disinfectant contact. Norton et al.64 provided step-by-step instructions for treatment with flocculant–disinfectant, which included a 5-min stirring step, a 5-min settling step, then filtering through a cloth before letting the filtered water sit for 20 min for disinfectant contact time. Respondents, all women, used a median of 11 (range: 0–48) flocculant–disinfectant sachets per week. Assuming 10 min of active time required, from stirring to filtering, that comes to 110 (range: 0–480) min per week spent actively treating water. Including the 20-min wait time, that increases to 330 (range: 0–1,440) min weekly spent treating and waiting for water before safe use. Rangel et al.44 reported three 30-s stirring and 5-min waiting periods before filtering through a cloth: ∼17 min of active time required. Ninety-four percent (94/100) of respondents were female and the reported median daily household drinking water consumption was 7L. Assuming once daily treatment, with each sachet treating 10L, that comes to 119 min of active time allocated to water treatment per week. Other estimates of daily household water volumes used were much higher. Altmann et al. estimated that families would need to purify 40L of water per day, and the intervention provided sufficient tablets for 3 months of daily treatment of this volume.56 Doocy and Burnham55 estimated 40 min as the total time required to treat water with PuR, including the stirring, filtering, and waiting steps, although this was not reported as too time consuming by participants in an IDP camp.
Institutional Intervention Settings
Although most studies were in households, a handful of interventions introduced in non-household settings achieved high adoption (Figure S5). Three chlorine POU interventions delivered at health facilities in combination with treatment for cholera,40 severe acute malnutrition,56 and pediatric HIV care52 resulted in adoption ranging from 94% to 99% observed at household follow-up visits. George et al.40 did a randomized controlled trial to evaluate a hospital-based intervention that included Aquatabs to reduce the spread of cholera from patients to household members in urban Bangladesh. The intervention included a week of households visits (94% adoption; 308/327 household visits), and data from 6- to 12-months later suggest that the intervention may have increased use of household water treatment overall, even if not specifically of chlorine products. Sugar et al.52 evaluated a program that distributed water storage containers, hypochlorite solution for drinking water treatment, soap, and insecticide-treated bed nets in a program designed to reduce diarrhea and malaria among children living with HIV in peri-urban Mombasa, Kenya. Adoption at household follow-up visits, 97% (1,314/1,350 visits) on average, was assessed by chlorine odor in stored water. Altmann et al.56 did a cluster randomized controlled trial to evaluate the benefits of a water, sanitation, and hygiene package added to clinic-based treatment of severe acute malnutrition in Chad. The 2-month multicomponent intervention included two home visits, and 98%–99% of households (1,373 observations across visits) had FCR ≥0.1mg/L at monthly visits.
One school-based program resulted in a moderate but sustained increase in chlorine POU product use relative to baseline. Blanton et al.58 evaluated a school-based program that delivered drinking water and handwashing infrastructure in schools in rural Kenya, PuR flocculant–disinfectant for drinking water treatment, WaterGuard for handwashing water (which children sometimes drank), and educational comic books and samples of PuR for children to take home to their parents. Use of either WaterGuard or PuR, as confirmed by FCR >0.1mg/L, was 21% (134/644 households) at 4 months and 18% (96/536 households) at the 13-month follow-up. The program did not include regular household visits or products beyond an initial sample of three free sachets of PuR, but it was in a setting with mass media promoting the products.
Humanitarian Intervention Settings
Two included studies reported high adoption in humanitarian settings.55,61 Colindres et al.61 interviewed 100 households that had received free PuR following a 2004 tropical storm in Haiti. Marketing of PuR was through the radio, community demonstrations, and word of mouth from community leaders and neighbors. Although nearly all (97%) of the 100 respondents said that “PuR-treated water appears, tastes, smells, and is healthier,” <25% stated that they would be willing to pay the product’s market price. Doocy and Burnham55 did a 12-wk trial of PuR with a water storage container in an IDP camp in Liberia. Additional free sachets were provided at weekly diarrhea monitoring visits; households were additionally visited weekly for unscheduled water quality testing. Across all 1,551 weekly water testing visit measures, 95% had FCR present, with the lowest adoption in the first week (90%). FCR was ≥0.5mg/L in 85% of visits. The study additionally included focus group discussions with participants, who reported that they preferred the taste of the chlorinated water over untreated water and that they noticed less diarrhea in their households.
Discussion
In this systematic review, we found a wide range in adoption of chlorine POU product use. On average, adoption declined over time, but the relatively short follow-up in most of the included studies limits our understanding of the long-term use of chlorine POU products. Notably, our search strategy selected for closely monitored trials vs. programs. The trials are more likely to have a short duration and intensive promotion, given the resources required for intervention studies. An important implication is that the observed levels of use described here are overestimates of the adoption likely to be observed long-term in programs that cannot continue the contact frequency of high-intensity interventions.
There was no standard definition for adoption of chlorine POU water treatment across the reviewed studies, but the proportion of households with FCR above a threshold on unannounced visits was the reported indicator most likely to capture both correct and consistent use. Although authors did not always explain the choice of FCR threshold that indicated adoption, 0.2 and 0.5mg/L align with widely used drinking water guidelines. World Health Organization Guidelines for Drinking Water recommend that a minimum 0.2mg/L FCR (and maximum 5mg/L) be present at the point of delivery.65 The Sphere Handbook, used in humanitarian response, also recommends FCR ≥0.2–0.5mg/L at the point of delivery for household water,66 although chlorine decay modeling suggests that 0.2mg/L may not be sufficient in refugee camps where high heat causes more rapid chlorine decay and water and sanitation conditions are especially poor.67 Using these guidelines, FCR as a metric of adoption may indicate that water is safely protected in typical settings, but FCR decays over time and thus underestimates usage. The observation by Quick et al.50 that adoption measured by total chlorine was higher than that measured by free chlorine suggests that, although not all households were dosing as instructed, more households may have been consistently using chlorine than would be suggested by FCR testing only.50 Therefore, compared with FCR, total chlorine may be a more appropriate indicator of use. Objective measures such as FCR and total chlorine are preferable to self-reported usage, which is subject to social desirability and courtesy bias and varies widely in its definition across studies (e.g., used “since yesterday”42 to “last two weeks”48). Given that correct and consistent water treatment is required to realize health benefits,10,11 self-reported usage defined as, for example, “during the last two weeks,” may be uninformative. Because of this nonstandardized measurement and reporting, claims of “high” adoption will provide little information without clearly defining the indicators that are used. In addition, when reporting product use, we assert that single time point measurements are more informative than measures pooled over the duration of a study because of the variability in adoption over time and because pooled measures across time points do not allow adoption to be linked to outcomes measured at single time points.
We found a positive association between contact frequency and adoption, suggesting that weekly contact between households and study staff is necessary to sustain high adoption. Studies that contacted participants once per month had a sample-size-weighted median adoption of <50%; adoption dropped off substantially to 11% among those studies with less frequent contact. This finding makes sense in the context of health behavior change theories.14 Each contact with study staff, for any reason, provides households with a reminder or nudge to action,68 increasing the likelihood of habit formation, and this has important implications for health interventions. A recent article that reviewed POU safe water interventions and health impacts found that interventions with demonstrated reductions in diarrheal illness had higher frequency of contact between participants and study staff at levels often considered infeasible at large scales.69 Efforts to replicate and scale household water treatment interventions that have been successful in trials must consider whether they have the field staff resources required to achieve weekly contact with participants; otherwise, our findings suggest that very low adoption levels should be expected.
We also noted higher adoption for tablet products, compared with liquid or flocculant–disinfectant products. Tablets have greater ease of use and convenience compared with liquid products,70 which may require measuring out the correct dose and require more product for dosing use because they are typically diluted to around 1%. Flocculant–disinfectants, because they require separate mixing and filtering steps, also require more effort for use than tablets. In settings where high contact frequency is possible, or in humanitarian, emergency, or outbreak situations, where we found that adoption was typically higher, these results suggest that tablet products may be more effective in achieving high levels of use, compared with liquid or flocculant–disinfectant products.
The evidence to date suggests it is unrealistic to rely solely on household-level treatment to realize the benefits of safe water at the necessary scales. The historical public health benefits of centrally treated piped water1 are often cited as evidence of the importance of safe water interventions. However, in this utility model, in which water is effectively treated at a centralized facility and then distributed through pressurized pipe networks to in-home taps, the responsibilities for correct, consistent, and sustained use are not on individuals in households. The in-effect 100% adoption provided by effective centralized systems contrasts starkly with the adoption observed in real-world evaluations of household water treatment products. At the same time, the infrastructure limitations that first motivated household water treatment approaches are changing. Since 2000, more than 1 billion people have gained access to piped water,3 and passive, in-line chlorination technologies are one example of safe water solutions that are increasingly compatible with this piped infrastructure. In urban Bangladesh, where researchers have generally observed low adoption to chlorine POU product use,42,71 a decentralized, passive, system-level chlorination technology had high acceptability and reduced child diarrhea by nearly 25%.72 This approach is closer to the centralized utility model in that the burden of treatment is not on individuals.
There is some demand for chlorine POU products, however, and it would be a mistake to dismiss the results of household water treatment trials as evidence that household water treatment should never be implemented. Chlorine POU provision at health facilities and in an IDP camp achieved ≥94% adoption, suggesting that settings in which health risks are front-of-mind may motivate increased use of chlorine POU products.40,52,55,56 Even with lower sustained adoption, chlorine POU may still be a worthwhile investment in some settings. Ahuja et al. calculated that a 20%–40% reduction in child diarrhea, on par with pooled effect estimates across studies with <100% adoption,12 makes chlorine POU a cost-effective health intervention. In urban Bangladesh, around half of households continued to use freely provided Aquatabs to treat their water for several months after promotional visits ended although water quality testing continued.59 In Kenya, where mass media promotion of household water treatment was ongoing and products were already widely available in markets, a school-based program to provide targeted education and promotion through students resulted in a sustained, although moderate, increase in use of chlorine POU products.58 Nearly all included studies provided chlorine entirely for free; of the few that followed up with households after encouraging them to purchase products, none found high sustained adoption. This makes sense in the context of demand for similar essential health products, for which demand drastically declines with increasing price and payment does not increase use after purchase.73 In some settings, the level of sustained demand is unclear because households may not have long-term access to the products that are so intensively promoted in shorter-term trials, although there may be other sustained and beneficial changes to household safe water behaviors.39 Factors such as continued communitywide messaging and access to products may be key to sustaining adoption, at least among households who do not view cost as a barrier to use. In rural Haiti, >50% of participants in a long-running, nongovernmental organization-supported safe water enterprise could easily purchase low-cost chlorine and had chlorine residual in stored water, compared with 10% of nonparticipants.74 Although household water treatment with chlorine products may not be a universal solution, it can play an important supporting role in providing safe water in settings where it is promoted and available.
One aspect that remains neglected in chlorine POU evaluations is the gendered time and labor cost of household water treatment. We found that the time required to treat the water was identified as a barrier to water treatment by respondents, the majority of whom were women who were targeted because of their roles as household water managers and primary caretakers of young children. The nonmonetary costs, particularly on mothers, of interventions designed to improve child well-being are often unacknowledged and implicitly set to zero.75 Although the burden of water fetching on women and girls is widely acknowledged and even quantified in global statistics,3 the gendered work of household water treatment receives little attention. In settings with “innumerable household tasks required for family survival,”37 the nonmonetary costs of household water treatment challenge the notion that chlorine POU treatment is simply the cost of a bottle of diluted bleach. The household burdens placed on women and girls in low-income settings are added to the everyday stresses of poverty, described by Mullainathan and Shafir as a “bandwidth tax”76 and further discussed in relation to safe water by Ray and Smith.77 In other words, when daily survival is a struggle, even an extra 30 min a day to chlorinate and wait for water can be burdensome. These are tasks that behavioral economists have alluded to as small hassles, seemingly minor but very real barriers in the everyday lives of the poor.78 These issues are not unique to chlorine POU products—other POU options such as boiling,79 solar disinfection,80 and filters81 all require time and labor for use and maintenance.
There are some limitations to our review. First, our inclusion criteria excluded some studies that are relevant for understanding the use of chlorine POU products, including large-scale programs bundled with antenatal care, disaster relief efforts, and social marketing campaigns (see the “Methods” section). Second, we did not address user preferences for chlorine POU when other POU options are available, nor did we examine (relative) adoption of other POU methods. Burt et al. did not report adoption for individual chlorine products and was therefore excluded from our review, but study respondents ranked and preferred both boiling and pot filters over WaterGuard and PuR, although self-reported adoption of all POU methods was high (average 85% and 91% across two sites).24 Luoto et al. observed very low adoption (<30% self-reported) across all POU products, but use was slightly higher for siphon filters compared with Aquatabs, WaterGuard, and PuR.42 The results from these studies indicate that non-chlorine products may be preferred over chlorine products, when available, but also that if adoption of chlorine POU is very low, adoption of non-chlorine POU is likely to be similar, and vice versa.
Our review has several strengths. First, we designed a broad search strategy to capture the loosely defined construct of adoption of household water treatment. Although high adoption is an important determinant of health impact, it is not measured or reported in any standardized way, in contrast to the increasingly standardized primary health outcomes that are common across these studies. Our approach allowed us to systematically identify available adoption data in the literature. Second, in studies where they were available, we extracted multiple adoption data points and frequency of contact between participants and study staff. This allowed us to observe changing product use over time within studies and to link adoption with intensity of behavior promotion and staff visits. Third, we extracted and emphasized the available data on the gendered burden of POU adoption, showing that so-called low-cost chlorination products are as low cost as they are in part because no value is assigned to intra-household care work.
We were motivated to conduct this systematic review in part because recent large-scale trials that included chlorine POU interventions had small or no effects on child health outcomes that have been linked to safe water consumption. At the same time, the historical public health benefit of chlorinating water supplies is undisputed. A key difference between these two modes of water access is the reliance on systems vs. households to implement the treatment, and although there is a nonzero sustained demand for chlorine POU products, the evidence to date suggests that this approach will not achieve the widespread public health benefits of system-level safe water solutions. Where appropriate infrastructure exists, the safe water community should enhance efforts toward evaluating, implementing, and maintaining system-level treatment options. The effectiveness of chlorination for safe water depends as much on the mode of delivery as it does on the disinfection efficacy of the chlorine itself.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank F. Goddard for his valuable contributions to the review protocol and C.D. Elmera for excellent research assistance.
Y.S.C. was supported by the National Science Foundation Graduate Research Fellowship Program under grant DGE 1752814. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
==== Refs
References
1. Cutler D, Miller G. 2005. The role of public health improvements in health advances: the twentieth-century United States. Demography 42 (1 ):1–22, PMID: , 10.1353/dem.2005.0002.15782893
2. White GC. 2010. White’s Handbook of Chlorination and Alternative Disinfectants. Hoboken, NJ: Wiley.
3. United Nations Children’s Fund, World Health Organization. 2019. Progress on Household Drinking Water, Sanitation and Hygiene 2000–2017: Special Focus on Inequalities. https://apps.who.int/iris/bitstream/handle/10665/329370/9789241516235-eng.pdf [accessed 10 July 2021].
4. Mintz E, Bartram J, Lochery P, Wegelin M. 2001. Not just a drop in the bucket: expanding access to point-of-use water treatment systems. Am J Public Health 91 (10 ):1565–1570, PMID: , 10.2105/ajph.91.10.1565.11574307
5. Schmidt WP, Cairncross S. 2009. Household water treatment in poor populations: is there enough evidence for scaling up now? Environ Sci Technol 43 (4 ):986–992, PMID: , 10.1021/es802232w.19320147
6. Clasen T, Bartram J, Colford J, Luby S, Quick R, Sobsey M. 2009. Comment on “Household water treatment in poor populations: is there enough evidence for scaling up now?” Environ Sci Technol 43 (14 ):5542–5544, PMID: , 10.1021/es9008147.19708394
7. Luby SP, Rahman M, Arnold BF, Unicomb L, Ashraf S, Winch PJ, et al. 2018. Effects of water quality, sanitation, handwashing, and nutritional interventions on diarrhoea and child growth in rural Bangladesh: a cluster randomised controlled trial. Lancet Glob Health 6 (3 ):e302–e315, PMID: , 10.1016/S2214-109X(17)30490-4.29396217
8. Null C, Stewart CP, Pickering AJ, Dentz HN, Arnold BF, Arnold CD, et al. 2018. Effects of water quality, sanitation, handwashing, and nutritional interventions on diarrhoea and child growth in rural Kenya: a cluster-randomised controlled trial. Lancet Glob Health 6 (3 ):e316–e329, PMID: , 10.1016/S2214-109X(18)30005-6.29396219
9. Humphrey JH, Mbuya MNN, Ntozini R, Moulton LH, Stoltzfus RJ, Tavengwa NV, et al. 2019. Independent and combined effects of improved water, sanitation, and hygiene, and improved complementary feeding, on child stunting and anaemia in rural Zimbabwe: a cluster-randomised trial. Lancet Glob Health 7 (1 ):e132–e147, PMID: , 10.1016/S2214-109X(18)30374-7.30554749
10. Brown J, Clasen T. 2012. High adherence is necessary to realize health gains from water quality interventions. PLoS One 7 (5 ):e36735, PMID: , 10.1371/journal.pone.0036735.22586491
11. Enger KS, Nelson KL, Rose JB, Eisenberg JNS. 2013. The joint effects of efficacy and compliance: a study of household water treatment effectiveness against childhood diarrhea. Water Res 47 (3 ):1181–1190, PMID: , 10.1016/j.watres.2012.11.034.23290123
12. Arnold BF, Colford JM Jr. 2007. Treating water with chlorine at point-of-use to improve water quality and reduce child diarrhea in developing countries: a systematic review and meta-analysis. Am J Trop Med Hyg 76 (2 ):354–364, PMID: , 10.4269/ajtmh.2007.76.354.17297049
13. Clasen TF, Alexander KT, Sinclair D, Boisson S, Peletz R, Chang HH, et al. 2015. Interventions to improve water quality for preventing diarrhoea. Cochrane Database Syst Rev 2015 (10 ):CD004794, PMID: , 10.1002/14651858.CD004794.pub3.26488938
14. Dreibelbis R, Winch PJ, Leontsini E, Hulland KR, Ram PK, Unicomb L, et al. 2013. The Integrated Behavioural Model for Water, Sanitation, and Hygiene: a systematic review of behavioural models and a framework for designing and evaluating behaviour change interventions in infrastructure-restricted settings. BMC Public Health 13 (1 ):1015, PMID: , 10.1186/1471-2458-13-1015.24160869
15. Brown J, Hayashi MAL, Eisenberg JNS. 2019. The critical role of compliance in delivering health gains from environmental health interventions. Am J Trop Med Hyg 100 (4 ):777–779, PMID: , 10.4269/ajtmh.18-0624.30675842
16. Fiebelkorn AP, Person B, Quick RE, Vindigni SM, Jhung M, Bowen A, et al. 2012. Systematic review of behavior change research on point-of-use water treatment interventions in countries categorized as low- to medium-development on the human development index. Soc Sci Med 75 (4 ):622–633, PMID: , 10.1016/j.socscimed.2012.02.011.22497845
17. Reygadas F, Gruber JS, Dreizler L, Nelson KL, Ray I. 2018. Measuring user compliance and cost effectiveness of safe drinking water programs: a cluster-randomized study of household ultraviolet disinfection in rural Mexico. Am J Trop Med Hyg 98 (3 ):824–834, PMID: , 10.4269/ajtmh.17-0440.29363448
18. Parvez SM, Azad R, Rahman M, Unicomb L, Ram PK, Naser AM, et al. 2018. Achieving optimal technology and behavioral uptake of single and combined interventions of water, sanitation hygiene and nutrition, in an efficacy trial (WASH benefits) in rural Bangladesh. Trials 19 (1 ):358, PMID: , 10.1186/s13063-018-2710-8.29976251
19. Moher D, Liberati A, Tetzlaff J, Altman DG, PRISMA Group. 2009. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ 339 :b2535, PMID: , 10.1136/bmj.b2535.19622551
20. Wolf J, Hunter PR, Freeman MC, Cumming O, Clasen T, Bartram J, et al. 2018. Impact of drinking water, sanitation and handwashing with soap on childhood diarrhoeal disease: updated meta-analysis and meta-regression. Trop Med Int Health 23 (5 ):508–525, PMID: , 10.1111/tmi.13051.29537671
21. World Bank. 2023. World Bank country and lending groups—country classification. https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups.
22. Fewtrell L, Kaufmann RB, Kay D, Enanoria W, Haller L, Colford JM Jr. 2005. Water, sanitation, and hygiene interventions to reduce diarrhoea in less developed countries: a systematic review and meta-analysis. Lancet Infect Dis 5 (1 ):42–52. [Database], PMID: , 10.1016/S1473-3099(04)01253-8.15620560
23. United Nations General Assembly. 2001. Road Map Towards the Implementation of the United Nations Millennium Declaration: Report of the Secretary-General. A/56/326. New York, NY: United Nations.
24. Burt Z, Njee RM, Mbatia Y, Msimbe V, Brown J, Clasen TF, et al. 2017. User preferences and willingness to pay for safe drinking water: experimental evidence from rural Tanzania. Soc Sci Med 173 :63–71, PMID: , 10.1016/j.socscimed.2016.11.031.27923155
25. Ray I. 2007. Women, water, and development. Annu Rev Environ Resour 32 (1 ):421–449, 10.1146/annurev.energy.32.041806.143704.
26. Loharikar A, Russo E, Sheth A, Menon M, Kudzala A, Tauzie B, et al. 2013. Long-term impact of integration of household water treatment and hygiene promotion with antenatal services on maternal water treatment and hygiene practices in Malawi. Am J Trop Med Hyg 88 (2 ):267–274, PMID: , 10.4269/ajtmh.2012.11-0375.23243106
27. Matanock A, Anderson T, Ayers T, Likicho L, Wamimbi R, Lu X, et al. 2016. Integrating water treatment into antenatal care: impact on use of maternal health services and household water treatment by mothers—rural Uganda, 2013. Am J Trop Med Hyg 94 (5 ):1150–1156, PMID: , 10.4269/ajtmh.15-0356.27001758
28. Sheth AN, Russo ET, Menon M, Wannemuehler K, Weinger M, Kudzala AC, et al. 2010. Impact of the integration of water treatment and handwashing incentives with antenatal services on hygiene practices of pregnant women in Malawi. Am J Trop Med Hyg 83 (6 ):1315–1321, PMID: , 10.4269/ajtmh.2010.10-0211.21118942
29. Lantagne DS, Clasen TF. 2012. Use of household water treatment and safe storage methods in acute emergency response: case study results from Nepal, Indonesia, Kenya, and Haiti. Environ Sci Technol 46 (20 ):11352–11360, PMID: , 10.1021/es301842u.22963031
30. Lantagne D, Clasen T. 2013. Effective use of household water treatment and safe storage in response to the 2010 Haiti earthquake. Am J Trop Med Hyg 89 (3 ):426–433, PMID: , 10.4269/ajtmh.13-0179.23836571
31. Mong Y, Kaiser R, Ibrahim D, Rasoatiana , Razafimbololona L, Quick RE. 2001. Impact of the safe water system on water quality in cyclone-affected communities in Madagascar. Am J Public Health 91 (10 ):1577–1579, PMID: , 10.2105/ajph.91.10.1577.11574310
32. van de Mortel TF. 2008. Faking it: social desirability response bias in self-report research. Aust J Adv Nurs 25 :40–48.http://hdl.handle.net/10072/56416 [accessed 10 July 2021].
33. Arnold BF, Khush RS, Ramaswamy P, Rajkumar P, Durairaj N, Ramaprabha P, et al. 2015. Reactivity in rapidly collected hygiene and toilet spot check measurements: a cautionary note for longitudinal studies. Am J Trop Med Hyg 92 (1 ):159–162, PMID: , 10.4269/ajtmh.14-0306.25385856
34. Adair JG. 1984. The Hawthorne effect: a reconsideration of the methodological artifact. J Appl Psychol 69 (2 ):334–345, 10.1037/0021-9010.69.2.334.
35. Albert J, Luoto J, Levine D. 2010. End-user preferences for and performance of competing POU water treatment technologies among the rural poor of Kenya. Environ Sci Technol 44 (12 ):4426–4432, PMID: , 10.1021/es1000566.20446726
36. Geremew A, Mengistie B, Mellor J, Lantagne DS, Alemayehu E, Sahilu G. 2019. Consistent point-of-use water chlorination among households using unimproved water sources and treatment preference in eastern Ethiopia. Int J Environ Health Res 29 (6 ):686–701, PMID: , 10.1080/09603123.2019.1569210.30694087
37. Luby SP, Mendoza C, Keswick BH, Chiller TM, Hoekstra RM. 2008. Difficulties in bringing point-of-use water treatment to scale in rural Guatemala. Am J Trop Med Hyg 78 (3 ):382–387, PMID: , 10.4269/ajtmh.2008.78.382.18337330
38. Chiller TM, Mendoza CE, Lopez MB, Alvarez M, Hoekstra RM, Keswick BH, et al. 2006. Reducing diarrhoea in Guatemalan children: randomized controlled trial of flocculant–disinfectant for drinking-water. Bull World Health Organ 84 (1 ):28–35, PMID: , 10.2471/blt.04.016980.16501712
39. George CM, Jung DS, Saif-Ur-Rahman KM, Monira S, Sack DA, Rashid MU, et al. 2016. Sustained uptake of a hospital-based handwashing with soap and water treatment intervention (Cholera-Hospital-Based Intervention for 7 days [CHoBI7]): a randomized controlled trial. Am J Trop Med Hyg 94 (2 ):428–436, PMID: , 10.4269/ajtmh.15-0502.26728766
40. George CM, Monira S, Sack DA, Rashid MU, Saif-Ur-Rahman KM, Mahmud T, et al. 2016. Randomized controlled trial of hospital-based hygiene and water treatment intervention (CHoBI7) to reduce cholera. Emerg Infect Dis 22 (2 ):233–241, PMID: , 10.3201/eid2202.151175.26811968
41. Crump JA, Otieno PO, Slutsker L, Keswick BH, Rosen DH, Hoekstra RM, et al. 2005. Household based treatment of drinking water with flocculant-disinfectant for preventing diarrhoea in areas with turbid source water in rural western Kenya: cluster randomised controlled trial. BMJ 331 (7515 ):478, PMID: , 10.1136/bmj.38512.618681.E0.16046440
42. Luoto J, Najnin N, Mahmud M, Albert J, Islam MS, Luby S, et al. 2011. What point-of-use water treatment products do consumers use? Evidence from a randomized controlled trial among the urban poor in Bangladesh. PLoS One 6 (10 ):e26132, PMID: , 10.1371/journal.pone.0026132.22028817
43. Potgieter N, Becker PJ, Ehlers MM. 2009. Evaluation of the CDC safe water-storage intervention to improve the microbiological quality of point-of-use drinking water in rural communities in South Africa. Water SA 35 (4 ), 10.4314/wsa.v35i4.76810.
44. Rangel JM, Lopez B, Mejia MA, Mendoza C, Luby S. 2003. A novel technology to improve drinking water quality: a microbiological evaluation of in-home flocculation and chlorination in rural Guatemala. J Water Health 1 (1 ):15–22, PMID: , 10.2166/wh.2003.0003.15384269
45. Reller ME, Mendoza CE, Lopez MB, Alvarez M, Hoekstra RM, Olson CA, et al. 2003. A randomized controlled trial of household-based flocculant–disinfectant drinking water treatment for diarrhea prevention in rural Guatemala. Am J Trop Med Hyg 69 (4 ):411–419, PMID: , 10.4269/ajtmh.2003.69.411.14640502
46. Shaheed A, Rathore S, Bastable A, Bruce J, Cairncross S, Brown J. 2018. Adherence to point-of-use water treatment over short-term implementation: parallel crossover trials of flocculation–disinfection sachets in Pakistan and Zambia. Environ Sci Technol 52 (11 ):6601–6609, PMID: , 10.1021/acs.est.8b00167.29733647
47. Sobsey MD, Handzel T, Venczel L. 2003. Chlorination and safe storage of household drinking water in developing countries to reduce waterborne disease. Water Sci Technol 47 (3 ):221–228, PMID: , 10.2166/wst.2003.0199.12639033
48. Opryszko MC, Majeed SW, Hansen PM, Myers JA, Baba D, Thompson RE, et al. 2010. Water and hygiene interventions to reduce diarrhoea in rural Afghanistan: a randomized controlled study. J Water Health 8 (4 ):687–702, PMID: , 10.2166/wh.2010.121.20705980
49. Quick RE, Venczel LV, González O, Mintz ED, Highsmith AK, Espada A, et al. 1996. Narrow-mouthed water storage vessels and in situ chlorination in a Bolivian community: a simple method to improve drinking water quality. Am J Trop Med Hyg 54 (5 ):511–516, PMID: , 10.4269/ajtmh.1996.54.511.8644907
50. Quick RE, Kimura A, Thevos A, Tembo M, Shamputa I, Hutwagner L, et al. 2002. Diarrhea prevention through household-level water disinfection and safe storage in Zambia. Am J Trop Med Hyg 66 (5 ):584–589, PMID: , 10.4269/ajtmh.2002.66.584.12201595
51. Quick RE, Venczel LV, Mintz ED, Soleto L, Aparicio J, Gironaz M, et al. 1999. Diarrhoea prevention in Bolivia through point-of-use water treatment and safe storage: a promising new strategy. Epidemiol Infect 122 (1 ):83–90, PMID: , 10.1017/s0950268898001782.10098789
52. Sugar NR, Schilling KA, Kim S, Ahmed A, Ngui Muyanga D, Sivapalasingam S, et al. 2017. Integrating household water treatment, hand washing, and insecticide-treated bed nets into pediatric HIV care in Mombasa, Kenya: impact on diarrhea and malaria risk. J Acquir Immune Defic Syndr 76 (3 ):266–272, PMID: , 10.1097/QAI.0000000000001520.28787328
53. Mellor JE, Kallman E, Oyanedel-Craver V, Smith JA. 2015. Comparison of three household water treatment technologies in San Mateo Ixtatán, Guatemala. J Environ Eng 141 (5 ):04014085, 10.1061/(ASCE)EE.1943-7870.0000914.
54. Clasen T, Saeed TF, Edmondson P, Boisson S, Shipin O. 2007. Household water treatment using sodium dichloroisocy anurate (NaDCC) tablets: a randomized, controlled trial to assess microbiological effectiveness in Bangladesh. Am J Trop Med Hyg 76 (1 ):187–192, PMID: , 10.4269/ajtmh.2007.76.187.17255252
55. Doocy S, Burnham G. 2006. Point-of-use water treatment and diarrhoea reduction in the emergency context: an effectiveness trial in Liberia. Trop Med Int Health 11 (10 ):1542–1552, PMID: , 10.1111/j.1365-3156.2006.01704.x.17002728
56. Altmann M, Altare C, van der Spek N, Barbiche JC, Dodos J, Bechir M, et al. 2018. Effectiveness of a household water, sanitation and hygiene package on an outpatient program for severe acute malnutrition: a pragmatic cluster-randomized controlled trial in Chad. Am J Trop Med Hyg 98 (4 ):1005–1012, PMID: , 10.4269/ajtmh.17-0699.29488461
57. Murray AL, Napotnik JA, Rayner JS, Mendoza A, Mitro B, Norville J, et al. 2020. Evaluation of consistent use, barriers to use, and microbiological effectiveness of three prototype household water treatment technologies in Haiti, Kenya, and Nicaragua. Sci Total Environ 718 :134685, PMID: , 10.1016/j.scitotenv.2019.134685.31839311
58. Blanton E, Ombeki S, Oluoch GO, Mwaki A, Wannemuehler K, Quick R. 2010. Evaluation of the role of school children in the promotion of point-of-use water treatment and handwashing in schools and households—Nyanza province, western Kenya, 2007. Am J Trop Med Hyg 82 (4 ):664–671, PMID: , 10.4269/ajtmh.2010.09-0422.20348516
59. Pickering AJ, Crider Y, Amin N, Bauza V, Unicomb L, Davis J, et al. 2015. Differences in field effectiveness and adoption between a novel automated chlorination system and household manual chlorination of drinking water in Dhaka, Bangladesh: a randomized controlled trial. PLoS One 10 (3 ):e0118397, PMID: , 10.1371/journal.pone.0118397.25734448
60. Jain S, Sahanoon OK, Blanton E, Schmitz A, Wannemuehler KA, Hoekstra RM, et al. 2010. Sodium dichloroisocyanurate tablets for routine treatment of household drinking water in periurban Ghana: a randomized controlled trial. Am J Trop Med Hyg 82 (1 ):16–22, PMID: , 10.4269/ajtmh.2010.08-0584.20064989
61. Colindres RE, Jain S, Bowen A, Mintz E, Domond P. 2007. After the flood: an evaluation of in-home drinking water treatment with combined flocculent-disinfectant following Tropical Storm Jeanne—Gonaives, Haiti, 2004. J Water Health 5 (3 ):367–374, PMID: , 10.2166/wh.2007.032.17878551
62. Boisson S, Stevenson M, Shapiro L, Kumar V, Singh LP, Ward D, et al. 2013. Effect of household-based drinking water chlorination on diarrhoea among children under five in Orissa, India: a double-blind randomised placebo-controlled trial. PLoS Med 10 (8 ):e1001497, PMID: , 10.1371/journal.pmed.1001497.23976883
63. Tsai FJ, Wu M, Lin CP. 2020. Does a free-trial approach increase purchase and use of a household water treatment and safe storage product in rural Haiti? Am J Trop Med Hyg 102 (3 ):518–525, PMID: , 10.4269/ajtmh.19-0606.31971158
64. Norton DM, Rahman M, Shane AL, Hossain Z, Kulick RM, Bhuiyan MI, et al. 2009. Flocculant–disinfectant point-of-use water treatment for reducing arsenic exposure in rural Bangladesh. Int J Environ Health Res 19 (1 ):17–29, PMID: , 10.1080/09603120802272219.19241244
65. World Health Organization. 2017. Guidelines for Drinking Water Quality: Fourth Edition Incorporating the First Addendum. Geneva, Switzerland: World Health Organization.
66. Sphere Association. 2018. The Sphere Handbook: Humanitarian Charter and Minimum Standards in Humanitarian Response. 4th ed. Geneva, Switzerland: Sphere Association. https://spherestandards.org/wp-content/uploads/Sphere-Handbook-2018-EN.pdf [accessed 10 July 2021].
67. Ali SI, Ali SS, Fesselet JF. 2021. Evidence-based chlorination targets for household water safety in humanitarian settings: recommendations from a multi-site study in refugee camps in South Sudan, Jordan, and Rwanda. Water Res 189 :116642, PMID: , 10.1016/j.watres.2020.116642.33246215
68. Thaler RH, Sunstein CR. 2009. Nudge: Improving Decisions about Health, Wealth and Happiness. New York, NY: Penguin Books.
69. Pickering AJ, Null C, Winch PJ, Mangwadu G, Arnold BF, Prendergast AJ, et al. 2019. The WASH benefits and SHINE trials: interpretation of WASH intervention effects on linear growth and diarrhoea. Lancet Glob Health 7 (8 ):e1139–e1146, PMID: , 10.1016/S2214-109X(19)30268-2.31303300
70. Clasen T, Edmondson P. 2006. Sodium dichloroisocyanurate (NaDCC) tablets as an alternative to sodium hypochlorite for the routine treatment of drinking water at the household level. Int J Hyg Environ Health 209 (2 ):173–181, PMID: , 10.1016/j.ijheh.2005.11.004.16387550
71. Najnin N, Leder K, Qadri F, Forbes A, Unicomb L, Winch PJ, et al. 2017. Impact of adding hand-washing and water disinfection promotion to oral cholera vaccination on diarrhoea-associated hospitalization in Dhaka, Bangladesh: evidence from a cluster randomized control trial. Int J Epidemiol 46 (6 ):2056–2066, PMID: , 10.1093/ije/dyx187.29025064
72. Pickering AJ, Crider Y, Sultana S, Swarthout J, Goddard FG, Anjerul Islam S, et al. 2019. Effect of in-line drinking water chlorination at the point of collection on child diarrhoea in urban Bangladesh: a double-blind, cluster-randomised controlled trial. Lancet Glob Health 7 (9 ):e1247–e1256, PMID: , 10.1016/S2214-109X(19)30315-8.31402005
73. Dupas P. 2014. Getting essential health products to their end users: subsidize, but how much? Science 345 (6202 ):1279–1281, PMID: , 10.1126/science.1256973.25214612
74. Harshfield E, Lantagne D, Turbes A, Null C. 2012. Evaluating the sustained health impact of household chlorination of drinking water in rural Haiti. Am J Trop Med Hyg 87 (5 ):786–795, PMID: , 10.4269/ajtmh.2012.12-0010.22987657
75. Evans DK, Jakiela P, Knauer HA. 2021. The impact of early childhood interventions on mothers. Science 372 (6544 ):794–796, PMID: , 10.1126/science.abg0132.34016770
76. Mullainathan S, Shafir E. 2013. Scarcity: Why Having Too Little Means So Much. New York, NY: Henry Holt and Co.
77. Ray I, Smith KR. 2021. Towards safe drinking water and clean cooking for all. Lancet Glob Health 9 (3 ):e361–e365, PMID: , 10.1016/S2214-109X(20)30476-9.33444550
78. Bertrand M, Mullainathan S, Shafir E. 2006. Behavioral economics and marketing in aid of decision making among the poor. J Public Policy Mark 25 (1 ):8–23, 10.1509/jppm.25.1.8.
79. Rosa G, Miller L, Clasen T. 2010. Microbiological effectiveness of disinfecting water by boiling in rural Guatemala. Am J Trop Med Hyg 82 (3 ):473–477, PMID: , 10.4269/ajtmh.2010.09-0320.20207876
80. Rainey RC, Harding AK. 2005. Acceptability of solar disinfection of drinking water treatment in Kathmandu Valley, Nepal. Int J Environ Health Res 15 (5 ):361–372, PMID: , 10.1080/09603120500289168.16416753
81. Pérez-Vidal A, Diaz-Gómez J, Castellanos-Rozo J, Usaquen-Perilla OL. 2016. Long-term evaluation of the performance of four point-of-use water filters. Water Res 98 :176–182, PMID: , 10.1016/j.watres.2016.04.016.27105031
82. Luby S, Agboatwalla M, Raza A, Sobel J, Mintz E, Baier K, et al. 2001. A low-cost intervention for cleaner drinking water in Karachi, Pakistan. Int J Infect Dis 5 (3 ):144–150, PMID: , 10.1016/S1201-9712(01)90089-X.11724671
83. Macy JT, Quick RE. 1998. Evaluation of a novel drinking water treatment and storage intervention in Nicaragua. Rev Panam Salud Publica 3 (2 ):135–136, PMID: , 10.1590/s1020-49891998000200017.9542451
84. Mengistie B, Berhane Y, Worku A. 2013. Household water chlorination reduces incidence of diarrhea among under-five children in rural Ethiopia: a cluster randomized controlled trial. PLoS One 8 (10 ):e77887, PMID: , 10.1371/journal.pone.0077887.24194899
85. Solomon ET, Robele S, Kloos H, Mengistie B. 2020. Effect of household water treatment with chlorine on diarrhea among children under the age of five years in rural areas of Dire Dawa, eastern Ethiopia: a cluster randomized controlled trial. Infect Dis Poverty 9 (1 ):64, PMID: , 10.1186/s40249-020-00680-9.32513277
86. Ercumen A, Naser AM, Unicomb L, Arnold BF, Colford JM Jr, Luby SP. 2015. Effects of source- versus household contamination of tubewell water on child diarrhea in rural Bangladesh: a randomized controlled trial. PLoS One 10 (3 ):e0121907, PMID: , 10.1371/journal.pone.0121907.25816342
|
PMC009xxxxxx/PMC9888235.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12600
10.1289/EHP12600
Response to Letter
Response to “Comment on ‘Maternal Exposure to Per- and Polyfluoroalkyl Substances (PFAS) and Male Reproductive Function in Young Adulthood: Combined Exposure to Seven PFAS’”
https://orcid.org/0000-0002-4250-7948
Hærvig Katia Keglberg 1
Petersen Kajsa Ugelvig 1
Hougaard Karin Sørig 2 3
Lindh Christian 4
Ramlau-Hansen Cecilia Høst 5
Toft Gunnar 6
Giwercman Aleksander 7
Høyer Birgit Bjerre 8
Flachs Esben Meulengracht 1
Bonde Jens Peter 1 2
Tøttenborg Sandra Søgaard 1 2
1 Department of Occupational and Environmental Medicine, Copenhagen University Hospital–Bispebjerg and Frederiksberg, Copenhagen, Denmark
2 Department of Public Health, Faculty of Health Sciences, University of Copenhagen, Copenhagen, Denmark
3 National Research Centre for the Working Environment, Copenhagen, Denmark
4 Division of Occupational and Environmental Medicine, Department of Laboratory Medicine, Lund University, Lund, Sweden
5 Department of Public Health, Research Unit for Epidemiology, Aarhus University, Aarhus, Denmark
6 Steno Diabetes Center Aarhus, Aarhus University Hospital, Aarhus, Denmark
7 Molecular Reproductive Medicine, Department of Translational Medicine, Lund University, Malmo, Sweden
8 Department of Regional Development, Region of Southern Denmark, Vejle, Denmark
Address correspondence to Katia Keglberg Hærvig, Department of Occupational and Environmental Medicine, Copenhagen University Hospital–Bispebjerg and Frederiksberg, Bispebjerg Bakke 23, Entrance 20F, 1st Floor, 2400 Copenhagen NV, Denmark. Telephone 45 27 38 19 62. Email: [email protected]
31 1 2023
1 2023
131 1 01800415 12 2022
23 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
A.G. has received research grants from Ferring Pharmaceuticals and personal fees from Besins Healthcare Nordic and Sandoz, unrelated to the original work that is the subject of this letter. The other authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP12457
==== Body
pmcIn his letter to the editor regarding our paper,1 Abraham raises an important discussion about co-exposure to per- and polyfluoroalkyl substances (PFAS) and other persistent organic pollutants, in particular dioxins.2 In our paper we acknowledged that “there might be factors, such as other chemicals or lifestyle, that might be associated with PFAS exposure which also have an impact on reproductive function.”1 Dioxins potentially constitute such co-exposure, and, if associated with semen quality, the observed inverse associations between PFAS and semen quality could be at least partly due to dioxins.
As we did not assess dioxin levels in our sample, we relied on other studies to elucidate the extend of covariation. Abraham states that he found dioxins significantly correlated with PFAS in blood sampled from 74 mothers 11 months after giving birth in the late 1990s.2 We have not identified other studies reporting correlations between PFAS and dioxins, but correlations between perfluorooctane sulfonic acid (PFOS) and 2,2′,4,4′,5,5′-hexachlorobiphenyl (PCB-153)—a chemical with kinetic properties similar to dioxins, for example, long half-life and accumulation in adipose tissue—were 0.62 in blood of 1,250 Greenlandic, Polish, and Ukrainian pregnant females sampled in 2002–2004.3 Stratification by country generally revealed that correlations were strongest for the Greenlandic sample: 0.52 for PFOS and polychlorinated biphenyl (PCB)-153 vs. 0.20 for Poland and 0.23 for Ukraine.3 Because PFAS bind mainly to proteins, whereas PCBs are lipid-soluble, a high correlation is not expected. We would expect correlations of PFAS and dioxins in the Fetal Programming of Semen Quality cohort (which provided the data used in our study) similar to the PFOS–PCB-153 correlation observed in Ukraine and Poland.
Based on these small-to-moderate correlations, we have little reason to believe that PFAS and dioxins would be highly correlated in our sample. Although we acknowledge the potency of dioxins, the lack of an established association between prenatal dioxin exposure and adult semen quality lessens our concern that our findings could originate from dioxin exposure. In the one study on this topic we were able to identify, Mocarelli et al. compared sperm concentration in sons of women who were exposed to either high levels of dioxins following the 1976 Seveso accident or background dioxin levels.4 Overall, sperm concentration was lower in the 39 sons born to highly exposed mothers, compared with the 58 sons born to less-exposed mothers. However, after stratification on breastfeeding, sperm concentration was lower only in breastfed sons of highly exposed mothers, not in formula-fed sons. This suggests that early postnatal exposure through breastfeeding rather than prenatal exposure drove the association. Furthermore, another study by our group indicated that prenatal exposure to organochlorine compounds, including a subgroup analysis of dioxin-like PCBs, was not associated with semen quality in a cohort of 178 males.5
We urge other investigators to present data on correlations between PFAS and dioxins. We are currently conducting analyses on several groups of persistent and nonpersistent chemicals hoping to further advance our knowledge about health and reproductive effects of multipollutant exposures during fetal life.
==== Refs
References
1. Hærvig KK, Petersen KU, Hougaard KS, Lindh C, Ramlau-Hansen CH, Toft G, et al. 2022. Maternal exposure to per- and polyfluoroalkyl substances (PFAS) and male reproductive function in young adulthood: combined exposure to seven PFAS. Environ Health Perspect 130 (10 ):107001, PMID: , 10.1289/EHP10285.36197086
2. Abraham K. 2023. Comment on “maternal exposure to per- and polyfluoroalkyl substance (PFAS) and male reproductive function in young adulthood: combined exposure to seven PFAS”. Environ Health Perspect 131 (1 ):018003, 10.1289/EHP12457.36719215
3. Lenters V, Portengen L, Smit LAM, Jönsson BAG, Giwercman A, Rylander L, et al. 2015. Phthalates, perfluoroalkyl acids, metals and organochlorines and reproductive function: a multipollutant assessment in Greenlandic, Polish and Ukrainian men. Occup Environ Med 72 (6 ):385–393, PMID: , 10.1136/oemed-2014-102264.25209848
4. Mocarelli P, Gerthoux PM, Needham LL, Patterson DG Jr, Limonta G, Falbo R, et al. 2011. Perinatal exposure to low doses of dioxin can permanently impair human semen quality. Environ Health Perspect 119 (5 ):713–718, PMID: , 10.1289/ehp.1002134.21262597
5. Vested A, Ramlau-Hansen CH, Olsen SF, Bonde JP, Støvring H, Kristensen SL, et al. 2014. In utero exposure to persistent organochlorine pollutants and reproductive health in the human male. Reproduction 148 (6 ):635–646, PMID: , 10.1530/REP-13-0488.25190505
|
PMC009xxxxxx/PMC9888239.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12457
10.1289/EHP12457
Letter to the Editor
Comment on “Maternal Exposure to Per- and Polyfluoroalkyl Substances (PFAS) and Male Reproductive Function in Young Adulthood: Combined Exposure to Seven PFAS”
https://orcid.org/0000-0003-1895-9909
Abraham Klaus 1
1 German Federal Institute for Risk Assessment, Berlin, Germany
Address correspondence to Klaus Abraham, German Federal Institute for Risk Assessment, Max-Dohrn-Str. 8–10, 10589 Berlin, Germany. Email: [email protected]
31 1 2023
1 2023
131 1 01800317 11 2022
23 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The author declares he has no potential conflicts of interest to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP10285
==== Body
pmcHærvig et al. report on inverse associations between maternal exposure to per- and polyfluoroalkyl substances (PFAS) during early pregnancy and parameters of semen quality in young adulthood, with a quite impressive number of participants (n=864).1 However, I was surprised that Hærvig et al.1 did not discuss co-exposures to other persistent organic pollutants (POPs), in particular polychlorinated dibenzo-p-dioxins and dibenzofurans (dioxins), given that a high correlation can be expected between internal dioxin and PFAS levels.
Reduced sperm concentration in Western societies has been discussed for decades, but clear evidence with respect to causal factors is still missing. One of the hypothesized factors is an impact of POPs during critical windows of susceptibility for developmental effects in the perinatal period, for example, on Sertoli cells.2 Dioxins, in particular, have been suspected agents ever since results began emerging from studies of cohorts exposed to very high levels of 2,3,7,8-tetrachlorodibenzo-p-dioxin following the 1976 Seveso accident3,4; a causal relationship is supported by animal data.5 In 2018, the European Food Safety Authority used data from a study of high ambient exposure in a Russian industrial city6 to derive a tolerable weekly intake for dioxins below mean consumption levels observed in several countries.7
Both PFAS and dioxins accumulate in the food chain, and consumption of foods of animal origin especially leads to daily low-level background exposures, resulting in their accumulation in the human body.7,8 Therefore, a high correlation of PFAS and dioxin levels can be expected in the mother, fetus, and newborn, and most especially in the infant at the end of the first year of life, after several months of breastfeeding. However, published data are missing.
In our study9 of mother–child pairs when the children were 11 months of age, Spearman’s rank correlation coefficients with plasma levels of dioxin International Toxicity Equivalents (I-TEq) were calculated to be 0.77 for perfluorooctanoic acid (PFOA) and 0.54 for perfluorooctane sulfonate (PFOS) (n=74, p<0.0001 in both cases). The correlation of levels of PFOA and dioxin I-TEq is displayed in Figure 1. The corresponding coefficients in the mothers were lower, but still highly significant (p<0.0001). Levels of other lipophilic POPs, including polychlorinated biphenyls and legacy organochlorine pesticides, were also found to have a high correlation with levels of PFOA and PFOS.9 Notably, our study included maternal blood samples collected in the late 1990s, similar to the time frame of the samples used by Hærvig et al.1 Due to regulatory measures,10 the patterns and levels of POPs measured in blood change continuously, and it is unclear whether the same correlations would be observed today.
Figure 1. Scatterplot of plasma levels of PFOA and toxic equivalents of dioxins (as I-TEq) in 11-month-old children (n=21 formula-fed, n=53 breast-fed for at least 4 months). Spearman’s rank correlation coefficient r=0.77. Based on data originally reported by Abraham et al.9 Does not include data for 27 children from a dioxin hotspot. Note: I-TEq, International Toxicity Equivalents; PFOA, perfluorooctanoic acid.
Figure 1 is a scatterplot, plotting perfluorooctanoic acid (micrograms per liter), ranging from 0 to 40 in increments of 10 (y-axis) across dioxin International Toxicity Equivalents (picograms per gram fat), ranging from 0 to 60 in increments of 10 (x-axis).
These data reinforce that associations between PFAS levels and parameters of semen quality require careful interpretation with respect to possible causal relationships and the critical window of exposure. The current evidence, including that from animal studies,5 suggests that the observed diminished semen quality may also be attributable to dioxins that co-occur with PFAS, either singly or in the mixture, or may be related to lifestyle factors altogether.11
==== Refs
References
1. Hærvig KK, Petersen KU, Hougaard KS, Lindh C, Ramlau-Hansen CH, Toft G, et al. 2022. Maternal exposure to per- and polyfluoroalkyl substances (PFAS) and male reproductive function in young adulthood: combined exposure to seven PFAS. Environ Health Perspect 130 (10 ):107001, PMID: , 10.1289/EHP10285.36197086
2. Sharpe RM, McKinnell C, Kivlin C, Fisher JS. 2003. Proliferation and functional maturation of Sertoli cells, and their relevance to disorders of testis function in adulthood. Reproduction 125 (6 ):769–784, PMID: , 10.1530/rep.0.1250769.12773099
3. Mocarelli P, Gerthoux PM, Patterson DG Jr, Milani S, Limonta G, Bertona M, et al. 2008. Dioxin exposure, from infancy through puberty, produces endocrine disruption and affects human semen quality. Environ Health Perspect 116 (1 ):70–77, PMID: , 10.1289/ehp.10399.18197302
4. Mocarelli P, Gerthoux PM, Needham LL, Patterson DG Jr, Limonta G, Falbo R, et al. 2011. Perinatal exposure to low doses of dioxin can permanently impair human semen quality. Environ Health Perspect 119 (5 ):713–718, PMID: , 10.1289/ehp.1002134.21262597
5. Faiad W, Soukkarieh C, Murphy DJ, Hanano A. 2022. Effects of dioxins on animal spermatogenesis: a state-of-the-art review. Front Reprod Health 4 :1009090, PMID: , 10.3389/frph.2022.1009090.36339774
6. Mínguez-Alarcón L, Sergeyev O, Burns JS, Williams PL, Lee MM, Korrick SA, et al. 2017. A longitudinal study of peripubertal serum organochlorine concentrations and semen parameters in young men: the Russian Children’s Study. Environ Health Perspect 125 (3 ):460–466, PMID: , 10.1289/EHP25.27713107
7. EFSA CONTAM Panel (EFSA Panel on Contaminants in the Food Chain), Knutsen HK, Alexander J, Barregård L, Bignami M, Brüschweiler B, et al. 2018. Risk for animal and human health related to the presence of dioxins and dioxin-like PCBs in feed and food. EFSA J 16 (11 ):e05333, PMID: , 10.2903/j.efsa.2018.5333.32625737
8. EFSA CONTAM Panel, Schrenk D, Bignami M, Bodin L, Chipman JK, Del Mazo J, et al. 2020. Risk to human health related to the presence of perfluoroalkyl substances in food. EFSA J 18 (9 ):e06223, PMID: , 10.2903/j.efsa.2020.6223.32994824
9. Abraham K, Mielke H, Fromme H, Völkel W, Menzel J, Peiser M, et al. 2020. Internal exposure to perfluoroalkyl substances (PFASs) and biological markers in 101 healthy 1-year-old children: associations between levels of perfluorooctanoic acid (PFOA) and vaccine response. Arch Toxicol 94 (6 ):2131–2147, PMID: , 10.1007/s00204-020-02715-4.32227269
10. Fiedler H, Kallenborn R, de Boer J, Sydnes LK. 2019. The Stockholm Convention: a tool for the global regulation of persistent organic pollutants. Chem Int 41 (2 ):4–11, 10.1515/ci-2019-0202.
11. Leisegang K, Dutta S. 2021. Do lifestyle practices impede male fertility? Andrologia 53 (1 ):e13595, PMID: , 10.1111/and.13595.32330362
|
PMC009xxxxxx/PMC9888258.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36719212
EHP11257
10.1289/EHP11257
Research
Accumulation of Black Carbon Particles in Placenta, Cord Blood, and Childhood Urine in Association with the Intestinal Microbiome Diversity and Composition in Four- to Six-Year-Old Children in the ENVIRONAGE Birth Cohort
https://orcid.org/0000-0003-3607-6376
Van Pee Thessa 1
Hogervorst Janneke 1
Dockx Yinthe 1
Witters Katrien 1
Thijs Sofie 1
Wang Congrong 1
Bongaerts Eva 1
Van Hamme Jonathan D. 2
Vangronsveld Jaco 1 3
Ameloot Marcel 4
Raes Jeroen 5 6
https://orcid.org/0000-0002-3583-3593
Nawrot Tim S. 1 7
1 Centre for Environmental Sciences, Hasselt University, Diepenbeek, Belgium
2 Department of Biological Sciences, Thompson Rivers University, Kamloops, British Columbia, Canada
3 Department of Plant Physiology and Biophysics, Faculty of Biology and Biotechnology, Maria Curie-Skłodowska University, Lublin, Poland
4 Biomedical Research Institute, Hasselt University, Diepenbeek, Belgium
5 Department of Microbiology and Immunology, Rega Instituut, KU Leuven–University of Leuven, Leuven, Belgium
6 Center for Microbiology, VIB, Leuven, Belgium
7 Department of Public Health and Primary Care, Leuven University, Leuven, Belgium
Address correspondence to Tim Nawrot, Agoralaan building D, 3590, Diepenbeek, Belgium. Telephone: +3211268382. Email: [email protected]
31 1 2023
1 2023
131 1 01701015 3 2022
28 11 2022
22 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
The gut microbiome plays an essential role in human health. Despite the link between air pollution exposure and various diseases, its association with the gut microbiome during susceptible life periods remains scarce.
Objectives:
In this study, we examined the association between black carbon particles quantified in prenatal and postnatal biological matrices and bacterial richness and diversity measures, and bacterial families.
Methods:
A total of 85 stool samples were collected from 4- to 6-y-old children enrolled in the ENVIRonmental influence ON early AGEing birth cohort. We performed 16S rRNA gene sequencing to calculate bacterial richness and diversity indices (Chao1 richness, Shannon diversity, and Simpson diversity) and the relative abundance of bacterial families. Black carbon particles were quantified via white light generation under femtosecond pulsed laser illumination in placental tissue and cord blood, employed as prenatal exposure biomarkers, and in urine, used as a post-natal exposure biomarker. We used robust multivariable-adjusted linear models to examine the associations between quantified black carbon loads and measures of richness (Chao1 index) and diversity (Shannon and Simpson indices), adjusting for parity, season of delivery, sequencing batch, age, sex, weight and height of the child, and maternal education. Additionally, we performed a differential relative abundance analysis of bacterial families with a correction for sampling fraction bias. Results are expressed as percentage difference for a doubling in black carbon loads with 95% confidence interval (CI).
Results:
Two diversity indices were negatively associated with placental black carbon [Shannon: −4.38% (95% CI: −8.31%, −0.28%); Simpson: −0.90% (95% CI: −1.76%, −0.04%)], cord blood black carbon [Shannon: −3.38% (95% CI: −5.66%, −0.84%); Simpson: −0.91 (95% CI: −1.66%, −0.16%)], and urinary black carbon [Shannon: −3.39% (95% CI: −5.77%, −0.94%); Simpson: −0.89% (95% CI: −1.37%, −0.40%)]. The explained variance of black carbon on the above indices varied from 6.1% to 16.6%. No statistically significant associations were found between black carbon load and the Chao1 richness index. After multiple testing correction, placental black carbon was negatively associated with relative abundance of the bacterial families Defluviitaleaceae and Marinifilaceae, and urinary black carbon with Christensenellaceae and Coriobacteriaceae; associations with cord blood black carbon were not statistically significant after correction.
Conclusion:
Black carbon particles quantified in prenatal and postnatal biological matrices were associated with the composition and diversity of the childhood intestinal microbiome. These findings address the influential role of exposure to air pollution during pregnancy and early life in human health. https://doi.org/10.1289/EHP11257
Supplemental Material is available online (https://doi.org/10.1289/EHP11257).
M.A. and T.S.N. declare competing financial interests: Aspects of the work mentioned in the paper are the subject of a patent application (Method for detecting and quantifying black carbon particles, US20190025215A1) filed by Hasselt University (Hasselt, Belgium) and KU Leuven (Leuven, Belgium). The remaining authors declare no competing interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Ambient air pollution accounts for over 4.2 million premature deaths each year and is recognized as the most important environmental cause of disease.1 The EU Directive 2008/50/EC states that there is no identifiable threshold for exposure to particulate matter (PM) with an aerodynamic diameter ≤2.5μm (PM2.5) below which it is not harmful to human health.2 One of the most toxic components of PM2.5 is believed to be combustion-derived PM, including black carbon particles, which are formed during incomplete fuel combustion and to which hazardous substances, such as heavy metals and polycyclic aromatic hydrocarbons (PAHs), can bind.3,4 After inhalation, black carbon particles smaller than 1μm can bypass the lung–blood barrier5 and translocate to distal body sites, as substantiated by their presence in urine,6 placental tissue,7 and cord blood.8 Quantified black carbon loads in these biological matrices correlate well with modeled prenatal and postnatal air pollution exposure and are therefore employed as individual internal exposure biomarkers.6–8
The human gut microbiome comprises 10–100 trillion symbiotic microbial cells,9 mainly belonging to the bacterial phyla Firmicutes and Bacteroidetes.10 The gut microbiome evolves during infancy to reach an adultlike state at approximately 3 y of life.11 During early life, bacteria are indispensable for, among other things, shaping the host immune system and mucosal integrity.12 Later on, the microbiome sustains human health via processes such as energy production and guardianship against pathogen colonization.12,13 Therefore, a healthy indigenous bacterial microbiome is essential, and intestinal dysbiosis, i.e., bacterial community composition imbalance, has been implicated in the pathogenesis of several disorders, including diabetes,14 obesity,15 cognitive deficits,16 and hypertension.17 As such, investigating factors that alter intestinal bacterial richness and diversity is paramount. Diet, medication, socioeconomic status, and sex are well-known determinants.18–20 Yet, these factors were calculated to explain in total 16% of the interindividual variation in intestinal bacterial composition, implying that over 80% of the variation remains unexplained.21 These findings suggest that additional factors, e.g., environmental factors like air pollution exposure, might play a role.
Studies addressing the impact of air pollution on the microbiome are scarce. Various animal studies found negative associations between air pollution exposure and the intestinal microbiome as summarized by Vallès et al.22 Human studies exist as well: Mariani et al.23 examined the impact of short-term PM exposure in adults on the nasal microbiome and found inverse associations with alpha diversity indices. A study24 involving 8-y-old primary school children with asthma in China reported that 5-d smog exposure was associated with a decrease in the relative abundance of the fecal bacterial families Bifidobacteriaceae and Erypsipelotrichaceae and an increase in Streptococcaceae, Rikenellaceae, and Porphyromonadaceae. Prior-year residential concentrations of freeway traffic-related air pollution were linked to a decrease in the relative abundance of Bacteroidaceae and an increase in Coriobacteriaceae in the feces of 17- to 19-y-old overweight and obese U.S. adolescents.25 A study26 in type 2 adults with diabetes with an average age of 52 y found negative associations of the prior 2-y average residential PM2.5 and PM10 exposure with alpha diversity indices of fecal microbiota. Together, the above studies suggest an influence of air pollution exposure on the gut bacterial diversity and composition.
Despite the emerging evidence, the effects of air pollution exposure on the gut microbiota in healthy children during susceptible life periods, i.e., fetal development and early childhood, remain uninvestigated. Here, we present a study within the ENVIRONAGE birth cohort framework (ENVIRonmental influence ON early AGEing),27 where we examined the association of air pollution with fecal bacterial richness and diversity, and the relative abundance of bacterial taxa. The primary objective was to investigate whether the placental and cord blood black carbon load (prenatal exposure biomarker) and urinary black carbon load (postnatal exposure biomarker) were associated with fecal bacterial richness and diversity in 4- to 6-y-old children.
Material and Methods
Study Population
The ENVIRONAGE birth cohort recruits mother–newborn pairs at arrival for delivery in the East Limburg Hospital (ZOL; Genk, Belgium) and follows them longitudinally.27 In total, 1,596 mother–child pairs are included in the cohort, and recruitment still continues. Written informed consent is obtained from all participating mothers, and the study is approved by the Ethical Committees of Hasselt University and East-Limburg Hospital (EudraCT B37120107805) and complies with the Helsinki Declaration. At the first antenatal visit, maternal body mass index (BMI) was determined by dividing the measured weight in kilograms by the measured height in square meters. The conception date was estimated based on the first day of the mother’s last menstrual period combined with the first ultrasonographic examination. After delivery, detailed lifestyle and sociodemographic information about the mother and child were gathered via questionnaires (e.g., maternal age and education, parity, descent, smoking habits, and antibiotic use during pregnancy) and medical records (e.g., newborn sex, mode of delivery, and day of delivery). Parity was categorized as mothers having their first, second, or third or more child. Descent was classified as European when two or more grandparents were of European descent. Maternal education was coded as “low” when the mother did not obtain a high school diploma, “middle” when the mother obtained a high school diploma, and “high” when the mother obtained a college or university degree.28 After approximately 4 and 10 y, mother–child pairs are contacted again to participate in the follow-up phase, in which anthropometric, cognitive, and cardiovascular examinations are performed and questionnaires regarding lifestyle, use of medication, and behavior are administered. In addition, a nonquantitative food frequency questionnaire detailing the child’s daily intake of, e.g., fruit, vegetable, and soda consumption over the prior 3 months (never, <1d/wk, 1 d/wk, 2 d/wk, 3–4 d/wk, 5–6/d/wk, one time per day, multiple times per day) is filled in by the mother.
For this study, mother–child pairs were contacted when the child reached the age of 4 to 6 y and were asked to agree to a house visit by a study employee, in which, among other biological samples and measurements, a stool and urine sample were collected from the child. In addition, questionnaire data (e.g., child’s age, antibiotic use in the month before the house visit, in-house smoking, and maternal occupation) was gathered, and the child’s anthropometrics (height and weight) were measured. Maternal occupational levels were coded using the Standard Occupational Classification: sales and customer service occupations, process, plant and machine operatives, and elementary occupations were coded as “low”; administrative and secretarial occupations, skilled trades occupations, and caring, leisure, and other service were coded as “middle”; and managers, directors, senior officials, professional occupations and associate professional and technical occupations were coded as “high.”29 Last, based on the home address of the mothers, median annual neighborhood income was defined using Belgian census-tract data (FOD Economie/DG Statistiek) as previously described.29 Written informed consent was obtained from the parents and oral permission from the child at the start of the house visit. Participant recruitment was carried out in two phases: spring 2017 and spring 2018. For this study, only mother–child pairs who already participated in the 4-y follow-up study up to 1 y before the house visit or who were going to participate within 1 y after the house visit, mother–child pairs who did not (plan to) move between the house visit study and the 4-y follow-up study, and mother–child pairs who had no major renovations planned during the house visit study, were eligible for inclusion. In total, 284 eligible mother–child pairs were identified, of which we succeeded in contacting 233, and 157 agreed to participate in the house visit study, where 96 children provided a stool sample (success rate of stool sample collection was 61.1%). The main reasons that participants did not provide a stool sample were the collection of a stool sample within a limited time frame (only 2 d) and the “yuck factor.” Three stool samples were excluded due to improper storage, four samples because of an insufficient DNA quality, and four due to a too-low number of sequence reads. As a result, the number of included participants amounted to 85, of which 36 (42%) were recruited in 2017 and 49 (58%) in 2018. Among the 85 participants, 63 participants had placental tissue, whereas cord blood and urine were each available from 80 participants (Figure 1). The overlap in availability for the three biological samples is depicted in Figure S1.
Figure 1. Participant flow chart depicting the selection of participants enrolled in the ENVIRONAGE birth cohort for arriving at the final study sample size. Note: Only mother–child pairs who already participated in the 4-y follow-up study up to 1 y before the house visit or who were going to participate within 1 y after the house visit, mother–child pairs who did not (plan to) move between the house visit study and the 4-y follow-up study and who had no major renovations planned during the house visit study were eligible for inclusion in this study. ASV, amplicon sequence variant; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Figure 1 is a flowchart with seven steps. Step 1: There are 1596 mother-child pairs enrolled in the birth cohort, of which 1312 are not eligible. Step 2: There are 284 mother-child pairs, 51 of which are unable to communicate with one another. Step 3: There are 233 pairs, and of those, 76 pairs do not want to participate in the house visit study. Step 4: There are 157 pairs, 66 of which do not have a stool sample. Step 5: There are 91 pairs, 2 of which have poor D-N-A quality. Step 6: There are 89 pairs, 4 of which have an incorrect ASV read number. Step 7: There are 85 pairs available, with 63 pairs of placental tissue, 80 pairs of cord blood, and 80 pairs of urine.
Sample Collection and Processing
At birth, fresh placental tissue was collected within 10 min after delivery. Four biopsies were taken at standardized sites: one in each quadrant of the fetal side across the middle region of the placenta, approximately 4cm away from the umbilical cord and 1cm below the chorion-amniotic membrane to avoid membrane contamination. Biopsies were stored at −80°C until further use.30 Because a previous study showed no differences in black carbon load among the four biopsies in three women in this cohort, only one biopsy was used for further examination within this study.7 For black carbon quantification, frozen biopsies were fixed in 4% formaldehyde on ice at least 24 h before being dehydrated and paraffin-embedded.7 Additionally, 4-μm sections were cut using a microtome (Leica Microsystems) and mounted on histological glass slides. Per biopsy, five slides were prepared. Umbilical cord blood was also gathered within 10 min after delivery in BD Vacutainer plastic whole blood tubes, spray-coated with K2 EDTA (BD) and stored at −25°C.31 For black carbon measurements, cord blood samples were thawed at room temperature, vortexed (VWR International), and 100μL was pipetted into imaging chambers fabricated in-house.8 Two imaging chambers were prepared per participant. Imaging chambers were constructed by placing a glass coverslip (24×24mm; no. 1.5, VWR) on a microscopic glass slide (75×25mm; VWR) merged with 100μm thick double-sided tape (product no. 4959, Tesa SE). The imaging chambers were air-sealed to prevent drying.
Two days before the house visit, the parent(s) collected one stool sample from their child in a designated sterile stool container (VWR) and stored it in their home freezer at −20°C. At the day of the house visit, the parent(s) also collected a urine sample and kept it in the refrigerator. After the examinations, stool and urine samples were taken, transported on ice, and stored at −80°C and −25°C, respectively, until further analysis. For black carbon analysis, urine aliquots were thawed on ice, homogenized for 30 min in a thermomixer (Eppendorf SE) at room temperature, and 100μL was pipetted into the imaging chambers fabricated in-house (described above). Urinary osmolality was determined using 150μL urine and a Knauer Osmometer (K-7400S) (Knauer).
Black Carbon Measurements
Black carbon particles were quantified in placental tissue, cord blood, and urine using white light generation under femtosecond pulsed illumination, allowing label-free detection.32 All images of cord blood were gathered at room temperature using a Zeiss LSM880 NLO scan head mounted to the rear port of an inverted laser-scanning microscope (Zeiss Axio Observer.Z1 motorized stand; Carl Zeiss) equipped with a two-photon femtosecond pulsed laser (810 nm, 120 fs, 80 MHz, MaiTai DeepSee; SpectraPhysics) and an Plan-ApoChromat 20×/0.8 M27 air objective (Carl Zeiss). All images of placental tissue and urine were collected at room temperature using a Zeiss LMS510 META NLO scan head mounted on an inverted laser-scanning microscope (Zeiss Axiovert 200M; Carl Zeiss) equipped with the same ultrashort pulsed laser using a Plan-Neofluar 10×per 0.3 M27 air objective for placental tissue and an EC Plan Neofluar 20×per 0.5 objective for urine (Carl Zeiss). For placental tissue, five 4-μm-thick sections were imaged entirely. The field view of the resulting tile scans ranged between 2,700×2,700μm2 and 4,500×4,500μm2, depending on the size of our tissue. This approach corresponds to either 9 images with a 3,888×3,888 pixel resolution or 25 images with a 6,480×6,480 pixel resolution, both recorded with a 2.51μs pixel dwell time. For cord blood, the resulting tile scans had a field of view of 4,250.96×4,250.96μm2 containing 100 images with a 5,120×5,120 pixel resolution, recorded with a 1.54μs pixel dwell time at three different locations within two imaging chambers. For urine, the resulting tile scans had a field of view of containing 9 images with a 1,536×1,536 pixel resolution and were recorded with a 1,60μs pixel dwell time at five different locations in the imaging chamber. Cord blood and urine images were taken 5μm above the coverslip.
For each image, two emission channels were employed (i.e., 450–650 nm for channel 1, and 400–410 nm for channel 2). In each channel, the number of black carbon particles was calculated using a peak-finding algorithm in MATLAB (MATLAB 2010; MathWorks, Inc.). This program counts pixels above a certain threshold value, i.e., 0.5% and 45% lower than the highest intensity value of channel 1 and channel 2, respectively. The detected pixels in both channels are compared, and only the matching pixels are identified as black carbon particle. For placental tissue, the effectively imaged placental area was determined in the imaging originating from channel 1 using Fiji (ImageJ). Based on MATLAB and Fiji outputs, the total relative number of black carbon particles per cubic nanometer of tissue or mL fluid was defined. We calculated Pearson correlation coefficients to assess the correlation between placental, cord blood, and urinary black carbon loads. In addition, to examine whether the measured black carbon load reflected well the participants’ exposure to ambient airborne black carbon, we also calculated Pearson correlation coefficients between the measured and modeled data. Placental and cord blood black carbon loads were compared with ambient airborne black carbon exposure averaged over pregnancy. Similarly, the urinary black carbon load was correlated to the average airborne black carbon exposure over the month, 6 months, and year preceding the house visit. Modeled air pollution data was generated as follows: residential black carbon exposure levels (micrograms per cubic meter) were interpolated for each mother’s (during pregnancy) and each child’s (during early life) residential address using a spatiotemporal interpolation method that considers land-cover data obtained from satellite images (CORINE land-cover data set) and pollution data from fixed monitoring stations in combination with a dispersion model. This model provides daily interpolated exposure values in a high-resolution receptor grid using data from the Belgian telemetric air quality networks, point sources, and line sources. Overall model performance was evaluated by leave-one-out cross-validation including 14 monitoring points for black carbon, resulting in a spatiotemporal explanatory variance of over 74% in Flanders.33 Daily air pollutants concentrations during pregnancy and the year preceding the house visit, taking into account address changes, were calculated.27
Intestinal Microbiome 16S rRNA V3-V4 Amplicon Sequencing
Stool samples were used as a proxy for the gut microbiome.34 Bacterial DNA was extracted from 200mg of stool employing the E.Z.N.A. Stool DNA Kit (Omega Bio-Tek Inc.) according to the manufacturer’s instructions. The extracted DNA was eluted in an elution buffer (10 mM Tris/HCl, pH 8.5) and stored at −20°C after checking the quantity and quality spectrophotometrically (Nanodrop ND-1000 Spectrophotometer; Isogen Life Sciences). Due to insufficient DNA quality, four samples were omitted from further analysis.
Amplification of 16S rRNA Amplicon and Preparation of 16S Library
The bacterial V3-V4 16S rRNA gene region was amplified using primers that incorporate Ion Torrent sequencing adaptors and Ion Xpress barcodes [amplification PCR: 341F (5′-TAC GGG AGG CAG CAG-3′) and 806R (5′-GGA CTA CVS GGG TAT CTA AT-3′) primers (Alpha DNA); index PCR: sequencing adaptor (underlined, underlined and bold) Ion Xpress barcoded (bold) 341F (5′-CCA TCT CAT CCC TGC GTG TCT CCG ACT CAG CTA AGG TAA CGA TTA CGG GAG GCA GCA G-3′) with P1 (underlined) adapted 806R (5′-CCA CTA CGC CTC CGC TTT CCT CTC TAT GGG CAG TCG GTG ATG GAC TAC VSG GGT ATC TAA T-3′)]. Both amplicon and index amplification were achieved on a T100 Thermal Cycler (Bio-Rad) via the polymerase chain reaction (PCR) programs in the Tables S1, S2, and S3. Amplified products from each round were purified using AMPure XP beads (Beckman Coulter) and a magnetic rack, quantified with the Quant-iT dsDNA HS Assay Kit (Thermo Fisher Scientific), and visualized on agarose gels (1.5% agarose gel, 1.5h, 90V). Barcoded amplicons were pooled in equimolar amounts, and the library dilution factor was determined using an Ion Library Quantitation Kit. An Ion 510 & 520 & 530 Kit-Chef on an Ion Chef system was used for sequencing template preparation, and sequencing was performed on an Ion 530 chip using 400 bp paired-end chemistry.
Sequencing Data Analysis
Sequencing data were received as a set of Ion Torrent-sequenced FASTQ files. Sequences were demultiplexed using the Ion Torrent software, and subsequently underwent quality trimming and primers removing using DADA2 1.10.1.35 Parameters for length trimming were set to keep the first 230 bases of the forward read, maxN=0, MaxEE=(2), trimLeft=15, and truncQ=2. Reads were de-replicated and error rates were inferred using the DADA2 default parameters. Sequence variants were inferred using the adjusted parameters for Ion Torrent-sequences: dada (homopolymer_gap_penalty=−1, band_size=32). After removal of chimeras via the removeBimeraDenovo() function, an amplicon sequence variant (ASV) table was built and taxonomy assigned using the assignTaxonomy function and the SILVA v138 training set,36,37 and alternatively using DECIPHER38 for taxonomic classification with IDTaxa function and the SILVA_SSU_r138_2019 database. The resulting ASVs and taxonomy tables were combined with the metadata file into a phyloseq object (Phyloseq, version 1.26.1).39 Contaminants were removed from the dataset using the package Decontam (version 1.2.1), applying the prevalence method with a 0.5 threshold value.40 Four samples were omitted from the analyses due to an insufficient number of reads (<10,000). Relative taxa abundances at family level were computed by normalizing the number of sequencing reads per ASV for the overall number of sequencing reads per stool sample. The relative abundance of a bacterial family thus represents what percentage (ranging from 0% to 100%) of the microbiome that is made up of that specific family. In a log-log model, a percentage change of, e.g., 235% per doubling in exposure would mean a 2.35-fold increase. Rarefaction analysis was performed with ranacapa 0.1.0.41 Based on the ASV table, alpha diversity was assessed by calculating Chao1 richness index, Shannon diversity index, and Simpson diversity index. Chao1 richness estimates the total richness, i.e., the number of expected species, based on the number of observed species, considering that low abundance species might be missed. In addition, Shannon and Simpson diversity take into account both richness and evenness. The Shannon diversity index focuses most on species richness and reflects the degree of uncertainty in predicting where randomly selected species will belong and ranges from one (single dominant specie) to the total number of all species (all species having equal abundance). A larger value indicates a greater diversity. On the other hand, the Simpson diversity index places greater emphasis on species evenness and ranges between 0 and 1. It reflects the probability that two bacteria randomly selected will belong to different species; hence, a larger value reflects a greater diversity.42
Statistical Analyses
All statistical analyses were performed using R Statistical Software (version 4.0.5; R Foundation for Statistical Computing). Descriptive statistics of the lifestyle characteristics are presented in Table 1 for all included participants (n=85) and compared to the entire birth cohort (Table S4). Continuous variables are expressed as median±interquartile range (IQR) and categorical variables as total number (n) and percentage (%). To improve normality of the distributions, we log-transformed black carbon loads and richness and diversity indices. First, robust linear regression models were fitted between the modeled black carbon exposure data (i.e., exposure during pregnancy and the month, 6 months, and year before stool sample collection) and the Chao1 richness, Shannon diversity and Simpson diversity indices, while accounting for the following covariables based on previous associations between the covariable and either the fecal microbiome or air pollution: parity (first, second, or third and more),43 season of delivery (winter, spring, summer, or autumn),44 sequencing batch (first or second),45 child’s age (continuous),46 sex (male or female),47 weight (continuous),48 height (continuous),49 and maternal education (low, middle, or high)20,50 as a proxy for socioeconomic status.51–53 Afterward, (Partial) Spearman correlations coefficients (r) and coefficients of determination (R2) were calculated to evaluate the correlations between the different richness and diversity indices and black carbon loads while accounting for the same covariables. Subsequently, the standardized R2 values of black carbon loads were compared to the standardized R2 values of other covariables.
Table 1 Anthropometric and lifestyle characteristics of the participating mother–child pairs (n=85) enrolled in the ENVIRONAGE birth cohort.
Characteristics Participants (n=85)
Median±IQR Total number (%)
Child characteristics
Sex — —
Male — 40 (47.1%)
Female — 45 (52.9%)
Age (y) 4.8±0.8 —
Weight (kg) 18.3±3.0 —
Height (cm) 107.0±6.8 —
Descent — —
European — 82 (96.5%)
Non-European — 3 (3.5%)
Gestational duration (d) 280.0±11.0 —
Season of delivery — —
Winter — 23 (27.1%)
Spring — 13 (16.5%)
Summer — 24 (28.2%)
Autumn — 24 (28.2%)
Antibiotic use in the month before sample collection — —
No — 77 (90.6%)
Yes — 8 (9.4%)
In-house smoke exposure — —
No — 83 (97.6%)
Yes — 2 (2.4%)
Vegetable intakea — —
Never — 0 (0%)
<1d/wk — 1 (1.4%)
1 d/wk — 1 (1.4%)
2 d/wk — 1 (1.4%)
3–4 d/wk — 8 (10.8%)
5–6 d/wk — 14 (18.9%)
1 time/d — 40 (54.1%)
Multiple times/d — 9 (12.2%)
Fruit intakea — —
Never — 1 (1.4%)
<1d/wk — 0 (0%)
1 d/wk — 2 (2.7%)
2 d/wk — 2 (2.7%)
3–4 d/wk — 10 (13.5%)
5–6 d/wk — 8 (10.8%)
1 time/d — 23 (31.1%)
Multiple times/d — 28 (37.8%)
Soda intakea — —
Never — 31 (41.9%)
<1d/wk — 15 (20.3%)
1 d/wk — 9 (12.2%)
2 d/wk — 7 (9.5%)
3–4 d/wk — 3 (4.1%)
5–6 d/wk — 1 (1.4%)
1 time/d — 8 (10.8%)
Multiple times/d — 0 (0%)
Mother Characteristics
Age at delivery (y) 30.0±5.0 —
BMI (kg/m2) 22.6±3.7 —
Smoking during pregnancy — —
No — 79 (92.9%)
Yes — 6 (7.1%)
Antibiotic use during pregnancy — —
No — 74 (87.1%)
Yes — 11 (12.9%)
Parity — —
First child — 44 (51.8%)
Second child — 34 (40.0%)
Third or following child — 7 (8.2%)
Education level — —
Low — 2 (2.4%)
Middle — 23 (27.1%)
High — 60 (70.5%)
Occupation level — —
Low — 9 (10.6%)
Middle — 32 (37.6%)
High — 44 (51.8%)
Median annual neighborhood income (Euro) 25,981.4±4,011.8 —
Mode of delivery — —
Vaginal — 84 (98.8%)
Cesarean section — 1 (1.2%)
Note: Continuous covariables are expressed as median±IQR and categorical covariables are described as total number (n) and percentage (%). Maternal educational level was coded “low” if the participant did not obtain a high school diploma, “middle” if the participant obtained a high school diploma, and “high” if the participant obtained a college or university degree. Maternal occupational levels were coded using the Standard Occupational Classification: sales and customer service occupations, process, plant and machine operatives, and elementary occupations were coded “low”; administrative and secretarial occupations, skilled trades occupations and caring, leisure and other service were coded “middle”; and managers, directors, senior officials, professional occupations and associate professional and technical occupations were coded “high.” Descent was based on the native country of the newborn’s grandparents and described as European when two or more grandparents were European, or non-European when at least three grandparents were of non-European origin. —, no data; BMI, body mass index; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort; IQR, interquartile range.
a Data available for 74 participants.
Next, robust multiple linear regression models were fitted to assess the effect size of black carbon loads in placental tissue, cord blood, and urine on the Chao1 richness index, Shannon diversity index, and Simpson diversity index, respectively, while accounting for the same covariables. A multiexposure robust linear regression model incorporating both prenatal (cord blood black carbon) and postnatal (urinary black carbon) exposure was fitted to assess the relative effect size on each of the richness and diversity indices. This multiexposure model was also adjusted for the aforementioned covariables. Cord blood black carbon was employed as prenatal exposure biomarker since cord blood samples were available for a larger number of participants compared to placental tissue. Robust models and partial correlations were used because of the small sample size to reduce the effect of influential cases. Results are presented as a percentage change (%) in index for a doubling in black carbon load. In a sensitivity analysis, we assessed whether the mode of delivery (vaginal or cesarean section),54 smoking during pregnancy (yes or no),55 antibiotic use during pregnancy (yes or no),56 in-house smoking during childhood (yes or no),57 antibiotic use in the child during the month before stool sampling (yes or no),58 descent (European or non-European),59 BMI z-score (continuous) instead of weight and height separately,60 fruit, vegetables, and soda intake as proxies for diet (never, <1d/wk, 1 d/wk, 2 d/wk, 3–4 d/wk, 5–6 d/wk, one time per day, multiple times per day),61 maternal occupation (low, middle, high)26 instead of maternal education, or adjustment for neighborhood income (continuous)62 together with maternal education affected the observed associations between the diversity measures and black carbon loads. Last, raw family counts were used to perform a differential relative abundance analysis at the family level using the “Analysis of Compositions of Microbiomes with Bias Correction” (ANCOM-BC) R package (version 1.0.55).63 Multiple testing was corrected by restricting the false discovery rate as lower than 0.10. All other options remained as default. All reported p-values were two-tailed and a p ≤0.5 was used to define statistical significance.
Results
Population Characteristics
Table 1 shows the anthropometric and lifestyle characteristics of the participating mother–child pairs. In total, 85 children were included, of which almost half were male (47%). On average±IQR, children were 5±1y of age with a mean weight of 18±3kg and mean height of 107±7cm. Almost all of them were of European descent (97%), and half of them were the first-born (52%). Eight children (9%) took antibiotics in the month before stool sample collection, whereas only two (2%) were exposed to in-house smoke. Most children ate vegetables once a day (54%), ate fruit multiple times a day (39%), and never drank soda (42%). The average±IQR maternal age at delivery was 30±5y, with a mean prepregnancy BMI of 23±4 kg/m2. The gestational duration was on average±IQR 280±11d, with 23 children born in winter (27%), 13 children in spring (17%), 24 children in summer (28%), and 24 children in autumn (28%). Only one mother gave birth via a cesarean section (1%). Six mothers (7%) smoked during pregnancy, and 11 mothers (13%) took antibiotics. Most mothers obtained a college or university degree (71%) and had an occupation classified as middle or high using the Standard Occupational Classification (89%). The average median annual neighborhood income was approximately 26,000±4,000 Euro.
Black Carbon Measurements in Biological Matrices and Modeled Values
Black carbon particles were identified in all three biological matrices (Figure 2). The median±IQR loads in placental tissue, cord blood, and urine were 2.25×104±1.25×104 particles per cubic millimeter tissue, 5.80×104±2.82×104 particles per milliliter cord blood, and 1.58×105±1.16×105 particles per milliliter urine, respectively (Table 2). Pearson correlation coefficients were calculated between the black carbon values in the three biological matrices: at birth, placental and cord blood black carbon were positively correlated (r=0.39, p=0.002), whereas no significant correlations were observed between black carbon in placenta or cord blood at birth and urine sampled 4 y later (r=0.23, p=0.10; r=0.09, p=0.44, respectively). In addition to black carbon measurements, modeled air pollution values were also employed. The median±IQR modeled black carbon exposure during pregnancy and the month, 6 months, and year preceding the house visit study were 0.89±0.32 μg/m3, 0.70±0.23 μg/m3, 0.99±0.18 μg/m3, and 0.96±0.20 μg/m3, respectively. We found that the modeled black carbon exposure during pregnancy was significantly correlated with black carbon particles quantified in placental tissue (r=0.48, p<0.0001) and cord blood (r=0.44, p<0.0001). Modeled black carbon exposure in the month, 6 months, and year preceding the house visit study were correlated with black carbon particles quantified in urine samples normalized for osmolality (month: r=0.32, p=0.004; 6 months: r=0.25, p=0.03; year: r=0.18, p=0.11) (Figure 3; Table S5).
Figure 2. Evidence of black carbon particles in (A) placental tissue, (B) cord blood, and (C) urine. White light generation originating from black carbon particles (yellow, indicated with a white arrow) under femtosecond pulsed laser illumination (excitation 810 nm, 120 fs, 80 MHz) was observed. Images represent the overlap of channel 1 (green, emission 450–650 nm) and channel 2 (red, emission 400–410 nm). All samples were collected in the ENVIRONAGE birth cohort and images were randomly selected from different participants. Scale bar: 50μm. Note: ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Figures 2A to 2C are stained tissues, namely, placental tissue, cord blood, and urine, depicting the black carbon particles under femtosecond pulsed laser illumination at 50 micrometers.
Table 2 Detailed information on the distribution of the placental, cord blood, and urinary BC load, quantified via the white light technique.
Placental BC load (n=63) Cord blood BC load (n=80) Urinary BC load (n=80)
Mean 22,488 58,041 177,731
Median 23,162 54,535 157,834
Standard deviation 8,733 25,211 116,419
25th percentile 15,875 42,849 92,171
75th percentile 28,410 71,090 234,356
Minimum value 5,208 11,686 426
Maximum value 44,788 151,919 539,538
Note: Samples were collected in the ENVIRONAGE birth cohort framework: placental tissue and cord blood were collected at birth and urine during the house visit study. Placental BC load is expressed as number of particles per cubic millimeter tissue and cord blood and urinary BC load are expressed as number of particles per milliliter fluid. Urinary black carbon is normalized for osmolality. BC, black carbon; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Figure 3. Correlation graphs between (A) placental black carbon and residential black carbon exposure averaged over the entire pregnancy (n=63), (B) cord blood black carbon and residential black carbon exposure averaged over the entire pregnancy (n=80), and (C), (D), and (E) urinary black carbon normalized for osmolality (n=80) and residential black carbon exposure averaged over the (C) preceding month, (D) preceding 6 months, and (E) preceding year of the house visit. See Table S5 for corresponding numeric data. Participants are enrolled in the ENVIRONAGE birth cohort. Note: ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Figures 3A to 3E are correlation graphs, plotting Black Carbon particles per millimeter cubed placental tissues, ranging from 10000 to 50000 in increments of 10000; Black Carbon particles per millimeter cord blood, ranging from 50000 to 150000 in increments of 50000; Black Carbon particles per millimeter urine normalized for osmolality, ranging from 0 to 600000 in increments of 200000; Black Carbon particles per millimeter urine normalized for osmolality, ranging from 0 to 600000 in increments of 200000; Black Carbon particles per millimeter urine normalized for osmolality, ranging from 0 to 600000 in increments of 200000 (y-axis) across Modeled Black carbon exposure during the entire pregnancy (microgram per meter cubed), ranging from 0.0 to 2.0 in increments of 0.5; Modeled Black carbon exposure during the entire pregnancy (microgram per meter cubed), ranging from 0.0 to 2.0 in increments of 0.5; Modeled Black carbon exposure during the month preceding the house visit (microgram per meter cubed), ranging from 0.0 to 1.5 in increments of 0.5; Modeled Black carbon exposure during the six-months preceding the house visit (microgram per meter cubed), ranging from 0.0 to 2.0 in increments of 0.5; and Modeled Black carbon exposure during the year preceding the house visit (microgram per meter cubed), ranging from 0.0 to 0.5 in increments of 0.5 (x-axis).
Sequencing Data and Alpha Diversity of the Intestinal Microbiome
Sequence data of the 16s rRNA gene hypervariable V3-V4 region were analyzed to obtain a median±IQR of 84,132±95,725 reads per sample and a median±IQR of 168±87 ASVs. Rarefaction analysis indicated that all samples had been sufficiently sequenced (Figure S2). Bacterial relative abundance levels were calculated at the family level (Figure 4). The most dominant bacterial families were Lachnospiraceae (29%), Bacteroidaceae (23%), and Ruminococcaceae (18%), together accounting for approximately 70% of the total bacterial taxa abundance. Lachnospiraceae and Ruminococcaceae belong to the phylum Firmicutes, whereas Bacteroidaceae is part of the phylum Bacteroidetes. Next, fecal microbiome richness and diversity indices were calculated for each sample. The species-richness measure Chao1 had a median value±IQR of 168±87. The species-diversity measures Shannon and Simpson, which combine richness and evenness estimates, had a median value±IQR of 3.8±0.62 and 0.95±0.03, respectively.
Figure 4. Overview of the relative abundance (percentage) of the 10 most abundant bacterial families in relation to all other taxa. Height of bars represents the relative abundance. Families are ranked in increasing order from bottom to top. Participants were enrolled in the ENVIRONAGE birth cohort. n=85. ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Figure 4 is a stacked bar graph titled family relative abundance, plotting percentage, ranging from 0 to 100 percent in increments of 25 (y-axis) across bacterial families (x-axis) for Lachnospiraceae, Bacteroidaceae, Ruminococcaceae, Others, Oscillospiraceae, Prevotellaceae, Rikenellaceae, Veillonellaceae, Bifidobacteriaceae, Eubacterium, and Tannerellaceae.
Association between Black Carbon Exposure/Loads and Intestinal Microbiome Alpha Diversity
Modeled exposure to black carbon particles during the entire pregnancy and the month, 6 months, or year prior to stool sample collection were not associated with the alpha diversity of the intestinal microbiome (Table S6).
Both before and after adjustment for parity, season of delivery, batch, child’s age, sex, weight and height, and maternal education, significant negative correlations were observed between placental, cord blood, and urinary black carbon, and the Shannon and Simpson diversity indices (Figure 5; Table S7). Placental black carbon explained on average 13% of the variation in the Shannon and 17% of the variation in Simpson diversity indices. The explained variance of these indices by the black carbon load of cord blood were 6% and 8%, respectively, and by the urinary black carbon were 10% and 8%, respectively. Moreover, when comparing the standardized R2 values of black carbon loads, we found that black carbon in all three biological matrices explained on average as much variation in the Shannon diversity index as antibiotic use during the previous month or soda intake during the previous 3 months, whereas the explained variance of black carbon on the Simpson diversity index was five times higher in comparison with the same two covariables (Figure 6; Table S8).
Figure 5. (Partial) Spearman correlation coefficients with 95% CI between Chao1 richness index, Shannon diversity index, and Simpson diversity index and placental black carbon load, cord blood black carbon, or urinary black carbon normalized for osmolality, based on corrected (partial/adjusted) and uncorrected models. Partial Spearman and adjusted coefficient of determination models were adjusted for parity, season of delivery, batch, age, sex, weight, height, and maternal education per categories included in Table 1. See Table S7 for corresponding numeric data. Participants were enrolled in the ENVIRONAGE birth cohort. * indicates p ≤0.05. p-Values were calculated using pairwise (partial) Spearman correlation. Placenta n=63, cord blood n=80, and urine n=80. Note: BC, black carbon; CI, confidence interval; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Figure 5 is a set of six error bar graphs titled Placental black carbon, Cord blood black carbon, and Urinary black carbon. Each title has two error bar graphs, plotting (Partial) Spearman correlation coefficients, ranging as negative 0.8 to 0.2 in increments of 0.2 (y-axis) across Chao, Shannon, Simpson (x-axis) for adjusted and unadjusted, respectively.
Figure 6. Percentage of variance (R2) in Shannon or Simpson diversity explained by different covariables. Covariables were categorized as depicted in Table 1. See Table S8 for corresponding numeric data. Participants were enrolled in the ENVIRONAGE birth cohort. Note: ENVIRONAGE: ENVIRonmental influence ON early AGEing birth cohort.
Figure 6 is a horizontal bar graph, plotting Black carbon placenta, Black carbon cord blood, Black carbon urine, Age, Sex, Antibiotic use during the previous month, Maternal education, Soda intake, Fruit intake, Vegetable intake, Ethnicity, and In house smoke exposure (y-axis) across explained variance (percentage), ranging from 0 to 15 in increments of 5 (x-axis) for Simpson diversity and Shannon diversity.
The robust multiple linear regression models confirmed our previous findings and evaluated the effect size of the association between black carbon loads and richness and diversity indices (Table 3). Overall, bacterial diversity indices Shannon and Simpson were inversely associated with both prenatal black carbon exposure (placental and cord blood black carbon) and postnatal exposure (black carbon load of urine). Each doubling in placental black carbon was associated with a 4.38% lower (95% CI: −8.31%, −0.28%; p=0.04) Shannon diversity index and a 0.90% lower (95% CI: −1.76%, −0.04%; p=0.04) Simpson diversity index. Each doubling in cord blood black carbon was associated with a 3.38% lower (95% CI: −5.66%, −0.84%; p=0.05) Shannon index, and a 0.91% lower (95% CI: −1.66%, −0.16%; p=0.02) Simpson diversity index. Last, for each doubling in urinary black carbon, the Shannon diversity index was 3.39% lower (95% CI: −5.77%, −0.94%; p=0.009), and the Simpson diversity index was 0.89% lower (95% CI: −1.37%, −0.40%; p<0.0001). No statistically significant associations were found in black carbon loads with the Chao1 richness index (Table 3). Additionally, the multiexposure model showed that each doubling in cord blood black carbon was associated with a 0.85% lower (95% CI: −1.59%, −0.10%; p=0.03) Simpson diversity index, whereas the association with the Shannon diversity index did not remain statistically significant (−2.61%, 95% CI: −6.17%, 1.10%; p=0.16). On the other hand, each doubling in urinary black carbon was associated with a 3.51% lower (95% CI: −5.95%, −1.00%; p=0.006) Shannon diversity index and a 1.05% lower (95% CI: −1.56%, −0.54%; p<0.0001) Simpson diversity index (Table 4).
Table 3 Overview of the associations between the bacterial Chao1 richness and Shannon and Simpson diversity indices and placental, cord blood, and urinary BC load.
Placental BC (n=63) Cord blood BC (n=80) Urinary BC (n=80)
Percentage change p-Value Percentage change p-Value Percentage change p-Value
Chao1 index −6.63 (−17.02, 5.08) 0.26 0.52 (−9.18, 11.26) 0.92 −2.45 (−8.95, 4.51) 0.48
Shannon index −4.38 (−8.31, −0.28) 0.04 −3.38 (−5.66, −0.84) 0.05 −3.39 (−5.77, −0.94) 0.009
Simpson index −0.90 (−1.76, −0.04) 0.04 −0.91 (−1.66, −0.16) 0.02 −0.89 (−1.37, −0.40) <0.0001
Note: Effects are expressed as percentage change and 95% CI for a doubling in black carbon load. Robust linear regression models were adjusted for parity, season of delivery, sequencing batch, age, sex, weight, and height of the child, and maternal education per categories included in Table 1. Urinary black carbon is normalized for osmolality. Participants were enrolled in the ENVIRONAGE birth cohort. BC, black carbon; CI, confidence interval; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Table 4 Overview of the results of multiexposure models considering the associations between the bacterial Chao1 richness and Shannon and Simpson diversity indices and both the cord blood and urinary BC load.
Cord blood BC (n=76) Urine BC (n=76)
Percentage change p-Value Percentage change p-Value
Chao1 index 1.62 (−8.19, 12.48) 0.76 −1.61 (−8.25, 5.50) 0.65
Shannon index −2.61 (−6.17, 1.10) 0.16 −3.51 (−5.95, −1.00) 0.006
Simpson index −0.85 (−1.59, −0.10) 0.03 −1.05 (−1.56, −0.54) <0.0001
Note: Effects are expressed as percentage change and 95% CI for a doubling in BC load. Robust linear regression models were adjusted for parity, season of delivery, sequencing batch, age, sex, weight and height of the child, and maternal education per categories included in Table 1. Urinary black carbon is normalized for osmolality. In addition, n=76 because either only cord blood or urine was available for four participants. Participants were enrolled in the ENVIRONAGE birth cohort. BC, black carbon; CI, confidence interval; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
In sensitivity analyses, we examined whether the main findings of the robust multiple linear regression models remained after correcting for smoking during pregnancy (n=6); antibiotic use during pregnancy (n=11); antibiotic use 1 month before stool sampling (n=8); BMI z-score instead of weight and height separately; fruit, vegetable, and soda intake (data available for 74 participants); maternal occupation instead of maternal education, or neighborhood income together with maternal education; or excluding children exposed to in-house smoke (n=2), mothers who gave birth via a cesarean section (n=1), or children of non-European descent (n=3). Correction for these variables or exclusion of these participants did not significantly change the effect estimates (Table S9).
Raw family counts were used as input to the ANCOM-BC R package to examine the relationship between black carbon loads and relative abundance at the family level (Table 5). Results are expressed as a percentage change in bacterial family per doubling in black carbon load. Within the model, we accounted for the same covariables that were accounted for in previous models. The associations between placental, cord blood, and urinary black carbon and all bacterial families are depicted in Table S10. After multiple testing correction via the false discovery rate, two bacterial families were inversely associated with placental black carbon, Defluviitaleaceae (−73.7%; q=0.09), Marinifilaceae (−96.9%; q=0.08), and two bacterial families were negatively associated with the urinary black carbon load: Christensenellaceae (−85.8%; q=0.03) and Coriobacteriaceae (−80.7%; q=0.08). The associations found with cord blood black carbon and bacterial families did not survive multiple testing.
Table 5 Results of the relative abundance analysis examining the association between placental, cord blood and urinary BC and bacterial families computed with the ANCOM-BC R package.
Matrix Family p-Value q-Value Percentage difference
Placenta Anaerovoracaceae 0.02 0.29 −45.81
Placenta Christenellaceae 0.04 0.36 −85.18
Placenta Defluviitaleaceae 0.003 0.09 −73.70
Placenta Marinifilaceae 0.001 0.08 −96.92
Placenta Muribaculaceae 0.02 0.29 −96.03
Placenta Oscillospiraceae 0.009 0.21 −52.02
Cord blood Anaerovoraceae 0.05 0.54 −45.81
Cord blood Coriobacteriaceae 0.03 0.58 445.66
Urine Christensenellaceae 0.0005 0.03 −85.84
Urine Coriobacteriaceae 0.0003 0.08 −80.67
Urine Coriobacteriales incertae sedis 0.05 0.42 −48.14
Urine Enterobacteriaceae 0.05 0.42 233.09
Urine Methanobacteriaceae 0.05 0.42 −57.80
Urine Rikenellaceae 0.05 0.42 −59.88
Note: Only bacterial families with a p ≤0.05 are shown. Results are expressed as percentage change per doubling in BC load: for instance, a −45.81% change would mean a 0.46-fold decrease per doubling in BC load. ANCOM-BC log-log models were corrected for parity, season of delivery, batch, age, sex, weight, height of the child, and maternal education per categories included in Table 1. Urinary black carbon was normalized for osmolality. p ≤0.05 and q ≤0.10 are considered statistically significant. Families that remained statistically significant after multiple testing correction via the false discovery rate are indicated in gray. Participants were enrolled in the ENVIRONAGE birth cohort. Placenta n=63, cord blood n=80, and urine n=80. BC, black carbon; ENVIRONAGE, ENVIRonmental influence ON early AGEing birth cohort.
Discussion
The key finding of our study is that the load of black carbon particles in prenatal tissues (placenta and cord blood) and in child urine was associated with lower fecal bacterial diversity and lower relative abundance of specific bacterial taxa in children age 4–6 y. Statistically significant negative correlations and associations were observed between the placental, cord blood, and urinary black carbon loads and the Shannon and Simpson diversity indices. High richness and diversity measures are important indicators of a healthy intestinal microbiome in humans, because a wide array of gut microbes are associated with a more capable and resilient gut microbiome, resulting in an improved health status.53,64,65 If air pollution exposure lowers gut bacterial diversity indices, this exposure could have detrimental effects on gut health. Within this study, fecal samples were used as a proxy for the gut microbiome.34 Our findings might have a public health impact because we found that between 6%–17% of the interindividual variation in intestinal bacterial composition at the species level could be explained by prenatal and postnatal measures of internal black carbon levels. To our knowledge, this is the first study that linked a measure of internal ambient air pollution particles with differences in the childhood gut microbiome. Furthermore, human studies examining the effect of prenatal air pollution exposure on the intestinal microbiome are lacking.
For every doubling in internal black carbon load in placental tissue, cord blood, and urine, the Shannon and Simpson indices decreased approximately 4% and 1%, respectively. Due to the large interindividual variation in black carbon load (e.g., ranging from 5,208 to 44,788 for placental black carbon), multiple doublings are necessary to compare doublings for low- and high-exposed children (e.g., four doublings to go from low- to high-exposed for placental black carbon). Thus, diversity indices differed considerably among low- and high-exposed participants. Additionally, other studies on the association between air pollution exposure and the intestinal microbiome did not find significant associations with bacterial diversity indices.24,25 The findings of our study are in line with the results of a previous study53 examining the effect of previous-year outdoor ozone (O3) exposure on the fecal bacterial richness and diversity in 101 adolescents living in Southern California. They noted negative associations between O3 exposure and bacterial evenness (p<0.001) and the Shannon diversity index (p<0.001) at the species level. They reported that up to 11.2% of the interindividual variation in bacterial composition at the species level could be explained by O3 concentrations. Similarly, negative associations were observed between PM2.5 and PM10 exposure during the preceding 3.5 y of life and intestinal alpha diversity of fecal samples in Chinese adults.26
In the current study, the bacterial families that negatively correlated with placental and urinary black carbon differed from each other. This differential suggests that prenatal and postnatal exposure to black carbon may exert different effects on the gut microbiome. An inverse association was observed between the placental black carbon load and the bacterial families Defluviitaleaceae and Marinifilaceae after correction for multiple testing. Both families have previously been linked to disorders, despite the lack of functional information. For instance, in a study by Liu et al.,66 64 patients with hyperlipidemia, which is characterized by elevated blood cholesterol and triglycerides levels and forms a major risk factor for coronary heart disease, ischemic stroke, and peripheral artery disease, were divided into two groups: group one, in which statin (a cholesterol-lowering medicine) treatment was successful, and group two, in which treatment failed. The relative abundance of the intestinal bacterial family Defluviitaleaceae was higher in men and women belonging to the first group in comparison with men and women in the second group. This finding suggests a modulating effect of Defluviitaleaceae on drug efficiency and accordingly the treatment of hyperlipidemia. Additionally, 41 inactive adults with celiac disease showed enriched intestinal Defluviitaleaceae levels accompanied by a reduction in resting heart rate after a 12-wk intervention with high-intensity interval training and lifestyle education.67 Furthermore, the bacterial family Marinifilaceae was negatively associated with black carbon particles in placental tissue in this study. Marinifilaceae has been indicated as a key actor in gut health by Ge et al.68 Specifically, mice on a high-fat diet received two hypoglycemic compounds to resolve their lipid metabolism disorder. After treatment, a statistically significant increase in the relative abundance of four intestinal bacterial families, including Marinifilaceae, was observed, linking this bacterium to a healthy intestinal flora. In addition, a study of the Cameron County Hispanic Cohort69 (n=217) in South Texas reported that the gut microbiome of Hispanic adults with liver fibrosis (n=28) in comparison with healthy controls was enriched with immunogenic commensals and depleted of, among other bacterial families, Marinifilaceae.
Using urinary black carbon load to reflect childhood exposure, we found that a higher load was associated with a lower relative abundance of the bacterial families Christensenellaceae and Coriobacteriaceae after false discovery rate correction. Both Christensenellaceae and Coriobacteriaceae have been associated with gut health,70 because patients suffering from Crohn’s disease and ulcerative colitis have been reported to harbor significantly lower levels of them.71–73 Coriobacteriaceae maintain host health by assisting in glucose, bile salt, and steroid metabolism and the activation of dietary polyphenols.74–76 A study conducted by Zhao et al.74 reported increased Coriobacteriaceae levels in stool samples in response to physical activity, whereas significantly lower levels were observed in mucosal–luminal interface samples from type two diabetes patients in comparison with healthy controls.76 Christensenellaceae is also involved in metabolic health, as demonstrated by Goodrich et al.77 and Fu et al.78 They found negative correlations with BMI and low-density lipoproteins and positive correlations with high-density lipoproteins. Additionally, Christensenellaceae has been found to be depleted in individuals suffering from metabolic syndrome, characterized by visceral fat, dyslipidemia, impaired glucose metabolism, increased risk for type 2 diabetes, and cardiovascular disease.79 Thus, higher intestinal Christensenellaceae levels have been linked to a lower cardiometabolic risk score.70,79 Together, these studies highlight the importance of a stable, indigenous gut microbiome to maintain host health.
Prenatal and postnatal exposure to black carbon particles may influence intestinal bacterial growth via, among other mechanisms, systemic inflammation. Black carbon exposure has been associated with markers of systemic inflammation, e.g., increased white blood cell count and pro-inflammatory cytokines such as interleukin-6, in both children80 and adults,81 including pregnant women. Subsequently, these pro-inflammatory mediators could impact intestinal bacterial growth, preferentially depleting beneficial gut flora and promoting the growth of otherwise dormant bacteria with potential pathogenic properties.82–84 For instance, male C57BL/6j mice long-term intratracheally instilled with diesel exhaust particles had higher circulating levels of interleukin-1β in serum accompanied by a higher relative abundance of Helicobacteraceae, Campylobacterales, Campylobacteria, Desulfovibrionaceae, Duslfovibrionales, Polyangiaceae, Myxococcales, and Deltaproteobacteria and lower relative abundance of Deferribacteraceae, Deferribacterales, and Deferribacteres.85 In addition, there may be a maternal–fetal efflux of pro-inflammatory mediators,86 and therefore we hypothesize that inflammation could impact the child’s gut microbiome in the womb. Prenatal black carbon exposure might also impact the gut microbiome via other pathways. For a long time, the “sterile womb paradigm” was an accepted dogma, stating that the human body is only colonized with microorganisms during and after birth.87 Depending on the route of delivery, these pioneering microbes are predominantly of vaginal, cutaneous, or oral origin.88,89 If air pollution exposure during pregnancy could influence the maternal vaginal, skin, and/or oral microbiome, this exposure might indirectly impact the infant’s intestinal microbiome. In addition, maternal skin bacteria such as Staphylococcus, Streptococcus, Lactobacillus, and Bifidobacterium might also be transferred during breastfeeding.90 Yet, recent studies challenged this dogma by discovering bacteria in placental tissue, amniotic fluid, and meconium.91–93 This “in utero colonization theory” leaves open the possibility of a maternal–fetal efflux of commensal bacteria,87 providing another framework of how prenatal air pollution exposure could more directly influence an infant’s intestinal microbiome. However, the presence of bacteria in meconium samples is also debated, because studies94,95 reported that the bacteria found in the majority of the “dogma-challenging” studies originated from contamination from lab reagents and the environment. During childhood, black carbon particles may also be transported into the gut after mucociliary clearance of particles from the airways or systemic uptake after inhalation.96–98 Once in the intestines, PM may alter bacterial growth by various mechanisms, such as gut inflammation, disruption of tight junction proteins, and oxidative stress.99–102 The majority of these mechanisms have been exclusively investigated in in vitro studies and in vivo animal models. For instance, Mutlu et al. showed that exposure to high doses of urban airborne PM was associated with a) oxidant-dependent gastrointestinal epithelial cell death, b) disruption of the tight junction protein Zonula occludens-1, c) an increase in the inflammatory markers interleukin-6, nuclear factor-kappa B, and tumor necrosis factor α, and d) an increase in gut permeability in in vitro (Caco-2 intestinal epithelial barrier model) and in vivo animal (male C57BL/6 mice) models.100,101 In addition, 126/SvEv mice gavaged with PM10 particles for 2 wk showed increased pro-inflammatory cytokine levels in intestinal tissue.99 As mentioned previously, black carbon particles are formed during the incomplete combustion of carbonaceous fuels,3,4 during which hazardous substances such as PAHs, including benzo(a)pyrene (BaP), can adhere to their surface. Besides a possible mutagenic and carcinogenic potential, multiple studies also reported a pro-inflammatory potential for BaP, potentially affecting intestinal bacterial growth.103,104 For instance, Khalil et al. demonstrated a significant increase in murine intestinal inflammation in association with a high-fat diet following BaP exposure.104 Additionally, C57BL/6 mice orally exposed to BaP showed a significant increment in inflammatory cell proliferation and crypt damage, leading to decreased levels of the intestinal bacterial genus Lactobacillus.103
To our knowledge, this study is the first to investigate the correlation between airborne particulate matter exposure of healthy young children during susceptible life periods and their gut bacterial richness and diversity of specific bacterial taxa. Other epidemiological studies examining the association between air pollution and the intestinal microbiome so far made use of spatial temporal modeled data.24–53 These air pollution models only consider exposure at the residential address, implying that individual and time-activity mobility patterns, e.g., commuting, hobbies, and work, are not considered. Consequently, exposure misclassification and a potential underestimation of the health risk can occur.6 As seen in our study, modeled black carbon exposure during pregnancy and the month, 6 months, or year before stool sample collection was not associated with the bacterial richness and diversity of the intestinal microbiome. To overcome these shortcomings, we employed the white light technique to assess individual and internal black carbon loads. The implication of this technique in population-based research is unique, allows the determination of a precise personal exposure measurement, and enables direct linkage with observed gut microbiome changes.32
We acknowledge some study limitations. First, only 85 children were included in the present study, resulting in a small sample size. Despite the small sample size, we were able to find statistically significant correlations and associations with diversity indices and the relative abundance of specific bacterial families. Due to the limited sample size of our study, our study population might not be completely representative for the general population: e.g., our study included only one mother giving birth via a cesarean section and had a slightly lower percentage of children of non-European descent (3.5% vs. 10.2%) and of mothers with low education levels (2.4% vs. 10.6%) in comparison with the entire birth cohort (Table S4). Moreover, the small sample size significantly limited our ability to test the influence of certain factors (e.g., cesarean section and descent) on the intestinal microbiome. Second, limited information regarding diet was available. Yet, we adjusted the robust linear regression models for self-reported fruit, vegetable, and soda intake, which did not significantly change our findings. Future population-based research is necessary to examine potential pathways (e.g., systemic oxidative stress and gut inflammation) that might underlie the observed associations between air pollution exposure and the intestinal microbiome because for now only in vivo animal and in vitro studies exist. In addition, metagenomic shotgun and pathway analyses could be performed to acquire more in-depth information on the microbiome composition and functionality.
Conclusion
Higher accumulation of black carbon particles in placental tissue, cord blood, and urine as internal biomarkers of prenatal and postnatal combustion-related air pollution exposure was associated with changes in intestinal bacterial diversity in young children. Black carbon loads in placental tissue were negatively associated with Defluviitaleaceae and Marinifilaceae, whereas urinary black carbon was negatively associated with Christensenellaceae and Coriobacteriaceae. These findings address the influential role of exposure to air pollution during pregnancy and early life in human health. Future studies are necessary to examine the mechanisms underlying the observed associations.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank all children and parents for taking part in the study and inviting them into their homes. The authors thank J.D.V.H. and B. McAmmond for sequencing data. This investigation is supported by the European Union research council Project ENVIRONAGE (ERC-2012-StG 310,890), Flemish Scientific Fund (G073315N/G048420N), and Methusalem T.V.P. holds a doctoral fellowship of the Research Foundation Flanders (FWO), grant number: 11C7421N.
==== Refs
References
1. Landrigan PJ, Fuller R, Acosta NJR, Adeyi O, Arnold R, Basu NN, et al. 2018. The Lancet Commission on pollution and health. Lancet 391 (10119 ):462–512, PMID: , 10.1016/S0140-6736(17)32345-0.29056410
2. European Union. 2008. Directive 2008/50/EC of the European Parliament and of the Council of 21 May 2008 on Ambient Air Quality and Cleaner Air for Europe, 169–212. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32008L0050&from=en [accessed 22 October 2021].
3. Janssen NA, Hoek G, Simic-Lawson M, Fischer P, van Bree L, ten Brink H, et al. 2011. Black carbon as an additional indicator of the adverse health effects of airborne particles compared with PM10 and PM2.5. Environ Health Perspect 119 (12 ):1691–1699, PMID: , 10.1289/ehp.1003369.21810552
4. Niranjan R, Thakur AK. 2017. The toxicological mechanisms of environmental soot (black carbon) and carbon black: focus on oxidative stress and inflammatory pathways. Front Immunol 8 :763, PMID: , 10.3389/fimmu.2017.00763.28713383
5. Chow JC, Watson JG, Mauderly JL, Costa DL, Wyzga RE, Vedal S, et al. 2006. Health effects of fine particulate air pollution: lines that connect. J Air Waste Manag Assoc 56 (10 ):1368–1380, PMID: , 10.1080/10473289.2006.10464545.17063860
6. Saenen ND, Bové H, Steuwe C, Roeffaers MBJ, Provost EB, Lefebvre W, et al. 2017. Children’s urinary environmental carbon load. A novel marker reflecting residential ambient air pollution exposure? Am J Respir Crit Care Med 196 (7 ):873–881, PMID: , 10.1164/rccm.201704-0797OC.28686472
7. Bove H, Bongaerts E, Slenders E, Bijnens EM, Saenen ND, Gyselaers W, et al. 2019. Ambient black carbon particles reach the fetal side of human placenta. Nat Commun 10 (1 ):3866, PMID: , 10.1038/s41467-019-11654-3.31530803
8. Bongaerts E, Lecante LL, Bove H, Roeffaers MBJ, Ameloot M, Fowler PA, et al. 2022. Maternal exposure to ambient black carbon particles and their presence in maternal and fetal circulation and organs: an analysis of two independent population-based observational studies. Lancet Planet Health 6 (10 ):e804–e811, PMID: , 10.1016/S2542-5196(22)00200-5.36208643
9. Ursell LK, Metcalf JL, Parfrey LW, Knight R. 2012. Defining the human microbiome. Nutr Rev 70 (suppl 1 ):S38–S44, PMID: , 10.1111/j.1753-4887.2012.00493.x.22861806
10. Mahowald MA, Rey FE, Seedorf H, Turnbaugh PJ, Fulton RS, Wollam A, et al. 2009. Characterizing a model human gut microbiota composed of members of its two dominant bacterial phyla. Proc Natl Acad Sci USA 106 (14 ):5859–5864, PMID: , 10.1073/pnas.0901529106.19321416
11. Derrien M, Alvarez AS, de Vos WM. 2019. The gut microbiota in the first decade of life. Trends Microbiol 27 (12 ):997–1010, PMID: , 10.1016/j.tim.2019.08.001.31474424
12. Ventura M, O’Toole PW, de Vos WM, van Sinderen D. 2018. Selected aspects of the human gut microbiota. Cell Mol Life Sci 75 (1 ):81–82, PMID: , 10.1007/s00018-017-2669-8.28986602
13. Jandhyala SM, Talukdar R, Subramanyam C, Vuyyuru H, Sasikala M, Reddy DN. 2015. Role of the normal gut microbiota. World J Gastroenterol 21 (29 ):8787–8803, PMID: , 10.3748/wjg.v21.i29.8787.26269668
14. Abdellatif AM, Sarvetnick NE. 2019. Current understanding of the role of gut dysbiosis in type 1 diabetes. J Diabetes 11 (8 ):632–644, PMID: , 10.1111/1753-0407.12915.30864231
15. Amabebe E, Robert FO, Agbalalah T, Orubu ESF. 2020. Microbial dysbiosis-induced obesity: role of gut microbiota in homoeostasis of energy metabolism. Br J Nutr 123 (10 ):1127–1137, PMID: , 10.1017/S0007114520000380.32008579
16. Proctor C, Thiennimitr P, Chattipakorn N, Chattipakorn SC. 2017. Diet, gut microbiota and cognition. Metab Brain Dis 32 (1 ):1–17, PMID: , 10.1007/s11011-016-9917-8.27709426
17. Yang T, Santisteban MM, Rodriguez V, Li E, Ahmari N, Carvajal JM, et al. 2015. Gut dysbiosis is linked to hypertension. Hypertension 65 (6 ):1331–1340, PMID: , 10.1161/HYPERTENSIONAHA.115.05315.25870193
18. Bajinka O, Tan Y, Abdelhalim KA, Özdemir G, Qiu X. 2020. Extrinsic factors influencing gut microbes, the immediate consequences and restoring eubiosis. AMB Express 10 (1 ):130, PMID: , 10.1186/s13568-020-01066-8.32710186
19. Haro C, Rangel-Zúñiga OA, Alcalá-Díaz JF, Gómez-Delgado F, Pérez-Martínez P, Delgado-Lista J, et al. 2016. Intestinal microbiota is influenced by gender and body mass index. PLoS One 11 (5 ):e0154090, PMID: , 10.1371/journal.pone.0154090.27228093
20. Lewis CR, Bonham KS, McCann SH, Volpe AR, D’Sa V, Naymik M, et al. 2021. Family SES is associated with the gut microbiome in infants and children. Microorganisms 9 (8 ):1608, PMID: , 10.3390/microorganisms9081608.34442687
21. Falony G, Joossens M, Vieira-Silva S, Wang J, Darzi Y, Faust K, et al. 2016. Population-level analysis of gut microbiome variation. Science 352 (6285 ):560–564, PMID: , 10.1126/science.aad3503.27126039
22. Valles Y, Francino MP. 2018. Air pollution, early life microbiome, and development. Curr Environ Health Rep 5 (4 ):512–521, PMID: , 10.1007/s40572-018-0215-y.30269309
23. Mariani J, Favero C, Spinazze A, Cavallo DM, Carugno M, Motta V, et al. 2018. Short-term particulate matter exposure influences nasal microbiota in a population of healthy subjects. Environ Res 162 :119–126, PMID: , 10.1016/j.envres.2017.12.016.29291434
24. Zheng P, Zhang B, Zhang K, Lv X, Wang Q, Bai X. 2020. The Impact of air pollution on intestinal microbiome of asthmatic children: a panel study. Biomed Res Int 2020 :5753427, PMID: , 10.1155/2020/5753427.33204702
25. Alderete TL, Jones RB, Chen Z, Kim JS, Habre R, Lurmann F, et al. 2018. Exposure to traffic-related air pollution and the composition of the gut microbiota in overweight and obese adolescents. Environ Res 161 :472–478, PMID: , 10.1016/j.envres.2017.11.046.29220800
26. Liu T, Chen X, Xu Y, Wu W, Tang W, Chen Z, et al. 2019. Gut microbiota partially mediates the effects of fine particulate matter on type 2 diabetes: evidence from a population-based epidemiological study. Environ Int 130 :104882, PMID: , 10.1016/j.envint.2019.05.076.31202028
27. Janssen BG, Madhloum N, Gyselaers W, Bijnens E, Clemente DB, Cox B, et al. 2017. Cohort Profile: the ENVIRonmental Influence on Early AGEing (ENVIRONAGE): a birth cohort study. Int J Epidemiol 46 (5 ):1386–1387, PMID: , 10.1093/ije/dyw269.28089960
28. Martens DS, Cox B, Janssen BG, Clemente DBP, Gasparrini A, Vanpoucke C, et al. 2017. Prenatal air pollution and newborns’ predisposition to accelerated biological aging. JAMA Pediatr 171 (12 ):1160–1167, PMID: , 10.1001/jamapediatrics.2017.3024.29049509
29. Martens DS, Janssen BG, Bijnens EM, Clemente DBP, Vineis P, Plusquin M, et al. 2020. Association of parental socioeconomic status and newborn telomere length. JAMA Netw Open 3 (5 ):e204057, PMID: , 10.1001/jamanetworkopen.2020.4057.32364595
30. Saenen ND, Vrijens K, Janssen BG, Roels HA, Neven KY, Vanden Berghe W, et al. 2017. Lower Placental leptin promoter methylation in association with fine particulate matter air pollution during pregnancy and placental nitrosative stress at birth in the ENVIRONAGE cohort. Environ Health Perspect 125 (2 ):262–268, PMID: , 10.1289/EHP38.27623604
31. Martens DS, Gouveia S, Madhloum N, Janssen BG, Plusquin M, Vanpoucke C, et al. 2017. Neonatal cord blood oxylipins and exposure to particulate matter in the early-life environment: an ENVIRONAGE birth cohort study. Environ Health Perspect 125 (4 ):691–698, PMID: , 10.1289/EHP291.27814242
32. Bove H, Steuwe C, Fron E, Slenders E, D’Haen J, Fujita Y, et al. 2016. Biocompatible label-free detection of carbon black particles by femtosecond pulsed laser microscopy. Nano Lett 16 (5 ):3173–3178, PMID: , 10.1021/acs.nanolett.6b00502.27104759
33. Lefebvre W, Vercauteren J, Schrooten L, Janssen S, Degraeuwe B, Maenhaut W, et al. 2011. Validation of the MIMOSA-AURORA-IFDM model chain for policy support: modeling concentrations of elemental carbon in Flanders. Atmos Environ 45 (37 ):6705–6713, 10.1016/j.atmosenv.2011.08.033.
34. Thomas V, Clark J, Dore J. 2015. Fecal microbiota analysis: an overview of sample collection methods and sequencing strategies. Future Microbiol 10 (9 ):1485–504, PMID: , 10.2217/fmb.15.87.26347019
35. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP. 2016. DADA2: high-resolution sample inference from Illumina amplicon data. Nat Methods 13 (7 ):581–583, PMID: , 10.1038/nmeth.3869.27214047
36. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P, et al. 2013. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res 41 (Database issue ):D590–D596, PMID: , 10.1093/nar/gks1219.23193283
37. Yilmaz P, Parfrey LW, Yarza P, Gerken J, Pruesse E, Quast C, et al. 2014. The SILVA and “all-species living tree project (LTP)” taxonomic frameworks. Nucleic Acids Res 42 (Database issue ):D643–D648, PMID: , 10.1093/nar/gkt1209.24293649
38. Wright ES. 2016. Using DECIPHER v2.0 to analyze big biological sequence data in R. R J 8 (1 ):352–359, 10.32614/RJ-2016-025.
39. McMurdie PJ, Holmes S. 2013. Phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One 8 (4 ):e61217, PMID: , 10.1371/journal.pone.0061217.23630581
40. Davis NM, Proctor DM, Holmes SP, Relman DA, Callahan BJ. 2018. Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data. Microbiome 6 (1 ):226, PMID: , 10.1186/s40168-018-0605-2.30558668
41. Kandlikar GS, Gold ZJ, Cowen MC, Meyer RS, Freise AC, Kraft NJB, et al. 2018. Ranacapa: an R package and Shiny web app to explore environmental DNA data with exploratory statistics and interactive visualizations. F1000Res 7 :1734, PMID: , 10.12688/f1000research.16680.1.30613396
42. Kim BR, Shin J, Guevarra R, Lee JH, Kim DW, Seol KH, et al. 2017. Deciphering diversity indices for a better understanding of microbial communities. J Microbiol Biotechnol 27 (12 ):2089–2093, PMID: , 10.4014/jmb.1709.09027.29032640
43. Berry ASF, Pierdon MK, Misic AM, Sullivan MC, O’Brien K, Chen Y, et al. 2021. Remodeling of the maternal gut microbiome during pregnancy is shaped by parity. Microbiome 9 (1 ):146, PMID: , 10.1186/s40168-021-01089-8.34176489
44. Chen R, Peng RD, Meng X, Zhou Z, Chen B, Kan H. 2013. Seasonal variation in the acute effect of particulate air pollution on mortality in the China Air Pollution and Health Effects Study (CAPES). Sci Total Environ 450–451 :259–265, PMID: , 10.1016/j.scitotenv.2013.02.040.23500824
45. Wang Y, LêCao KA. 2020. Managing batch effects in microbiome data. Brief Bioinform 21 (6 ):1954–1970, PMID: , 10.1093/bib/bbz105.31776547
46. Bosco N, Noti M. 2021. The aging gut microbiome and its impact on host immunity. Genes Immun 22 (5–6 ):289–303, PMID: , 10.1038/s41435-021-00126-8.33875817
47. Valeri F, Endres K. 2021. How biological sex of the host shapes its gut microbiota. Front Neuroendocrinol 61 :100912, PMID: , 10.1016/j.yfrne.2021.100912.33713673
48. Sepp E, Loivukene K, Julge K, Voor T, Mikelsaar M. 2013. The association of gut microbiota with body weight and body mass index in preschool children of Estonia. Microb Ecol Health Dis 24 (1 ), PMID: , 10.3402/mehd.v24i0.19231.24009544
49. Manor O, Dai CL, Kornilov SA, Smith B, Price ND, Lovejoy JC, et al. 2020. Health and disease markers correlate with gut microbiome composition across thousands of people. Nat Commun 11 (1 ):5206, PMID: , 10.1038/s41467-020-18871-1.33060586
50. Wang Y, Wang Y, Xu H, Zhao Y, Marshall JD. 2022. Ambient air pollution and socioeconomic status in China. Environ Health Perspect 130 (6 ):67001, PMID: , 10.1289/EHP9872.35674427
51. Ballon M, Botton J, Charles MA, Carles S, de Lauzon-Guillain B, Forhan A, et al. 2018. Socioeconomic inequalities in weight, height and body mass index from birth to 5 years. Int J Obes (Lond) 42 (9 ):1671–1679, PMID: , 10.1038/s41366-018-0180-4.30120430
52. Sherar LB, Griffin TP, Ekelund U, Cooper AR, Esliger DW, van Sluijs EM, et al. 2016. Association between maternal education and objectively measured physical activity and sedentary time in adolescents. J Epidemiol Community Health 70 (6 ):541–548, PMID: , 10.1136/jech-2015-205763.26802168
53. Fouladi F, Bailey MJ, Patterson WB, Sioda M, Blakley IC, Fodor AA, et al. 2020. Air pollution exposure is associated with the gut microbiome as revealed by shotgun metagenomic sequencing. Environ Int 138 :105604, PMID: , 10.1016/j.envint.2020.105604.32135388
54. Reyman M, van Houten MA, van Baarle D, Bosch A, Man WH, Chu M, et al. 2019. Impact of delivery mode-associated gut microbiota dynamics on health in the first year of life. Nat Commun 10 (1 ):4997, PMID: , 10.1038/s41467-019-13014-7.31676793
55. McLean C, Jun S, Kozyrskyj A. 2019. Impact of maternal smoking on the infant gut microbiota and its association with child overweight: a scoping review. World J Pediatr 15 (4 ):341–349, PMID: , 10.1007/s12519-019-00278-8.31290060
56. Zou ZH, Liu D, Li HD, Zhu DP, He Y, Hou T, et al. 2018. Prenatal and postnatal antibiotic exposure influences the gut microbiota of preterm infants in neonatal intensive care units. Ann Clin Microbiol Antimicrob 17 (1 ):9, PMID: , 10.1186/s12941-018-0264-y.29554907
57. Berkowitz L, Pardo-Roa C, Salazar GA, Salazar-Echegarai F, Miranda JP, Ramirez G, et al. 2019. Mucosal exposure to cigarette components induces intestinal inflammation and alters antimicrobial response in mice. Front Immunol 10 :2289, PMID: , 10.3389/fimmu.2019.02289.31608070
58. Ramirez J, Guarner F, Fernandez LB, Maruy A, Sdepanian VL, Cohen H. 2020. Antibiotics as major disruptors of gut microbiota. Front Cell Infect Microbiol 10 :572912, PMID: , 10.3389/fcimb.2020.572912.33330122
59. Dwiyanto J, Hussain MH, Reidpath D, Ong KS, Qasim A, Lee SWH, et al. 2021. Ethnicity influences the gut microbiota of individuals sharing a geographical location: a cross-sectional study from a middle-income country. Sci Rep 11 (1 ):2618, PMID: , 10.1038/s41598-021-82311-3.33514807
60. Thompson AL. 2012. Developmental origins of obesity: early feeding environments, infant growth, and the intestinal microbiome. Am J Hum Biol 24 (3 ):350–360, PMID: , 10.1002/ajhb.22254.22378322
61. Singh RK, Chang HW, Yan D, Lee KM, Ucmak D, Wong K, et al. 2017. Influence of diet on the gut microbiome and implications for human health. J Transl Med 15 (1 ):73, PMID: , 10.1186/s12967-017-1175-y.28388917
62. Bowyer RCE, Jackson MA, Le Roy CI, Ni Lochlainn M, Spector TD, Dowd JB, et al. 2019. Socioeconomic status and the gut microbiome: a TwinsUK cohort study. Microorganisms 7 (1 ):17, PMID: , 10.3390/microorganisms7010017.30641975
63. Lin H, Peddada SD. 2020. Analysis of compositions of microbiomes with bias correction. Nat Commun 11 (1 ):3514, PMID: , 10.1038/s41467-020-17041-7.32665548
64. Calle ML. 2019. Statistical analysis of metagenomics data. Genomics Inform 17 (1 ):e6, PMID: , 10.5808/GI.2019.17.1.e6.30929407
65. Dogra SK, Dore J, Damak S. 2020. Gut microbiota resilience: definition, link to health and strategies for intervention. Front Microbiol 11 :572921, PMID: , 10.3389/fmicb.2020.572921.33042082
66. Liu Y, Song X, Zhou H, Zhou X, Xia Y, Dong X, et al. 2018. Gut microbiome associates with lipid-lowering effect of rosuvastatin in vivo. Front Microbiol 9 :530, PMID: , 10.3389/fmicb.2018.00530.29623075
67. Warbeck C, Dowd AJ, Kronlund L, Parmar C, Daun JT, Wytsma-Fisher K, et al. 2021. Feasibility and effects on the gut microbiota of a 12-week high-intensity interval training plus lifestyle education intervention on inactive adults with celiac disease. Appl Physiol Nutr Metab 46 (4 ):325–336, PMID: , 10.1139/apnm-2020-0459.32961065
68. Ge XD, Chang CE, Chen HL, Liu TT, Chen LG, Huang Y, et al. 2020. Luteolin cooperated with metformin hydrochloride alleviates lipid metabolism disorders and optimizes intestinal flora compositions of high-fat diet mice. Food Funct 11 (11 ):10033–10046, PMID: , 10.1039/d0fo01840f.33135040
69. Kwan SY, Jiao J, Joon A, Wei P, Petty LE, Below JE, et al. 2022. Gut microbiome features associated with liver fibrosis in Hispanics, a population at high risk for fatty liver disease. Hepatology 75 (4 ):955–967, PMID: , 10.1002/hep.32197.34633706
70. Waters JL, Ley RE. 2019. The human gut bacteria Christensenellaceae are widespread, heritable, and associated with health. BMC Biol 17 (1 ):, PMID: , 10.1186/s12915-019-0699-4.31660948
71. Pascal V, Pozuelo M, Borruel N, Casellas F, Campos D, Santiago A, et al. 2017. A microbial signature for Crohn’s disease. Gut 66 (5 ):813–822, PMID: , 10.1136/gutjnl-2016-313235.28179361
72. Lee T, Clavel T, Smirnov K, Schmidt A, Lagkouvardos I, Walker A, et al. 2017. Oral versus intravenous iron replacement therapy distinctly alters the gut microbiota and metabolome in patients with IBD. Gut 66 (5 ):863–871, PMID: , 10.1136/gutjnl-2015-309940.26848182
73. Pittayanon R, Lau JT, Leontiadis GI, Tse F, Yuan Y, Surette M, et al. 2020. Differences in gut microbiota in patients with vs without inflammatory bowel diseases: a systematic review. Gastroenterology 158 (4 ):930–946.e1, PMID: , 10.1053/j.gastro.2019.11.294.31812509
74. Zhao X, Zhang Z, Hu B, Huang W, Yuan C, Zou L. 2018. Response of gut microbiota to metabolite changes induced by endurance exercise. Front Microbiol 9 :765, PMID: , 10.3389/fmicb.2018.00765.29731746
75. Clavel TLP, Charrier C. 2014. The Family Coriobacteriaceae. In: The Prokaryotes. Rosenberg E, DeLong EF, Lory S, Stackebrandt E, Thompson F, eds. Berlin, Germany: Springer, 10.1007/978-3-642-30138-4_343.
76. Liu H, Zhang H, Wang X, Yu X, Hu C, Zhang X. 2018. The family Coriobacteriaceae is a potential contributor to the beneficial effects of Roux-en-Y gastric bypass on type 2 diabetes. Surg Obes Relat Dis 14 (5 ):584–593, PMID: , 10.1016/j.soard.2018.01.012.29459013
77. Goodrich JK, Waters JL, Poole AC, Sutter JL, Koren O, Blekhman R, et al. 2014. Human genetics shape the gut microbiome. Cell 159 (4 ):789–799, PMID: , 10.1016/j.cell.2014.09.053.25417156
78. Fu J, Bonder MJ, Cenit MC, Tigchelaar EF, Maatman A, Dekens JA, et al. 2015. The Gut microbiome contributes to a substantial proportion of the variation in blood lipids. Circ Res 117 (9 ):817–824, PMID: , 10.1161/CIRCRESAHA.115.306807.26358192
79. He Y, Wu W, Wu S, Zheng HM, Li P, Sheng HF, et al. 2018. Linking gut microbiota, metabolic syndrome and economic status based on a population-level analysis. Microbiome 6 (1 ):172, PMID: , 10.1186/s40168-018-0557-6.30249275
80. Li X, Zhang X, Zhang Z, Han L, Gong D, Li J, et al. 2019. Air pollution exposure and immunological and systemic inflammatory alterations among schoolchildren in China. Sci Total Environ 657 :1304–1310, PMID: , 10.1016/j.scitotenv.2018.12.153.30677897
81. Tamagawa E, Bai N, Morimoto K, Gray C, Mui T, Yatera K, et al. 2008. Particulate matter exposure induces persistent lung inflammation and endothelial dysfunction. Am J Physiol Lung Cell Mol Physiol 295 (1 ):L79–L85, PMID: , 10.1152/ajplung.00048.2007.18469117
82. Gómez-Hurtado I, Santacruz A, Peiró G, Zapater P, Gutiérrez A, Pérez-Mateo M, et al. 2011. Gut microbiota dysbiosis is associated with inflammation and bacterial translocation in mice with CCl4-induced fibrosis. PLoS One 6 (7 ):e23037, PMID: , 10.1371/journal.pone.0023037.21829583
83. Xu W, Song Q. 2017. Abstract 19614: C-reactive protein re-shapes the composition of gut microbiota and causes obesity. Circulation 136 (suppl 1 ):A19614, 10.1161/circ.136.suppl_1.19614.
84. Armstrong H, Bording-Jorgensen M, Dijk S, Wine E. 2018. The complex interplay between chronic inflammation, the microbiome, and cancer: understanding disease progression and what we can do to prevent it. Cancers (Basel). 10 (3 ):83, PMID: , 10.3390/cancers10030083.29558443
85. Xu Y, Li Z, Liu Y, Pan B, Peng R, Shao W, et al. 2021. Differential roles of water-insoluble and water-soluble fractions of diesel exhaust particles in the development of adverse health effects due to chronic instillation of diesel exhaust particles. Chem Res Toxicol 34 (12 ):2450–2459, PMID: , 10.1021/acs.chemrestox.1c00199.34780166
86. Boyle AK, Rinaldi SF, Norman JE, Stock SJ. 2017. Preterm birth: inflammation, fetal injury and treatment strategies. J Reprod Immunol 119 :62–66, PMID: , 10.1016/j.jri.2016.11.008.28122664
87. Perez-Muñoz ME, Arrieta MC, Ramer-Tait AE, Walter J. 2017. A critical assessment of the “sterile womb” and “in utero colonization” hypotheses: implications for research on the pioneer infant microbiome. Microbiome 5 (1 ):48, PMID: , 10.1186/s40168-017-0268-4.28454555
88. Dunn AB, Jordan S, Baker BJ, Carlson NS. 2017. The maternal infant microbiome: considerations for labor and birth. MCN Am J Matern Child Nurs 42 (6 ):318–325, PMID: , 10.1097/NMC.0000000000000373.28825919
89. Backhed F, Roswall J, Peng Y, Feng Q, Jia H, Kovatcheva-Datchary P, et al. 2015. Dynamics and stabilization of the human gut microbiome during the first year of life. Cell Host Microbe 17 (5 ):690–703, PMID: , 10.1016/j.chom.2015.04.004.25974306
90. Moossavi S, Miliku K, Sepehri S, Khafipour E, Azad MB. 2018. The prebiotic and probiotic properties of human milk: implications for infant immune development and pediatric asthma. Front Pediatr 6 :197, PMID: , 10.3389/fped.2018.00197.30140664
91. Collado MC, Rautava S, Aakko J, Isolauri E, Salminen S. 2016. Human gut colonisation may be initiated in utero by distinct microbial communities in the placenta and amniotic fluid. Sci Rep 6 :23129, PMID: , 10.1038/srep23129.27001291
92. Funkhouser LJ, Bordenstein SR. 2013. Mom knows best: the universality of maternal microbial transmission. PLoS Biol 11 (8 ):e1001631, PMID: , 10.1371/journal.pbio.1001631.23976878
93. Jiménez E, Marín ML, Martín R, Odriozola JM, Olivares M, Xaus J, et al. 2008. Is meconium from healthy newborns actually sterile? Res Microbiol 159 (3 ):187–193, PMID: , 10.1016/j.resmic.2007.12.007.18281199
94. Dos Santos SJ, Pakzad Z, Elwood CN, Albert AYK, Gantt S, Manges AR, et al. 2021. Early neonatal meconium does not have a demonstrable microbiota determined through use of robust negative controls with cpn60-based microbiome profiling. Microbiol Spectr 9 (2 ):e0006721, PMID: , 10.1128/Spectrum.00067-21.34585952
95. Kennedy KM, Gerlach MJ, Adam T, Heimesaat MM, Rossi L, Surette MG, et al. 2021. Fetal meconium does not have a detectable microbiota before birth. Nat Microbiol 6 (7 ):865–873, PMID: , 10.1038/s41564-021-00904-0.33972766
96. Beamish LA, Osornio-Vargas AR, Wine E. 2011. Air pollution: an environmental factor contributing to intestinal disease. J Crohns Colitis 5 (4 ):279–286, PMID: , 10.1016/j.crohns.2011.02.017.21683297
97. Bosch AJT, Rohm TV, Alasfoor S, Dervos T, Cavelti-Weder C. 2018. Air pollution–induced diabetes is mediated via the gastrointestinal tract. Diabetes 67 (suppl 1 ):2416–PUB, 10.2337/db18-2416-PUB.
98. Salim SY, Kaplan GG, Madsen KL. 2014. Air pollution effects on the gut microbiota: a link between exposure and inflammatory disease. Gut Microbes 5 (2 ):215–219, PMID: , 10.4161/gmic.27251.24637593
99. Kish L, Hotte N, Kaplan GG, Vincent R, Tso R, Gänzle M, et al. 2013. Environmental particulate matter induces murine intestinal inflammatory responses and alters the gut microbiome. PLoS One 8 (4 ):e62220, PMID: , 10.1371/journal.pone.0062220.23638009
100. Mutlu EA, Comba IY, Cho T, Engen PA, Yazici C, Soberanes S, et al. 2018. Inhalational exposure to particulate matter air pollution alters the composition of the gut microbiome. Environ Pollut 240 :817–830, PMID: , 10.1016/j.envpol.2018.04.130.29783199
101. Mutlu EA, Engen PA, Soberanes S, Urich D, Forsyth CB, Nigdelioglu R, et al. 2011. Particulate matter air pollution causes oxidant-mediated increase in gut permeability in mice. Part Fibre Toxicol 8 :19, PMID: , 10.1186/1743-8977-8-19.21658250
102. Renwick AG, Drasar BS. 1976. Environmental carcinogens and large bowel cancer. Nature 263 (5574 ):234–235, PMID: , 10.1038/263234a0.958475
103. Ribiere C, Peyret P, Parisot N, Darcha C, Dechelotte PJ, Barnich N, et al. 2016. Oral exposure to environmental pollutant benzo[a]pyrene impacts the intestinal epithelium and induces gut microbial shifts in murine model. Sci Rep 6 :31027, PMID: , 10.1038/srep31027.27503127
104. Khalil A, Villard P-H, Dao MA, Burcelin R, Champion S, Fouchier F, et al. 2010. Polycyclic aromatic hydrocarbons potentiate high-fat diet effects on intestinal inflammation. Toxicol Lett 196 (3 ):161–167, PMID: , 10.1016/j.toxlet.2010.04.010.20412841
|
PMC009xxxxxx/PMC9888265.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36719213
EHP10477
10.1289/EHP10477
Research
BPDE, the Migration and Invasion of Human Trophoblast Cells, and Occurrence of Miscarriage in Humans: Roles of a Novel lncRNA-HZ09
Dai Mengyuan 1 2 *
Huang Wenxin 1 2 *
Huang Xinying 1 2
Ma Chenglong 1 2
Wang Rong 1 2
Tian Peng 1
Chen Weina 1
Zhang Ying 1
Mi Chenyang 1
https://orcid.org/0000-0001-7845-3331
Zhang Huidong 1
1 Research Center for Environment and Female Reproductive Health, The Eighth Affiliated Hospital, Sun Yat-sen University, Shenzhen, China
2 Department of Toxicology, School of Public Health, Fujian Medical University, Fuzhou, China
Address correspondence to Huidong Zhang. Email: [email protected]
31 1 2023
1 2023
131 1 01700911 10 2021
30 9 2022
22 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Recurrent miscarriage (RM) affects 1%–3% of pregnancies. However, in almost 50% of cases, the cause is unknown. Increasing evidence have shown that benzo(a)pyrene [B(a)P], a representative of polycyclic aromatic hydrocarbons (PAHs), is correlated with miscarriage. However, the underlying mechanisms of B(a)P/benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide (BPDE)-induced trophoblast cell dysfunctions and miscarriage remain largely unknown.
Objective:
The objective was to discover the role(s) of a novel lncRNA, lnc-HZ09, in the regulation of BPDE-inhibited migration and invasion of trophoblast cells and the occurrence of miscarriage.
Method:
Human trophoblast cells were treated with 0, 0.25, 0.5, 1.0, or 1.5μM BPDE with or without corresponding lnc-HZ09 silencing or overexpression. Using these cells, we evaluated cell migration and invasion, the mRNA and protein levels of members of the PLD1/RAC1/CDC42 pathway, the regulatory roles of lnc-HZ09 in PLD1 transcription and mRNA stability, and lnc-HZ09 transcription and stability. Human villous tissues were collected from RM (n=15) group and their matched healthy control (HC, n=15) group. We evaluated the levels of BPDE-DNA adducts, lnc-HZ09, and the mRNA and protein expression of members of the PLD1/RAC1/CDC42 pathway, and correlated their relative expression levels. We further constructed 0, 0.05 or 0.2mg/kg B(a)P-induced mouse miscarriage model (each n=6), in which the mRNA and protein expression of members of the Pld1/Rac1/Cdc42 pathway were measured.
Results:
We identified a novel lnc-HZ09. Human trophoblast cells treated with lnc-HZ09 exhibited less cell migration and invasion. In addition, the levels of this lncRNA were higher in villous tissues from women with recurrent miscarriage than those from healthy individuals. SP1-mediated PLD1 mRNA levels were lower, and HuR-mediated PLD1 mRNA stability was less in trophoblast cells overexpressing lnc-HZ09. However, trophoblast cells treated with MSX1 had higher levels of lnc-HZ09, and METTL3-mediated m6A methylation on lnc-HZ09 resulted in greater lnc-HZ09 RNA stability. In BPDE-treated human trophoblast cells and in RM villous tissues, MSX1-mediated lnc-HZ09 transcription and METTL3-mediated lnc-HZ09 stability were both greater. In our mouse miscarriage model, B(a)P-treated mice had lower mRNA and protein levels of members of the Pld1/Rac1/Cdc42 pathway.
Discussion:
These results suggest that in human trophoblast cells, BPDE exposure up-regulated lnc-HZ09 level, suppressed PLD1/RAC1/CDC42 pathway, and inhibited migration and invasion, providing new insights in understanding the causes and mechanisms of unexplained miscarriage. https://doi.org/10.1289/EHP10477
Supplemental Material is available online (https://doi.org/10.1289/EHP10477).
* These authors contributed equally to this work.
All authors declare they have no actual or potential conflicts of interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Miscarriage, one of the most common adverse pregnancy outcomes, is defined as a termination of pregnancy due to fetus exclusion prior to 22 wk of gestation or <500g of fetal weight.1 Reportedly, there is an 11%–22% cumulative risk of miscarriages in 5–20 wk of gestation, and the risk of miscarriage is higher in early gestation (<14wk) relative to that in the later period.2,3 Recurrent miscarriage (RM) refers to two or more consecutive miscarriages with the same spouse, affecting 1%–3% of pregnancies.4,5 Many studies have indicated that various risk factors are associated with miscarriage in the first trimester, such as chromosomal abnormalities, antiphospholipid syndrome, congenital structural abnormalities of the uterus, type I diabetes, and thyroid dysfunction.1,6,7 However, in approximately 50% of RM patients, the clinical causes of miscarriage are completely unknown, and those are collectively classified as unexplained recurrent miscarriage.8,9 Therefore, it is particularly urgent to explore the unknown causes to precisely prevent and treat miscarriage.
Polycyclic aromatic hydrocarbons (PAHs) are a typical class of persistent organic pollutants, produced from incomplete pyrolysis and/or combustion of domestic or industrial coal, cigarettes, fossil fuel, wood, and food items.10 PAHs persist in the environment and biota for long periods and exert toxic effects on organisms largely through inhalation and diet.11 Increasing studies have reported that many PAHs are associated with skin, lung, and bladder cancers.10 Recent studies also suggest that PAHs act as endocrine disrupters and have adverse effects on reproductive health.12 B(a)P, a prototypical representative of PAHs, is metabolized to generate the ultimate metabolite, benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide (BPDE). In a small study of 29 women undergoing in vitro fertilization, higher levels of B(a)P were measured in the follicular fluid of women who self-reported exposure to mainstream smoke than in those who self-reported no exposure to cigarette smoke (1.32±0.68 ng/mL vs. 0.03±0.01 ng/mL).13 More B(a)P has also been found out in follicular fluid of nonpregnant women than those pregnant women in Lithuania during the period 1991–1995 (1.79±0.03 ng/mL vs. 0.08±0.03 ng/mL).14 A case–control study of miscarriage suggests that higher level of BPDE-DNA adducts in maternal blood is associated with a higher miscarriage risk.15 In mice, exposure to B(a)P in early pregnancy impaired murine uterine receptivity,16 suppressed embryo implantation,16 and endometrial stromal cell decidualization.17 Therefore, it is important to explore how B(a)P or BPDE affects miscarriage and specific pathways involved.
As one of the most important components in the placenta, human extravillous trophoblast (EVT) invades into the pregnant uterus to establish a maternal–fetal interface.18 Their normal proliferation, migration and invasion are essential for a successful pregnancy.19 Dysfunctions of human trophoblast cells may lead to impaired uterine spiral artery rebuilding and trophoblast-related adverse pregnancy outcomes, such as miscarriage.20 In our recent work, we found that human trophoblast cell lines HTR-8/SVneo21 and Swan 7120,22 cells treated with BPDE had higher levels of apoptosis and exhibited less proliferation, invasion, and migration. Human villous explants treated with BPDE also exhibited less migration and invasion.21 However, the underlying mechanisms of these results are still largely unknown.
The invasion and migration of mammalian cells are regulated by many signaling pathways. Phospholipase D hydrolyze 1 (PLD1) is involved in the regulation of various cellular biological processes, such as cell growth, proliferation, migration, and intracellular protein transportation.22 In addition, RAC1 (Ras related C3 botulinum toxin substrate 1) and CDC42 (cell division control protein 42) also regulate the dynamics of cell motility and proliferation.23 Moreover, it was reported that PLD1 could interact with RAC1 and CDC42 in RBL-2H3 cells.24 Therefore, whether the PLD1/RAC1/CDC42 pathway might regulate migration and invasion of human trophoblast cells and villous tissues should be further explored.
LncRNAs are noncoding RNA molecules with more than 200 ribonucleotides in length, which have exhibited important regulatory functions at both transcriptional and posttranscriptional levels.25 The expression of lncRNA is developmentally regulated, and lncRNAs are also tissue-specific and/or cell-type–specific. Studies suggest that trophoblast cell functions may be influenced by certain lncRNAs, such as lncRNA EPB41L4A-AS1, which induces metabolic reprogramming in trophoblast cells and placenta tissue26; lnc-SLC4A1-1, which alters trophoblast functions via activation of immune responses and regulation of the NF-κB/CXCL8 axis27; and lncRNA MEG8, which is involved in the regulation of early trophoblast cell function.27 Recently, we have performed transcriptome sequencing of unexplained RM and HC villous tissues and BPDE-treated trophoblast Swan 71 cells.28 Based on these sequencing data, we identified a number of novel lncRNAs of interest, including lnc-HZ01, which was found to regulate BPDE-suppressed trophoblast cell proliferation and miscarriage by forming a lnc-HZ01/MXD1 feedback loop29; lnc-HZ03, which up-regulated p53/SAT1 pathway to promote trophoblast apoptosis and affected miscarriage28; lnc-HZ04, which served as a ceRNA for miR-hz04 and up-regulated the IP3R1/CaMKII/SGCB pathway30; and lnc-HZ08, which regulated BPDE-induced trophoblast cell dysfunctions and was associated with miscarriage.31 These works illustrate that these lncRNAs are closely associated with the occurrence of miscarriage. However, a large number of novel lncRNAs have yet to be identified. How these novel lncRNAs regulate BPDE-suppressed migration and invasion of human trophoblast cells and regulate miscarriage is still largely unclear.
Therefore, in this study, we aimed to explore the signaling pathway by which BPDE inhibited migration and invasion of trophoblast cells and to identify the regulatory roles of a novel lncRNA in both BPDE-treated trophoblast cells and RM villous tissues. For these aims, we conducted in vitro cellular experiments, human tissue experiments, and in vivo mouse model experiments, and we expected to discover the mechanism of BPDE-inhibited invasion and migration of human trophoblast cells in the occurrence of miscarriage, as well as the regulatory roles of a novel lncRNA. Hopefully, this work might provide novel insights in understanding the causes and mechanism of unexplained miscarriage.
Materials and Methods
Chemicals
Anhydrous DMSO, corn oil, benzo(a)pyrene [B(a)P, purity 99%] and actinomycin D were from Sigma-Aldrich. Benzo(a)pyren-7, 8-dihydrodiol-9, 10-epoxide (BPDE, 99.9%) was from MRIGlobal (MRIGlobal). BPDE was dissolved in DMSO at 4 mM and stored at −80°C. B(a)P was dissolved in corn oil before use. All the vehicle control or cultures included an equal amount of DMSO (0.03%, v/v).
In Vitro Molecular and Cellular Biology Assays
Cell culture.
Swan 71 cells, first-trimester human trophoblast cells that have been immortalized by human telomerase, were constructed by the Gil Mor group at Yale University32 and were obtained from this group as a gift. HTR-8/SVneo cells, human first-trimester trophoblast cells that have been immortalized by SVneo virus, were purchased from Hunan Fenghui Biotechnology Co., Ltd. (CL0164). Swan 71 cells were cultured in DMEM/F12 medium (GIBCO, Invitrogen), and HTR-8/SVneo cells were cultured in RPMI 1640 medium (GIBCO), both of which were supplemented with 10% FBS (GIBCO), in a humidified atmosphere containing 5% CO2 at 37°C.
Cell transfection.
Empty vector pcDNA3.1 (Catalog no. V790-20) was purchased from Thermo Fisher Scientific Company. cDNAs that were used for construction of overexpression plasmid of lnc-HZ09 (pcDNA3.1-HZ09), METTL3 mRNA (pcDNA3.1-METTL3), MSX1 mRNA (pcDNA3.1-MSX1), SP1 mRNA (pcDNA3.1-SP1), or HuR mRNA (pcDNA3.1-HuR) were synthesized and constructed into pcDNA3.1 vector by Addgene (Table S1). The corresponding RNA sequences were obtained from National Center for Biotechnology Information (NCBI) database (Gene Bank, Homo sapiens, GRCh38.p14; sequences in Table S1). Empty vector pcDNA3.1 was used as a negative control. Si-HZ09, si-METTL3, si-MSX1, si-SP1, si-HuR, and si-NC (negative control) were customized by Thermo Fisher (sequences in Table S2). Swan 71 and HTR-8/SVneo cells (1×106 cells/well) were seeded in 6-well plates and cultured to 80% confluence. Trophoblast cells were transfected with 2.0μg/well plasmids or 50 nM siRNAs in Lipofectamine 3,000 (Invitrogen) medium for 24 h according to the manufacturer’s protocols. For all assays, cell quantification was performed using TC20 Automated Cell Counter (Bio-Rad Laboratories).
Cell migration and invasion.
Trophoblastic cells (1×106 cells/well) in six-well plates were transfected with si-NC, si-HZ09, si-PLD1, si-HuR, or si-METTL3, or transfected with pcDNA3.1, pcDNA3.1-HZ09, pcDNA3.1-PLD1, or pcDNA3.1-METTL3 using Lipofectamine 3,000 media for 24 h. Media was removed, cells were detached using trypsin, and the surviving cells were resuspended into DMEM/F12 or RPMI 1640 medium. For migration assays, 3×104 cells/well were seeded in 24-well transwell chambers (Corning) and cultured for 24 h. For invasion assays, 80μL aliquots of Matrigel (BD Biosciences) diluted in DMEM/F12 medium (dilution ratio 1:8) were coated on 24-well transwell chambers and were solidified at 37°C for 1 h. Cells (1×104 cells/well) were plated on the top of Matrigel matrix and cultured for 24 h. The bottom chamber contained medium containing 10% FBS as chemoattractant of human trophoblast cells. After 24 h, the whole chambers were fixed with 4% paraformaldehyde for 20 min, stained with crystal violet for 15 min, and then washed thrice with phosphate-buffered saline (PBS). For visualization, the cells on the bottom surface of membrane were photographed by Axio Observer 3 (Zeiss) at 200× magnification and counted in five random fields.
High-throughput mRNA sequencing and data processing.
Swan 71 cells (5×106 cells) in a 100-mm dish overexpressing lnc-HZ09 by transfecting with 2.0μg/well pcDNA3.1-HZ09 and equal number of control cells by transfecting with 2.0μg/well pcDNA3.1 were also used for mRNA sequencing. High-throughput mRNA sequencing was performed on HiSeq 2000 sequencing platform (BGI-Shenzhen) according to BGI commercial standard process (https://www.bgi.com/).28,29,31 Briefly, total RNAs were extracted by Trizol reagent (Thermo Fisher Scientific). The process included the removal of rRNA, synthesis of double-stranded cDNA, end repair, degradation of one strand, and enrichment of the other strand by polymerase chain reaction (PCR). The library quality was confirmed by sequencing. The differentially expressed mRNAs with differences >2-fold and p<0.05 were generated from read counts using the online bioinformatic platform Dr. Tom provided by BGI (biosys.bgi.com). Differentially mRNAs were searched in the NCBI database (Gene Bank, Homo sapiens, GRCh38.p14) to determine their genome loci. These differentially expressed mRNAs were used for gene ontology (GO) analysis (http://geneontology.org/) to generate GO plots.33,34
5′ and 3′ rapid amplification of cDNA ends (RACE) assays.
Total RNAs were isolated with Trizol reagent. RACE-ready first-strand cDNA were performed with 1μg total RNAs using the SMARTer RACE 5′/3′ Kit (Takara Bio) according to the manufacturer’s instructions. The 5′- and 3′-RACE PCR reactions (including PCR Master Mix, SeqAmp DNA Polymerase, 5′- or 3′-RACE-ready cDNA, 5′ or 3′ GSP, UPM, SeqAmp Buffer, and four dNTPs) were performed on LightCycler480 II (Roche) following the manufacturer’s instructions (Takara Bio). The PCR programs are shown in Table S3. RACE PCR products were separated on a 1% agarose gel and extracted with the NuceloSpin Gel and PCR Clean-Up Kit (Takara Bio). Then, the DNA products were subcloned into pGH-T vectors (GV0108-C, Shanghai Generay Biotech Co., Ltd.) using In-Fusion Snap Assembly Master Mix (Takara Bio) for 15 min at 50°C and then sequenced bidirectionally using gene specific primers (Sagene) (Table S4). RNA integrity was determined with a Bioanalyzer 2100 using RNA 6000 nanochips (Agilent). DNA quantity was assessed using a NanoDrop One C spectrophotometer (Thermo Fisher Scientific). The full-length nucleotide sequence of lnc-HZ09 is shown in Table S5.
Total RNA extraction and quantitative real-time polymerase chain reaction (RT-qPCR).
RNA was extracted from Swan 71 cells, HTR-8/SVneo cells, human villous tissues, or mouse placental tissues using Trizol (Invitrogen) containing DNase I (Life Technologies), according to the manufacturer’s protocols. RNA quality and quantity were assessed using a NanoDrop 2000 UV spectrophotometer (Thermo Fisher Scientific). RT-qPCR analysis was performed as described previously.28,29,31 Briefly, the isolated RNAs (800 ng) were converted into cDNAs using the first-strand cDNA synthesis kit (Invitrogen). Then, cDNAs were further amplified using a 20-μL SYBR qPCR system (including Maxima SYBR Green qPCR Mix, which includes Maxima Hot Start Taq DNA polymerase, SYBR Green I, and dNTPs) following the manufacturer’s instructions (Thermo Fisher Scientific) (conditions in Table S6). The amplification results were automatically analyzed using 2−ΔΔCt method with iQ5 real-time detection system (Bio-Rad Laboratories). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA was used as normalization internal standard for lncRNAs and mRNAs. The primer (Sangon Biotech) sequences are listed in Table S7.
Western blot analysis.
Western blot analysis was performed as described previously.28,29,31 Briefly, proteins were extracted using RIPA lysis buffer (Thermo Fisher Scientific) and quantified using Pierce BCA Protein Assay Kit (Pierce). Proteins (30μg) were separated using 6%–12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and were then transferred to equilibrated polyvinylidene fluoride (PVDF) membrane (Amersham Biosciences). The membranes were blocked with 5% bovine serum albumin (BSA; Sigma-Aldrich) at 25°C for 1 h. Then, the membranes were incubated with primary antibody at 4°C overnight. The primary antibodies included anti-PLD1 (dilution 1:100; sc-28,314, Santa Cruz Biotechnology), anti-RAC1 (dilution 1:1,000; ab155938, Abcam), anti-CDC42 (dilution 1:1,000; ab1429, Abcam), anti-SP1 (dilution 1:1,000; ab255289, Abcam), anti-HuR (dilution 1:1,000; ab200342, Abcam), anti-MSX1 (dilution 1:600; LS-C30725, LifeSpan Biosciences), anti-METTL3 (dilution 1:1,000; ab195352, Abcam), anti-GAPDH (dilution 1:1,000; ab8245, Abcam) and anti-β-tubulin (dilution 1:1,000; ab78078, Abcam). After washing thrice with Tris-buffered saline containing Tween-20 (TBST) for 10 min in each time, the membrane was incubated with secondary antibody in blocking solution at 25°C for 1 h. The secondary antibodies included goat anti-rabbit immunoglobulin G (IgG) (dilution 1:1,000; ab207995, Abcam) and goat antimouse IgG (dilution 1:1,000; ab207996, Abcam). The relative density of each protein band was analyzed by Image J with β-tubulin or GAPDH as the internal standard.
Fluorescence in situ hybridization (FISH).
Lnc-HZ09 in Swan 71 cells was detected by Cy3-labeled lnc-HZ09 probe (5′-CACGAGC-Cy3-TGCCCACGGTCT-Cy3-TCCTTT-3′) (Empire Genomics) according to FISH Kit procedure (RIBOBIO), as described previously.31 Briefly, Swan 71 cells (1×104 cells/well) were seeded in 35-mm confocal dishes (Cat. No. 80100; NEST) for 24 h; then they were washed with PBS and fixed by 4% paraformaldehyde (Sigma-Aldrich) for 15 min and treated with 1% Triton X-100 (Sigma-Aldrich) for 30 min, following incubation with 2μM FISH probe of lnc-HZ09 that was pretreated at 73°C for 5 min after washing with PBS thrice. Afterward, these fixed cells were further stained with DAPI for 20 min. After washing with PBS thrice, the emission fluorescence was recorded at 488 nm for DAPI and 570 nm for FISH probe on a confocal fluorescence microscope (N-STORM+A1R; Nikon). The relative levels of lnc-HZ09 in the cytoplasm and nucleus were quantified using 20 random cells with software Image J (Hue=200).
mRNA stability assays.
Swan 71 or HTR-8/SVneo cells (1×106 cells/well) with overexpression or knockdown of lnc-HZ09, HuR, or METTL3 were seeded in a 6-well plate for 12 h. Then, the cells were treated with 5μg/mL actinomycin D (Sigma-Aldrich) to block mRNA transcription. After 0, 1, 2, 3, 4, or 5 h, RNAs were extracted from cells, and the lnc-HZ09 or PLD1 mRNA were analyzed by RT-qPCR assays. GAPDH mRNA was used as the normalization internal standard.
Chromatin immunoprecipitation (ChIP) assays.
ChIP assays were performed with EZ-Magna ChIP Chromatin Immunoprecipitation Kit (Millipore) based on the manufacturer’s protocols.29,31 Briefly, Swan 71 or HTR-8/SVneo cells (1×107 cells) were transfected with pcDNA3.1-HZ09 or si-HZ09 or were treated with 1.0μM BPDE. Then, the cells were cross-linked with 1% formaldehyde at 37°C for 15 min and were quenched in 125 mM glycine for 5 min. DNA fragments ranged from 300 to 600 bp were generated after sonication using a Bioruptor (Diagenode SA). Subsequently, the antibody against MSX1 (dilution 1:200; sc-517256, Santa Cruz Biotechnology) or SP1 (dilution 1:200; ab227383, Abcam) was used for immunoprecipitation at 4°C overnight on an inverse rotator, with equal weight of IgG (dilution 1:200; ab172730, Abcam) as negative control. The promoter regions of lnc-HZ09 or PLD1 were immunoprecipitated, extracted, and amplified (800 ng) by qPCR (programs in Table S6). The specific qPCR primers are listed in Table S8.
RNA immunoprecipitation (RIP) assays.
RIP assays were performed using a Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore) based on the manufacturer’s protocols.29 Briefly, trophoblastic cells (1×107 cells) were lysed in lysis buffer containing RNase inhibitor and protease inhibitor. Cell lysates were incubated with magnetic beads attached with human HuR antibody (ab200342; Abcam) or mouse IgG (5873S; Cell Signaling Technology) as negative control. The immunoprecipitated RNAs (800 ng) were extracted by Trizol and analyzed by RT-qPCR (programs in Table S6) using primers (Sangon Biotech) listed in Table S8.
Methylated RNA immunoprecipitation (MeRIP).
N6-methyladenosine (m6A RNA methylation) modification on lnc-HZ09 was determined by MeRIP-qPCR assays, as described previously.29 Briefly, the purified and fragmented RNAs were incubated with Pierce Protein A/G Magnetic Beads (Thermo Fisher Scientific) coupled with 5μg anti-m6A antibody (ab208577; Abcam) or rabbit IgG as negative control for 2 h at 4°C with rotation. After washes, m6A-modified RNAs were eluted and detected by RT-qPCR (programs in Table S6) using the primers listed in Table S8. One tenth of the amount of total RNAs was used as input. The corresponding m6A modification level in each sample was calculated by normalizing against the level in input.
In vitro RNA pull-down assays.
In vitro RNA pull-down assays were performed as described previously.31 Briefly, lnc-HZ09, lnc-HZ09-AS (antisense sequence), PLD1 mRNA or PLD1 mRNA-AS was respectively in vitro transcribed from pGEM-T-HZ09, pGEM-T-HZ09-AS, pGEM-T-PLD1, or pGEM-T-PLD1-AS, all of which were customized and synthesized using pGEM-T empty vector (A3600; Promega), using TranscriptAid T7 High Yield Transcription Kit (Thermo Fisher Scientific) and were purified with GeneJET RNA Purification Kit (Thermo Fisher Scientific). Then, each of the RNAs was labeled with biotin using Biotin RNA labeling mix (Roche) and was mixed with lysate proteins in protein-RNA binding buffer for 1 h at 4°C. Lnc-HZ09-AS or PLD1 mRNA-AS was used as negative control (sequences in Table S9). Subsequently, the RNA-protein complexes were pulled down by streptavidin attached on magnetic beads using Pierce Magnetic RNA-Protein Pull-Down Kit (Thermo Fisher Scientific), and the proteins were analyzed by Western blotting.
Human Tissue Assays
Tissues collection and statement.
Villous tissues were collected from 15 patients with unexplained recurrent miscarriage (RM group) and 15 women who had elective abortion as healthy control (HC group) ages 25–30 y old and were treated at the West China Fourth Hospital (Chengdu, China) from December 2018 to December 2019, as described previously.28,29,31 All RM patients had ≥2 consecutive unexplained miscarriages. Any women with one of the following features was excluded35: a) eclampsia or preeclampsia; b) viral infectious disease (e.g., AIDS, syphilis, tuberculosis, gonorrhea); c) polycystic ovarian syndrome; d) uterine abnormalities or cervical incompetence; e) luteal phase defects, autoimmune abnormality, hyperandrogenemia, hyperprolactinemia, or antiphospholipid antibody syndrome; f) abnormal karyotype of the parents or abortus; g) the symptoms of endocrine or metabolic diseases (e.g., hyperthyroidism, diabetes, and hypothyroidism); and h) tuberculosis, HIV, HBV, HCV, or with positive results from γ-interferon release tests. All of the HC group lacked any of the eight features as described above and had previous pregnancies. A piece of villous tissue with dimensions of approximate 2×0.5×0.5 cm3 was manually dissected from the fetal side of the placenta and cleared of maternal decidua from these two groups at 6–10 wk of gestation. These samples were serially washed and immediately frozen in liquid nitrogen. Villous tissues were homogenized using Silica beads (107735; Merck) via shaking for a 1-min burst using a TissueLyser LT instrument (Qiagen). Approximately 30mg villous tissues were homogenized in 600 uL Trizol reagent (Invitrogen) for total RNA extraction and in RIPA lysis buffer (Thermo Fisher Scientific) containing a protease inhibitor cocktail for total protein separation. The experiment protocols have been authorized by the Medical Ethics Committee of the West China Fourth Hospital, Sichuan University. Written informed consent was signed before the study.
Determination of BPDE-DNA adduct levels in villous tissues.
Genomic DNA was isolated from human villous tissues using Tissue/Cell Genomic DNA extraction and Purification Kit (K1442-100; BioVision). The levels of BPDE-DNA adducts were assessed using BPDE-DNA adduct ELISA kit (STA-357; Cell Biolabs). Briefly, DNA samples (2μg/mL) were sonicated into fragments with 200–1,000 bp as determined by 1% agarose gel and then incubated with anti-BPDE antibody (dilution 1:1,000; 235601, Cell Biolabs) in 96-well plates for 2 h at room temperature. After washing with wash buffer (310806; Cell Biolabs), the secondary antibody (10902; Cell Biolabs) was added to each well for 1 h, 3,3′,5,5′-tetramethylbenzidine buffer was added, and then the mixture was incubated for 20 min at room temperature. After termination of reaction, the relative levels of BPDE-DNA adducts were determined by measuring the absorbance at 450 nm using a microplate reader (VL0L00D0; Thermo Fisher Scientific), with the Reduced DNA Standard (235602; Cell Biolabs) as absorbance blank. The amount of BPDE-DNA adducts was quantified using a BPDE-DNA standard curve. The results were expressed as nanograms of BPDE-DNA adducts per microgram of DNA.
In Vivo Mouse Model Assays
A B(a)P-induced miscarriage mouse model was constructed as described previously.28,29,31 Briefly, C57BL/6 mice were from the Charles River Company and were housed in the local laboratory animal center. Mice (6–8 wk old) were housed under standard environmental conditions (12 h light/dark cycle, 22°C). The mice had a 1-wk acclimation, during which time they received standard chow and tap water ad libitum. Female mice were mated with male mice overnight, and the appearance of vaginal plug was considered as the first day of pregnancy (D1), which was further validated by monitoring the increase in weight. Three groups of pregnant mice (each n=8) were daily given B(a)P (0, 0.05, or 0.2mg/kg, dissolved in corn oil) by oral gavage from D1 to D13, with an equal volume of corn oil as vehicle control. All mice were weighed daily, and they were euthanized by injection with Nembutal (100mg/kg) on D14 to collect uteri. A random placenta was collected from every mouse in each group. The placental tissues were manually dissected and then snap-frozen in liquid nitrogen and stored at −80°C for subsequent RT-qPCR and Western blotting analysis. The protocol was approved by the Medical Ethics Committee of West China Medical Center.
Software Prediction and Data Analysis
Coding Potential Assessment Tool [CPAT (https://wlcb.oit.uci.edu/cpat/)]36 was used to analyze the protein-coding potential of lncRNAs. The open reading frame (ORF) was predicted with the NCBI Open Reading Frame Finder [ORF finder (https://www.ncbi.nlm.nih.gov/orffinder/)]. Conserved Domain Database [CDD (https://www.ncbi.nlm.nih.gov/cdd/)] and Pfam 34.0 (http://pfam.xfam.org/)37 were used to analyze the conserved domains in lncRNAs. PROMO software [version 3.0.2 (http://alggen.lsi.upc.es/cgi-bin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3)]38,39 was used to predict the transcription factors. The proteins that might be regulated by PLD1 were analyzed using STRING [version 11.5 (https://cn.string-db.org/)].40 Last, the m6A methylation modification site on RNA was identified with SRAMP software [version 2.15 (http://www.cuilab.cn/sramp/)].41
All experiments were replicated thrice independently, and the data were presented as mean±standard deviation (SD; n=3). Statistical analysis was operated using SPSS software (version 24.0; SPSS Inc.). Independent-samples Student’s t-test was used between two groups; and one-way analysis of variance was used for more groups with Dunnett’s or LSD post hoc test. The correlation analysis of the relative expression levels was performed using Pearson analysis. The dotted box was used to visually distinguish the data points of RM and HC groups. All graphs were made with GraphPad Prism (version 8.0; GraphPad Inc.). Differences were considered significant when *p<0.05, **p<0.01, or ***p<0.001.
Results
Characterization of a Novel lnc-HZ09 Highly Expressed in RM Villous Tissues and in BPDE-Exposed Human Trophoblast Cells
In our previous work, we identified 22 novel lncRNAs that were significantly highly expressed in human Swan 71 cells after BPDE treatment and 10 novel lncRNAs that were significantly highly expressed in RM tissues relative to HC tissues by high-throughput transcriptome sequencing.28 In these sequencing data, a novel lncRNA, lnc-32238, was one of the most considerably highly expressed lncRNAs in BPDE-exposed trophoblast cells (Figure S1A) and in RM tissues relative to HC tissues (Figure S1B), implying that this lncRNA might regulate the dysfunctions of BPDE-exposed trophoblast cells and the occurrence of miscarriage in an unidentified regulatory approach. In the current work, we focused on this lncRNA. Lnc-32238 was further confirmed to be significantly highly expressed in BPDE-exposed Swan 71 and HTR-8/SVneo cells by RT-qPCR analysis (Figure 1A,B). This lncRNA was identified as a sense transcript with 365 nucleotides (nt) in length by rapid amplification of cDNA ends (RACE) assays (Figure S1C; Tables S4 and S5) and resided at chromosome 16 (chr 16: 3,220,699 - 3,221,017). Then, this lnc-32238 was termed as lnc-HZ09, and its sequence was submitted to NCBI with accession No. MW675687. The protein-coding potential of lnc-HZ09 was analyzed to be very weak using NCBI ORF finder and CPAT (coding probability 0.036<0.364).36 Additionally, lnc-HZ09 was found to have no conserved domains using CDD and Pfam.42 The data indicated that lnc-HZ09 might not encode a protein (Table. S10). Lnc-HZ09 was distributed to both the cytoplasm and nucleus of Swan 71 cells, as detected by FISH assays (Figure 1C).
Figure 1. Expression levels of a novel lnc-HZ09 and the migration and invasion of human trophoblast cells with overexpression or knockdown of lnc-HZ09. (A–B) RT-qPCR analysis (each n=3) of lnc-HZ09 expression levels in BPDE-treated Swan 71 (A) or HTR-8/SVneo (B) cells. (C) FISH analysis (each n=3) of the distribution of lnc-HZ09 (red) in the nucleus and cytoplasm of Swan 71 cells (scale bar, 50μm) and the relative levels of lnc-HZ09 foci. (D–G) Transwell assay analysis of the migration and invasion of Swan 71 (D–E) or HTR-8/SVneo (F–G) cells with overexpression (D and F) or knockdown (E and G) of lnc-HZ09 (scale bar, 200μm). The number of cells per view was quantified. The summary data of these bar charts were shown in Excel Table S1. The RNA level in the untreated cells was set as “1” in RT-qPCR assays. C–G show the representative data from three independent experiments. Data in (A–G) show mean±SD of three independent experiments. Two-tailed Student’s t-test for (D,F); one-way ANOVA analysis for (A,B,E,G); *p<0.05, **p<0.01, and ***p<0.001. Note: ANOVA, analysis of variance; BPDE, benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide; FISH, fluorescence in situ hybridization; HZ09, overexpression of lnc-HZ09; NC, negative control of siRNA; ns, nonsignificance; RT-qPCR, quantitative reverse transcription polymerase chain reaction; SD, standard deviation; si-HZ09, knockdown of lnc-HZ09; Vector, empty vector of pcDNA3.1.
Figures 1A and 1B are bar graphs titled Swan 71 and H T R-8 or SVneo, plotting ribonucleic acid relative expression, plotting 0 to 5 in unit increments and 12 to 18 in increments of 2, and 0 to 5 in unit increments and 12 to 15 in increments of 2 (y-axis) across Benzo(a)pyrene diol epoxide, micromolar, ranging from 0 to 0.5 in increments of 0.25, 0.5 to 1.5 in increments of 0.5 (x-axis) for Inc-HZ09, respectively. Figure 1C is a set of one stained tissue and one bar graph. The stained issues depict the Fluorescence in Situ hybridization analysis in 4′,6-diamidino-2-phenylindole, H Z 09, Merge, and Zoom. The bar graph, plotting HZ 09 relative expression, ranging 0 to 60 in increments 20 (y-axis) across Cytoplasm and Nucleus (x-axis). Figure 1D is a set of one stained tissue and one bar graph titled Swan 71. The stained tissue displays vector and HZ 09 (columns) and Migration and Invasion (rows). The bar graph, plotting cells or view, ranging 0 to 250 in increments of 50 (y-axis) across migration and invasion (x-axis) for Vector and HZ 09. Figures 1E and 1G are set of one stained tissue and one bar graph titled Swan 71 and H T R-8 or Vneo. The stained tissue displays negative control, si-HZ 09, and si2-HZ 09 (columns) and Migration and Invasion (rows). The bar graph, plotting cells or view, ranging 0 to 200 in increments of 100 (y-axis) across Migration and Invasion (x-axis) for negative control, si-HZ 09, and si2-HZ 09, respectively. Figure 1F is a set of one stained tissue and one bar graph titled THE-8 or SVneo. The stained tissue displays vector and HZ 09 (columns) and Migration and Invasion (rows). The bar graph, plotting cells or view, ranging 0 to 100 in increments of 50 (y-axis) across migration and invasion (x-axis) for Vector and HZ 09.
Measurement of Migration Ad Invasion of Human Trophoblast Cells with Overexpression or Silence of lnc-HZ09
Because lnc-HZ09 was highly expressed in BPDE-treated trophoblast cells, its roles in the regulation of trophoblast cell functions were explored. To identify this, lnc-HZ09 was overexpressed in human trophoblast Swan 71 cells by transfecting with pcDNA3.1-HZ09 (validation shown in Figure S1D); and these cells and the corresponding control cells were used for mRNA sequencing. In sequencing data, we found that 1,055 mRNAs were down-regulated, and 710 mRNAs were up-regulated with differences >2-fold and p-values <0.05 with overexpression of lnc-HZ09 (sequencing data in Excel Table S2). Subsequently, GO biological process analysis showed that cell migration might be significantly regulated by lnc-HZ09 overexpression (Figure S1E). To experimentally validate this, lnc-HZ09 was overexpressed by transfecting with pcDNA3.1-HZ09 or silenced by transfecting with its two distinct siRNAs (si1-HZ09 or si2-HZ09) in both Swan 71 and HTR-8/SVneo cells, and their efficiencies were validated by RT-qPCR analysis (Figure S1D, F–H). Migration and invasion of the respective cells were assessed. We found that lnc-HZ09-overexpressing cells demonstrated significantly less migration and invasion in comparison with cells transfected with empty vector. In contrast, lnc-HZ09-silenced cells demonstrated significantly greater migration and invasion (Figure 1D–G).
The Regulation of lnc-HZ09 on Trophoblast Cell Migration and Invasion through PLD1/RAC1/CDC42 Pathway
Next, the potential signaling pathway that might be regulated by lnc-HZ09 was further studied. First, the key molecules that were differentially expressed were discovered. In RNA sequencing data of BPDE-treated Swan 71 cells, human RM vs HC villous tissues, and lnc-HZ09-overexpressed cells, we identified nine down-regulated mRNAs with expression level difference >2-fold between experimental and control (e.g., BPDE-treated vs. untreated Swan 71 cells, RM vs. HC villous tissues, and lnc-HZ09-overexpressed vs. control Swan 71 cells) with p-values <0.05 in the intersection of these three sets of sequencing data (Figure 2A). RT-qPCR analysis further confirmed that they were all up-regulated in Swan 71 cells with lnc-HZ09 knockdown via siRNA (Figure S2A). Among them, PLD1 was one of the most considerably altered mRNAs (Figure S2A). String analysis24 showed that PLD1 might interact with RAC1 and CDC42 (Figure 2B), implying that this PLD1/RAC1/CDC42 pathway might regulate migration and invasion of trophoblast cells and might also be regulated by lnc-HZ09 in trophoblast cells.
Figure 2. Expression levels of members of PLD1/RAC1/CDC42 pathway regulated by lnc-HZ09 in human trophoblast cells. (A) The significantly down-regulated mRNAs in the intersection of mRNA sequencing data of BPDE-treated vs. untreated Swan 71 cells, lnc-HZ09-overexpressed vs control Swan 71 cells, and RM vs HC villous tissues. (B) String analysis of PLD1, RAC1 and CDC42. (C) Representative western blot analysis of the protein levels of PLD1, RAC1 and CDC42 in Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of lnc-HZ09, with GAPDH as internal standard. The relative intensity of each band was quantified and the mean±SD of three replicates was shown in Figure S3A. (D) Representative western blot analysis of the protein levels of SP1 and PLD1 in Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of SP1, with GAPDH as internal standard. The relative intensity of each band was quantified and the mean±SD of three replicates was shown in Figure S3G. (E–F) SP1 ChIP assay analysis (each n=3) of the relative enrichment of SP1 in the promoter region of PLD1 gene in Swan 71 (E) or HTR-8/SVneo (F) cells. (G) Representative western blot analysis of the protein levels of SP1 in Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of lnc-HZ09, with β-tubulin as internal standard. The relative intensity of each band was quantified and the mean±SD of three replicates was shown in Figure S3H. (H–I) SP1 ChIP assay analysis (each n=3) of the relative enrichment of SP1 in the promoter region of PLD1 gene in Swan 71 cells with overexpression (H) or knockdown (I) of lnc-HZ09. (J–K) The mRNA stability of PLD1 (each n=3) in Swan 71 (J) or HTR-8/SVneo (K) cells with overexpression or knockdown of lnc-HZ09. (L) Representative western blot analysis of the protein levels of PLD1 in Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of HuR, with β-tubulin as internal standard. The relative intensity of each band was quantified and the mean±SD of three replicates was shown in Figure S5E. (M–N) The mRNA stability of PLD1 (each n=3) in Swan 71 (M) or HTR-8/SVneo (N) cells with overexpression or knockdown of HuR. (O–R) RIP assay analysis (each n=3) of the relative levels of lnc-HZ09 (O and Q) or PLD1 mRNA (P and R) that was pulled down by HuR protein in Swan 71 (O and P) or HTR-8/SVneo (Q and R) cells. (S–V) RIP assay analysis (each n=3) of the relative levels of PLD1 mRNA that was pulled down by HuR protein in Swan 71 (S and T) or HTR-8/SVneo (U and V) cells with overexpression (S and U) or knockdown (T and V) of lnc-HZ09. (W) Representative western blot analysis of HuR that was pulled down by biotin-labeled lnc-HZ09 or PLD1 mRNA in Swan 71 or HTR-8/SVneo cells in pull-down assays. (X) Representative western blot analysis of HuR that was pulled down by biotin-labeled PLD1 mRNA in Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of lnc-HZ09 in pull-down assays. (Y) Representative western blot analysis of HuR that was pulled down by biotin-labeled lnc-HZ09 in Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of PLD1 in pull-down assays. The summary data of these bar charts and diagrams were shown in Excel Table S1. The expression level in NC or Vector group was set as “1” in all of mRNA stability assays; the levels of DNA or RNA pulled down by IgG were set as “1” in all of ChIP and RIP assays, respectively; and the band intensity in NC or Vector group was set as ‘100’ in all of western blot assays. C,D,G,L,W–Y show the representative data from three independent experiments. Data in (E–F, H–K, M–V) show mean±SD of three independent experiments. Two-tailed Student’s t-test for (E–F,H–I,O–V); *p<0.05, **p<0.01. Note: BPDE, benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide; ChIP, chromatin immunoprecipitation; GADPH, glyceraldehyde-3-phosphate dehydrogenase; HC, healthy control; HuR, overexpression of HuR; HZ09, overexpression of lnc-HZ09; NC, negative control of siRNA; ns, nonsignificance; PLD1, phospholipase D hydrolyze 1; RIP, RNA immunoprecipitation; RM, recurrent miscarriage; SD, standard deviation; si-HuR, knockdown of HuR; si-SP1, knockdown of SP1; SP1, overexpression of SP1; Vector, empty vector of pcDNA3.1; si-HZ09, knockdown of lnc-HZ09.
Figure 2A is a Venn diagram titled Benzo(a)pyrene diol epoxide treated versus untreated displays three circles. The circle on the top is labeled 250, the circle on the left, is labeled 1043, the circle on the right, is labeled 927. The intersection area is labeled 9. The intersection area represents Collagen alpha-1(12), Type 2 iodothyronine deiodinase, Myosin Heavy Chain 10, Phospholipase D1, Pleckstrin and Sec7 Domain Containing 3, Protein Tyrosine Phosphatase Non-Receptor Type 13, single-stranded D N A binding protein 2, and transforming growth factor beta-2. Figure 2B is a String analysis depicting Phospholipase D1, Ras-related C3 botulinum toxin substrate 1 and Cell division control protein 42 homolog. Figure 2C is a set of two western blots, displaying Vector (100), HZ 09 (31), Negative control (100), si1-HZ 09 (163), si2-HZ 09 (187) for Swan 71 and Vector (100), HZ 09 (74), Negative control (100), si1-HZ 09 (144), si2-HZ 09 (179) for H T R-8 or SVneo (columns) and Phospholipase D1, Lowercase beta-tubulin, specificity protein 1, and lowercase beta-tubulin (rows). Figure 2D is a western bolt, displaying Vector (100), HZ 09 (169), Negative control (100), si1-HZ 09 (53), si2-HZ 09 (51) columns and specificity protein 1, Phospholipase D1, and Glyceraldehyde 3-phosphate dehydrogenase, each under Swan 71 and H T R-8 or SVneo (rows). Figures 2E, 2F, 2O, 2P, 2Q, 2R are bar graphs titled chromatin immunoprecipitation in Swan 71, chromatin immunoprecipitation in H T R-8 or SVneo, receptor-interacting protein in Swan 71, receptor-interacting protein in swan 71, receptor-interacting protein in H T R-8 or SVneo, receptor-interacting protein in H T R-8 or SVneo, plotting Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 6 in increments of 2; Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 6 in increments of 2; ribonucleic acid relative expression, ranging from 0 to 6 in increments of 2; ribonucleic acid relative expression, ranging from 0 to 4 in unit increment; ribonucleic acid relative expression, ranging from 0 to 6 in increments of 2; and ribonucleic acid relative expression, ranging from 0 to 4 in unit increments (y-axis) across anti-Immunoglobulin G and anti- specificity protein 1; anti-Immunoglobulin G and anti- specificity protein 1; anti-Immunoglobulin G and anti-Human antigen R; anti-Immunoglobulin G and anti-Human antigen R; anti-Immunoglobulin G and anti-Human antigen R; and anti-Immunoglobulin G and anti-Human antigen R (x-axis) for HZ 09 and Phospholipase D1, respectively. Figures 2H, 2I, 2S, 2T, 2U, 2V are clustered bar graph titled chromatin immunoprecipitation in Swan 71, chromatin immunoprecipitation in Swan 71, receptor-interacting protein in swan 71, receptor-interacting protein in swan 71, receptor-interacting protein in H T R-8 or SVneo, and receptor-interacting protein in H T R-8 or SVneo, plotting Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 4 in unit increments; Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 8 in increments of 2; messenger ribonucleic acid relative expression, ranging from 0 to 4 in increments of 2; messenger ribonucleic acid relative expression, ranging from 0 to 4 in increments of 2; messenger ribonucleic acid relative expression, ranging from 0 to 3 in unit increments; and messenger ribonucleic acid relative expression, ranging from 0 to 4 in increments of 2 (y-axis) across anti-Immunoglobulin G and anti- specificity protein 1; anti-Immunoglobulin G and anti- specificity protein 1; anti-Immunoglobulin G and anti-Human antigen R; anti-Immunoglobulin G and anti-Human antigen R; anti-Immunoglobulin G and anti-Human antigen R; and anti-Immunoglobulin G and anti-Human antigen R (x-axis) for Vector and HZ 09; Negative control and si-HZ09; Vector and HZ 09; HZ 09; Negative control and si1-HZ09; Vector and HZ 09; and Negative control and si1-HZ09, respectively. Figure 2G is a western blot, displaying Vector (100), HZ 09 (34), Negative control (100), si1-HZ 09 (143), si2-HZ 09 (157) columns and specificity protein 1 and lowercase beta-tubulin, each under swan 71 and H T R-8 or SVneo (rows). Figures 2J and 2M, each have a set of two line graphs titled Swan 71, plotting Phospholipase D1 messenger ribonucleic acid remaining, ranging from 0.0 to 1.0 in increments of 0.5 and 0.0 to 1.0 in increments of 0.5 (y-axis) across Time (hour), ranging from 1 to 5 in unit increments (x-axis) for Vector, HZ 09, si1,2-HZ 09, and Negative control. Figures 2K and 2N, each have a set of two line graphs titled H T R-8 or SVneo, plotting Phospholipase D1 messenger ribonucleic acid remaining, ranging from 0.0 to 1.0 in increments of 0.5 and 0.0 to 1.0 in increments of 0.5 (y-axis) across Time (hour), ranging from 1 to 5 in unit increments (x-axis) for Vector, HZ 09, si1,2-HZ 09, and Negative control. Figure 2L is a western blot, displaying Vector (100), Human antigen R (147), Negative control (100), si1-HZ 09 (41), si2-HZ 09 (56) columns and Phospholipase D1 and lowercase beta-tubulin, each under swan 71 and H T R-8 or SVneo (rows). Figure 2W is a western blot titled pull-down by biotin-labeled HZ 09 or Phospholipase D1 messenger ribonucleic acid, displaying input, HZ 09 sense, HZ 09 antisense, input, Phospholipase D1 sense, and Phospholipase D1 antisense (columns) and Human antigen R, under Swan 71 and H T E-8 or SVneo. Figure 2X is a western blot titled pull-down by biotin-labeled Phospholipase D1 messenger ribonucleic acid, displaying input, Vector (Phospholipase D1 sense), HZ 09 (Phospholipase D1 antisense), input, Negative control (Phospholipase D1 sense), and si1-HZ 09 (Phospholipase D1 antisense) (columns) and Human antigen R, under Swan 71 and H T E-8 or SVneo. Figure 2Y is a western blot titled pull-down by biotin-labeled HZ 09, displaying input, Vector (HZ 09 sense), HZ 09 (HZ 09 antisense), input, Negative control (HZ 09 sense), and si1-HZ 09 (HZ 09 antisense) (columns) and Human antigen R, under Swan 71 and H T E-8 or SVneo.
Subsequently, the functions of this pathway were explored in human trophoblast cells. Cells (both Swan 71 and HTR-8/SVneo) with PLD1 overexpression had higher protein levels of RAC1 and CDC42, whereas cells with PLD1 knockdown had lower protein levels of RAC1 and CDC42 (Figure S2B–E). Furthermore, Swan 71 or HTR-8/SVneo cells withPLD1 overexpression had less migration and invasion, whereas cells with PLD1 knockdown had greater migration and invasion (Figure S2F–I). Subsequently, the role of lnc-HZ09 in regulation of this pathway was further explored. The mRNA and protein expression levels of PLD1, RAC1 and CDC42 were all lower in lnc-HZ09-overexpressed Swan 71 or HTR-8/SVneo cells and were all higher in lnc-HZ09-silenced cells compared to relevant controls (Figure 2C; S3A–B).
The Effects of lnc-HZ09 on SP1-Mediated PLD1 mRNA Transcription in Human Trophoblast Cells
We next explored how lnc-HZ09 regulated PLD1 expression. First, how lnc-HZ09 affected PLD1 mRNA transcription was studied. It has been reported that SP1 acted as a transcription factor to promote PLD1 mRNA transcription in hepatocytes.43 As analyzed by PROMO software, SP1 might recognize the promoter sequence of PLD1 (Figure S4). Experimentally, Swan 71 and HTR-8/SVneo cells with overexpression of SP1 had higher, whereas cells with SP1 knockdown had lower mRNA and protein expression levels of PLD1 (Figure 2D, S3C–G). SP1 ChIP assays further confirmed that SP1 could bind with the promoter region of PLD1 (Figure 2E–F), indicating that SP1 may act as a transcription factor to facilitate PLD1 transcription in human trophoblast cells. Moreover, cells with overexpression of lnc-HZ09 had lower, whereas cells with lnc-HZ09 knockdown had higher SP1 mRNA and protein levels (Figure 2G; Figure S3H). SP1 ChIP assays further showed that cells with lnc-HZ09 overexpression had greater, whereas cells with knockdown of lnc-HZ09 had lower occupancy of SP1 on the promoter region of PLD1 (Figure 2H,I; Figure S3J,K). Collectively, these results supported that lnc-HZ09 might suppress SP1-mediated PLD1 mRNA transcription.
The Effects of lnc-HZ09 on PLD1 mRNA Stability in Human Trophoblast Cells
Subsequently, we further investigated whether lnc-HZ09 might affect PLD1 mRNA stability. For both Swan 71 and HTR-8/SVneo cells, lnc-HZ09-overexpressed cells had lower, whereas lnc-HZ09-silenced cells had higher PLD1 mRNA stability (Figure 2J,K). As control, alteration of lnc-HZ09 did not affect the mRNA stability of GAPDH in both cells (Figure S5A,B).
It has been reported that HuR is an RNA binding protein, which could bind RNAs containing AUUU specific sequence and enhance their RNA stability.44 Here, we explored whether HuR protein could promote PLD1 mRNA stability in human trophoblast cells. We found that in both Swan 71 and HTR-8/SVneo cells, those with overexpression of HuR had higher, whereas those with HuR knockdown had lower PLD1 mRNA and protein levels (Figure 2L; S5C–F). Similarly, HuR-overexpressed cells had higher mRNA stability and those with HuR knockdown had lower mRNA stability (Figure 2M,N). As control, alteration of HuR did not affect GAPDH mRNA stability in both cells (Figure S5G,H). Furthermore, cells with overexpression of HuR exhibited greater, whereas cells with HuR knockdown exhibited less migration and invasion (Figure S5I–L). Notably, alteration of lnc-HZ09 did not affect the protein level of HuR in both trophoblast cells (Figure S5M,N).
Because HuR is an RNA binding protein, whether HuR could competitively bind with lnc-HZ09 or PLD1 mRNA was determined in human trophoblast cells. RIP assays showed that both PLD1 mRNA and lnc-HZ09 could be pulled down by HuR protein in both Swan 71 and HTR-8/SVneo cells (Figure 2O–R). Furthermore, cells with lnc-HZ09 overexpression had lower levels of PLD1 mRNA that were pulled down by HuR protein in both cells, whereas those with lnc-HZ09 knockdown had higher levels of PLD1 mRNA that were pulled down by HuR protein in both cells (Figure 2S–V). RNA pull-down assays further confirmed that HuR protein could be pulled down by biotin-labeled lnc-HZ09 or PLD1 mRNA but not by their antisense RNAs (Figure 2W). Moreover, cells with overexpression of lnc-HZ09 had lower levels of HuR protein that were pulled down by biotin-labeled PLD1 mRNA, whereas cells with knockdown of lnc-HZ09 had higher levels of HuR protein that were pulled down by biotin-labeled PLD1 mRNA (Figure 2X). Similarly, cells with PLD1 overexpression had lower levels of HuR protein pulled down by biotin-labeled lnc-HZ09, whereas cells with knockdown of PLD1 had higher levels of HuR protein pulled down by biotin-labeled lnc-HZ09 (Figure 2Y).
Regulatory Roles of MSX1 in lnc-HZ09 Transcription in Human Trophoblast Cells
We next explored what regulated lnc-HZ09 expression level in human trophoblast cells. First, the transcription of lnc-HZ09 was studied. It has been reported that MSX1 was a transcription factor that promoted mRNA transcription of fibroblast growth factor 9 in murine myoblast C2C12 cells.45 As identified by PROMO software, MSX1 might recognize the promoter region of lnc-HZ09 (Figure S6). Experimentally, in both Swan 71 and HTR-8/SVneo cells, cells with overexpression of MSX1 had higher expression level of lnc-HZ09, whereas those with knockdown of MSX1 had lower expression level of lnc-HZ09 (Figure 3A–B; Figure S7A). Moreover, MSX1 ChIP assays showed that MSX1 could bind to the promoter region of lnc-HZ09 (Figure 3C,D), indicating that MSX1 may act as a transcription factor to promote lnc-HZ09 transcription.
Figure 3. Expression levels of lnc-HZ09 regulated by MSX1 and m6A RNA methylation in human trophoblast cells. (A–B) RT-qPCR analysis (each n=3) of the levels of lnc-HZ09 in Swan 71 (A) or HTR-8/SVneo (B) cells with overexpression or knockdown of MSX1. (C–D) MSX1 ChIP assay analysis (each n=3) of the relative enrichment of MSX1 in the promoter region of lnc-HZ09 in Swan 71 (C) or HTR-8/SVneo (D) cells. (E–F) MeRIP assay analysis (each n=3) of the levels of m6A RNA methylation on lnc-XIST or lnc-HZ09 in Swan 71 (E) or HTR-8/SVneo (F) cells. (G–H) MeRIP assay analysis (each n=3) of the levels of m6A RNA methylation on lnc-HZ09 in Swan 71 (G) or HTR-8/SVneo (H) cells with overexpression or knockdown of METTL3. (I–J) RT-qPCR analysis (each n=3) of the levels of lnc-HZ09 in Swan 71 (I) or HTR-8/SVneo (J) cells with overexpression or knockdown of METTL3. (K–L) MeRIP assay analysis (each n=3) of the levels of m6A RNA methylation on lnc-HZ09 in Swan 71 (K) or HTR-8/SVneo (L) cells with DAA treatment. (M) RT-qPCR analysis (each n=3) of the levels of lnc-HZ09 in Swan 71 or HTR-8/SVneo cells with DAA treatment. (N) RT-qPCR analysis (each n=3) of the levels of lnc-HZ09 in Swan 71 or HTR-8/SVneo cells with overexpression of METTL3 or overexpression of METTL3 together with DAA treatment. (O) The RNA stability (each n=3) of lnc-HZ09 in Swan 71 cells with overexpression or knockdown of METTL3, or with overexpression of METTL3 together with DAA treatment. (P) Representative western blot analysis of PLD1 protein levels in Swan 71 cells with overexpression or knockdown of METTL3, or with overexpression of METTL3 together with DAA treatment, with GAPDH as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S7F. The summary data of these bar charts and diagrams were shown in Excel Table S1. The expression level in NC or Vector group was set as “1” in all of RT-qPCR and mRNA stability assays; the levels of DNA or RNA pulled down by IgG were set as “1” in all ChIP and MeRIP assays; and the band intensity in NC or Vector group was set as “100” in all western blot assays. (P) shows the representative data from three independent experiments. Data in (A–O) show mean±SD of three independent experiments. Two-tailed Student’s t-test for (A–M); one-way ANOVA analysis for (A,B,I,J,N). *p<0.05, **p<0.01, and ***p<0.001. Note: ANOVA, analysis of variance; ChIP, chromatin immunoprecipitation; DAA, 3-deazaadenosine; GADPH, glyceraldehyde-3-phosphate dehydrogenase; IgG, immunoglobulin G; MeRIP, methylated RNA immunoprecipitation; METTL3, overexpression of METTL3; MSX1, overexpression of MSX1; NC, negative control of siRNA; ns, nonsignificance; PLD1, phospholipase D hydrolyze 1; RT-qPCR, quantitative reverse transcription polymerase chain reaction; SD, standard deviation; si-METTL3, knockdown of METTL3; si-MSX1, knockdown of MSX1; Vector, empty vector of pcDNA3.1.
Figures 3A, 3B, 3I, 3J are bar graphs titled Swan 7, H T R-8 or SVneo, Swan 71, and H T R-8 or SVneo, plotting ribonucleic acid relative expression, ranging from 0.0 to 2.0 in increments of 0.5, 0.0 to 2.0 in increments of 0.5, 0.0 to 1.0 in increments of 0.5 and 75 to 85 in increments of 5; 0.0 to 1.0 in increments of 0.5 and 75 to 90 in increments of 5 (y-axis) across Vector, Homeobox protein M S X-1, negative control, si1-Homeobox protein MSX-1, and si2-Homeobox protein MSX-1; Vector, Homeobox protein M S X-1, negative control, si1-Homeobox protein MSX-1, and si2-Homeobox protein MSX-1; Vector, m(6)A methyltransferase, negative control, si1- m(6)A methyltransferase, and si2- m(6)A methyltransferase; and Vector, m(6)A methyltransferase, negative control, si1- m(6)A methyltransferase, and si2- m(6)A methyltransferase (x-axis) for Inc-HZ 09, respectively. Figures 3C and 3D are bar graphs, titled chromatin immunoprecipitation in Swan 71, chromatin immunoprecipitation in H T R-8 or SVneo, plotting Inc-HZ 09 promoter relative fold enrichment, ranging from 0 to 4 in unit increments (y-axis) across anti-Immunoglobulin G and anti- Homeobox protein M S X-1 (x-axis), respectively. Figures 3E and 3F are bar graphs, titled methylated RNA immunoprecipitation sequencing in Swan 71 and methylated RNA immunoprecipitation sequencing in H T R-8 or SVneo, plotting N 6-methyladenosine relative fold enrichment, ranging from 0 to 6 in increments of 2 (y-axis) across Inc- X Inactive Specific Transcript and Inc-HZ 09 (x-axis) for anti-Immunoglobulin G and anti-N 6-methyladenosine, respectively. Figures 3G and 3H, each are a set of two clustered bar graphs titled methylated RNA immunoprecipitation sequencing in Swan 71 and methylated RNA immunoprecipitation sequencing in H T R-8 or SVneo, plotting Inc-HZ 09 m(6)A methyltransferase relative fold enrichment, ranging from 0 to 10 in increments of 5 (y-axis) across anti-Immunoglobulin G and anti-N 6-methyladenosine (x-axis) for Vector, m(6)A methyltransferase, si1- m(6)A methyltransferase, negative control, respectively. Figures 3K and 3L is clustered graph titled methylated RNA immunoprecipitation sequencing in Swan 71 and methylated RNA immunoprecipitation sequencing in H T R-8 or SVneo, plotting Inc-HZ 09 m(6)A methyltransferase relative fold enrichment, ranging from 0 to 6 in increments of 2 (y-axis) across anti-Immunoglobulin G and anti-N 6-methyladenosine (x-axis) for Dimethyl sulfoxide and Diacetone alcohol, respectively. Figures 3M and 3N are bar graphs, Inc-HZ 09 ribonucleic acid relative fold enrichment, ranging from 0.0 to 1.0 in increments of 0.5 and 0 to 3 in unit increments (y-axis) across Swan 71 and H T R-8 or SVneo x-axis) for Dimethyl sulfoxide, Diacetone alcohol, Vector, methylated RNA immunoprecipitation sequencing, and methylated RNA immunoprecipitation sequencing plus Diacetone alcohol. Figure 3O is a set of two line graphs titled Swan 71, plotting Inc-HZ 09 ribonucleic acid remaining, ranging from 0.0 to 1.0 in increments of 0.5 (y-axis) across Time (hour), ranging from 0 to 5 in unit increments (x-axis) for methylated RNA immunoprecipitation sequencing, methylated RNA immunoprecipitation sequencing plus Diacetone alcohol, vector, and vector plus Dimethyl sulfoxide. Figure 3P is a set of two western blot titled Swan 71. On the left, the western blot displays 100, 36, 183, and 117 (columns) and Phospholipase D1, Glyceraldehyde 3-phosphate dehydrogenase, vector, methylated RNA immunoprecipitation sequencing, Dimethyl sulfoxide, and Diacetone alcohol (rows). On the right, the western blot displays negative control (100) and si- methylated RNA immunoprecipitation sequencing (239) (columns) and Phospholipase D1 and Glyceraldehyde 3-phosphate dehydrogenase (rows).
The Effects of METTL3-Mediated m6A RNA Methylation on lnc-HZ09 Stability in Human Trophoblast Cells
Subsequently, lnc-HZ09 RNA stability was investigated. It has been shown that m6A modification on lncRNAs may regulate the stability of lncRNAs.46 M6A modification site (5′-GGACU-3′41) was identified in lnc-HZ09 sequence using SRAMP software (Table S11). Subsequently, MeRIP assays confirmed the presence of m6A RNA modification on lnc-HZ09 (Figure 3E,F). LncRNA XIST containing m6A RNA modification47 was used as a positive control.
METTL3 is an important RNA methyltransferase to produce m6A RNA methylation on RNAs.48 Herein, in both Swan 71 and HTR-8/SVneo cells, those with overexpression of METTL3 had higher levels of m6A RNA modification on lnc-HZ09, whereas cells with knockdown of METTL3 had lower levels of m6A RNA modification on lnc-HZ09 (Figure 3G,H; Figure S7B) and expression level of lnc-HZ09 (Figure 3I,J), as determined by MeRIP assays and RT-qPCR analysis, respectively. The addition of 3-deazaadenosine (DAA), an inhibitor of m6A RNA methylation,49 resulted in lower m6A modification level on lnc-HZ09 (Figure 3K,L) and lower lnc-HZ09 expression level (Figure 3M). Furthermore, the up-regulation of lnc-HZ09 by overexpressing METTL3 was diminished by the addition of DAA (Figure 3N). Moreover, RNA stability assays also showed that, in Swan 71 and HTR-8/SVneo cells, cells overexpressing METTL3 had greater lnc-HZ09 stability, whereas cells with knockdown of METTL3 had lower lnc-HZ09 stability (Figure 3O; S7C). As control, alteration of METTL3 did not affect GAPDH mRNA stability in both cells (Figure S7D,E). Furthermore, the enhancement of lnc-HZ09 stability by overexpressing METTL3 was diminished by treating trophoblast cells with DAA (Figure 3O; Figure S7C).
Subsequently, the effects of METTL3 on PLD1 expression levels and trophoblast cell functions were also explored. In Swan 71 and HTR-8/SVneo cells, METTL3-overexpressed cells had lower PLD1 protein levels, whereas cells with knockdown of METTL3 or treated with DAA had higher PLD1 protein levels (Figure 3P; Figure S7F–H). Moreover, cells with overexpression of METTL3 had less migration and invasion, an effect that was mitigated by treating cells with DAA (Figure S7I). In contrast, cells with knockdown of METTL3 had greater migration and invasion of (Figure S7J).
The Effects of lnc-HZ09 on Migration and Invasion of BPDE-Exposed Human Trophoblast Cells through PLD1/RAC1/CDC42 Pathway
To discover the underlying mechanism of BPDE effects on migration and invasion of human trophoblast cells, the regulation of lnc-HZ09 on the PLD1/RAC1/CDC42 pathway was investigated in BPDE-exposed trophoblast cells. First, we found that the expression levels of mRNAs and proteins of members in this PLD1/RAC1/CDC42 pathway were all lower (Figure 4A; Figure S8A–D), whereas the levels of lnc-HZ09 were higher (Figure 1A–B), with increasing BPDE concentrations in BPDE-treated Swan 71 or HTR-8 cells. Of BPDE-treated Swan 71 or HTR-8/SVneo cells, those overexpressing lnc-HZ09 had less migration and invasion, whereas cells with knockdown of lnc-HZ09 had greater migration and invasion (Figure 4B–E; Figure S8E–F). In the PLD1/RAC1/CDC42 pathway, the levels of mRNAs and proteins were all lower with lnc-HZ09 overexpression and were all higher with lnc-HZ09 knockdown in BPDE-treated both trophoblast cells (Figure 4F; Figure S8G–N).
Figure 4. PLD1/RAC1/CDC42 pathway regulated by lnc-HZ09 in BPDE-exposed human trophoblast cells. (A) Representative western blot analysis of the protein levels of PLD1, RAC1 and CDC42 in BPDE-treated Swan 71 or HTR-8/SVneo cells, with GAPDH as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S8A,B. (B–E) Representative transwell assay analysis of the migration and invasion of 0.5μM BPDE-treated Swan 71 (B and C) or HTR-8/SVneo (D and E) cells with overexpression or knockdown of lnc-HZ09 (scale bar, 200μm). (F) Representative western blot analysis of the protein levels of SP1, PLD1, RAC1 and CDC42 in 0.5μM BPDE-treated Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of lnc-HZ09, with GAPDH as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S8G–J. (G) Representative western blot analysis of the protein levels of SP1 in BPDE-treated Swan 71 or HTR-8/SVneo cells, with GAPDH as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S8O. (H–I) SP1 ChIP assay analysis (each n=3) of the relative enrichment of SP1 in the promoter region of PLD1 gene in untreated or 0.5μM BPDE-treated Swan 71 (H) or HTR-8/SVneo (I) cells. (J–K) SP1 ChIP assay analysis (each n=3) of the relative enrichment of SP1 in the promoter region of PLD1 gene in 0.5μM BPDE-treated Swan 71 (J) or HTR-8/SVneo (K) cells with overexpression of lnc-HZ09. (L–M) The mRNA stability of PLD1 (each n=3) in 0.5μM BPDE-treated Swan 71 (L) or HTR-8/SVneo (M) cells with overexpression or knockdown of lnc-HZ09. (N) Representative western blot analysis of the protein levels of PLD1 in 0.5μM BPDE-treated Swan 71 or HTR-8/SVneo cells with overexpression or knockdown of HuR, with β-tubulin as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S9C. (O) The mRNA stability of PLD1 (each n=3) in 0.5μM BPDE-treated Swan 71 cells with overexpression or knockdown of HuR. (P) Representative western blot analysis of the protein levels of HuR in 0–1.5μM BPDE-treated Swan 71 or HTR-8/SVneo cells, with β-tubulin as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S9H. (Q–R) RIP assay analysis (each n=3) of the relative levels of PLD1 mRNA that was pulled down by HuR protein in 0.5μM BPDE-treated Swan 71 (Q) or HTR-8/SVneo (R) cells with overexpression of lnc-HZ09. (S) Representative western blot analysis of MSX1 protein levels in 0–1.5μM BPDE-treated Swan 71 or HTR-8/SVneo cells, with GAPDH as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S9L. (T) MSX1 ChIP assay analysis (each n=3) of the relative enrichment of MSX1 in the promoter region of lnc-HZ09 in untreated or 0.5μM BPDE-treated Swan 71 cells. (U) Representative western blot analysis of the protein levels of METTL3 in BPDE-treated Swan 71 or HTR-8/SVneo cells, with GAPDH as internal standard. The relative intensity of each band was quantified and their mean±SD of three replicates was shown in Figure S9O. (V) MeRIP assay analysis (each n=3) of the relative levels of m6A RNA methylation on lnc-HZ09 in untreated or 0.5μM BPDE-treated Swan 71 cells. (W) The RNA stability of lnc-HZ09 (each n=3) in untreated or 0.5μM BPDE-treated Swan 71 cells. (X) The mRNA stability of PLD1 (each n=3) in untreated or 0.5μM BPDE-treated Swan 71 cells. The summary data of these bar charts and diagram were shown in Excel Table S1. The RNA levels in NC or Vector group were set as “1” in all of mRNA stability assays; the DNA or RNA levels in IgG group were set as “1” in all of ChIP, RIP, and MeRIP assays; and the band intensity in NC or Vector group was set as “100” in all of western blot assays. (A–G,N,P,S,U) show the representative data from three independent experiments. Data in (H–M,O,Q,R,T,V–X) show mean±SD of three independent experiments. Two-tailed Student’s t-test for (H–K, Q–R, T, V); *p<0.05, **p<0.01, and ***p<0.001. Note: BPDE, benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide; ChIP, chromatin immunoprecipitation; GADPH, glyceraldehyde-3-phosphate dehydrogenase; HuR, overexpression of HuR; HZ09, overexpression of lnc-HZ09; IgG, immunoglobulin G; MeRIP, methylated RNA immunoprecipitation; NC, negative control of siRNA; ns, nonsignificance; PLD1, phospholipase D hydrolyze 1; RIP, RNA immunoprecipitation; SD, standard deviation; si-HuR, knockdown of HuR; si-HZ09, knockdown of lnc-HZ09; Vector, empty vector of pcDNA3.1.
Figure 4A is a set of two western blots, displaying 100, 85, 73, 1, and 17 under Swan 71 and 100, 80, 61, 22, and 11 under H T R-8 or SVneo (columns) and Phospholipase D1, Ras-related C3 botulinum toxin substrate 1, Cell division control protein 42 homolog, Glyceraldehyde 3-phosphate dehydrogenase (rows), respectively. The Benzo(a)pyrene diol epoxide, micromolar ranges from 0to 1.5 in increments of 0.5. Figure 4B is a stained tissue titled Benzo(a)pyrene diol epoxide treated Swan 71, displaying vector and HZ 09 (columns) and migration and invasion (rows). Figure 4C is a stained tissue titled Benzo(a)pyrene diol epoxide treated Swan 71, displaying negative control, si1-HZ 09, and si2-HZ 09 (columns) and migration and invasion (rows). Figure 4D is a stained tissue titled Benzo(a)pyrene diol epoxide treated H T R-8 or SVneo, displaying vector and HZ 09 (columns) and migration and invasion (rows). Figure 4E is a stained tissue titled Benzo(a)pyrene diol epoxide treated H T R-8 or SVneo, displaying negative control, si1-HZ 09, and si2-HZ 09 (columns) and migration and invasion (rows). Figure 4F is a set of two western blots titled Benzo(a)pyrene diol epoxide treated, displaying Vector (100), HZ 09 (72), Negative control (100), si1-HZ 09 (187), si2-HZ 09 (219), and Vector (100), HZ 09 (63), Negative control (100), si1-HZ 09 (136), si2-HZ 09 (184) (columns) and specificity protein 1, Phospholipase D1, Ras-related C3 botulinum toxin substrate 1, Cell division control protein 42 homolog, and Glyceraldehyde 3-phosphate dehydrogenase (rows). Figure 4G is a western blot, displaying 100, 93, 86, 28, and 15 (columns) and specificity protein 1 and Glyceraldehyde 3-phosphate dehydrogenase, each under Swan 71 and H T R-8 or SVneo (rows). The Benzo(a)pyrene diol epoxide, micromolar ranges from 0to 1.5 in increments of 0.5. Figures 4H, 4I, 4J, 4K, 4Q, 4R, 4T, 4V are clustered bar graphs chromatin immunoprecipitation in Swan 71, chromatin immunoprecipitation in H T R-8 or SVneo, chromatin immunoprecipitation in Benzo(a)pyrene diol epoxide-treated Swan 71, chromatin immunoprecipitation in Benzo(a)pyrene diol epoxide-treated H T R-8 or SVneo, receptor-interacting protein in Benzo(a)pyrene diol epoxide-treated Swan 71, receptor-interacting protein in Benzo(a)pyrene diol epoxide-treated H T R-8 or SVneo, chromatin immunoprecipitation in Swan 71, methylated RNA immunoprecipitation sequencing in Swan 71, plotting Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 4 in unit increments, Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 4 in unit increments, Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 4 in unit increments, Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 4 in unit increments, messenger ribonucleic acid relative expression, ranging from 0 to 4 in increments of 2, messenger ribonucleic acid relative expression, ranging from 0 to 4 in increments of 2, lnc-HZ09 promoter relative fold enrichment, ranging from 0 to 8 in increments of 2, and Lnc-HZ09 messenger ribonucleic acid relative fold enrichment, ranging from 0 to 10 in increments of 5 (y-axis) across anti-Immunoglobulin G and anti- specificity protein 1, anti-Immunoglobulin G and anti- specificity protein 1, anti-Immunoglobulin G and anti- specificity protein 1, anti-Immunoglobulin G and anti- specificity protein 1, anti-Immunoglobulin G and anti- Human antigen R, anti-Immunoglobulin G and anti- Human antigen R, anti-Immunoglobulin G and anti- Homeobox protein MSX-1, and anti-Immunoglobulin G and anti- N6-Methyladenosine (x-axis) for Dimethyl sulfoxide, Benzo(a)pyrene diol epoxide, vector, and HZ 09, respectively. Figures 4L, 4M, 4O, each are set of two line graphs titled Benzo(a)pyrene diol epoxide-treated Swan 71, Benzo(a)pyrene diol epoxide-treated H T R-8 or SVneo, and Benzo(a)pyrene diol epoxide-treated Swan 71, plotting Phospholipase D1 messenger ribonucleic acid remaining, ranging from 0.0 to 1.0 in increments of 0.5 (y-axis) across Time (hour), ranging from 0 to 5 in unit increments (x-axis) for Vector, HZ 09, Si1,2 HZ 09, negative control, and Human antigen R, respectively. Figure 4N is a western blot titled Benzo(a)pyrene diol epoxide treated, displaying Vector (100), HZ 09 (134), Negative control (100), si1-HZ 09 (47), si2-HZ 09 (13) (columns) and Phospholipase D1 and lowercase beta-tubulin, each under Swan 71 and H T R-8 or SVneo (rows). Figure 4P is a western blot, displaying 100, 92, 83, 45, and 23 (columns) and Human antigen R and Glyceraldehyde 3-phosphate dehydrogenase, each under Swan 71 and H T R-8 or SVneo (rows). The Benzo(a)pyrene diol epoxide, micromolar ranges from 0to 1.5 in increments of 0.5. Figure 4S is a western blot, displaying 100, 156, 198, 216, and 374 (columns) and Homeobox protein MSX-1 and Glyceraldehyde 3-phosphate dehydrogenase, each under Swan 71 and H T R-8 or SVneo (rows). The Benzo(a)pyrene diol epoxide, micromolar ranges from 0to 1.5 in increments of 0.5. Figure 4U is a western blot, displaying 100, 113, 187, 173 and 219 (columns) and m(6)A methyltransferase and Glyceraldehyde 3-phosphate dehydrogenase, each under Swan 71 and H T R-8 or SVneo (rows). The Benzo(a)pyrene diol epoxide, micromolar ranges from 0to 1.5 in increments of 0.5. Figures 4W and 4X are line graphs titled Swan 71, Lnc-HZ09 messenger ribonucleic acid remaining, ranging from 0.0 to 1.0 in increments of 0.5 and Phospholipase D1 messenger ribonucleic acid remaining, ranging from 0.0 to 1.0 in increments of 0.5 (y-axis) across Time (hour), ranging from 0 to 5 in unit increments (x-axis).
The Effects of lnc-HZ09 on PLD1 mRNA Transcription and mRNA Stability in BPDE-Exposed Human Trophoblast Cells
How lnc-HZ09 regulated PLD1 mRNA transcription was further investigated in BPDE-treated trophoblast cells. The mRNA and protein levels of the transcription factor SP1 were lower in BPDE-treated Swan 71 or HTR-8/SVneo cells (Figure 4G; Figure S8O–P). SP1 ChIP assays showed that the occupancy of SP1 on the promoter region of PLD1 was less after BPDE treatment in both cells (Figure 4H–I). In BPDE-treated Swan 71 or HTR-8/SVneo cells, cells with lnc-HZ09 overexpression had lower SP1 mRNA and protein levels, whereas cells with lnc-HZ09 knockdown had higher SP1 mRNA and protein levels (Figure 4F; Figure S8G–J,Q–R). Furthermore, SP1 ChIP assays showed that in BPDE-treated Swan 71 or HTR-8/SVneo cells, cells with overexpression of lnc-HZ09 had less occupancy of SP1 on the promoter region of PLD1, whereas cells with knockdown of lnc-HZ09 had greater occupancy of SP1 on the promoter region of PLD1 (Figure 4J–K; Figure S8S–T).
Furthermore, the effects of lnc-HZ09 on PLD1 mRNA stability were also studied in BPDE-treated trophoblast cells. In BPDE-treated trophoblast cells, cells with overexpression of lnc-HZ09 had lower, whereas cells with knockdown of lnc-HZ09 had greater PLD1 mRNA stability (Figure 4L–M). As control, alteration of lnc-HZ09 did not affect GAPDH mRNA stability in BPDE-treated both cells (Figure S9A,B). In BPDE-treated Swan 71 or HTR-8/SVneo cells, cells with overexpression of HuR had higher PLD1 mRNA and protein levels, whereas cells with knockdown of HuR had lower PLD1 mRNA and protein levels (Figure 4N; Figure S9C–F). Furthermore, BPDE-treated Swan 71 cells overexpressing HuR also had greater PLD1 mRNA stability; BPDE-treated Swan 71 cells with HuR knockdown had less PLD1 mRNA stability (Figure 4O). However, alteration of HuR did not affect GAPDH mRNA stability (Figure S9G). Moreover, cells (both Swan 71 and HTR8/SVneo) treated with BPDE had lower HuR protein expression levels (Figure 4P; Figure S9H). However, HuR levels were not affected by lnc-HZ09 in BPDE-treated trophoblast cells (Figure S9I,J). Furthermore, cells with overexpression of lnc-HZ09 had lower levels of PLD1 mRNA pulled down by HuR protein in BPDE-treated trophoblast cells, whereas those with knockdown of lnc-HZ09 had higher levels of PLD1 mRNA pulled down by HuR protein in BPDE-treated trophoblast cells (Figure 4Q,R; Figure S9K,L), suggesting lnc-HZ09 competed with PLD1 mRNA to bind with HuR.
Examination of lnc-HZ09 Transcription and Stability in BPDE-Exposed Human Trophoblast Cells
Subsequently, lnc-HZ09 transcription was investigated in BPDE-treated trophoblast cells. The mRNA and protein levels of MSX1 were higher in BPDE-treated trophoblast cells (Figure 4S; Figure S9M,N). ChIP assays showed that the occupancy of MSX1 on the promoter region of lnc-HZ09 was greater in BPDE-treated Swan 71 or HTR-8/SVneo cells (Figure 4T; Figure S9O).
Finally, lnc-HZ09 stability was also investigated in BPDE-treated trophoblast cells. The mRNA and protein levels of METTL3 (Figure 4U; Figure S9P,Q), as well as the levels of m6A RNA methylation on lnc-HZ09 (Figure 4V; Figure S9R), were all higher in BPDE-treated trophoblast cells. Lnc-HZ09 RNA stability was greater in BPDE-treated Swan 71 cells, whereas PLD1 mRNA stability was lower in BPDE-treated Swan 71 cells (Figure 4W,X). As control, GAPDH mRNA stability was not affected with BPDE treatment (Figure S9S,T).
The Correlation of lnc-HZ09 with PLD1/RAC1/CDC42 Pathway in RM Villous Tissues
Having known the regulatory roles of lnc-HZ09 in trophoblast cells, its roles in villous tissues were further explored. We collected villous tissue samples from unexplained the RM and the matched healthy control (HC) groups (each n=15). The parameters, such as body mass index, age, gestational days, did not show significant differences between the RM and HC groups (Table S12). However, the levels of BPDE-DNA adducts were significantly higher in RM group than those in HC group (Table S12; Figure S10A). We also found that lnc-HZ09 was significantly highly expressed in RM group relative to those in HC group (Figure 5A), and this trend was consistent with that observed in BPDE-treated human trophoblast cells, suggesting that lnc-HZ09 might simultaneously regulate BPDE-induced dysfunctions of human trophoblast cells and the occurrence of miscarriage. The mRNA and protein levels of PLD1/RAC1/CDC42 were all lower in RM tissues than those in HC tissues (Figure 5B,C; Figure S10B,C). Correlation analysis showed that the levels of PLD1, RAC1, and CDC42 were all negatively correlated with those of lnc-HZ09 in RM tissues (Figure 5D,E; Figure S10D–G). The locations of most data points in HC and RM groups were relatively separated, manifesting that this pathway was differentially regulated.
Figure 5. Regulation roles of lnc-HZ09 in human villous tissues. (A–B) RT-qPCR analysis of the levels of lnc-HZ09 (A) or PLD1 mRNA (B) in HC (healthy control, round) and RM (recurrent miscarriage, square) tissues (each n=15). (C) Western blot analysis of the protein levels of SP1, PLD1, RAC1, and CDC42 in HC and RM tissues (each n=10), with GAPDH as internal standard. The relative intensity of each band was quantified, and their levels were shown in Figure S10C. (D) The correlation between PLD1 mRNA levels and lnc-HZ09 levels in HC (round) and RM (square) tissues (each n=15). (E) The correlation between PLD1 protein levels and lnc-HZ09 levels in HC (round) and RM (square) groups (each n=10). (F) RT-qPCR analysis (each n=3) of the mRNA levels of SP1 in HC (round) and RM (square) tissues (each n=15). (G) The correlation between the protein levels of PLD1 and SP1 in HC (round) and RM (square) groups (each n=10). (H) SP1 ChIP assay analysis of the relative enrichment of SP1 in the promoter region of PLD1 gene in HC and RM tissues (each n=6). (I) The correlation between SP1 protein levels and lnc-HZ09 levels in HC (round) and RM (square) groups (each n=10). (J) RT-qPCR analysis of the mRNA levels of MSX1 in HC and RM tissues (each n=15). (K) Western blot analysis of the protein levels of HuR, MSX1, and METTL3 in HC and RM tissues (each n=10), with GAPDH as internal standard. The relative intensity of each band was quantified, and their levels were shown in Figure S10H,J,K. (L) MSX1 ChIP assay analysis of the relative enrichment of MSX1 in the promoter region of lnc-HZ09 in HC and RM tissues (each n=6). (M) The correlation between the levels of lnc-HZ09 and the protein levels of MSX1 in HC (round) and RM (square) groups (each n=10). (N) RT-qPCR analysis (each n=3) of the mRNA levels of METTL3 in HC and RM tissues (each n=15). (O) MeRIP assay analysis (each n=3) of the levels of m6A RNA methylation on lnc-HZ09 in HC and RM tissues (each n=6). The summary data of these bar charts and scatter plots were shown in Excel Table S1. The DNA or RNA level in IgG group was set as “1” in all of ChIP and MeRIP assays and the middle intensity value was set to “100” in all of western blot assays. (C–E, G, I, K, M) shows representative data from three independent experiments. Data in (H, L, O) show mean±SD of six independent experiments. Two-tailed Student’s t-test for (A–B,F,H,J,L,N–O); Pearson analysis for (D,E,G,I,M). *p<0.05, **p<0.01, and ***p<0.001. Note: ChIP, chromatin immunoprecipitation; GADPH; glyceraldehyde-3-phosphate dehydrogenase; HC, healthy control group; IgG, immunoglobulin G; MeRIP, methylated RNA immunoprecipitation; n, the number of biologically independent samples; ns, nonsignificance; PLD1, phospholipase D hydrolyze 1; RM, recurrent miscarriage group.
Figures 5A, 5B, 5F, 5N are error bar graphs titled Inc-HZ 09, Phospholipase D1 messenger ribonucleic acid, specificity protein 1, m(6)A methyltransferase, plotting ribonucleic acid relative expression, ranging from 0 to 3 in unit increments, ribonucleic acid relative expression, ranging from 0 to 2 in unit increments, messenger ribonucleic acid relative expression, ranging from 0 to 2 in unit increments, and messenger ribonucleic acid relative expression, ranging from 0 to 10 in increments of 5 (y-axis) across H C and R M (x-axis), respectively. Figures 5D, 5E, 5G, 5I, 5M are graphs titled Phospholipase D1 messenger ribonucleic acid-Inc-HZ 09, Phospholipase D1 Protein-Inc-HZ 09, Phospholipase D1 Protein- specificity protein 1 protein, specificity protein 1-IncHZ 09, In HZ 09- Homeobox protein MSX-1 protein, plotting Phospholipase D1 messenger ribonucleic acid-relative expression, ranging from 0.0 to 2.5 in increments of 0.5; Phospholipase D1 protein relative expression, ranging from 0 to 300 in increments of 100, Phospholipase D1 protein relative expression, ranging from 0 to 800 in increments of 200, specificity protein 1 protein relative expression, ranging from 0 to 200 in increments of 100, Homeobox protein MSX-1 protein relative expression, ranging from 0 to 250 in increments of 200 (y-axis) across Inc-HZ 09 relative expression, ranging from 0 to 4 in unit increments; Inc-HZ 09 relative expression, ranging from 0.5 to 2.5 in increments of 0.5; specificity protein 1 protein relative expression, ranging from 0 to 250 in increments of 50; Inc-HZ 09 relative expression, ranging from 0.5 to 2.5 in increments of 0.5, HZ 09 relative expression, ranging from 0.0 to 2.5 in increments of 0.5 (x-axis), respectively. Figure 5C is a set of two western blots. On the top, the western blot displays 124, 147, 89, 100, 163 under H C and 94, 36, 21, 92, 24 under R M (columns) and specificity protein 1, Phospholipase D1, Ras-related C3 botulinum toxin substrate 1, Cell division control protein 42 homolog, and Glyceraldehyde 3-phosphate dehydrogenase (rows). At the bottom, 73, 193, 96, 125, 159 under H C and 14, 19, 23, 148, 100 under P M (columns) and specificity protein 1, Phospholipase D1, Ras-related C3 botulinum toxin substrate 1, Cell division control protein 42 homolog, and Glyceraldehyde 3-phosphate dehydrogenase (rows). Figures 5H, 5L, 5O is a clustered bar graph, chromatin immunoprecipitation in tissues, chromatin immunoprecipitation in tissues, and methylated RNA immunoprecipitation sequencing in tissues, plotting Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 4 in increment of 2; Inc-HZ 09 promoter relative fold enrichment, ranging from 0 to 6 in increment of 2, Inc-HZ 09 m(6)A methyltransferase relative fold enrichment, ranging from 0 to 9 in increments of 3 (y-axis) across anti-Immunoglobulin G and anti- specificity protein 1, anti-Immunoglobulin G and Homeobox protein MSX-1, and anti-Immunoglobulin G and anti-N6-Methyladenosine (x-axis) for H C and R M, respectively. Figure 5K is a set of two western blots. On the left, the western blot displays 196, 153, 108, 100, 93 under H C and 21, 46, 130, 31, 87 under R M (columns) and Human antigen R, Homeobox protein MSX-1, m(6)A methyltransferase and Glyceraldehyde 3-phosphate dehydrogenase (rows). On the right, 92, 100, 126, 131, 149 under H C and 74, 79, 43, 25, 98 under R M (columns) and Human antigen R, Homeobox protein MSX-1, m(6)A methyltransferase and Glyceraldehyde 3-phosphate dehydrogenase (rows).
The Effects of lnc-HZ09 on PLD1 mRNA Transcription and mRNA Stability in RM Villous Tissues
Subsequently, the effects of lnc-HZ09 on PLD1 mRNA transcription were studied in villous tissues. The mRNA and protein levels of PLD1 and its transcription factor SP1 were all expressed at a low level in RM tissues relative to those in HC tissues (Figure 5B,C,F; Figure S10C). The levels of PLD1 and SP1 were positively correlated in RM tissues (Figure 5G). SP1 ChIP assays showed that the occupancy of SP1 on the promoter region of PLD1 was lower in RM tissues relative to that in HC tissues (Figure 5H), suggesting that SP1-mediated PLD1 transcription was suppressed in RM tissues. Furthermore, the levels of lnc-HZ09 were negatively correlated with the levels of SP1 in RM tissues (Figure 5I).
Then, the effects of lnc-HZ09 on PLD1 mRNA stability were also studied in villous tissues. The levels of HuR were lower in RM tissues in comparison with those in HC tissues (Figure 5K; Figure S10H). The protein levels of HuR and PLD1 were positively correlated in RM tissues (Figure S10I).
Measurement of lnc-HZ09 Transcription and RNA Stability in RM Tissues
In tissues, the mRNA and protein levels of MSX1 were higher in RM tissues relative to HC tissues (Figure 5J,K; Figure S10J). MSX1 ChIP assays showed that the occupancy of MSX1 on the promoter region of lnc-HZ09 was greater in RM tissues relative to HC tissues (Figure 5L). The levels of MSX1 were positively correlated with the levels of lnc-HZ09 in RM tissues (Figure 5M).
Lnc-HZ09 stability was also studied in villous tissues. The mRNA and protein levels of METTL3 (Figure 5K,N; Figure S10K), as well as the levels of m6A RNA methylation on lnc-HZ09 (Figure 5O), were all higher in RM tissues relative to those in HC tissues. The levels of METTL3 were positively correlated with the levels of lnc-HZ09 in RM tissues (Figure S10L).
Evaluation of Sp1-Mediated Pld1 mRNA Transcription in Placental Tissues of Mice with B(a)P-Induced Miscarriage
To investigate the mechanism by which B(a)P induced miscarriage in vivo, we constructed a mouse model by treating pregnant mice with 0, 0.05, or 0.2mg/kg B(a)P to induce miscarriage, as described previously.28,29,31 Considering that B(a)P is metabolized into BPDE in organisms, we treated mice with B(a)P directly. After sequence alignment, we found that PLD1, RAC1, CDC42, and SP1 genes are evolutionarily conservative in human, rhesus, elephants, dogs, and mice (Figure S11A–D; Table S13), implying that this migration and invasion pathway might be conserved among these species. However, lnc-HZ09 sequence was conserved in only human and rhesus, but not in mice (Figure S11E). Using this miscarriage model, we found that the mRNA and protein levels of murine Pld1, Rac1 and Cdc42 were all lower in placental tissues with increasing B(a)P concentrations (Figure 6A–C,E; Figure S11F). This change trend was consistent with that found in BPDE-treated human trophoblast cells and in RM tissues.
Figure 6. Pld1/Rac1/Cdc42 pathway in placental tissues of mice with BaP-induced miscarriage. (A–D) RT-qPCR analysis of the mRNA levels of Pld1 (A), Rac1 (B), Cdc42 (C), and Sp1 (D) in each B(a)P-treated mouse group (each n=8). (E) Western blot analysis of the protein levels of Sp1, Pld1, Rac1, and Cdc42 in each B(a)P-treated mouse group (each n=6), with GAPDH as internal standard. The relative intensity of each band was quantified and their levels were shown in Figure S11F. (F) Sp1 ChIP assay analysis of the relative enrichment of Sp1 in the promoter region of Pld1 gene in control and 0.2mg/kg B(a)P-treated mouse groups (each n=6). (G) The correlation between the protein levels of Pld1 and Sp1 in control (round) and 0.2mg/kg B(a)P-treated (square) groups (each n=6). (H) The proposed regulation mechanism of lnc-HZ09. Lnc-HZ09 suppressed SP1-mediated PLD1 mRNA transcription and also reduced PLD1 mRNA stability by competitively restraining the binding of PLD1 mRNA with HuR, a protein that could stabilize PLD1 mRNA. BPDE exposure promoted MSX1-mediated lnc-HZ09 transcription and also enhanced lnc-HZ09 RNA stability by up-regulating its m6A methylation modification. Thus, BPDE exposure might up-regulate lnc-HZ09 expression level, suppress PLD1/RAC1/CDC42 pathway, inhibit the migration and invasion, and might further induce miscarriage. The summary data of these bar charts and scatter plots were shown in Excel Table S1. The DNA level in IgG group was set as “1” in ChIP assay, and the middle intensity value was set as “100” in western blot assay. (A–E,G) shows representative data from three independent experiments. Data in (F) show mean±SD of six independent experiments. Two-tailed Student’s t-test for (F); one-way ANOVA analysis for (A–D), Pearson analysis for (G). (H) was generated by Microsoft Office PowerPoint. *p<0.05, **p<0.01, and ***p<0.001. Note: B(a)P, benzo(a)pyrene; BPDE, benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide; ChIP, chromatin immunoprecipitation; GADPH, glyceraldehyde-3-phosphate dehydrogenase; IgG, immunoglobulin G; n, the number of biologically independent samples; ns, nonsignificance; RT-qPCR, quantitative reverse transcription polymerase chain reaction; SD, standard deviation.
Figures 6A to 6D are error bar graphs, plotting messenger ribonucleic acid relative expression, ranging from 0.0 to 1.0 in increments of 05; 0.0 to 1.0 in increments of 0.5; 0 to 2 in unit increments and 0 to 2 in unit increments (y-axis) across biofilm-associated protein (milligram per kilogram), ranging from 0 to 0.05 in increments of 0.05 and 0.05 to 0.2 in increments of 0.15 (x-axis) for Phospholipase D1, Ras-related C3 botulinum toxin substrate 1, Cell division control protein 42 homolog, and specificity protein 1. Figure 6E is a set of two western blots, plotting specificity protein 1, Phospholipase D1, Ras-related C3 botulinum toxin substrate 1, Cell division control protein 42 homolog, and Glyceraldehyde 3-phosphate dehydrogenase (rows) across biofilm-associated protein (milligram per kilogram), ranging from 0 to 0.05 in increments of 0.05 and 0.05 to 0.2 in increments of 0.15 (columns). Figure 6F is a clustered bar graph titled chromatin immunoprecipitation in tissues, plotting Phospholipase D1 promoter relative fold enrichment, ranging from 0 to 6 in increments of 2 (y-axis) across Anti- Immunoglobulin G and Anti- specificity protein 1 (x-axis) for 0 and biofilm-associated protein. Figure 6H is a schematic flowchart with eight steps. Step 1: Biofilm-associated protein or Benzo(a)pyrene diol epoxide interacts with Msh Homeobox 1 to activate promoter and transcription, and biofilm-associated protein or Benzo(a)pyrene diol epoxide interacts with m(6) to activate transcription. A methyltransferase leads to Inc-HZ09. Step 2: Benzo(a)pyrene diol epoxide exposed trophoblast cells to Inc-HZ09 (to varying degrees) results in degradation. Step 3: The interaction of Inc-HZ09 and specificity protein 1 results in promoter and transcription. Step 4: The promoter and transcription lead to phospholipase D1 messenger ribonucleic acid. Step 5: The messenger ribonucleic acid leads to phospholipase D1, Ras-related C3 botulinum toxin substrate 1, or cell division control protein 42 homolog. Step 6: Invasion and migration are aided by the phospholipase D1 or Ras-related C3 botulinum toxin substrate 1 or cell division control protein 42 homolog. Step 7: Miscarriage occurs as a result of the invasion and migration. Step 8: The phospholipase D1 messenger ribonucleic acid (more or less) leads to degradation. Figure 6G is a graph titled phospholipase D1 protein relative expression, ranging from 0 to 150 in increments of 50 (y-axis) across specificity protein 1 protein relative expression (x-axis).
Furthermore, the transcription of Pld1 was also explored in B(a)P-treated mouse model. The mRNA and protein levels of murine transcription factor Sp1 were lower with increasing B(a)P concentrations (Figure 6D,E). Moreover, Sp1 ChIP assays showed that the occupancy of Sp1 on the promoter region of Pld1 was lower in B(a)P-treated mouse group relative to those in the untreated group (Figure 6F). The levels of Sp1 were positively correlated with the levels of Pld1 in 0.2mg/kg group (Figure 6G). The locations of most data points in these groups were relatively separated.
Discussion
Increasing attention has been paid to environmental carcinogens and human reproductive health. B(a)P, one of the most widely spread and unavoidably environmental carcinogens, could induce various adverse pregnancy outcomes.50 Emerging studies have revealed that lncRNAs may regulate the occurrence of miscarriage.26,51 In our recent works, we identified several novel lncRNAs that regulated trophoblast cell proliferation, apoptosis, and other cell functions by different pathways.28,29 However, evidence that lncRNAs regulate trophoblast invasion and migration under B(a)P exposure conditions and thus affect miscarriage is still lacking. In this work, we performed in vitro cellular experiments, human tissue experiments, and mouse model experiments, and we found that lnc-HZ09 was highly expressed in BPDE-treated human trophoblast cells and in RM tissues relative to HC tissues (Figure 6H). The mRNA and protein levels of members of the PLD1/RAC1/CDC42 pathway were lower in BPDE-treated human trophoblast cells, in RM relative to HC tissues, and in the placental tissues of B(a)P-treated mice. All these results suggest that lnc-HZ09 suppressed the invasion and migration of trophoblast cells by down-regulating the PLD1/RAC1/CDC42 pathway in BPDE-exposed human trophoblast cells and in RM tissues. However, by searching the NCBI database, we did not find the murine counterpart of lnc-HZ09 in mouse systems, implying that lnc-HZ09 might have specific epigenetic regulation roles in human system.
Regulation Mechanisms of lnc-HZ09
The regulation mechanisms of lnc-HZ09 in human trophoblast cells were proposed (Figure 6H). Based on these data, we hypothesize that MSX1, a transcription factor of lnc-HZ09, promoted lnc-HZ09 transcription. Furthermore, we suggest that METTL3 promoted m6A methylation modification on lnc-HZ09 and enhanced its RNA stability. Thus, both MSX1 and METTL3 may positively regulate lnc-HZ09 expression level. Subsequently, we hypothesize that lnc-HZ09 suppressed SP1 expression, which was a transcription factor of PLD1 and thus inhibited SP1-mediated PLD1 transcription and reduced PLD1 expression level. In addition, lnc-HZ09 and PLD1 mRNA may competitively bind with HuR, which is an RNA binding protein and could maintain RNA stability. We posit that lnc-HZ09 impaired the binding of PLD1 mRNA with HuR and thus reduced PLD1 mRNA stability and its expression level. Therefore, we suggest that lnc-HZ09 finally down-regulated PLD1 expression level, which further suppressed the PLD1/RAC1/CDC42 pathway and inhibited the migration and invasion of human trophoblast cells. Once trophoblast cells are exposed to environmental B(a)P or BPDE, the expression levels of MSX1 and METTL3 would be increased, which subsequently increased the lnc-HZ09 level, down-regulated the PLD1/RAC1/CDC42 pathway, suppressed the migration and invasion of trophoblast cells, and might further induce miscarriage.
Roles of lnc-HZ09 in Regulation of PLD1 Expression
In general, PLD1 is abnormally up-regulated in various cancers and is related to tumor malignancy, maintenance of self-renewal of cancer stem cells and resistance to radiotherapy and chemotherapy.52 PLD1 was shown to promote the invasion, migration, and proliferation of glioblastoma cell lines.53 In glioblastoma, abnormally elevated transcription factor SP1 enhanced PLD1 transcription and increased PLD1 expression level, which further enhanced the drug resistance of glioblastoma tumor cells.53 It has been proposed that early embryonic development and tumor metastasis might have similar biological manifestations.54 In our work, we found that lnc-HZ09 suppressed SP1-mediated PLD1 transcription in human RM tissues and in BPDE-treated human trophoblast cells, which might possibly be similar in cancer cell lines. Moreover, lnc-HZ09 also restrained the binding of PLD1 mRNA with HuR and reduced PLD1 mRNA stability. Finally, due to the regulatory roles of lnc-HZ09, PLD1 transcription was inhibited and its degradation was increased, and thus the expression level of PLD1 was ultimately decreased. However, the actual regulatory mechanisms of lnc-HZ09 on PLD1 expression should be further explored in a specific cancer cell line.
The Roles of m6A Modification in Regulation of Miscarriage
M6A modification regulates various physiological and pathological processes.55 It has been reported that linc1281 played an important role in proper differentiation of embryonic stem cells by acting as a ceRNA to attenuate the functions of let-7 miRNAs.56 The m6A enrichment on linc1281 was required for this linc1281-mediated ceRNA model. In our recent work, we found that m6A modification on lnc-HZ01 enhanced lnc-HZ01 RNA stability, further regulated trophoblast cell proliferation, and was associated with the occurrence of miscarriage.29 In this work, we found that lnc-HZ09 also contained m6A modification, which was associated with greater lnc-HZ09 stability and higher expression level. In BPDE-treated trophoblast cells and RM tissues, the levels of m6A modification on lnc-HZ09 are higher, which was associated with higher lnc-HZ09 expression level, and less migration and invasion of trophoblast cells.
The Upstream Effects of Environment Carcinogens on Trophoblast-Related Adverse Pregnancy Outcomes
To study the possible causes and mechanism of unexplained RM, we collected villous tissue samples from RM and HC groups, with exclusion of the known causes of miscarriage. We found that the levels of BPDE-DNA adducts were significantly higher in the RM group than those in the HC group. This effect is possibly because B(a)P is a ubiquitous environmental carcinogen,57,58 and some women may inevitably intake more B(a)P. It has been reported that female smokers have significantly higher levels of BaP (1.32±0.68 ng/mL) in their follicular fluid in comparison with their nonsmoking counterparts (0.03±0.01 ng/mL).13 The levels of BPDE-DNA adducts in RM and HC villous tissues detected in this study also agreed with a previous case–control study of miscarriage, in which 2.2-fold more BPDE-DNA adducts were detected in maternal blood of the miscarriage group relative to the health control group.14 In this work, our data showed that BPDE exposure might up-regulate the lnc-HZ09 level, inhibit the PLD1/RAC1/CDC42 pathway, suppress migration and invasion, and finally induce miscarriage. Notably, it would likely be the suppressed invasion and migration of trophoblast cells rather than BPDE or B(a)P exposure that directly induces miscarriage. When the PLD1/RAC1/CDC42 pathway is inhibited, the upstream BPDE exposure might become less vital for the ultimate miscarriage. In a larger sense, not only BPDE or B(a)P but other environmental factors or pathways might also suppress trophoblast cell invasion and migration. For example, bisphenol A and para-nonylphenol suppressed the migration and invasion of HTR-8/SVneo cells.59 Heavy metal cadmium (Cd)-induced apoptosis and inhibited migration and invasion of HTR-8/SVneo cells in a dose-dependent manner.60 Besides of miscarriage, dysfunctions of trophoblast cells might also induce other trophoblast-related adverse pregnancy outcomes, such as eclampsia, preeclampsia, intrauterine growth restriction, and gestational diabetes. Epidemiological studies have shown that PAHs exposure might be correlated with these adverse pregnancy outcomes.15,62 BPDE inhibited the invasion and migration of trophoblast cells, which implies that BPDE might be one possible compound that induces these adverse pregnancy outcomes. In a larger sense, other environmental carcinogens, such as air pollutants,63 toxic metals (Cd and palladium),64,65 or trihalomethanes in drinking water,63 may also lead to human trophoblast cell dysfunctions, which suggests that these environmental carcinogens might also induce these trophoblast-related adverse pregnancy outcomes. It is a long stream line from the upstream environmental carcinogen exposure to the ultimate adverse pregnancy outcomes, providing multiple potential biotargets or biomarkers for diagnosis and treatment against these trophoblast-related adverse pregnancy outcomes.
The Limitation of This Work
In this work, we collected villous tissues (n=15) to verify the cellular results. More samples should be collected to exclude the possible demographic differences. Because lnc-HZ09 counterpart was not identified in mouse system, only the Pld1/Rac1/Cdc42 pathway was studied in the mouse model. However, the epigenetic regulation on this pathway in mouse models should be further explored. In these tissue samples, we evaluated the internal exposure level of B(a)P by detecting BPDE-DNA adducts in genomic DNA in RM and HC villous tissues, which might reflect the total external exposures of B(a)P. However, we did not consider a single external environmental exposure, such as smoking, in this work. Although the B(a)P level was detected to be higher in the RM group relative to the HC group, we cannot exclude the possibility that other carcinogens or genetic differences might also induce miscarriage. Other carcinogens that might induce miscarriage should be further explored.
Conclusion
RM is a global problem that produces severe challenges to socioeconomic development. Although many studies have indicated that various risk factors may induce miscarriage in the first trimester, about half of RMs have unknown clinical causes.66 In this work, we have identified a novel lnc-HZ09, which suppressed the migration and invasion of human trophoblast cells and affected the occurrence of miscarriage. In mechanism, this lnc-HZ09 suppressed SP1-mediated PLD1 mRNA transcription and reduced PLD1 mRNA stability. BPDE exposure promoted MSX1-mediated lnc-HZ09 transcription and also increased lnc-HZ09 stability by up-regulating its m6A methylation modification. Thus, BPDE exposure up-regulated the lnc-HZ09 level, suppressed the PLD1/RAC1/CDC42 pathway, inhibited migration and invasion, and might further induce miscarriage. In summary, this work provided novel insights in the roles of lnc-HZ09 in regulation of BPDE-induced dysfunctions of human trophoblast cells and the occurrence of unexplained miscarriage.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors acknowledge K. Inman, Science Editor of Environmental Health Perspectives, for careful and comprehensive revision of this manuscript. The authors acknowledge financial support from the Natural Science Foundation of China (NSFC No. 82073589), the Fundamental Research Funds for the Central Universities, Shenzhen Science and Technology Program (No. JCYJ20210324114814038) and Futian Health Care Research Project (No. FTWS2021009). The authors also acknowledge the facility supports by Central Laboratory of the Eighth Affiliated Hospital at Sun Yat-sen University.
M.D., W.H., and H.Z. designed this study. M.D. and W.H. performed most of the experiments. M.D., W.H., and H.Z. wrote the draft manuscript. X.H. and C.M. were responsible for animal experiments. R.W. and P.T. contributed to the tissue preparation. W.C., Y.Z., and C.M. conducted partial experiments.
==== Refs
References
1. Jurkovic D, Overton C, Bender-Atik R. 2013. Diagnosis and management of first trimester miscarriage. BMJ 346 :f3676, PMID: , 10.1136/bmj.f3676.23783355
2. Ammon Avalos L, Galindo C, Li DK. 2012. A systematic review to calculate background miscarriage rates using life table analysis. Birth Defects Res A Clin Mol Teratol 94 (6 ):417–423, PMID: , 10.1002/bdra.23014.22511535
3. Savitz DA, Hertz-Picciotto I, Poole C, Olshan AF. 2002. Epidemiologic measures of the course and outcome of pregnancy. Epidemiol Rev 24 (2 ):91–101, PMID: , 10.1093/epirev/mxf006.12762085
4. Pereza N, Ostojic S, Kapovic M, Peterlin B. 2017. Systematic review and meta-analysis of genetic association studies in idiopathic recurrent spontaneous abortion. Fertil Steril 107 (1 ):150–159.e2, PMID: , 10.1016/j.fertnstert.2016.10.007.27842992
5. Vaiman D. 2015. Genetic regulation of recurrent spontaneous abortion in humans. Biomed J 38 (1 ):11–24, PMID: , 10.4103/2319-4170.133777.25179715
6. Chan YY, Jayaprakasan K, Tan A, Thornton JG, Coomarasamy A, Raine-Fenning NJ. 2011. Reproductive outcomes in women with congenital uterine anomalies: a systematic review. Ultrasound Obstet Gynecol 38 (4 ):371–382, PMID: , 10.1002/uog.10056.21830244
7. Temple R, Aldridge V, Greenwood R, Heyburn P, Sampson M, Stanley K. 2002. Association between outcome of pregnancy and glycaemic control in early pregnancy in type 1 diabetes: population based study. BMJ 325 (7375 ):1275–1276, PMID: , 10.1136/bmj.325.7375.1275.12458245
8. ESHRE Capri Workshop Group. 2008. Genetic aspects of female reproduction. Hum Reprod Update 14 :293–307, PMID: , 10.1093/humupd/dmn009.18385259
9. Regan L, Rai R. 2000. Epidemiology and the medical causes of miscarriage. Baillieres Best Pract Res Clin Obstet Gynaecol 14 (5 ):839–854, PMID: , 10.1053/beog.2000.0123.11023804
10. Bostrom CE, Gerde P, Hanberg A, Jernström B, Johansson C, Kyrklund T, et al. 2002. Cancer risk assessment, indicators, and guidelines for polycyclic aromatic hydrocarbons in the ambient air. Environ Health Perspectives 110 (suppl 3 ):451–488, PMID: , 10.1289/ehp.110-1241197.12060843
11. Alegbeleye OO, Opeolu BO, Jackson VA. 2017. Polycyclic aromatic hydrocarbons: a critical review of environmental occurrence and bioremediation. Environ Manage 60 (4 ):758–783, PMID: , 10.1007/s00267-017-0896-2.28573478
12. Sun K, Song Y, He F, Jing M, Tang J, Liu R. 2021. A review of human and animals exposure to polycyclic aromatic hydrocarbons: health risk and adverse effects, photo-induced toxicity and regulating effect of microplastics. Sci Total Environ 773 :145403, PMID: , 10.1016/j.scitotenv.2021.145403.33582342
13. Neal MS, Zhu J, Holloway AC, Foster WG. 2007. Follicle growth is inhibited by benzo-[a]-pyrene, at concentrations representative of human exposure, in an isolated rat follicle culture assay. Hum Reprod 22 (4 ):961–967, PMID: , 10.1093/humrep/del487.17218370
14. Ptashekas J, Ciuniene E, Barkiene M, Zurlyte I, Jonauskas G, Sliachtic N, et al. 1996. Environmental and health monitoring in Lithuanian cities: exposure to heavy metals and benz(a)pyrene in Vilnius and Siauliai residents. J Environ Pathol Toxicol Oncol 15 (2–4 ):135–141, PMID: .9216796
15. Wu J, Hou H, Ritz B, Chen Y. 2010. Exposure to polycyclic aromatic hydrocarbons and missed abortion in early pregnancy in a Chinese population. Sci Total Environ 408 (11 ):2312–2318, PMID: , 10.1016/j.scitotenv.2010.02.028.20219237
16. Zhao Y, Chen X, Liu X, Ding Y, Gao R, Qiu Y, et al. 2014. Exposure of mice to benzo(a)pyrene impairs endometrial receptivity and reduces the number of implantation sites during early pregnancy. Food Chem Toxicol 69 :244–251, PMID: , 10.1016/j.fct.2014.04.021.24769007
17. Li X, Shen C, Liu X, He J, Ding Y, Gao R, et al. 2017. Exposure to benzo[a]pyrene impairs decidualization and decidual angiogenesis in mice during early pregnancy. Environ Pollut 222 :523–531, PMID: , 10.1016/j.envpol.2016.11.029.28043741
18. Ji L, Brkić J, Liu M, Fu G, Peng C, Wang Y-L. 2013. Placental trophoblast cell differentiation: physiological regulation and pathological relevance to preeclampsia. Mol Aspects Med 34 (5 ):981–1023, PMID: , 10.1016/j.mam.2012.12.008.23276825
19. Ball E, Bulmer JN, Ayis S, Lyall F, Robson SC. 2006. Late sporadic miscarriage is associated with abnormalities in spiral artery transformation and trophoblast invasion. J Pathol 208 (4 ):535–542, PMID: , 10.1002/path.1927.16402350
20. Zong S, Li C, Luo C, Zhao X, Liu C, Wang K, et al. 2016. Dysregulated expression of IDO may cause unexplained recurrent spontaneous abortion through suppression of trophoblast cell proliferation and migration. Sci Rep 6 :19916, PMID: , 10.1038/srep19916.26814137
21. Ye Y, Jiang S, Zhang C, Cheng Y, Zhong H, Du T, et al. 2020. Environmental pollutant benzo[a]pyrene induces recurrent pregnancy loss through promoting apoptosis and suppressing migration of extravillous trophoblast. BioMed Res Int 2020 :8983494, PMID: , 10.1155/2020/8983494.33123590
22. Jenkins GM, Frohman MA. 2005. Phospholipase D: a lipid centric review. Cell Mol Life Sci 62 (19–20 ):2305–2316, PMID: , 10.1007/s00018-005-5195-z.16143829
23. Burridge K, Wennerberg K. 2004. Rho and Rac take center stage. Cell 116 (2 ):167–179, PMID: , 10.1016/s0092-8674(04)00003-0.14744429
24. Powner DJ, Hodgkin MN, Wakelam MJ. 2002. Antigen-stimulated activation of phospholipase D1b by Rac1, ARF6, and PKCalpha in RBL-2H3 cells. Mol Biol Cell 13 (4 ):1252–1262, PMID: , 10.1091/mbc.01-05-0235.11950936
25. Costa FF. 2010. Non-coding RNAs: meet thy masters. Bioessays 32 (7 ):599–608, PMID: , 10.1002/bies.200900112.20544733
26. Zhu Y, Liu Q, Liao M, Diao L, Wu T, Liao W, et al. 2019. Overexpression of lncRNA EPB41L4A-AS1 induces metabolic reprogramming in trophoblast cells and placenta tissue of miscarriage. Mol Ther Nucleic Acids 18 :518–532, PMID: , 10.1016/j.omtn.2019.09.017.31671345
27. Huang Z, Du G, Huang X, Han L, Han X, Xu B, et al. 2018. The enhancer RNA lnc-SLC4A1-1 epigenetically regulates unexplained recurrent pregnancy loss (URPL) by activating CXCL8 and NF-kB pathway. EBioMedicine 38 :162–170, PMID: , 10.1016/j.ebiom.2018.11.015.30448228
28. Liang T, Xie J, Zhao J, Huang W, Xu Z, Tian P, et al. 2021. Novel lnc-HZ03 and miR-hz03 promote BPDE-induced human trophoblastic cell apoptosis and induce miscarriage by upregulating p53/SAT1 pathway. Cell Biol Toxicol 37 (6 ):951–970, PMID: , 10.1007/s10565-021-09583-3.33566220
29. Xu Z, Tian P, Guo J, Mi C, Liang T, Xie J, et al. 2021. Lnc-HZ01 with m6A RNA methylation inhibits human trophoblast cell proliferation and induces miscarriage by up-regulating BPDE-activated lnc-HZ01/MXD1 positive feedback loop. Sci Total Environ 776 :145950, PMID: , 10.1016/j.scitotenv.2021.145950.33647641
30. Huang W, Dai M, Qiu T, Liang T, Xie J, Mi C, et al. 2021. Novel lncRNA-HZ04 promotes BPDE-induced human trophoblast cell apoptosis and miscarriage by upregulating IP3 R1/CaMKII/SGCB pathway by competitively binding with miR-hz04. FASEB J 35 (9 ):e21789, PMID: , 10.1096/fj.202100376RR.34383983
31. Xie J, Liang T, Zhao J, Xu Z, Tian P, Wang R, et al. 2022. Lnc-HZ08 regulates BPDE-induced trophoblast cell dysfunctions by promoting PI3K ubiquitin degradation and is associated with miscarriage. Cell Biol Toxicol 38 (2 ):291–310, PMID: , 10.1007/s10565-021-09606-z.33864160
32. Straszewski-Chavez SL, Abrahams VM, Alvero AB, Aldo PB, Ma Y, Guller S, et al. 2009. The isolation and characterization of a novel telomerase immortalized first trimester trophoblast cell line, Swan 71. Placenta 30 (11 ):939–948, PMID: , 10.1016/j.placenta.2009.08.007.19766308
33. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. 2000. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25 (1 ):25–29, PMID: , 10.1038/75556.10802651
34. Gene Ontology Consortium. 2021. The Gene Ontology resource: enriching a GOld mine. Nucleic Acids Res 49 :D325–D334, PMID: , 10.1093/nar/gkaa1113.33290552
35. Tian F-J, He X-Y, Wang J, Li X, Ma X-L, Wu F, et al. 2018. Elevated tristetraprolin impairs trophoblast invasion in women with recurrent miscarriage by destabilization of HOTAIR. Mol Ther Nucleic Acids 12 :600–609, PMID: , 10.1016/j.omtn.2018.07.001.30195796
36. Wang L, Park HJ, Dasari S, Wang S, Kocher J-P, Li W. 2013. CPAT: Coding-Potential Assessment Tool using an alignment-free logistic regression model. Nucleic Acids Res 41 (6 ):e74, PMID: , 10.1093/nar/gkt006.23335781
37. Mistry J, Chuguransky S, Williams L, Qureshi M, Salazar GA, Sonnhammer ELL, et al. 2021. Pfam: the protein families database in 2021. Nucleic Acids Res 49 (D1 ):D412–D419, PMID: , 10.1093/nar/gkaa913.33125078
38. Farre D. 2003. Identification of patterns in biological sequences at the ALGGEN server: PROMO and MALGEN. Nucleic Acids Res 31 (13 ):3651–3653, PMID: , 10.1093/nar/gkg605.12824386
39. Messeguer X, Escudero R, Farré D, Núñez O, Martínez J, Albà MM. 2002. PROMO: detection of known transcription regulatory elements using species-tailored searches. Bioinformatics 18 (2 ):333–334, PMID: , 10.1093/bioinformatics/18.2.333.11847087
40. Szklarczyk D, Gable AL, Nastou KC, Lyon D, Kirsch R, Pyysalo S, et al. 2021. The STRING database in 2021: customizable protein-protein networks, and functional characterization of user-uploaded gene/measurement sets. Nucleic Acids Res 49 (D1 ):D605–D612, PMID: , 10.1093/nar/gkaa1074.33237311
41. Zhou Y, Zeng P, Li YH, Zhang Z, Cui Q. 2016. SRAMP: prediction of mammalian N6-methyladenosine (m6A) sites based on sequence-derived features. Nucleic Acids Res 44 (10 ):e91, PMID: , 10.1093/nar/gkw104.26896799
42. Finn RD, Coggill P, Eberhardt RY, Eddy SR, Mistry J, Mitchell AL, et al. 2016. The Pfam protein families database: towards a more sustainable future. Nucleic Acids Res 44 (D1 ):D279–D285, PMID: , 10.1093/nar/gkv1344.26673716
43. Gao S, Murakami M, Ito H, Furuhata A, Yoshida K, Tagawa Y, et al. 2009. Mutated ras induced PLD1 gene expression through increased Sp1 transcription factor. Nagoya J Med Sci 71 (3–4 ):127–136, PMID: .19994725
44. Guo J, Li R, Xu Z, Tian P, Wang R, Li Y, et al. 2021. Upregulated lnc-HZ02 and miR-hz02 inhibited migration and invasion by downregulating the FAK/SRC/PI3K/AKT pathway in BPDE-treated trophoblast cells. J Biochem Mol Toxicol 35 (6 ):1–13, 10.1002/jbt.22757.
45. Yang Y, Zhu X, Jia X, Hou W, Zhou G, Ma Z, et al. 2020. Phosphorylation of Msx1 promotes cell proliferation through the Fgf9/18-MAPK signaling pathway during embryonic limb development. Nucleic Acids Res 48 (20 ):11452–11467, PMID: , 10.1093/nar/gkaa905.33080014
46. Gu C, Wang Z, Zhou N, Li G, Kou Y, Luo Y, et al. 2019. Mettl14 inhibits bladder TIC self-renewal and bladder tumorigenesis through N6-methyladenosine of Notch1. Mol Cancer 18 (1 ):168, PMID: , 10.1186/s12943-019-1084-1.31760940
47. Patil DP, Chen C-K, Pickering BF, Chow A, Jackson C, Guttman M, et al. 2016. m(6)a RNA methylation promotes XIST-mediated transcriptional repression. Nature 537 (7620 ):369–373, PMID: , 10.1038/nature19342.27602518
48. Liu J, Yue Y, Han D, Wang X, Fu Y, Zhang L, et al. 2014. A METTL3-METTL14 complex mediates mammalian nuclear RNA N6-adenosine methylation. Nat Chem Biol 10 (2 ):93–95, PMID: , 10.1038/nchembio.1432.24316715
49. Fustin J-M, Doi M, Yamaguchi Y, Hida H, Nishimura S, Yoshida M, et al. 2013. RNA-methylation-dependent RNA processing controls the speed of the circadian clock. Cell 155 (4 ):793–806, PMID: , 10.1016/j.cell.2013.10.026.24209618
50. da Silva Moreira S, de Lima Inocêncio LC, Jorge BC, Reis ACC, Hisano H, Arena AC. 2021. Effects of benzo(a)pyrene at environmentally relevant doses on embryo-fetal development in rats. Environ Toxicol 36 (5 ):831–839, PMID: , 10.1002/tox.23085.33350577
51. Che D, Huang W, Fang Z, Li L, Wu H, Pi L, et al. 2019. The lncRNA CCAT2 rs6983267 G allele is associated with decreased susceptibility to recurrent miscarriage. J Cell Physiol 234 (11 ):20577–20583, PMID: , 10.1002/jcp.28661.30982978
52. Brown HA, Thomas PG, Lindsley CW. 2017. Targeting phospholipase D in cancer, infection and neurodegenerative disorders. Nat Rev Drug Discov 16 (5 ):351–367, PMID: , 10.1038/nrd.2016.252.28209987
53. Kang DW, Hwang WC, Noh YN, Kang Y, Jang Y, Kim J-A, et al. 2021. Phospholipase D1 is upregulated by vorinostat and confers resistance to vorinostat in glioblastoma. J Cell Physiol 236 (1 ):549–560, PMID: , 10.1002/jcp.29882.32869317
54. Murray MJ, Lessey BA. 1999. Embryo implantation and tumor metastasis: common pathways of invasion and angiogenesis. Semin Reprod Endocrinol 17 (3 ):275–290, PMID: , 10.1055/s-2007-1016235.10797946
55. Jiang X, Liu B, Nie Z, Duan L, Xiong Q, Jin Z, et al. 2021. The role of m6A modification in the biological functions and diseases. Signal Transduct Target Ther 6 (1 ):74, PMID: , 10.1038/s41392-020-00450-x.33611339
56. Yang D, Qiao J, Wang G, Lan Y, Li G, Guo X, et al. 2018. N6-Methyladenosine modification of lincRNA 1281 is critically required for mESC differentiation potential. Nucleic Acids Res 46 (8 ):3906–3920, PMID: , 10.1093/nar/gky130.29529255
57. Reizer E, Csizmadia IG, Palotas AB, Viskolcz B, Fiser B. 2019. Formation mechanism of benzo(a)pyrene: one of the most carcinogenic polycyclic aromatic hydrocarbons (PAH). Molecules 24 (6 ):1040, PMID: , 10.3390/molecules24061040.30884744
58. Shi Q, Godschalk RWL, van Schooten FJ. 2017. Inflammation and the chemical carcinogen benzo[a]pyrene: Partners in crime. Mutat Res Rev Mutat Res 774 :12–24, PMID: , 10.1016/j.mrrev.2017.08.003.29173495
59. Spagnoletti A, Paulesu L, Mannelli C, Ermini L, Romagnoli R, Cintorino M, et al. 2015. Low concentrations of bisphenol a and para-Nonylphenol affect extravillous pathway of human trophoblast cells. Mol Cell Endocrinol 412 :56–64, PMID: , 10.1016/j.mce.2015.05.023.26027920
60. Liao Y, Zheng H, Wu L, He L, Wang Y, Ou Y, et al. 2021. Cadmium cytotoxicity and possible mechanisms in human trophoblast HTR-8/SVneo cells. Environ Toxicol 36 (6 ):1111–1124, PMID: , 10.1002/tox.23110.33559965
61. Robert S, Gicquel T, Bodin A, Lagente V, Boichot E. 2016. Characterization of the MMP/TIMP imbalance and collagen production induced by IL-1β or TNF-α release from human hepatic stellate cells. PLoS One 11 (4 ):e0153118, PMID: , 10.1371/journal.pone.0153118.27046197
62. Wu X, Li M, Chen SQ, Li S, Guo F. 2018. Pin1 facilitates isoproterenol-induced cardiac fibrosis and collagen deposition by promoting oxidative stress and activating the MEK1/2ERK1/2 signal transduction pathway in rats. Int J Mol Med 41 (3 ):1573–1583, PMID: , 10.3892/ijmm.2017.3354.29286102
63. Manka SW, Bihan D, Farndale RW. 2019. Structural studies of the MMP-3 interaction with triple-helical collagen introduce new roles for the enzyme in tissue remodelling. Sci Rep 9 :18785, PMID: , 10.1038/s41598-019-55266-9.31827179
64. Gardner DK, Lane M. 2003. Blastocyst transfer. Clin Obstet Gynecol 46 (2 ):231–238, PMID: , 10.1097/00003081-200306000-00005.12808377
65. Zigrino P, Brinckmann J, Niehoff A, Lu Y, Giebeler N, Eckes B, et al. 2016. Fibroblast-derived MMP-14 regulates collagen homeostasis in adult skin. J Invest Dermatol 136 (8 ):1575–1583, PMID: , 10.1016/j.jid.2016.03.036.27066886
66. Muyayalo KP, Li ZH, Mor G, Liao AH. 2018. Modulatory effect of intravenous immunoglobulin on Th17/Treg cell balance in women with unexplained recurrent spontaneous abortion. Am J Reprod Immunol 80 (4 ):e13018, PMID: , 10.1111/aji.13018.29984444
|
PMC009xxxxxx/PMC9891133.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36723383
EHP10857
10.1289/EHP10857
Research
Perfluoroalkyl Acids in Follicular Fluid and Embryo Quality during IVF: A Prospective IVF Cohort in China
Zeng Xiao-Wen 1 2 *
Bloom Michael S. 3 *
Wei Fu 4 5 *
Liu Liling 5
Qin Jie 5
Xue Lintao 5
Wang Shikai 5
Huang Guolan 5
Teng Min 5
He Bing 5
Mao Xianbao 5
Chu Chu 1 2
Lin Shao 6 7
https://orcid.org/0000-0002-2578-3369
Dong Guang-Hui 1 2
Tan Weihong 5
1 Department of Occupational and Environmental Health, School of Public Health, Sun Yat-sen University, Guangzhou, China
2 Guangdong Provincial Engineering Technology Research Center of Environmental and Health Risk Assessment, Sun Yat-sen University, Guangzhou, China
3 Department of Global and Community Health, George Mason University, Fairfax, Virginia, USA
4 Department of Anatomy and Embryology. Leiden University Medical Center, Leiden, The Netherlands
5 Department of Reproductive Medicine and Genetics Center, The People’s Hospital of Guangxi Zhuang Autonomous Region, Nanning, Guangxi, China
6 Department of Environmental Health Sciences, University at Albany, State University of New York, Rensselaer, 12 Albany, NY, USA
7 Department of Epidemiology and Biostatistics, University at Albany, State University of New York, Rensselaer, 12 Albany, NY, USA
Address correspondence to Guang-Hui Dong, Guangzhou Key Laboratory of Environmental Pollution and Health Risk Assessment, Department of Occupational and Environmental Health, School of Public Health, Sun Yat-sen University, 74 Zhongshan 2nd Rd., Yuexiu District, Guangzhou 510080, China. Telephone: 86 20 87333409. Email: [email protected]. And, Weihong Tan, Department of Reproductive Medicine and Genetics Center. The People’s Hospital of Guangxi Zhuang Autonomous Region, Nanning, Guangxi, China. Email: [email protected]. And, Shao Lin, Departments of Environmental Health Sciences and Epidemiology and Biostatistics, University at Albany, State University of New York, One University Place, Rensselaer, Albany, NY 12144 USA. Email: [email protected]
1 2 2023
2 2023
131 2 02700223 12 2021
16 10 2022
23 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Perfluoroalkyl acids (PFAA) have been measured in ovarian follicular fluid from women using in vitro fertilization (IVF), although associations between follicular fluid PFAA and IVF outcomes have been inconsistent.
Objectives:
We investigated the association between follicular fluid PFAA and embryo quality in women undergoing IVF.
Methods:
We prospectively enrolled 729 women undergoing IVF treatment in Guangxi province, China, from July 2018 to December 2018. We measured 32 PFAA, including branched isomers, in follicular fluid using ultra-performance liquid chromatography coupled to tandem mass spectrometry. We applied restricted cubic splines, linear regression, and log-binominal regression models to investigate associations between follicular fluid PFAA and embryo quality, adjusting for confounding variables and investigated oocyte maturity as an intervening variable using causal mediation analysis. We further estimated the overall effect of the PFAA mixture on outcomes using Bayesian kernel machine regression (BKMR).
Results:
We detected 8 of 32 measured PFAA in >85% of follicular fluid samples. Higher PFAA concentrations were associated with fewer high-quality embryos from IVF. The high-quality embryo rates at the 50th percentile of linear perfluoro-1-octanesulfonate acid (n-PFOS), all branched PFOS isomers (Br-PFOS) and linear perfluoro-n-octanoic acid (n-PFOA) were −6.34% [95% confidence interval (CI): −9.45, −3.32%], −16.78% (95% CI: −21.98, −11.58%) and −8.66% (95% CI: −11.88, −5.43%) lower, respectively, than the high quality embryo rates at the reference 10th percentile of PFAA. Oocyte maturity mediated 11.76% (95% CI: 3.18, 31.80%) and 14.28% (95% CI: 2.95, 31.27%) of the n-PFOS and n-PFOA associations, respectively. The results of the BKMR models showed a negative association between the PFAA mixture and the probability of high-quality embryos, with branched PFOS isomers having posterior inclusion probabilities of 1 and accounting for the majority of the association.
Discussion:
Exposure to higher PFAA concentrations in follicular fluid was associated with poorer embryo quality during IVF. Branched PFOS isomers may have a stronger effect than linear PFOS isomers. More studies are needed to confirm these findings and to directly estimate the effects on pregnancy and live-birth outcomes. https://doi.org/10.1289/EHP10857
Supplemental Material is available online (https://doi.org/10.1289/EHP10857).
* These authors contributed equally to this work.
All authors declare no conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Infertility, defined as 12 months of unprotected heterosexual intercourse without a successful pregnancy according to the International Committee for Monitoring Assisted Reproductive Technology,1 affects approximately 48.5 million reproductive-age couples worldwide.2 More advanced reproductive-age,3 unhealthy lifestyle factors (e.g., irregular sleeping mode, etc.), and psychosocial stress4 are likely to contribute to lower fecundity. However, growing evidence also suggests that exposure to environmental endocrine-disrupting chemicals (EDCs) may affect human reproductive health by altering the hormonal milieu that plays an important role in the pathogenesis of infertility.5
Perfluoroalkyl acids (PFAA) are EDCs and are a large family of synthesized chemicals, which have been applied in myriad industries since the 1940s, such as in the manufacture of nonstick pans, textile coatings, aqueous film-forming firefighting foams, and food packaging.6 Dietary exposure, drinking water contamination, and inhalation are the dominant PFAA exposure routes in the general human population.7 Perfluorooctane sulfonate (PFOS) and perfluorooctanoic acid (PFOA), recognized as persistent organic pollutants and human health risks by the Stockholm Convention [United Nations Environment Program (UNEP)] (http://chm.pops.int/TheConvention/ThePOPs/TheNewPOPs), are the most widely studied PFAA to date, although PFOS and PFOA production has been restricted or banned in most Western nations, leading to declining levels in human blood.8 However, China continues to produce PFOS, PFOA, and other PFAA, so exposure to PFAA in Chinese populations is likely to grow.9
Experimental studies have demonstrated that PFAA have the potential to impair female fertility by altering hypothalamic–pituitary–ovarian regulation (as reviewed by Ding et al.10). However, the findings of observational studies of PFAA exposure and female fecundability, the probability of conceiving a pregnancy in any “at risk” menstrual cycle, have been inconsistent (as reviewed by Bach et al.11 and Negri et al.12). For example, in the Danish National Birth Cohort, Fei et al. found strong associations between greater maternal plasma PFOS and PFOA concentrations and longer waiting time to pregnancy, a measure of fecundability,13 although concerns were expressed about the impact of parity on the results.14,15 In contrast, Bach et al. later reported null associations in a smaller subpopulation from the same cohort.16
EDCs, including PFAA, have been measured in ovarian follicular fluid, which bathes the developing oocyte.17 Endocrine disruptor–associated changes in follicular hormone activities may lead to diminished oocyte quality, which may further influence embryo development and lead to poorer reproductive outcomes.18 However, few epidemiological studies have investigated the potential relationship between follicular fluid PFAA and oocyte maturation and embryo quality.5 Given its highly invasive nature, follicular fluid is generally obtained from women using in vitro fertilization (IVF) treatment for infertility, so previous studies have been limited by small numbers of participants.19–24
To further clarify the potential impact of follicular fluid PFAA on female fertility, we conducted a hypothesis-generating study of associations between follicular fluid PFAA and IVF outcomes in a large IVF cohort from China.
Methods
Study Participants and Data Collection
Participants were enrolled into the Guangxi In Vitro Fertilization and the Environment Study (GIVES), a prospective investigation of environmental factors and IVF outcomes among couples from an infertility center in Guangxi province, China. Guangxi province is the centralized region for the Zhuang ethnic minority located in south China. The Zhuang population accounts for approximately 36% of Guangxi province residents, with 6.5% from other ethnic minority groups, such as Yao, Miao, Dong, and other ethnicities, and 57.5% are of the Chinese ethnic Han majority.25 Heterosexual couples initiating a first IVF cycle at the Reproductive Medicine Center, the People’s Hospital of Guangxi Zhuang Autonomous Region, were enrolled into GIVES from July 2018 to December 2018 (n=736). Of the enrolled subjects, no exclusions were applied. Participants completed a comprehensive self-administered study questionnaire, and women agreed to provide follicular fluid samples for analysis.
All participants provided informed consent prior to enrollment. This study was approved by the Ethics Committee Board of the People’s Hospital of Guangxi Zhuang Autonomous Region (No. 2018-42).
Clinical Protocols and Outcomes
Initial infertility diagnoses were evaluated by a physician according to Society for Assisted Reproductive Technology (SART) definitions.26 Based on the infertility evaluation and other clinical history, including baseline sex-hormone measurements, women underwent one of three ovarian controlled hyperstimulation protocols: a) long gonadotrophin releasing hormone (GnRH) agonist protocol (start at luteal or follicular phase); b) GnRH-antagonist protocol; or c) other protocols (including short GnRH agonist protocol, mild stimulation with addition of clomiphene citrate (CC)/letrozole to gonadotropins, modified natural cycle, and progestin for luteinizing hormone (LH) peak suppression). Serum estradiol (E2) was assayed, and follicle size was monitored throughout ovarian stimulation using transvaginal ultrasound. When the two largest follicles reached 18mm diameter or ≥3 follicles reached 17mm, ovulation was triggered by the administration of 0.5mL 250μg OVIDREL (Recombinant Human Choriogonadotropin alfa Solution for Injection; Merck Serono). Oocytes were retrieved 36 h later by transvaginal needle aspiration under transvaginal ultrasound guidance.
Retrieved oocytes in metaphase-II (MII) arrest (i.e., “mature” oocytes) were fertilized by incubation with fresh sperm using conventional IVF or by intracytoplasmic sperm injection (ICSI) for couples with severe male factor infertility. Rescue ICSI was performed if fertilization failure occurred at early cumulus cell removal after insemination of 4–6 h. Fertilization was confirmed 16–18 h after insemination by the appearance of an oocyte with two pronuclei (i.e., “2PN”). Embryo quality was evaluated by an embryologist according to morphology and the number of blastomeres on the second and third days post fertilization based on the SART guidelines.26 A high-quality embryo, suitable for transfer, was defined as an embryo with: a) no multinucleated blastomeres, b) 7–9 blastomeres on day 3, and c) <20% anucleated fragments (Alpha Scientists in Reproductive Medicine ESHRE Special Interest Group of Embryology, 2011).27 All other embryos were classified as low quality. Oocyte maturity (MII rate) was calculated as the number of MII oocytes divided by the total number of retrieved oocytes. High-quality embryo rate was calculated as the number of high-quality embryos divided by the number of 2PN zygotes per woman. The presence of high-quality embryos was defined as >1 high-quality embryo available for transfer per oocyte retrieval cycle. Each woman contributed one IVF cycle (all women initiating their first IVF cycle) to the study.
Follicular Fluid Sample Collection and PFAA Measurement
Pooled follicular fluid (approximately 5mL in total) was aspirated from two to four follicles from an individual woman during oocyte retrieval and collected in a 15mL Falcon tube. Samples were examined for visual evidence of blood contamination, and only clear fluid was analyzed.28 The sample was centrifuged at 1,500×g for 20 min and the supernatant was stored in polypropylene tubes at −80°C until analysis. Follicular fluid specimens were collected from 729 women (follicular fluid was not collected from 7 women).
We analyzed 32 PFAA concentrations in follicular fluid using ultra-performance liquid chromatography coupled with triple quadrupole tandem mass spectrometry (UPLC-MS/MS) with electrospray ionization in a negative mode (Agilent 6495B, Agilent Technologies). All PFAA standards were purchased from Wellington Laboratories (Guelph, Ontario, Canada). The abbreviations for the measured PFAA are listed in Table S1. The method was based on minor modifications to our previously described method.29 In brief, 0.2mL follicular fluid was mixed with 2mL of 0.1M formic acid, followed by spiking with 0.5 ng of mass labeled PFAA internal standards and extracted using a solid phase extraction cartridge (200mg/6 cc; Oasis-HLB). The extracts were centrifuged at 12,000×g for 10 min at 4°C and analyzed using UPLC-MS/MS. Procedural blanks (saline) were prepared and included in each interval of 20 samples to monitor for method contamination. Solvent blanks containing methanol and Milli-Q water (3:7 v/v) were prepared and included after every 12 samples to monitor for background contamination. Duplicate injections and calibration check standards were run after every 20 samples to ensure the precision and accuracy of each run. We also checked for PFAA contamination of the sterile needles used for oocyte retrieval by measuring the concentration of PFAA in saline run through the needles. The method detection limit (MDL) was defined as the mean concentrations of PFAA plus three times the standard deviation (SD) of procedural blanks.30 Values below the MDL were imputed as MDL/2 prior to analysis.31 The MDL of each PFAA is listed in Table S2.
In this study, we focused on follicular fluid PFAA with >85% detection rate to minimize bias associated with imputing values below the detection limits, according to the standardized data quality assessment guidelines outlined by the U.S. Environmental Protection Agency.32 Considering the potential for PFAA isomer–specific associations with health outcomes and having insufficient data available to assess the risks, we also incorporated all the branched PFOS isomers into the analysis. In total, we included nine PFAA in the analysis (Table S2), including linear sodium perfluoro-1-octanesulfonate (n-PFOS), all branched PFOS isomers (Br-PFOS), linear perfluoro-n-octanoic acid (n-PFOA), perfluoro-n-decanoic acid (PFDA), potassium perfluoro-1-hexanesulfonate (PFHxS), Sodium perfluro-1-heptanesulfonate (PFHpS), perfluoro-n-nonanoic acid (PFNA), perfluoro-n-undecanoic acid (PFUnDA), and perfluoro-n-tetradecanoic acid (PFTrDA). The concentration of Br-PFOS was comprised of the sum of all branched PFOS isomer concentrations.
Potential Confounding Variables
We identified potential confounding variables in this study a priori based on a direct acyclic graph (DAG; Figure S1), with reference to the previous literature.11,19 The minimally sufficient adjustment set of variables for estimating unbiased associations of follicular fluid PFAA with embryo quality included: women’s ages, socioeconomic status, prepregnancy body mass index (BMI), ovarian stimulation protocol, parity, infertility diagnosis, and seafood consumption. Information on age (years), family income in Chinese Yuan (CNY) (<5,000 CNY/month; 5,000–10,000 CNY/month; >10,000 CNY/month), prepregnancy BMI (kilograms per square meter), ovarian stimulation protocol (GnRH agonist or GnRH antagonist), and female infertility factors (tubal, ovulation disorder, poor ovarian response, advanced reproductive age, or “other”) were collected from the medical record and the self-administered study questionnaire. Regular seafood consumption was defined as a positive answer to the question “Eat seafood at least one time per week.” Approximately 14.4% participants did not respond to the seafood consumption question (n=105) and were assigned as a negative answer (no regular seafood consumption) for this question. There were no other missing data. In this study, we categorized the ethnicity into Han majority group and ethnic minority groups; the latter included Zhuang, Yao, Miao, Dong, and other minority ethnicities. This information was collected from participant’s medical record.
Statistical Analyses
Descriptive analyses.
Continuous data were presented as the mean±SD for normally distributed data or the median [quartile 1 (Q1), quartile 3, (Q3)]. Continuous PFAA concentrations were log10 transformed to reduce the influence of outliers and to normalize the skewed distributions. We also estimated pairwise Spearman correlations between PFAA concentrations in follicular fluid (Table S3).
Single PFAA predictor models.
We first estimated associations between individual follicular fluid PFAA as predictors of embryo quality with and without adjustment for confounders. Restricted cubic splines (RCS) were applied to check the linearity of associations. We chose models with a 3-knot RCS function, because the Akaike Information Criterion (AIC) value was lowest in comparison with models with 4-knot or 5-knot RCS functions for most PFAA (Table S4).33 We constructed multivariable linear regression models to estimate the difference (β) and 95% confidence interval (95% CI) for the difference in the high-quality embryo rate associated with a one log-unit increase in follicular fluid PFAA concentrations in crude models without confounder adjustment and in main models adjusted for the aforementioned confounding variables. We also calculated the estimated differences (95% CI) in high-quality embryo rate at the 25th, 50th, and 75th percentile of PFAA concentration in follicular fluid against the reference values (the 10th percentile of PFAA concentration) using RCS model.
We next used log-binominal regression models to estimate associations between the presence of ≥1 high-quality embryo (yes/no) and the concentrations of PFAA in follicular fluid by the relative risk (RR) and 95% CI. We applied the COPY method to avoid convergence problems in the log-binomial regression models.34
We additionally stratified the main analyses by parity (nulliparous women vs. parous women), age (<35 y old vs. >35 y old), and infertility diagnosis (only male factor, only female factor, and both female and male factor) because previous studies suggested that these factors may modify the associations.5,35 We further stratified the association by ethnicity (Han population vs. minority population) to explore potential disparities for the large Zhuang minority population residing in Guangxi province.
We identified oocyte maturity (MII rate) as a potential mediator using the DAG (Figure S1) and subsequently found an association between follicular fluid PFAA and MII rate. Thus, we used a counterfactual-based causal mediation analysis,36 employing the CAUSALMED procedure in SAS (version 9.4; SAS Institute Inc.), to investigate oocyte maturity (MII rate) as an intervening variable to explain the association between PFAA exposure and embryo quality. We employed 1,000 bootstrap samples to calculate the 95% CIs for the estimates.
Multiple PFAA predictors models.
Considering the interrelationships among follicular fluid PFAA concentrations, we next estimated the potential joint effect estimates of a PFAA mixture on embryo quality using Bayesian kernel machine regression (BKMR).37 The BKMR model allows for assessment of the independent associations of individual PFAA mixture components with an outcome, in addition to the overall combined mixed PFAA exposure.38,39 In the current study, we entered only follicular fluid PFAA with detection frequencies >85%, as well as branched PFOS isomers, into BKMR models. Briefly, we applied two BKMR models to investigate the exposure–response relationship between follicular PFAA exposure (log transformed) and the high-quality embryo rate: a) estimating the overall effect estimates of an increase in a PFAA mixture on the high-quality embryo rate when fixing individual PFAA at their median levels and b) estimating the dose–response relationship between each individual PFAA and the high-quality embryo rate when fixing the other PFAA at their 25th, 50th, and 75th percentiles. We examined the overall effect of PFAA mixtures on high-quality embryo rate via fitting the BKMR models, which fixed all PFAA mixtures at the same percentile concentration (in the range of 25th to 75th percentile, with fifth percentile increments) and compared with all PFAA when they were fixed at their 50th percentile. To implement a hierarchical variable selection method by Markov chain Monte Carlo algorithm, we categorized PFAA groups by perfluorocarboxylic acid and perfluoroalkyl sulfonates, respectively. We then calculated the group-posterior inclusion probabilities (PIPs) and conditional-PIPs, a measure of variable importance ranging from 0 to 1, to determine the major PFAA contributing to the main effect using 0.5 as a threshold. All models were run at 10,000 iterations using the Markov chain Monte Carlo algorithm and adjusted for the aforementioned confounding variables.38
All data analyses were conducted using SAS (version 9.4), with the exception of the BKMR model, which was conducted using R software (version 4.0.2; R Development Core Team) with the “bkmr” package. All hypothesis testing was 2-sided. Statistical significance was set at p<0.05 following correction for multiple hypothesis testing using the false discovery rate (FDR).40
Results
Table 1 shows the characteristics of 729 study participants. Women’s mean age was 32.7±4.5 y. Minority ethnicities accounted for 41.8% of participants; 35.8% were Zhuang ethnicity. Only seven women currently smoked tobacco. Infertility was attributed to female factor for most couples (78.1%) and 13.0% to male factor infertility; two women had unexplained infertility. Almost three-quarters of the women were nulliparous (72.2%). Seventy-nine percent of couples used conventional IVF, with 16.5% ICSI only. The long GnRH-agonist ovarian stimulation protocol (75.2%) was used most frequently. A mean of 11.4 oocytes were retrieved from women. The MII oocyte rate and high-quality embryos rate were 81.9% and 43.8%, respectively.
Table 1 Sociodemographic and clinical factors among women undergoing in vitro fertilization, GIVES (n=729).
Variables Values
Sociodemographic factors
Age (y) 32.7±4.5
Prepregnancy BMI (kg/m2) 22.0±3.3
Ethnicity —
Han 424 (58.2)
Minority 305 (41.8)
Zhuang minority 261 (35.8)
Other minority 44 (6.0)
Family income (CNY/month) —
<5,000 CNY 192 (26.3)
5,000−10,000 CNY 250 (34.3)
>10,000 CNY 125 (17.2)
Unknown 162 (22.2)
Cigarette smoker 7 (0.01)
Regular seafood consumption 159 (21.8)
Parity —
Nulliparous 526 (72.2)
Parous 203 (27.9)
Clinical factors
Initial infertility diagnosis —
Only male infertility factor 95 (13.0)
Only female infertility factor 569 (78.1)
Both male and female infertility factor 63 (8.6)
Unexplained 2 (0.3)
Ovarian stimulation protocol —
Long GnRH agonist 548 (75.2)
GnRH antagonist protocol 163 (22.4)
Othersa 18 (2.5)
Fertilization procedure —
Conventional IVF 579 (79.4)
ICSI 120 (16.5)
Rescued ICSIb 30 (4.1)
Peak estradiol level [E2 (pg/mL)] 2,747±1,393
Total oocytes retrieved 11.4±6.1
MII oocytes retrieved 9.3±5.2
MII oocyte rate (%)c 81.9±16.8
Fertilization outcome
Oocyte fertilization rate (%)d 71.7±22.2
High quality embryos —
Conventional IVF 3.0±2.6
ICSI 2.8±2.9
Rescued ICSI 2.7±2.3
High-quality embryos rate (%)e 43.8±28.3
Presence of high-quality embryo (>1 high-quality embryo) 623 (85.5)
Note: Values are represented as mean + SD, or number (%). Percentages may not sum to 100% due to number rounding. Approximately 14.4% participants (n=105) did not response to the seafood consumption question and were assigned as no regular seafood consumption participants. —, no data; 2PN, two pronuclei; BMI, body mass index; CC, clomiphene citrate; CNY, Chinese yuan; GIVES, Guangxi In Vitro Fertilization and the Environment Study; GnRH, gonadotrophin releasing hormone; ICSI, intracytoplasmic sperm injection; IVF, in vitro fertilization; LH, luteinizing hormone; MII, metaphase II; SD, standard deviation.
a Included short GnRH agonist protocol, mild stimulation with addition of CC/letrozole to gonadotropins, modified natural cycle, and progestin for LH peak suppression.
b If fertilization failure was determined to have occurred at early cumulus cell removal after insemination of 4–6 h, rescue ICSI was performed.
c Value was calculated as the number of MII oocytes divided by the total number of retrieved oocytes.
d Value was calculated as the number of 2PN zygotes divided by the total number of oocytes retrieved for conventional IVF, or as the number of 2PN zygotes divided by the number of MII oocytes number for ICSI.
e Value was calculated as the number of high-quality embryos divided by the number of 2PN zygotes.
There were eight PFAA (n-PFOS, n-PFOA, PFDA, PFHxS, PFHpS, PFNA, PFUnDa, and PFTrDA) with detection frequencies >85% in follicular fluid samples (Table 2; Table S2). Table 2 presents the concentration of PFAA in follicular fluid samples, with the highest median concentration measured for n-PFOS (1.70 ng/mL), followed by n-PFOA (1.09 ng/mL), total branched PFOS (Br-PFOS, 0.47 ng/mL), PFNA (0.39 ng/mL), and PFUnDa (0.30 ng/mL). In general, each one of PFAA was highly and positively correlated with the others (r=0.141–0.920, p<0.001; Table S3).
Table 2 Concentrations of PFAA (nanograms per milliliter) in follicular fluid (n=729).
PFAA Detection frequency (%) Mean±SD Median (Q1, Q3)
n-PFOS 100.0 2.08±1.71 1.70 (1.04, 2.61)
Br-PFOS — 0.60±0.46 0.47 (0.21, 0.84)
1m-PFOS 75.6 0.06±0.05 0.04 (0.02, 0.07)
3+4+5m-PFOS 72.4 0.43±0.40 0.32 (0.06, 0.64)
iso-PFOS 29.6 0.11±0.09 0.06 (0.06, 0.14)
Σ2m-PFOS 16.3 0.007±0.019 0.004 (0.004, 0.004)
n-PFOA 98.9 1.31±1.07 1.09 (0.59, 1.71)
PFBA 42.5 0.73±0.97 0.15 (0.15, 1.05)
PFBS 4.7 0.01±0.05 0.01 (0.01, 0.01)
PFDA 100.0 0.33±0.17 0.29 (0.22, 0.38)
PFDS 0.0 0.01±0.00 0.01 (0.01, 0.01)
PFHpA 0.7 0.03±0.00 0.03 (0.03, 0.03)
PFHpS 91.6 0.05±0.05 0.04 (0.03, 0.07)
PFHxA 1.5 0.03±0.01 0.03 (0.03, 0.03)
PFHxS 100.0 0.17±0.16 0.13 (0.09, 0.19)
PFNA 100.0 0.44±0.27 0.39 (0.29, 0.51)
PFNS 0.1 0.01±0.00 0.01 (0.01, 0.01)
PFPeA 0.1 0.11±0.01 0.11 (0.11, 0.11)
PFPeS 1.9 0.01±0.01 0.01 (0.01, 0.01)
PFUnDA 100.0 0.35±0.19 0.30 (0.23, 0.42)
PFDoDA 75.5 0.02±0.02 0.02 (0.01, 0.03)
PFTrDA 93.7 0.08±0.07 0.07 (0.04, 0.11)
PFTeDA 11.1 0.006±0.003 0.005 (0.005, 0.005)
Note: —, no data; 1m-PFOS, potassium perfluoro-1-methylheptanesulfonate; 3m-PFOS, potassium perfluoro-3-methylheptanesulfonate; 3+4+5m-PFOS, sum of 3m, 4m, and 5m-PFOS; Σm2-PFOS, sum of all dimethyl PFOS isomers; 4m-PFOS, potassium perfluoro-4-methylheptanesulfonate; 5m-PFOS, potassium perfluoro-5-methylheptanesulfonate; BMI, body mass index; Br-PFOS, sum of all branched PFOS isomers; CI, confidence interval; iso-PFOS, potassium perfluoro-6-methylheptanesulfonate; n-PFOA, linear perfluoro-n-octanoic acid; n-PFOS, linear sodium perfluoro-1-octanesulfonate; PFAA, perfluoroalkyl acid; PFBA, perfluoro-n-butanoic acid; PFBS, potassium perfluoro-1-butanesulfonate; PFDA, perfluoro-n-decanoic acid; PFDoDA, perfluoro-n-dodecanoic acid; PFDS, sodium perfluoro-1-decanesulfonate; PFHpA, perfluoro-n-heptanoic acid; PFHpS, sodium perfluoro-1-heptanesulfonate; PFHxA, perfluoro-n-hexanoic acid; PFHxS, potassium perfluoro-1-hexanesulfonate; PFNA, perfluoro-n-nonanoic acid; PFNS, sodium perfluoro-1-nonanesulfonate; PFPeA, perfluoro-n-pentanoic acid; PFPeS, sodium perfluoro-1-pentanesulfonate; PFTeDA, perfluoro-n-tetradecanoic acid; PFTrDA, perfluoro-n-tridecanoic acid; PFUnDA, perfluoro-n-undecanoic acid.
We found a lower high-quality embryo rate in association with greater follicular fluid PFAA exposure. For example, the high-quality embryo rates at the 50th percentile of n-PFOS, Br-PFOS, n-PFOA, and PFHxS were −6.34% (95% CI: −9.45, −3.32%), −16.78% (95% CI: −21.98, −11.58%), −8.66% (95% CI: −11.88, −5.43%), and −10.12% (95% CI: −14.52, −5.72%) lower, respectively, than the high-quality embryo rates at the reference 10th percentile of PFAA (Table 3; Figure S2). The negative associations were consistent in both crude and confounder-adjusted regression models (Table S5). When we further dichotomized the high embryo quality rate by the presence of ≥1 high-quality embryo as the outcome, we detected associations between higher concentrations of PFAA in follicular fluid and lower likelihoods for the presence of a high-quality embryo using log-binominal regression models adjusted for confounders, particularly for n-PFOS (RR=0.97; 95% CI: 0.96, 0.98), Br-PFOS (RR=0.88; 95% CI: 0.86, 0.91), PFOA (RR=0.96; 95% CI: 0.95, 0.98), PFHxS (RR=0.91; 95% CI: 0.88, 0.94), PFDA (RR=0.91; 95% CI: 0.86, 0.98), and PFUnDA (RR=0.93; 95% CI: 0.88, 1.00) (Table 4).
Table 3 Estimated differences in high-quality embryo rate (percentage) at the 25th, 50th, and 75th percentiles of log-PFAA concentrations (nanograms per milliliter) measured in follicular fluid against the 10th percentile (reference concentration).a
PFAA β (95% CI)
25th vs. 10th 50th vs. 10th 75th vs. 10th
n-PFOS −3.68 (−5.90, −1.45) −6.34 (−9.45, −3.32) −8.97 (−12.80, −5.13)
Br-PFOS −7.60 (−10.60, −4.61) −16.78 (−21.98, −11.58) −23.36 (−28.53, −18.18)
n-PFOA −4.89 (−7.17, −2.60) −8.66 (−11.88, −5.43) −11.39 (−15.40, −7.37)
PFHxS −5.47 (−8.10, −2.84) −10.12 (−14.52, −5.72) −12.94 (−17.89, −7.98)
PFDA −0.57 (−3.37, 2.21) −3.63 (−9.79, 2.52) −6.12 (−11.78, −0.46)
PFHpS −1.20 (−2.96, 0.55) −2.21 (−5.38, 0.97) −2.38 (−6.34, 1.57)
PFNA −1.25 (−3.76, 1.25) −2.71 (−7.06, 1.63) −4.22 (−9.16, 0.72)
PFUnDA −0.99 (−3.18, 1.19) −2.56 (−6.29, 1.18) −5.43 (−9.85, −1.01)
PFTrDA 1.01 (−1.13, 3.15) 0.72 (−2.47, 3.92) −1.28 (−5.44, 2.88)
Note: β, regression coefficient; BMI, body mass index; Br-PFOS, sum of all branched PFOS isomers; CI, confidence interval; n-PFOA, linear perfluoro-n-octanoic acid; n-PFOS, linear sodium perfluoro-1-octanesulfonate; PFAA, perfluoroalkyl acid; PFDA, perfluoro-n-decanoic acid; PFHpS, sodium perfluoro-1-heptanesulfonate; PFHxS, sodium perfluoro-1-heptanesulfonate; PFNA, perfluoro-n-nonanoic acid; PFTrDA, perfluoro-n-tridecanoic acid; PFUnDA, perfluoro-n-undecanoic acid.
a Models adjusted for women’s age, family income, prepregnancy BMI, infertility diagnosis, parity, stimulation protocol, and regular seafood consumption.
Table 4 Probabilities (95% CIs) of a high-quality embryo per log-unit increase in follicular fluid PFAA concentrations (nanograms per milliliter) using log-binomial regression models (n=729).
PFAA RRs (95% CIs)a p-Value
n-PFOS 0.97 (0.96, 0.98) <0.001
Br-PFOS 0.88 (0.86, 0.91) <0.001
n-PFOA 0.96 (0.95, 0.98) <0.001
PFHxS 0.91 (0.88, 0.94) <0.001
PFDA 0.91 (0.86, 0.98) 0.009
PFHpS 1.00 (0.95, 1.03) 0.819
PFNA 0.96 (0.90, 1.03) 0.241
PFUnDA 0.93 (0.88, 1.00) 0.026
PFTrDA 1.01 (0.97, 1.05) 0.535
Note: BMI, body mass index; Br-PFOS, sum of all branched PFOS isomers; CI, confidence interval; n-PFOA, linear perfluoro-n-octanoic acid; n-PFOS, linear sodium perfluoro-1-octanesulfonate; PFAA, perfluoroalkyl acid; PFDA, perfluoro-n-decanoic acid; PFHpS, sodium perfluoro-1-heptanesulfonate; PFHxS, potassium perfluoro-1-hexanesulfonate; PFNA, perfluoro-n-nonanoic acid; PFTrDA, perfluoro-n-tridecanoic acid; PFUnDA, perfluoro-n-undecanoic acid; RR, relative risk.
a Models adjusted for women’s age, family income, prepregnancy BMI, infertility diagnosis, parity, stimulation protocol, and regular seafood consumption.
We found associations for n-PFOS, PFDA, PFHpS, PFNA, and PFUnDA among nulliparous (Table S6). However, we found no evidence of modification when we stratified the associations by maternal age (Table S7) and ethnicity (Table S8). When we restricted our analyses in participants with only female factor of infertility diagnosis (n=569), we found results consistent with our main findings of associations between higher follicular fluid PFAA and lower odds of a high-quality embryo (Table S9). We also found that higher concentrations of PFAA, particularly n-PFOS, Br-PFOS, n-PFOA, and PFHxS, were associated with lower MII rate, adjusted for confounding variables (Table S10). The causal mediation analysis model suggested that the MII rate may partially mediate the relationship between the presence of a high-quality embryo and follicular fluid n-PFOS (11.76%; 95% CI: 3.18, 31.80%), n-PFOA (14.28%; 95% CI: 2.95, 31.27%), and PFHxS (8.13%; 95% CI: 1.53, 20.24%) (Table 5).
Table 5 Causal mediation analysis of oocyte maturation in the association between PFAA and the presence of high-quality embryos.a
PFAA Total effect (95% CI)b Direct effect (95% CI)b Indirect effect (95% CI)b Proportion mediated (95% CI)b
n-PFOS −0.054 (−0.077, −0.031)c −0.047 (−0.070, −0.022)c −0.007 (−0.014, −0.002)c 11.76 (3.18, 31.80)c
Br-PFOS −0.072 (−0.092, −0.050)c −0.066 (−0.087, −0.043)c −0.006 (−0.014, 0.000) 8.38 (−0.75, 20.17)
n-PFOA −0.049 (−0.071, −0.028)c −0.042 (−0.064, −0.022)c −0.007 (−0.014, −0.002)c 14.28 (2.95, 31.27)c
PFHxS −0.083 (−0.127, −0.046)c −0.077 (−0.125, −0.041)c −0.007 (−0.016, −0.001)c 8.13 (1.53, 20.24)c
PFDA −0.089 (−0.147, −0.024)c −0.082 (−0.142, −0.017)c −0.007 (−0.018, −0.0001) 7.40 (−0.028, 27.57)
PFHpS −0.004 (−0.043, 0.034) −0.007 (−0.044, 0.033) 0.002 (−0.003, 0.010) −53.61 (−3,614, −19.25)
PFNA −0.043 (−0.105, 0.011) −0.038 (−0.098, 0.017) −0.005 (−0.015, 0.001) 10.76 (−22.68, 186.4)
PFUnDA −0.068 (−0.130, −0.010)c −0.062 (−0.121, −0.003)c −0.006 (−0.018, −0.0001) 9.21 (−1.53, 44.67)
PFTrDA 0.013 (−0.024, 0.052) 0.010 (−0.025, 0.050) 0.002 (−0.002, 0.010) 17.26 (−7.26, 2,026)
Note: BMI, body mass index; Br-PFOS, sum of all branched PFOS isomers; CI, confidence interval; n-PFOA, linear perfluoro-n-octanoic acid; n-PFOS, linear sodium perfluoro-1-octanesulfonate; PFAA, perfluoroalkyl acid; PFDA, perfluoro-n-decanoic acid; PFHpS, sodium perfluoro-1-heptanesulfonate; PFHxS, sodium perfluoro-1-heptanesulfonate; PFNA, perfluoro-n-nonanoic acid; PFTrDA, perfluoro-n-tridecanoic acid; PFUnDA, perfluoro-n-undecanoic acid.
a Models adjusted for women’s age, family income, prepregnancy BMI, infertility diagnosis, parity, stimulation protocol, and regular seafood consumption.
b Bootstrap bias-corrected 95% CI at 1,000 bootstrap replicates.
c p<0.05.
Figure 1 shows the overall association of the PFAA mixture with the difference in the high-quality embryo rate using confounder-adjusted BKMR models. Consistent with our findings from the single-PFAA predictor models, the overall effect estimates of the PFAA mixture suggested that greater concentrations of follicular fluid PFAA were monotonically associated with a lower high-quality embryo rate (Figure 1A; Table S11). For example, we found a lower posterior mean estimate of a high-quality embryo rate (−7.25%; 95% CI: −9.56, −4.94%) when the PFAA mixture concentration was fixed at the 75th percentile of the exposure distribution in comparison with being fixed at the 50th percentile (Table S11). We further characterized the contribution of individual follicular fluid PFAA exposure components to the overall PFAA mixture association and found that branched PFOS isomers were significantly associated with a lower high-quality embryo rate while holding all other PFAA at the 25th, 50th, and 75th percentiles (Figure 1B and Table S12). Specifically, the PIPs ranging from 0 to 1, indicating the importance of each individual PFAA (from unimportant to most important), showed that Br-PFOS was the dominant contributor (cond-PIP=1) (Table S13).
Figure 1. The relationship of exposure to a PFAA mixture (A) or a single PFAA (B) and the estimated difference in high-quality embryo rate (See Table S11 and Table S12 for numeric data.). Note: (A) Overall effect of the PFAA mixture on the estimated difference (95% CI) in high-quality embryo rate when all component PFAA were fixed at a specific quantile (ranging from 25th to 75th percentiles) compared to all component PFAA fixed at their median values. (B) The effect of a single PFAA (95% CI) on the estimated difference (95% CI) in high-quality embryo rate, with all other component PFAA fixed at a specific quantile (25th, 50th, and 75th percentiles). CI, confidence interval; PFAA, perfluoroalkyl acid.
Figure 1A is an error bar graph, plotting Overall effect on the high-quality embryo rate, ranging from negative 10 to 10 in increments of 5 (y-axis) across Perfluoroalkyl acids mixture quantile, ranging from thirtieth to seventieth in increments of 10 (x-axis). Figure 1B is an error bar graph, plotting lowercase n-perfluorooctanesulfonic acid; branched isomers of perfluorooctanesulfonic acid; perfluorooctanoic acid, lowercase n-perfluorohexanesulfonic acid; Perfluoroheptanesulfonic acid; Nonadecafluorodecanoic acid; Perfluorononanoic acid; Perfluoroundecanoic acid; and Perfluorotridecanoic acid (y-axis) across Estimated difference in the high-quality embryo rate, ranging from negative 20 to 10 in increments of 10 (x-axis) for fixed quantile, including twenty-fifth, fiftieth, and seventy-fifth.
Discussion
In this large prospective cohort study of infertile couples undergoing IVF treatment, exposure to individual follicular fluid PFAA (n-PFOS, Br-PFOS, n-PFOA, PFHxS, PFDA and PFUnDA) was inversely associated with embryo quality during IVF. Furthermore, associations of n-PFOS, n-PFOA, and PFHxS were partially mediated by the MII-oocyte rate. A mixture of nine follicular fluid PFAA was associated with a lower high-quality embryo rate, and branched PFOS isomers appeared to dominate the association. Thus, exposure to PFAA may further lower IVF success rates.
PFAA Levels in Follicular Fluid
A growing body of literature describes the worldwide distribution of PFAA in human blood and other biospecimens, including blood, urine, hair, nails, and semen.10,11 The few studies that have measured ovarian follicular fluid showed that PFOS and PFOA were the most prevalent PFAA (Table S14). The range of follicular fluid PFAA concentrations in our study was comparable to other studies. For example, the median follicular fluid linear PFOS (1.70 ng/mL) and PFOA (1.09 ng/mL) concentrations in this study were similar to the geometric mean follicular fluid PFOS (1.8 ng/mL) and PFOA (1.9 ng/mL) concentrations in women without polycystic ovary syndrome (PCOS) (n=59) from a United Kingdom fertility clinic,21 and the mean PFOS (4.80 ng/mL) and PFOA (2.40 ng/mL) concentrations in women (n=97) from an Australian IVF clinic.20 However, the median follicular fluid PFAA concentrations in our study were lower than those reported in women seeking IVF treatment (4.54 ng/mL for PFOS and 3.38 ng/mL for PFOA, n=28) at the Peking University People’s Hospital in China41 and were also lower than those in women who underwent the IVF treatment in Yantai, China (4.77 ng/mL for n-PFOS and 6.37 ng/mL for n-PFOA, n=124).24 The discrepancies in follicular fluid PFAA concentrations may be related to different sources of environmental PFAA exposure (e.g., PFAA-contaminated drinking water exposure vs. exposure via dust inhalation), different dietary sources of PFAA exposure (e.g., seafood consumption),42 as well as different sociodemographic factors across different IVF patient populations.
PFAAs and Assisted Reproduction
Among studies evaluating the association between PFAAs and female infertility, only five have examined PFAA concentrations in follicular fluid, a direct indicator of exposure in the oocyte’s microenvironment.20–24 Other studies used serum or plasma PFAA concentrations as an exposure indicator. Governini et al. found a correlation between higher follicular fluid PFAAs with lower fertilization rate in a pilot study (n=16).23 However, the authors did not provide detailed information of PFAA levels in the samples. In a U.S.-based study (n=36), McCoy et al. detected negative relationships of PFDA and PFUnDA concentrations with blastocyst conversion rate (the number of day-6 blastocysts per cultured day-3 embryo) although no statistically significant associations with measures of ovarian response to gonadotropin simulation during IVF were found.19 In our study, we also detected statistically significant associations between higher follicular fluid PFDA and PFUnDA concentrations and poorer fertilization outcomes, measured as a lower high-quality embryo rate. A more recent study showed that both maternal and paternal plasma PFOA concentrations were negatively associated with IVF outcomes.43 In a Belgian study (n=38), Petro et al. observed contradictory findings using a principal component analysis in which overall follicular fluid PFAAs were associated with an unexpected higher fertilization rate after adjusting for covariates and other EDCs.22 No relationship between eight different follicular fluid PFAAs and fertilization rate was observed among Australian women.20 Hong et al. also reported no association between PFAA concentrations in follicular fluid and IVF clinical outcomes in China (n=124).24 All of the above published studies were based on small sample sizes. Ours is the largest study to date to assess associations of follicular fluid PFAAs with fertility outcomes. In addition, our study comprehensively profiled follicular fluid PFAA concentrations including linear and branched PFOS in follicular fluid, which provided new evidence on the potential health impacts of exposure to PFAA isomers.
We found that increased follicular fluid PFAAs were associated with a lower oocyte maturity rate, which is a marker of the efficiency of ovarian stimulation and triggering in IVF treatment.44 Our findings further suggested that oocyte maturity rate may partially mediate the adverse associations between greater PFAAs and poorer embryo quality. Recent in vitro and in vivo studies have shown that PFAAs can impair oocyte maturation, but the epidemiological evidence to date is limited.10 Similar to our results, Ma et al. reported that higher maternal serum PFOA concentrations were significantly associated with fewer mature oocytes and fewer good-quality embryos among 97 women using IVF.43 Still, additional epidemiological studies are needed to more definitively characterize the potential adverse association between PFAA exposure and fertility and the potential mediating role of oocyte quality.
Recent literature reviews suggest that health risks may differ for exposure to linear and branched PFAA isomers.45,46 Although a large number of studies have reported adverse effects of exposure to total PFAA in early life, including differences in hormones, very few have investigated associations with PFAA isomers to date.47 For example, recent studies have shown that exposure to branched PFAA were associated with subclinical maternal hypothyroidism47 and with lower estradiol levels in young men.48 We found branched PFOS isomers to be the major contributor to the association between a mixture of follicular fluid PFAA and a lower high-quality embryo rate in the current study. On the contrary, another recent study from China observed no relationship of exposure to branched PFOS isomers in follicular fluid with embryo quality and implantation in women who underwent IVF treatment.24 However, the number of participants in the aforementioned study was smaller (n=124 vs. n=729), and the concentration of branched PFOS was also lower than that in our study (< limit of quantitation vs. 0.47 ng/mL). PFAA isomer-specific epidemiological and toxicological studies will be needed to more definitively characterize potential differences in reproductive health risks given broad exposure to PFAA isomers.
Biologic Mechanisms
The results of experimental studies suggest that PFAA have adverse effects on ovarian folliculogenesis by altering oocyte development.10,49,50 Exposure to PFAA induced apoptosis in human ovarian granulosa cells51 and disrupted steroidogenic secretion in both granulosa and porcine theca cells with or without gonadotropic stimulation.52 Using an ex vivo and transcript sequencing approach, Khan et al. found that PFOS was the active component in a mixture of PFOS, PFOA, and PFNA that altered normal ovarian function in Atlantic cod fish.53 Neonatal exposure to high-dose PFOS or PFOA (0.1, 1, and 10mg/kg/d) reduced follicle numbers in female rats, although no altered ovarian morphology was observed.54 Recent studies showed that PFOS55 and PFNA56 could impede oocyte maturation by damaging mitochondrial functions in vitro, inducing oxidative stress and causing apoptosis of oocytes. In an in vitro bovine oocyte model, Hallberg et al. also found that exposure to human-relevant PFOS concentrations altered early embryo development, which may have negative consequences for further development.57
A plausible mechanism for the relationship between PFAA exposure and lower high-quality embryo rate might be attributed to activation of peroxisome proliferator–activated receptors (PPARs)58 that play a crucial role in gamete function and oocyte development.59 PFAA binding to PPARs could interact with steroidogenesis response elements and have deleterious impacts on follicular and subsequently embryo development. In ovo exposure to environmentally relevant levels of PFOS suppressed transcription of genes involved in PPAR-mediated transcription.60 Exposure of bovine oocytes to 0.01μg/mL–100μg/mL PFHxS during in vitro maturation affected pathways downstream of PPARγ and estrogen signaling and decreased oocyte developmental competence, leading to compromised embryo quality and development.49 Sant et al. reported that PFOS induced apoptosis and oxidative stress in zebrafish embryos, probably by activating potential nuclear factor-erythroid 2-related factor 2 (Nrf2)–PPAR crosstalk.61 Another study showed that a mixture of linear and branched PFOS elicited a greater transcriptional response than linear PFOS alone elicited, including PPARγ signaling, in chicken embryo hepatocytes, suggesting that the isomer-specific toxicological properties of PFOS should be considered.62 However, the PFAA doses used in animal studies were much higher than the PFAA concentrations encountered by most human populations, which may activate different toxicity pathways and toxic effects. Experimental studies using PFAA doses equivalent to human background exposures and using PFAA mixtures are required in future studies to elucidate the potential effects of PFAA on human fertility more clearly.10
Strengths and Limitations
Our study has some major strengths and novel features. First, we measured PFAA concentrations in the follicle microenvironment, which may more closely approximate the biologically effective dose to a developing oocyte than blood PFAA. In addition, our sample size of 729 follicular fluids is the largest reported to date, far eclipsing the sample sizes of previous studies, which measured no more than 100 follicular fluids. Second, we applied PFOS isomer-specific analysis to capture isomer-specific health associations, because crude measures of “total-PFOS” may have masked associations with branched isomers in previous work.47 Third, we collected a large number of covariates, allowing us to adjust for confounding variables and to assess potential biases using sensitivity analyses, including the possibility for confounding and reverse causation by parity.11 Fourth, we used a mediation analysis to investigate successful completion of meiosis 1 (i.e., MII-oocyte rate) as a potential biological mechanism driving the associations between PFAAs and embryo quality. Finally, we used BKMR models, which do not constrain the form or nature of associations a priori, to assess the overall effect of the PFAA mixture on embryo quality and to identify the major contributor of PFAA effects in the exposure–response relationship.
However, our findings should be interpreted as hypothesis-generating, given some limitations. First, we collected and pooled two to four follicular fluid specimens from each participant, which may have misclassified exposure for some women if there is significant follicle-to-follicle variability in PFAA concentrations. However, plasma/serum and follicular fluid PFAA were strongly correlated in previous studies19,21,23 with no difference22 or only modestly greater plasma relative to follicular fluid PFAA concentrations,19 suggesting that follicular fluid PFAA reflect the blood compartment. Still, follicle-to-follicle variability may have misclassified exposure in some follicles, and the pooled exposure estimate may have underestimated the associations with embryo quality. A future investigation using a “one-follicle, one-oocyte” design will be necessary for a more definitive result. We also did not incorporate seminal PFAA concentrations from the male partner into the current analysis, which may have further misclassified exposure or introduced unmeasured copollutant confounding. A recent study showed that the seminal PFOS and PFOA concentrations were associated with a lower percentage of progressive sperm and higher percentage of DNA fragmentation in men (n=664), suggesting deleterious effects of PFAA exposure on sperm quality.63 This finding may indicate a potential contribution of PFAA exposure from seminal PFAA concentrations on embryo fertilization. We plan to validate our findings using male partners’ semen PFAA data in the next step. Second, we cannot rule out the possible contamination of blood PFAA in follicular fluid, which may introduce exposure measurement error. However, any occult blood contamination was unlikely to have been differential by embryo quality outcome and will most likely bias the results toward the null hypothesis. Third, IVF patients tend to be highly selected, often comprising older couples actively trying to conceive and with higher socioeconomic status than the general population,64 which may restrict generalizability of the study results. However, IVF patients may also be a vulnerable population, with heightened sensitivity to environmental chemical exposures providing sentinel indicators of reproductive toxicity.65
Conclusions
This study found significant adverse associations between follicular fluid PFAA and embryo quality. Oocyte maturity may partially mediate the associations. Important PFAA isomeric associations with poorer embryo quality were also suggested, which may subsequently impact pregnancy and live birth from IVF. However, a future follow-up study of the associations between follicular fluid PFAA with pregnancy and live-birth outcomes is needed to estimate the impact directly. These findings may have important public health implications that help contribute to understanding potential environmental risk factors for unfavorable IVF outcomes and offer regulators and policy makers additional evidence to manage use of PFAA.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This project was supported by the National Key Research and Development Program of China (2018YFE0106900), the National Natural Science Foundation of China (82073503, 81872582, 81703179), the Guangxi Natural Science Foundation (2018GXNSFBA138047), the Guangxi Key Research and Development Plan (GUIKEAB18050024), and Guangdong Natural Science Foundation (2021B1515020015).
==== Refs
References
1. Zegers-Hochschild F, Nygren KG, Adamson GD, de Mouzon J, Lancaster P, Mansour R, et al. 2006. The International Committee Monitoring Assisted Reproductive Technologies (ICMART) glossary on ART terminology. Fertil Steril 86 (1 ):16–19, PMID: , 10.1016/j.fertnstert.2006.04.018.16762350
2. Mascarenhas MN, Flaxman SR, Boerma T, Vanderpoel S, Stevens GA. 2012. National, regional, and global trends in infertility prevalence since 1990: a systematic analysis of 277 health surveys. PLoS Med 9 (12 ):e1001356, PMID: , 10.1371/journal.pmed.1001356.23271957
3. Steiner AZ, Pritchard D, Stanczyk FZ, Kesner JS, Meadows JW, Herring AH, et al. 2017. Association between biomarkers of ovarian reserve and infertility among older women of reproductive age. JAMA 318 (14 ):1367–1376, PMID: , 10.1001/jama.2017.14588.29049585
4. Evers JLH. 2002. Female subfertility. Lancet 360 (9327 ):151–159, PMID: , 10.1016/S0140-6736(02)09417-5.12126838
5. Karwacka A, Zamkowska D, Radwan M, Jurewicz J. 2019. Exposure to modern, widespread environmental endocrine disrupting chemicals and their effect on the reproductive potential of women: an overview of current epidemiological evidence. Hum Fertil (Camb) 22 (1 ):2–25, PMID: , 10.1080/14647273.2017.1358828.28758506
6. ITRC (Interstate Technology and Regulatory Council). Per- and Polyfluoroalkyl Substances Technical and Regulatory Guidance. https://pfas-1.itrcweb.org/wp-content/uploads/2022/09/PFAS-Guidance-Document-9-2022.pdf [accessed 10 October 2022].
7. Houde M, Martin JW, Letcher RJ, Solomon KR, Muir DCG. 2006. Biological monitoring of polyfluoroalkyl substances: a review. Environ Sci Technol 40 (11 ):3463–3473, PMID: , 10.1021/es052580b.16786681
8. U.S. Centers for Disease Control and Prevention. 2019. Fourth National Report on Human Exposure to Environmental Chemicals. Updated Tables. Atlanta, GA: Centers for Disease Control and Prevention; National Center for Environmental Health; Division of Laboratory Sciences. http://www.cdc.gov/exposurereport [accessed 10 October 2022].
9. Lim TC, Wang B, Huang J, Deng S, Yu G. 2011. Emission inventory for PFOS in China: review of past methodologies and suggestions. ScientificWorldJournal 11 :1963–1980, PMID: , 10.1100/2011/868156.22125449
10. Ding N, Harlow SD, Randolph JF Jr, Loch-Caruso R, Park SK. 2020. Perfluoroalkyl and polyfluoroalkyl substances (PFAS) and their effects on the ovary. Hum Reprod Update 26 (5 ):724–752, PMID: , 10.1093/humupd/dmaa018.32476019
11. Bach CC, Vested A, Jørgensen KT, Bonde JPE, Henriksen TB, Toft G. 2016. Perfluoroalkyl and polyfluoroalkyl substances and measures of human fertility: a systematic review. Crit Rev Toxicol 46 (9 ):735–755, PMID: , 10.1080/10408444.2016.1182117.27268162
12. Negri E, Metruccio F, Guercio V, Tosti L, Benfenati E, Bonzi R, et al. 2017. Exposure to PFOA and PFOS and fetal growth: a critical merging of toxicological and epidemiological data. Critical Reviews in Toxicology 47 (6 ):482–508, PMID: , 10.1080/10408444.2016.1271972.28617200
13. Fei C, McLaughlin JK, Lipworth L, Olsen J. 2009. Maternal levels of perfluorinated chemicals and subfecundity. Hum Reprod 24 (5 ):1200–1205, PMID: , 10.1093/humrep/den490.19176540
14. Fei C, Weinberg CR, Olsen J. 2012. Commentary: perfluorinated chemicals and time to pregnancy: a link based on reverse causation? Epidemiology 23 (2 ):264–266, PMID: , 10.1097/EDE.0b013e3182467608.22317809
15. Whitworth KW, Haug LS, Baird DD, Becher G, Hoppin JA, Skjaerven R, et al. 2012. Perfluorinated compounds and subfecundity in pregnant women. Epidemiology 23 (2 ):257–263, PMID: , 10.1097/EDE.0b013e31823b5031.22081060
16. Bach CC, Liew Z, Bech BH, Nohr EA, Fei C, Bonefeld-Jorgensen EC, et al. 2015. Perfluoroalkyl acids and time to pregnancy revisited: an update from the Danish National Birth Cohort. Environ Health 14 :59, PMID: , 10.1186/s12940-015-0040-9.26148742
17. Gosden RG, Hunter RH, Telfer E, Torrance C, Brown N. 1988. Physiological factors underlying the formation of ovarian follicular fluid. J Reprod Fertil 82 (2 ):813–825, PMID: , 10.1530/jrf.0.0820813.3283348
18. Petro EML, Leroy JLMR, Covaci A, Fransen E, De Neubourg D, Dirtu AC, et al. 2012. Endocrine-disrupting chemicals in human follicular fluid impair in vitro oocyte developmental competence. Hum Reprod 27 (4 ):1025–1033, PMID: , 10.1093/humrep/der448.22267834
19. McCoy JA, Bangma JT, Reiner JL, Bowden JA, Schnorr J, Slowey M, et al. 2017. Associations between perfluorinated alkyl acids in blood and ovarian follicular fluid and ovarian function in women undergoing assisted reproductive treatment. Sci Total Environ 605–606 :9–17, PMID: , 10.1016/j.scitotenv.2017.06.137.28651210
20. Kim YR, White N, Bräunig J, Vijayasarathy S, Mueller JF, Knox CL, et al. 2020. Per- and poly-fluoroalkyl substances (PFASs) in follicular fluid from women experiencing infertility in Australia. Environ Res 190 :109963, PMID: , 10.1016/j.envres.2020.109963.32745751
21. Heffernan AL, Cunningham TK, Drage DS, Aylward LL, Thompson K, Vijayasarathy S, et al. 2018. Perfluorinated alkyl acids in the serum and follicular fluid of UK women with and without polycystic ovarian syndrome undergoing fertility treatment and associations with hormonal and metabolic parameters. Int J Hyg Environ Health 221 (7 ):1068–1075, PMID: , 10.1016/j.ijheh.2018.07.009.30037723
22. Petro EML, D’Hollander W, Covaci A, Bervoets L, Fransen E, De Neubourg D, et al. 2014. Perfluoroalkyl acid contamination of follicular fluid and its consequence for in vitro oocyte developmental competence. Sci Total Environ 496 :282–288, PMID: , 10.1016/j.scitotenv.2014.07.028.25089690
23. Governini L, Orvieto R, Guerranti C, Gambera L, De Leo V, Piomboni P. 2011. The impact of environmental exposure to perfluorinated compounds on oocyte fertilization capacity. J Assist Reprod Genet 28 (5 ):415–418, PMID: , 10.1007/s10815-011-9548-2.21344220
24. Hong A, Zhuang L, Cui W, Lu Q, Yang P, Su S, et al. 2022. Per- and polyfluoroalkyl substances (PFAS) exposure in women seeking in vitro fertilization-embryo transfer treatment (IVF-ET) in China: blood-follicular transfer and associations with IVF-ET outcomes. Sci Total Environ 838 (pt 3 ):156323, PMID: , 10.1016/j.scitotenv.2022.156323.35636536
25. Zhou F, Tang J, Li P, Liao B, Qin C. 2020. Distribution of cerebral artery stenosis and risk factors in ethnic Zhuang and Han patients with ischemic stroke in Guangxi province. Ann Palliat Med 9 (2 ):256–263, PMID: , 10.21037/apm.2020.02.32.32156133
26. SART (Society for Assisted Reproductive Technology). 2019. National Summary Report: All SART Member Clinics. https://www.sartcorsonline.com/rptCSR_PublicMultYear.aspx?ClinicPKID=0 [accessed 10 October 2022].
27. Alpha Scientists in Reproductive Medicine and ESHRE Special Interest Group of Embryology. 2011. The Istanbul consensus workshop on embryo assessment: proceedings of an expert meeting. Hum Reprod 26 (6 ):1270–1283, PMID: , 10.1093/humrep/der037.21502182
28. Levay PF, Huyser C, Fourie FL, Rossouw DJ. 1997. The detection of blood contamination in human follicular fluid. J Assist Reprod Genet 14 (4 ):212–217, PMID: , 10.1007/BF02766112.9130069
29. Chu C, Zhou Y, Li Q-Q, Bloom MS, Lin S, Yu Y-J, et al. 2020. Are perfluorooctane sulfonate alternatives safer? New insights from a birth cohort study. Environ Int 135 :105365, PMID: , 10.1016/j.envint.2019.105365.31830731
30. U.S. EPA (U.S. Environmental Protection Agency). 2016. Definition and Procedure for the Determination of the Method Detection Limit, Revision 2. EPA 821-R-16R006. Washington, DC: Office of Science and Technology, U.S. Environmental Protection Agency.
31. Lubin JH, Colt JS, Camann D, Davis S, Cerhan JR, Severson RK, et al. 2004. Epidemiologic evaluation of measurement data in the presence of detection limits. Environ Health Perspect 112 (17 ):1691–1696, PMID: , 10.1289/ehp.7199.15579415
32. U.S. EPA. 2000. Guidance for Data Quality Assessment–Quality Practical Methods for Data Analysis. EPA QA/G-9, QA00 version. Washington, DC: Office of Environmental Information, U.S. Environmental Protection Agency.
33. Desquilbet L, Mariotti F. 2010. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med 29 (9 ):1037–1057, PMID: , 10.1002/sim.3841.20087875
34. Deddens JA, Petersen MR. 2008. Approaches for estimating prevalence ratios. Occup Environ Med 65 (7 ):501–506, PMID: , 10.1136/oem.2007.034777.18562687
35. Bach C, Matthiesen N, Olsen J, Henriksen T. 2018. Conditioning on parity in studies of perfluoroalkyl acids and time to pregnancy: an example from the Danish National Birth Cohort. Environ Health Perspect 126 (11 ):117003, PMID: , 10.1289/EHP1493.30417653
36. Valeri L, Vanderweele TJ. 2013. Mediation analysis allowing for exposure-mediator interactions and causal interpretation: theoretical assumptions and implementation with SAS and SPSS macros. Psychol Methods 18 (2 ):137–150, PMID: , 10.1037/a0031034.23379553
37. Bobb JF, Claus Henn B, Valeri L, Coull BA. 2018. Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression. Environ Health 17 (1 ):67, PMID: , 10.1186/s12940-018-0413-y.30126431
38. Preston EV, Webster TF, Claus Henn B, McClean MD, Gennings C, Oken E, et al. 2020. Prenatal exposure to per- and polyfluoroalkyl substances and maternal and neonatal thyroid function in the project viva cohort: a mixtures approach. Environ Int 139 :105728, PMID: , 10.1016/j.envint.2020.105728.32311629
39. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16 (3 ):493–508, PMID: , 10.1093/biostatistics/kxu058.25532525
40. Benjamini Y, Hochberg Y. 1995. Controlling the false discovery rate: of practical and powerful approach two multiple testing. J R Stat Soc Series B Stat Methodol 57 (1 ):289–300, 10.1111/j.2517-6161.1995.tb02031.x.
41. Kang Q, Gao F, Zhang X, Wang L, Liu J, Fu M, et al. 2020. Nontargeted identification of per- and polyfluoroalkyl substances in human follicular fluid and their blood-follicle transfer. Environ Int 139 :105686, PMID: , 10.1016/j.envint.2020.105686.32278886
42. Domingo JL, Nadal M. 2017. Per- and polyfluoroalkyl substances (PFASs) in food and human dietary intake: a review of the recent scientific literature. J Agric Food Chem 65 (3 ):533–543, PMID: , 10.1021/acs.jafc.6b04683.28052194
43. Ma X, Cui L, Chen L, Zhang J, Zhang X, Kang Q, et al. 2021. Parental plasma concentrations of perfluoroalkyl substances and in vitro fertilization outcomes. Environ Pollut 269 :116159, PMID: , 10.1016/j.envpol.2020.116159.33279270
44. ESHRE Special Interest Group of Embryology and Alpha Scientists in Reproductive Medicine 2017. The Vienna consensus: report of an expert meeting on the development of ART laboratory performance indicators. Hum Reprod Open 2017 (2 ):hox011, PMID: , 10.1093/hropen/hox011.31486806
45. Liu Y, Li A, An Q, Liu K, Zheng P, Yin S, et al. 2021. Prenatal and postnatal transfer of perfluoroalkyl substances from mothers to their offspring. Crit Rev Environ Sci Technol 52 (14 ):2510–2537, 10.1080/10643389.2021.1886556.
46. Schulz K, Silva MR, Klaper R. 2020. Distribution and effects of branched versus linear isomers of PFOA, PFOS, and PFHxS: a review of recent literature. Sci Total Environ 733 :139186, PMID: , 10.1016/j.scitotenv.2020.139186.32474294
47. Reardon AJF, Khodayari Moez E, Dinu I, Goruk S, Field CJ, Kinniburgh DW, et al. 2019. Longitudinal analysis reveals early-pregnancy associations between perfluoroalkyl sulfonates and thyroid hormone status in a Canadian prospective birth cohort. Environ Int 129 :389–399, PMID: , 10.1016/j.envint.2019.04.023.31150980
48. Luo K, Liu X, Nian M, Wang Y, Qiu J, Yu H, et al. 2021. Environmental exposure to per- and polyfluoroalkyl substances mixture and male reproductive hormones. Environ Int 152 :106496, PMID: , 10.1016/j.envint.2021.106496.33744484
49. Hallberg I, Kjellgren J, Persson S, Örn S, Sjunnesson Y. 2019. Perfluorononanoic acid (PFNA) alters lipid accumulation in bovine blastocysts after oocyte exposure during in vitro maturation. Reprod Toxicol 84 :1–8, PMID: , 10.1016/j.reprotox.2018.11.005.30502403
50. Domínguez A, Salazar Z, Arenas E, Betancourt M, Ducolomb Y, González-Márquez H, et al. 2016. Effect of perfluorooctane sulfonate on viability, maturation and gap junctional intercellular communication of porcine oocytes in vitro. Toxicol In Vitro 35 :93–99, PMID: , 10.1016/j.tiv.2016.05.011.27233358
51. Zhou Y, Li H, Lin C, Mao Y, Rao J, Lou Y, et al. 2020. Perfluorooctanoic acid (PFOA) inhibits the gap junction intercellular communication and induces apoptosis in human ovarian granulosa cells. Reprod Toxicol 98 :125–133, PMID: , 10.1016/j.reprotox.2020.09.005.32971237
52. Chaparro-Ortega A, Betancourt M, Rosas P, Vázquez-Cuevas FG, Chavira R, Bonilla E, et al. 2018. Endocrine disruptor effect of perfluorooctane sulfonic acid (PFOS) and perfluorooctanoic acid (PFOA) on porcine ovarian cell steroidogenesis. Toxicol In Vitro 46 :86–93, PMID: , 10.1016/j.tiv.2017.09.030.28982594
53. Khan EA, Zhang X, Hanna EM, Yadetie F, Jonassen I, Goksøyr A, et al. 2021. Application of quantitative transcriptomics in evaluating the ex vivo effects of per- and polyfluoroalkyl substances on Atlantic cod (Gadus morhua) ovarian physiology. Sci Total Environ 755 (pt 1 ):142904, PMID: , 10.1016/j.scitotenv.2020.142904.33138996
54. Du G, Hu J, Huang Z, Yu M, Lu C, Wang X, et al. 2019. Neonatal and juvenile exposure to perfluorooctanoate (PFOA) and perfluorooctane sulfonate (PFOS): advance puberty onset and kisspeptin system disturbance in female rats. Ecotoxicol Environ Saf 167 :412–421, PMID: , 10.1016/j.ecoenv.2018.10.025.30368134
55. Chen J, Miao Y, Gao Q, Cui Z, Xiong B. 2021. Exposure to perfluorooctane sulfonate in vitro perturbs the quality of porcine oocytes via induction of apoptosis. Environ Pollut 284 :117508, PMID: , 10.1016/j.envpol.2021.117508.34261219
56. Jiao X, Liu N, Xu Y, Qiao H. 2021. Perfluorononanoic acid impedes mouse oocyte maturation by inducing mitochondrial dysfunction and oxidative stress. Reprod Toxicol 104 :58–67, PMID: , 10.1016/j.reprotox.2021.07.002.34246765
57. Hallberg I, Persson S, Olovsson M, Sirard M-A, Damdimopoulou P, Rüegg J, et al. 2021. Perfluorooctane sulfonate (PFOS) exposure of bovine oocytes affects early embryonic development at human-relevant levels in an in vitro model. Toxicology 464 :153028, PMID: , 10.1016/j.tox.2021.153028.34762985
58. Wolf CJ, Schmid JE, Lau C, Abbott BD. 2012. Activation of mouse and human peroxisome proliferator-activated receptor-alpha (PPARα) by perfluoroalkyl acids (PFAAs): further investigation of C4–C12 compounds. Reprod Toxicol 33 (4 ):546–551, PMID: , 10.1016/j.tox.2021.153028.22107727
59. Maradonna F, Carnevali O. 2018. Lipid metabolism alteration by endocrine disruptors in animal models: an overview. Front Endocrinol (Lausanne) 9 :654, PMID: , 10.3389/fendo.2018.00654.30467492
60. Jacobsen AV, Nordén M, Engwall M, Scherbak N. 2018. Effects of perfluorooctane sulfonate on genes controlling hepatic fatty acid metabolism in livers of chicken embryos. Environ Sci Pollut Res Int 25 (23 ):23074–23081, PMID: , 10.1007/s11356-018-2358-7.29860686
61. Sant KE, Sinno PP, Jacobs HM, Timme-Laragy AR. 2018. Nrf2a modulates the embryonic antioxidant response to perfluorooctanesulfonic acid (PFOS) in the zebrafish, Danio rerio. Aquat Toxicol 198 :92–102, PMID: , 10.1016/j.aquatox.2018.02.010.29524743
62. O’Brien JM, Austin AJ, Williams A, Yauk CL, Crump D, Kennedy SW. 2011. Technical-grade perfluorooctane sulfonate alters the expression of more transcripts in cultured chicken embryonic hepatocytes than linear perfluorooctane sulfonate. Environ Toxicol Chem 30 (12 ):2846–2859, PMID: , 10.1002/etc.700.21994020
63. Pan Y, Cui Q, Wang J, Sheng N, Jing J, Yao B, et al. 2019. Profiles of emerging and legacy per-/polyfluoroalkyl substances in matched serum and semen samples: new implications for human semen quality. Environ Health Perspect 127 (12 ):127005, PMID: , 10.1289/EHP4431.31841032
64. Männistö J, Mentula M, Bloigu A, Gissler M, Heikinheimo O, Niinimäki M. 2019. Induced abortion and future use of IVF treatment: a nationwide register study. PLoS One 14 (11 ):e0225162, PMID: , 10.1371/journal.pone.0225162.31725766
65. Dodge LE, Williams PL, Williams MA, Missmer SA, Toth TL, Calafat AM, et al. 2015. Paternal urinary concentrations of parabens and other phenols in relation to reproductive outcomes among couples from a fertility clinic. Environ Health Perspect 123 (7 ):665–671, PMID: , 10.1289/ehp.1408605.25767892
|
PMC009xxxxxx/PMC9891135.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36722980
EHP11248
10.1289/EHP11248
Research
Transportation Noise and Risk of Tinnitus: A Nationwide Cohort Study from Denmark
https://orcid.org/0000-0003-3201-9606
Cantuaria Manuella Lech 1 2 3
Pedersen Ellen Raben 1
Poulsen Aslak Harbo 2
Raaschou-Nielsen Ole 2 4
Hvidtfeldt Ulla Arthur 2
Levin Gregor 4
Jensen Steen Solvang 4
Schmidt Jesper Hvass 3 5 6
Sørensen Mette 2 7
1 The Mærsk McKinney Møller Institute, University of Southern Denmark, Odense, Denmark
2 Work, Environment and Cancer, Danish Cancer Society Research Center, Copenhagen, Denmark
3 Department of Clinical Research, University of Southern Denmark, Odense, Denmark
4 Department of Environmental Science, Aarhus University, Roskilde, Denmark
5 Research Unit for ORL – Head and Neck Surgery and Audiology, Odense University Hospital, Odense, Denmark
6 OPEN, Odense Patient Data Explorative Network, Odense University Hospital, Odense, Denmark
7 Department of Natural Science and Environment, Roskilde University, Roskilde, Denmark
Address correspondence to Manuella Lech Cantuaria, Campusvej 55, 5230, Odense M, Denmark. Telephone: +45 2721-1181. Email: [email protected]
01 2 2023
2 2023
131 2 02700114 3 2022
13 12 2022
20 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
There is a growing body of evidence linking residential exposure to transportation noise with several nonauditory health outcomes. However, auditory outcomes, such as tinnitus, are virtually unexplored.
Objectives:
We aimed to investigate the association between residential transportation noise and risk of incident tinnitus.
Methods:
We conducted a nationwide cohort study including all residents in Denmark age ≥30y, of whom 40,692 were diagnosed with tinnitus. We modeled road traffic and railway noise at the most (Ldenmax) and least (Ldenmin) exposed façades of all Danish addresses from 1990 until 2017. For all participants, we calculated 1-, 5-, and 10-y time-weighted mean noise exposure and retrieved detailed information on individual- and area-level socioeconomic covariates. We conducted analyses using Cox proportional hazards models.
Results:
We found positive associations between exposure to road traffic noise and risk of tinnitus, with hazard ratios of 1.06 [95% confidence interval (CI): 1.04, 1.08] and 1.02 (95% CI: 1.01, 1.03) per 10-dB increase in 10-y Ldenmin and Ldenmax, respectively. Highest risk estimates were found for women, people without a hearing loss, people with high education and income, and people who had never been in a blue-collar job. The association with road Ldenmin followed a positive, monotonic exposure–response relationship. We found no association between railway noise and tinnitus.
Discussion:
To our knowledge, this is the first study to show that residential exposure to road traffic noise may increase risk of tinnitus, suggesting noise may negatively affect the auditory system. If confirmed, this finding adds to the growing evidence of road traffic noise as a harmful pollutant with a substantial health burden. https://doi.org/10.1289/EHP11248
Supplemental Material is available online (https://doi.org/10.1289/EHP11248).
J.H.S. declares research collaboration with the companies Oticon, GN Hearing, Interacoustics, and WSA. All other authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Transportation noise is increasingly recognized as harmful to human health, being a growing source of concern among the general population. In Europe, more than 100 million people are exposed to transportation noise above the Environmental Noise Directive threshold of 55 dB.1 Noise is defined as unwanted sound, being often unpleasant and/or disruptive to the listener. Transportation noise exposure is believed to be detrimental to human health through stress reactions with activation of the hypothalamic–pituitary–adrenal axis, followed by increased levels of stress hormones.2,3 In addition, nighttime noise exposure can impact sleep quality and duration, which are crucial for physiological and mental restoration.2,3
In an extensive review of health effects of transportation noise, the World Health Organization (WHO) concluded that there is high-quality evidence for an association between road traffic noise and ischemic heart diseases.4 Since the WHO review, several observational studies have provided evidence suggesting that road traffic noise is also a risk factor for diabetes, stroke, and cardiovascular mortality.5–11 Despite emerging knowledge, the extent of health impacts from transportation noise is still not clear; e.g., the WHO stated in 2018 that there was a lack of studies investigating transportation noise and hearing-related outcomes, such as tinnitus, even though these are considered critical outcomes for the development of guidelines regarding health effects of noise.4
Tinnitus is a very common disorder characterized by the perception of sound in the ears or head in the absence of an external sound source.12 Epidemiological studies worldwide have reported the tinnitus prevalence to range between 5% and 43%.13 However, there is no standard criterion for tinnitus diagnosis, and the heterogeneity of the disease in terms of severity and impact is substantial.13 Although many people can habituate to it, others are severely affected by the disorder even after seeking medical treatment.12,14
Hearing loss and other otological conditions affecting the middle ear are main risk factors for tinnitus. However, although cochlear damage is often the origin of tinnitus, the central nervous system is believed to play an important role in the onset and persistence of the disorder.12,15 Tinnitus is considered a stressor per se, leading to increased physiological arousal and psychological distress.14,16 Nevertheless, several studies also suggest a reverse mechanism, where stressful situations and sleep disturbances precede tinnitus occurrence and contribute to the transition from mild to severe symptoms.14,15,17 As stress and sleep disturbance are proposed key mechanisms behind the harmful effects of noise,2–4 we hypothesize that transportation noise can affect onset and severity of tinnitus. However, to the extent of our knowledge, no longitudinal studies have investigated the effect of residential transportation noise on tinnitus or other auditory outcomes.
This nationwide cohort study aimed to investigate the association between residential road traffic and railway noise exposure, measured both at the least and the most exposed façades, and risk of incident tinnitus. The study was based on individual-level information on hearing-related diagnoses, address history, and socioeconomic factors.
Methods
Study Design and Participants
We performed a nationwide cohort study, including all residents age ≥30y living in Denmark between 1 January 2000 and 31 December 2017, and born after 1920. All Danish residents were followed across health and administrative registers using the Danish unique personal identification number.18 We identified all Danish addresses in the Building and Housing Register,19 and, by linking them with the Civil Registration System,20 we retrieved address history for all study participants from 10 y before enrollment until censoring.
The study was conducted in accordance with principles of the Declaration of Helsinki and approved by the local authorities (record number: 2018-DCRC-0055). Because the study was entirely based on data from the Danish national registers, it did not require patients’ consent and approval from ethical committees.
Outcome
We followed all cohort members up for an incident diagnosis of tinnitus by linking their personal identification numbers to the Danish National Patient Register (DNPR).21 Tinnitus was defined as a primary (i.e., main cause of the visit) or secondary (i.e., coexisting) diagnosis for outpatient records according to the International Classification of Diseases (ICD) 8 code 781.31 or ICD10 code H93.1. Cohort members with a diagnosis of tinnitus before baseline were excluded.
Noise Exposure Assessment
We calculated road traffic and railway noise at the center of all façades of all residential buildings in Denmark, using precise geocoded data for location and floor (corresponding to the address height) of each address.22 We further selected the lowest and highest noise level for each address, which corresponded to noise at the least and most exposed façade, respectively. Noise levels were calculated as the equivalent continuous A-weighted sound pressure level (LAeq) for day, evening, and night and represented as Lden. An A-weighted scale aims to mimic human hearing responses by emphasizing the main frequencies perceived by the human ear when sound pressure levels are calculated. All noise values below 35 dB were set to 35 dB, because transportation noise levels below this threshold are likely not perceived, due to other noise sources. Road traffic noise was modeled for the years 1995, 2000, 2005, 2010, and 2015 using the Nordic prediction method.23 Input variables included annual average daily traffic, vehicle distribution, road type, and travel speed.24 Railway noise was modeled for the years 1997 and 2012 for all addresses within 1,000m of a rail line, using the Nord2000 models.25 Input variables included annual average daily train lengths, travel speed, and train types. Both models considered screening from terrains, noise barriers, berms, and buildings, as well as ground absorption and noise reflections. We used linear interpolation to quantify exposure in intermediary years.
Covariates
We retrieved information on a variety of covariates available on registers at Statistics Denmark, which were selected with basis on the current literature and a review of plausible mechanisms. These included yearly individual-level information on: a) civil status: married/cohabiting/registered partnership; widowed; divorced; single; b) highest attained education: mandatory; secondary, vocational, or other short further education after high school; medium or long-term higher education (Bachelor’s, Master’s, or PhD degree); c) occupational status: blue-collar (employment that requires low-level skills); low-level white-collar (managers with 0–4 employees or with employment that requires intermediate-level skills); high-level white-collar (managers/directors with ≥5 employees or with employment that requires high-level skills); unemployed; retired; d) disposable income (in quintiles, based on the yearly distribution among Danish adults between 25 and 70 y of age and standardized by calendar year and sex-specific categories); and e) country of origin: Denmark; immigrant or descendent of individuals from other Western country (i.e., European Union country, Andorra, Australia, Canada, Iceland, Liechtenstein, Monaco, New Zealand, Norway, San Marino, Switzerland, United States, and Vatican City State); immigrant or descendent of individuals from a non-Western country (all other countries).
We also collected address-level data on population density and generated neighborhood (parish) socioeconomic status variables by aggregating national data for the 2,160 parishes of Denmark (i.e., the proportion of inhabitants in each parish with low income (lowest quartile), who were unemployed, with manual labor, with only basic education, with a criminal record, and who belonged to single-parent families).26 Given potential beneficial health effects when living close to green areas (e.g., stress reduction and restoration),27,28 we also considered “access to green areas” as a covariate in the present study. For each address, we calculated the proportion of forests, recreational areas, and open nature areas within a 150-m and 1,000-m radius buffer (hereinafter referred as high-quality green areas), based on land-use categories extracted from a nationwide land-use and land-cover map for Denmark.29
Last, we retrieved from the DNPR all participants who had been referred to hospital examinations and treatment for hearing loss, other hearing loss–related diseases, and outer and middle ear diseases during and before the study period. Hearing loss diagnosis was defined according to the following ICD10 codes: H80, H810, H833, H838, H839, H90, H91, H93 (excluding H931, H932, H933B), and DH94. Outer and middle ear diseases were defined by the following ICD10 codes: H60, H61, H62, H65, H66, H67, H68, H69, H70, H71, H72, H73, H74, and H75.
Statistical Analyses
Hazard ratios (HR) were estimated using Cox regression models with age as underlying time scale. Participants were enrolled on 1 January 2000 or when they turned 30 y of age, whichever came last, and censored at age of tinnitus diagnosis, death, migration, missing address, or end of follow-up (31 December 2017), whichever came first. We included road traffic and railway noise exposures as time-weighted means calculated for running 1-, 5-, and 10-y periods, taking into consideration the full address history for each study participant.
We calculated linear associations between road traffic and railway noise at the most (Ldenmax) and least (Ldenmin) exposed façades (per 10 dB) and incident tinnitus, using: a) a basic model, adjusted by age (underlying time scale), sex, and calendar year; and b) a fully adjusted model, including all covariates. All covariates, apart from sex and region of origin, were included in the model as a time-varying variables. Individual-level covariates were updated yearly, whereas area-level covariates were changed every time a person changed address. In a sensitivity analysis, we calculated adjusted HRs considering only primary diagnoses of tinnitus and excluding all persons with a previous diagnosis for outer and middle ear diseases.
Using the fully adjusted model and a total of nine exposure categories, we explored the joint effect of 10-y mean road traffic Ldenmax (<55, 55–60, and ≥60dB) and Ldenmin (<40, 40–50, and ≥50 dB) in relation to tinnitus. The lowest exposure category was used as reference. Additionally, we investigated exposure–response associations between road traffic and railway noise and tinnitus, using 3 dB categories of 10-y mean Ldenmax and Ldenmin. Reference categories were <45 dB for road traffic Ldenmax and <40 dB for road traffic Ldenmin, railway Ldenmax, and railway Ldenmin. Reference categories were chosen based on the distribution of the exposure variables and to maintain consistency with previous studies.10,30
We evaluated possible modification of the association between 10-y mean road traffic Ldenmax and Ldenmin and risk of tinnitus by including interaction terms between the exposure variable and different covariates: sex, education, blue-collar occupation, income, high-quality green space in 150m, previous hearing loss diagnosis recorded in the DNPR, and previous diagnosis of cardiovascular comorbidity (stroke, hypertension, and ischemic heart diseases).
The assumption of linearity of road traffic and railway noise in relation to tinnitus was tested by log likelihood ratio tests comparing models with and without a quadratic term. We observed no deviation from linearity (p=0.08 and 0.58 for road traffic Ldenmax and Ldenmin, respectively). We used Pearson coefficients to inspect correlation between road traffic and railway Ldenmax and Ldenmin.
Analyses were performed in SAS (version 9.4; SAS Institute Inc.). All participants with incomplete address history and/or missing information on covariates were excluded from the analyses.
Results
The study base included 4.1 million Danish residents. Of those, we excluded 12,476 individuals with prevalent tinnitus, 496,241 with incomplete address history, and 71,387 with incomplete information on covariates. The study population consisted of 3,520,926 individuals with a mean follow-up of 14.1 y and 40,692 incident cases of tinnitus.
Baseline sociodemographic characteristics for people exposed to road traffic Ldenmax above and below 55 dB are presented in Table 1 and Table S1. The histograms for road traffic and railway Ldenmax and Ldenmin at baseline are shown in Figure S1 and Figure S2. Road traffic Ldenmax ranged from 35.0 to 90.1 dB, with median, mean, and interquartile ranges (IQR) of 57.6, 56.6, and 10.8 dB, respectively. For road traffic Ldenmin, the values ranged from 35.0 to 81.7 dB, with median of 45.2 dB, mean of 45.5 dB, and IQR of 8.3 dB. Railway Ldenmax among exposed persons ranged from 35.0 to 86.0 dB, with median of 54.5 dB, mean of 54.1 dB, and IQR of 11.8 dB; railway Ldenmin among exposed ranged from 35.0 to 81.5 dB, with median, mean, and IQR of 41.1, 44.8, and 8.5 dB, respectively. Road traffic and railway noise were correlated with each other, with Pearson correlation coefficients between road traffic Ldenmax and road traffic Ldenmin and railway Ldenmax and railway Ldenmin of 0.49, 0.28, and 0.22, respectively (Table S2).
Table 1 Baseline characteristics of the study population (Denmark, 2000–2017) according to road traffic noise exposure at the most exposed façade.
Baseline characteristics Entire population (N=3,520,926) >55dB road traffic noise (n=2,164,287) ≤55dB road traffic noise (n=1,356,639)
Sex [% (men)] 49.1 49.5 48.5
Age (mean±standard deviation) 46.5±14.7 43.8±14.7 50.7±13.7
Country of origin (%)
Denmark 99.0 98.9 99.2
Other Western country 0.4 0.4 0.4
Non-Western country 0.6 0.7 0.4
Civil status (%)
Married or cohabiting 73.2 69.7 78.9
Widow(er) 4.6 4.1 5.5
Divorced 5.9 6.1 5.4
Single 16.3 20.1 10.2
Individual income (%)a
Q1 20.3 20.8 19.4
Q2 21.0 21.9 19.7
Q3 21.0 21.9 19.6
Q4 19.8 19.5 20.3
Q5 17.9 15.9 21.0
Occupational status (%)
Blue-collar 40.0 40.9 38.5
Low-level white-collar 17.7 18.4 16.6
High-level white-collar 12.4 12.8 11.7
Unemployed 6.0 6.9 4.7
Retired 23.9 21.0 28.5
Highest attained education (%)
Mandatory education 32.8 30.9 36.0
Secondary or vocational education 47.4 47.7 46.8
Medium or long education 19.8 21.4 17.3
High-quality green space
≥15% in 150-m radius 19.6 18.5 21.3
≥20% in 1,000-m radius 24.3 22.6 26.9
Area-level factors (mean±standard deviation)b
% of population with low income (1st quartile) 4.7±2.3 5.0±2.5 4.2±1.9
% unemployed in population 1.6±0.6 1.6±0.6 1.6±0.6
% of population in manual labor 14.7±4.0 13.9±4.1 15.9±3.5
% of population with only basic education 12.1±3.8 11.6±3.9 13.0±3.5
% population with criminal record 0.5±0.3 0.5±0.4 0.5±0.3
% single-parent families 5.2±1.8 5.2±1.8 5.1±1.8
Note: Data were complete for all variables. The corresponding number of persons for each category is shown on Table S1. dB, decibel.
a Individual income quintiles were standardized by calendar year and sex.
b Based on the 2,160 parishes available in Denmark.
Road traffic noise was positively associated with tinnitus for all exposure windows, with substantially higher HRs for Ldenmin in comparison with Ldenmax. With the fully adjusted model, a 10-dB increase in 10-y mean Ldenmin and Ldenmax was associated with a 6% (HR=1.06; 95% CI: 1.04, 1.08) and 2% (HR=1.02; 95% CI: 1.01, 1.03) higher risk of tinnitus, respectively (Table 2). When only primary tinnitus diagnoses were considered, HRs per 10 dB were 1.07 and 1.06 for road Ldenmin and Ldenmax, respectively (Table S3). The HRs were nearly the same when individuals with a previous diagnosis for outer and middle ear diseases were excluded (Table S3). No association was found between railway noise and tinnitus (Table 2; Table S4; Figure S3).
Table 2 Associations between 1-, 5-, and 10-y averaged residential exposure to road traffic and railway noise (linear, per 10 dB) at the most (Ldenmax) and least (Ldenmin) exposed façade and risk of incident tinnitus (40,692 cases, including both primary and secondary tinnitus diagnoses). Results were derived from cox proportional hazards models.
Noise exposure per 10 dB Basic modela HR (95% CI) Fully adjusted modelb HR (95% CI)
Road traffic, Ldenmax
1-y exposure 1.02 (1.01, 1.03) 1.01 (1.00, 1.02)
5-y exposure 1.03 (1.01, 1.04) 1.02 (1.00, 1.03)
10-y exposure 1.03 (1.02, 1.04) 1.02 (1.01, 1.03)
Road traffic, Ldenmin
1-y exposure 1.05 (1.04, 1.07) 1.04 (1.03, 1.06)
5-y exposure 1.06 (1.04, 1.07) 1.05 (1.03, 1.07)
10-y exposure 1.06 (1.05, 1.08) 1.06 (1.04, 1.08)
Railway, Ldenmax
1-y exposure 1.03 (1.00, 1.06) 1.01 (0.98, 1.04)
5-y exposure 1.03 (1.00, 1.06) 1.00 (0.98, 1.03)
10-y exposure 1.02 (0.99, 1.04) 0.99 (0.97, 1.02)
Railway, Ldenmin
1-y exposure 1.08 (1.04, 1.12) 1.02 (0.98, 1.06)
5-y exposure 1.06 (1.02, 1.10) 1.00 (0.96, 1.04)
10-y exposure 1.05 (1.02, 1.09) 0.99 (0.96, 1.03)
Note: CI, confidence interval; dB, decibel; HR, hazard ratio.
a Adjusted for age (underlying time scale, sex, and calendar year).
b Adjustment for age (underlying time scale), sex, calendar year, civil status, income, country of origin, occupational status, education, proportion of high-quality green areas within 150 and 1,000m buffers, and a number of area-level socioeconomic variables: percentage of population with low income, with only basic education, who are unemployed, with manual labor, who are single-parent and with a criminal record, as well as mutual road traffic and railway noise adjustment. All covariates, apart from sex and region of origin, were included in the model as time-varying variables.
We found that the association between road Ldenmin and tinnitus followed a monotonic exposure–response relationship across the entire exposure range, whereas for Ldenmax an increase in risk was observed only in the low-exposure range, followed by a leveling off in risk at higher exposures (Figure 1; Table S4). Accordingly, when investigating combined exposure to road Ldenmax and Ldenmin, we observed higher HRs as Ldenmin exposure increased, whereas no clear tendencies were found across higher Ldenmax categories (Table 3).
Figure 1. Associations between 10-y mean exposure to road traffic noise at the most (A) and least (B) exposed façades and risk of tinnitus using the fully adjusted model. The vertical bars show hazard ratios with 95% confidence interval at the median of the exposure categories compared with the reference category. Reference category was <45 dB for Ldenmax and <40 dB for Ldenmin. Risk estimates and number of cases for each exposure category are shown in Table S3.
Figures 1A and 1B are error bar graphs, plotting hazard ratio, ranging from 0.95 to 1.30 in increments of 0.05 (y-axis) across road traffic noise at the most exposed façade (10 year, decibel), ranging from 40 to 75 in increments of 5 and 35 to 65 in increments of 5 (x-axis).
Table 3 Associations between categories combining residential exposure to road traffic at the most and least exposed façade and risk of incident tinnitus (40,692 cases, including both primary and secondary tinnitus diagnoses). Results were derived from cox proportional hazards models.
Road traffic noise, Ldenmax
Road traffic noise, Ldenmin <55 dB 55–60 dB ≥60 dB
<40 dB n cases=6,213 n cases=1,477 n cases=1,461
Ref 0.99 (0.94, 1.05) 0.97 (0.92, 1.03)
40–50 dB n cases=11,205 n cases=4,797 n cases=6,400
1.03 (1.00, 1.07) 1.02 (0.98, 1.06) 1.05 (1.01, 1.09)
≥50 dB n cases=1,425 n cases=3,184 n cases=4,530
1.04 (0.98, 1.10) 1.07 (1.03, 1.12) 1.08 (1.03, 1.12)
Note: Results are given in hazard ratio (95% confidence interval) and were based on the fully adjusted model, i.e., adjusted for age (underlying time scale), railway noise, sex, calendar year, civil status, income, country of origin, occupational status, education, proportion of high-quality green areas within 150 and 1,000m buffers, and a number of area-level socioeconomic variables: percentage of population with low income, with only basic education, who are unemployed, with manual labor, who are single-parent and with a criminal record. All covariates, apart from sex and region of origin, were included in the model as time-varying variables. dB, decibel; Ref, reference.
As shown in Figure 2, the positive associations between road traffic noise and tinnitus were present only among persons with no previous diagnosis of hearing loss. In contrast, we found negative associations among people with hearing loss. The association was only observed for women, and we found stronger associations among people with higher vs. lower education and income, among people who had never worked in a blue-collar occupation vs. those who had, and among people without vs. with cardiovascular comorbidity. We found no modification of the noise–tinnitus association in relation to surrounding green space (Figure 2 and Table S5).
Figure 2. Effect modification analysis of associations between 10-y mean road traffic noise (continuous, per 10 dB, using the fully adjusted model) at the most and least exposed façade and risk of tinnitus by: hearing loss diagnosis, sex, education, income, green space (150m), occupation, and comorbidity. Risk estimates and number of cases in each modifier subgroup are shown in Table S5.
Figure 2 is a set of two forest plots titled Road traffic noise and tinnitus, plotting effect modifier with number of cases (bottom to top), ranging as Comorbidity: 8240 cases under yes and 32452 cases under no; Occupation: 22766 cases under Blue collar and 15179 cases under never blue collar; Green space (150 meters): 32313 cases under less than 15 percent and 8379 cases under greater than or equal to 15 percent; Income: 9785 cases of low (Quarter 1), 23938 cases of medium (Quarter 2 to Quarter 4), and 6969 cases High (Quarter 5); Education: 13176 cases of Low, 19327 cases of Medium, and 8189 cases of High; Sex: 23764 cases of Men and 16928 cases of Women; and Hearing loss: 32913 cases of yes and 7779 cases of no (y-axis) across hazard ratios (95 percent confidence intervals), ranging from 0.95 to 1.15 in increments of 0.05 and 0.95 to 1.30 in increments of 0.05 (x-axis) for most exposed façade (level begin subscript day, evening, night end subscript maximum) and least exposed façade (level begin subscript day, evening, night end subscript minimum).
Discussion
In this large cohort study, we found residential exposure to road traffic noise to be associated with increased risk of tinnitus, especially noise at the least exposed façade (Ldenmin). Highest risk estimates were found for women and for people without a hearing loss, people with high education and income, and people who had never been in a blue-collar job. Associations between road Ldenmin and tinnitus followed a monotonic exposure–response relationship throughout the exposure range. Railway noise was not associated with increased risk of tinnitus.
We are not aware of previous studies investigating associations between residential exposure to transportation noise and tinnitus. Several epidemiological studies, however, have looked at occupational and leisure noise in relation to tinnitus and consistently shown increased risk of tinnitus following repeated noise exposure.31,32 Noise levels of 85 dB over an 8-h period is typically considered the threshold of which noise can damage the auditory system, likely causing permanent hearing loss and tinnitus.31 Residential exposure to road traffic and railway noise is not likely to reach this level for such a long period of time (hours) and therefore not expected to cause permanent changes in the cochlea. However, transportation noise is a known environmental stressor.3 A recent scoping review concluded that stress increases the distress caused by and the loudness of tinnitus,33 and studies have indicated that stress may play a major role in the development, maintenance, and worsening of tinnitus, especially among people who have a strong negative emotional reaction toward the condition.14,33 In support, we found stronger noise–tinnitus associations when considering only primary diagnoses of tinnitus, which likely represent the more severe cases because these patients have been referred for additional treatment or counseling. One of the most discussed mechanistic models posits that high levels of arousal or stress sustain negative thoughts in relation to tinnitus, therefore reducing the ability of the individual to habituate to the symptoms.34 Also, increased levels of stress hormones may affect the limbic, reticular, and auditory systems, possibly causing or worsening the disorder.35 These are therefore potential underlying mechanisms behind the observed noise–tinnitus associations.
Most studies on transportation noise and health are based on noise estimations at the most exposed façade. However, assessing noise at the least exposed façade may be of significant relevance and a proxy for nighttime exposure, because buildings usually have a quiet side where people in noisy conditions would likely place a bedroom.36 In the present study, we observed higher risk estimates when noise was modeled for the least exposed façade in comparison with the most exposed façade, and we found a clear exposure–response relationship for Ldenmin. Having difficulties in initiating and maintaining sleep is one of the most frequent reports by tinnitus patients, often originating as a reaction to the annoyance experienced by tinnitus patients.32 However, studies have also shown sleeping difficulties to precede tinnitus distress, therefore aggravating tinnitus symptoms.37 The bidirectional relationship between tinnitus and sleeping problems may be explained by a common neurobiological mechanism related to hyperarousal of the sympathetic nervous system, which can be reinforced by a vicious cycle, where sleep disturbance would worsen tinnitus symptoms and more severe symptoms would contribute to impaired sleep.38 In agreement, tinnitus has been found to be louder and more distressing during the night and in the early morning.39 It is therefore possible that noise-induced stress arousals in the middle of the night, together with sleep interruptions, may increase tinnitus patients’ awareness of tinnitus and subsequent level of distress when they try to resume sleep. These people may consequently be more prone to seek medical help, enabling us to identify them in the Danish health registers.
Most tinnitus sufferers are also hearing impaired.12,40,41 In our study only 19% of the identified cases were not previously diagnosed with hearing loss. We observed a positive association between road traffic noise and tinnitus only among people without a hearing loss diagnosis. We found higher risk estimates when considering only primary diagnoses of tinnitus, which likely correspond to people without hearing loss, who would approach the ear, nose, throat (ENT) doctor solely due to their tinnitus symptoms when the symptoms become bothersome to them. In contrast, tinnitus for many hearing loss patients will likely be registered as a secondary diagnosis (in the hospital registry) when they attend a hearing clinic for investigation of their original hearing problem. When hearing loss is present, the cause of tinnitus and hearing loss is usually the same, unless stated otherwise.42,43 In addition, König et al. observed a clear relation between tinnitus intensity and the degree of hearing loss.44 We therefore speculate that the severity of tinnitus among individuals with hearing impairment would probably be much more related to their degree of hearing loss and therefore not be as affected by noise-induced stressful events in comparison with the individuals with other causal onsets of tinnitus. Furthermore, individuals with hearing impairment hear sounds with a reduced acuity and may be likely less disturbed by transportation noise.
Another interesting finding of our study was a negative association between road traffic noise and tinnitus among people who were previously diagnosed with hearing loss. A potential explanation is that external low-level noise may mask tinnitus sounds.41,45,46 This masking is possibly the case for individuals with mild to moderate hearing impairment who may still be able to hear the noise but with not such high intensity to cause stress reactions and/or wake them up, especially at night when no hearing aid is used.
We found the association between road traffic noise and tinnitus to be present only among women and to be higher among people with high income and education and among people who have never been in a blue-collar job, generally indicating socioeconomic status as an important effect modifier of tinnitus–noise associations. Women and people of higher socioeconomic position may be more likely to seek a doctor with their medical problems, and they may be more persistent in demanding examination and treatment.47,48 These characteristics would make them more likely to be captured as cases in the hospital register. Furthermore, these are groups that were likely less exposed to occupational noise during their life courses.49,50 These findings may also be explained by a much lower proportion of hearing loss diagnoses in these groups and the fact that we observed associations only among people without a hearing loss diagnosis, therefore drawing the risk estimates toward the null. In agreement, the effect modification analysis showed higher HRs among individuals without cardiovascular comorbidity, which generally consists of younger individuals who also are less likely to have a hearing loss. Additionally, previous research revealed higher degrees of tinnitus distress, tension, and perceived stress among female patients,51 which may also explain higher risk estimates among women in comparison with men.
We did not find railway noise to be associated with tinnitus. Railway noise is usually perceived as less annoying than road traffic noise,4,52 which could explain why this noise source is not sufficient to aggravate tinnitus symptoms. Besides, the Danish railway system consists mostly of passenger trains, which usually do not operate during the night, thus not causing sleep disturbances to the same extent as road traffic sources.
The use of high-quality Danish registers enabled us to identify many tinnitus cases and conduct a nationwide prospective study with a long follow-up time, which is a major strength of the study. Besides a very large study population, we had access to detailed and time-varying individual- and area-level sociodemographic and socioeconomic information, as well as precise address location and history for each study participant. Our analysis relied on validated exposure models to estimate noise from two transportation sources at both the most and least exposed façade, the latter better reflecting nighttime exposure.22,23
Diagnosing tinnitus imposes many challenges, because there is no objective test to confirm the occurrence of the disorder. Moreover, tinnitus intensity and severity are highly heterogeneous, with some people being much more affected by the condition than others.12 Because our study is based on register-based diagnoses, we believe our analyses are rather limited to more severe and bothersome tinnitus cases, because these patients would be more likely to seek medical help for their tinnitus (i.e., primary diagnosis) and/or complain about the condition even if tinnitus was not the primary cause of the visit (i.e., secondary diagnosis). Therefore, we expect our population to have a large number of people with an undiagnosed tinnitus, especially those who are not significantly bothered by the condition and possibly those of lower socioeconomic position. This outcome misclassification is believed to be nondifferential with regard to noise exposure and would thus in most situations drive the risk estimates toward the null.
Our study presents other limitations. The exposure assessment was limited to home addresses and did not consider individual preventive measures regarding, e.g., window quality and bedroom disposition, therefore hindering the estimation of indoor noise exposure, and other noise sources (e.g., from neighbors, community life, and construction sites). We also lacked detailed information on occupational noise exposure, such as type and exposure duration, which is a well-known risk factor for tinnitus and potentially a confounder on the exposure–disease pathway. Even though our models were adjusted for occupational status, we were not able to differentiate the available classes (i.e., blue-collar, low- and high-level white-collar, unemployed, and retired) into different job functions, or to capture the full picture of current and past exposure history. Similarly, although we used detailed socioeconomic information, measured by individual (e.g., disposable income and highest attained education) and address-level covariates, we cannot rule out potential residual confounding. Finally, our findings were limited to the Danish population, which represents specific characteristics related to ethnicity, genetics, and the presence and distribution of various noise sources. Therefore, our findings should be generalized with caution, and more studies are needed to test the consistency of our results in other study settings, including different population characteristics and geographical locations.
To the best extent of our knowledge, this is the first study investigating the association between residential exposure to transportation noise and risk of tinnitus. Our study, which covered an entire country, showed consistent associations between road traffic noise and tinnitus, especially when noise was measured at the least exposed façade. No association was found for railway noise. The hypothesized underlying mechanisms behind the observed associations include noise-induced stress reactions and disturbance of sleep, which would increase people’s awareness of tinnitus, likely exacerbating the condition. Further mechanistic studies, preferably including information on indoor exposure, are needed to confirm the noise–tinnitus pathways hereby proposed. Although causality remains uncertain, these findings suggest that transportation noise may also affect the auditory system. Last, this study adds to the evidence of road traffic noise as a harmful pollutant with a growing health burden.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by the William Demant Foundation (grant number: 18-0964).
==== Refs
References
1. European Environment Agency. 2020. Environmental Noise in Europe – 2020. https://www.eea.europa.eu/publications/environmental-noise-in-europe/ [accessed 2 February 2022].
2. Babisch W. 2002. The noise/stress concept, risk assessment and research needs. Noise Health 4 (16 ):1–11, PMID: .12537836
3. Basner M, Babisch W, Davis A, Brink M, Clark C, Janssen S, et al. 2014. Auditory and non-auditory effects of noise on health. Lancet 383 (9925 ):1325–1332, PMID: , 10.1016/S0140-6736(13)61613-X.24183105
4. World Health Organization. 2018. Environmental Noise Guidelines for the European Region. Copenhagen, Denmark: WHO Regional Office for Europe. https://www.euro.who.int/__data/assets/pdf_file/0008/383921/noise-guidelines-eng.pdf [accessed 2 February 2022].
5. Roswall N, Pyko A, Ögren M, Oudin A, Rosengren A, Lager A, et al. 2021. Long-term exposure to transportation noise and risk of incident stroke: a pooled study of nine Scandinavian cohorts. Environ Health Perspect 129 (10 ):1–10, PMID: , 10.1289/EHP8949.34605674
6. Vienneau D, Eze IC, Probst-Hensch N, Roosli M. 2019. Association between transportation noise and cardio-metabolic diseases: an update of the WHO meta-analysis. In: Proceedings of the 23rd International Congress on Acoustics Integrating 4th EAA Euroregion 2019. 9–13 September 2019. Aachen, Germany. 10.18154/RWTH-CONV-239440. http://publications.rwth-aachen.de/record/769851/files/769851.pdf [accessed 3 February 2022].
7. Münzel T, Sørensen M, Daiber A. 2021. Transportation noise pollution and cardiovascular disease. Nat Rev Cardiol 18 (9 ):619–636, PMID: , 10.1038/s41569-021-00532-5.33790462
8. Hegewald J, Schubert M, Lochmann M, Seidler A. 2021. The burden of disease due to road traffic noise in Hesse, Germany. Int J Environ Res Public Health 18 (17 ):9337, PMID: , 10.3390/ijerph18179337.34501923
9. Hao G, Zuo L, Weng X, Fei Q, Zhang Z, Chen L, et al. 2022. Associations of road traffic noise with cardiovascular diseases and mortality: longitudinal results from UK Biobank and meta-analysis. Environ Res 212 (pt A ):113129, PMID: , 10.1016/j.envres.2022.113129.35358546
10. Thacher JD, Poulsen AH, Hvidtfeldt UA, Raaschou-Nielsen O, Brandt J, Geels C, et al. 2021. Long-term exposure to transportation noise and risk for type 2 diabetes in a nationwide cohort study from Denmark. Environ Health Perspect 129 (12 ):1–11, PMID: , 10.1289/EHP9146.34855467
11. Thacher JD, Hvidtfeldt UA, Poulsen AH, Raaschou-Nielsen O, Ketzel M, Brandt J, et al. 2020. Long-term residential road traffic noise and mortality in a Danish cohort. Environ Res 187 :109633–109638, 10.1016/j.envres.2020.109633.32442789
12. Baguley D, McFerran D, Hall D. 2013. Tinnitus. Lancet 382 (9904 ):1600–1607, PMID: , 10.1016/S0140-6736(13)60142-7.23827090
13. McCormack A, Edmondson-Jones M, Somerset S, Hall D. 2016. A systematic review of the reporting of tinnitus prevalence and severity. Hear Res 337 :70–79, PMID: , 10.1016/j.heares.2016.05.009.27246985
14. Betz LT, Mühlberger A, Langguth B, Schecklmann M. 2017. Stress reactivity in chronic tinnitus. Sci Rep 7 :41521–41529, PMID: , 10.1038/srep41521.28134346
15. Rauschecker JP, Leaver AM, Mühlau M. 2010. Tuning out the noise: limbic-auditory interactions in tinnitus. Neuron 66 (6 ):819–826, PMID: , 10.1016/j.neuron.2010.04.032.20620868
16. Heinecke K, Weise C, Schwarz K, Rief W. 2008. Physiological and psychological stress reactivity in chronic tinnitus. J Behav Med 31 (3 ):179–188, PMID: , 10.1007/s10865-007-9145-0.18193350
17. Mazurek B, Boecking B, Brueggemann P. 2019. Association between stress and tinnitus – new aspects. Otol Neurotol 40 (4 ):e467–e473, PMID: , 10.1097/MAO.0000000000002180.30870382
18. Schmidt M, Pedersen L, Sørensen HT. 2014. The Danish Civil Registration System as a tool in epidemiology. Eur J Epidemiol 29 (8 ):541–549, PMID: , 10.1007/s10654-014-9930-3.24965263
19. Christensen G. 2011. The Building and Housing Register. Scand J Public Health 39 (suppl 7 ):106–108, PMID: , 10.1177/1403494811399168.21775366
20. Pedersen CB. 2011. The Danish Civil Registration System. Scand J Public Health 39 (suppl 7 ):22–25, PMID: , 10.1177/1403494810387965.21775345
21. Lynge E, Sandegaard JL, Rebolj M. 2011. The Danish National Patient Register. Scand J Public Health 39 (suppl 7 ):30–33, PMID: , 10.1177/1403494811401482.21775347
22. Thacher JD, Poulsen AH, Raaschou-Nielsen O, Jensen A, Hillig K, Roswall N, et al. 2020. High-resolution assessment of road traffic noise exposure in Denmark. Environ Res 182 :109051–109059, PMID: , 10.1016/j.envres.2019.109051.31896468
23. Bendtsen H. 1999. The Nordic prediction method for road traffic noise. Sci Total Environ 235 (1–3 ):331–338, 10.1016/S0048-9697(99)00216-8.
24. Jensen SS, Plejdrup MS, Hillig K. 2019. GIS-Based National Road and Traffic Database 1960–2020. Aarhus, Denmark: Aarhus University, DCE – Danish Centre for Environment and Energy. https://dce2.au.dk/pub/TR151.pdf [accessed 8 February 2022].
25. Plovsing B, Kragh J. 2006. Nord2000. Comprehensive Outdoor Sound Propagation Model. Part 1: Propagation in an Atmosphere without Significant Refraction. Lyngby, Denmark: DELTA Danish Electronics, Light & Acoustics. http://www.magasbakony.hu/Val/Nord2000_homogeneous_atmosphere_Part_1.pdf [accessed 8 February 2022].
26. Hvidtfeldt UA, Sørensen M, Geels C, Ketzel M, Khan J, Tjønneland A, et al. 2019. Long-term residential exposure to PM2.5, PM10, black carbon, NO2, and ozone and mortality in a Danish cohort. Environ Int 123 :265–272, PMID: , 10.1016/j.envint.2018.12.010.30551059
27. D’Alessandro D, Buffoli M, Capasso L, Fara GM, Rebecchi A, Capolongo S, et al. 2015. Green areas and public health: improving wellbeing and physical activity in the urban context. Epidemiol Prev 39 (4 suppl 1 ):8–13, PMID: .26499409
28. Nieuwenhuijsen MJ, Khreis H, Triguero-Mas M, Gascon M, Dadvand P. 2017. Fifty shades of green: pathway to healthy urban living. Epidemiology 28 (1 ):63–71, PMID: , 10.1097/EDE.0000000000000549.27525811
29. Levin G, Jepsen MR, Blemmer MK. 2012. BASEMAP: Technical Documentation of a Model for Elaboration of a Land-Use and Land-Cover Map for Denmark. Aarhus, Denmark: Aarhus University, DCE – Danish Centre for Environment and Energy. https://www2.dmu.dk/Pub/TR11.pdf [accessed 10 February 2022].
30. Sørensen M, Poulsen AH, Hvidtfeldt UA, Münzel T, Thacher JD, Ketzel M, et al. 2021. Transportation noise and risk of stroke: a nationwide prospective cohort study covering Denmark. Int J Epidemiol 50 (4 ):1147–1156, PMID: , 10.1093/ije/dyab024.33755127
31. Themann CL, Masterson EA. 2019. Occupational noise exposure: a review of its effects, epidemiology, and impact with recommendations for reducing its burden. J Acoust Soc Am 146 (5 ):3879–3905, PMID: , 10.1121/1.5134465.31795665
32. Bhatt JM, Lin HW, Bhattacharyya N. 2016. Tinnitus epidemiology: prevalence, severity, exposures, and treatment in the United States. JAMA Otolaryngol Head Neck Surg 142 (10 ):959–965, PMID: , 10.1001/jamaoto.2016.1700.27441392
33. Elarbed A, Fackrell K, Baguley DM, Hoare DJ. 2021. Tinnitus and stress in adults: a scoping review. Int J Audiol 60 (3 ):171–182, PMID: , 10.1080/14992027.2020.1827306.33000672
34. McKenna L, Handscomb L, Hoare DJ, Hall DA. 2014. A scientific cognitive-behavioral model of tinnitus: novel conceptualizations of tinnitus distress. Front Neurol 5 :196–15, PMID: , 10.3389/fneur.2014.00196.25339938
35. Al-Mana D, Ceranic B, Djahanbakhch O, Luxon LM. 2008. Hormones and the auditory system: a review of physiology and pathophysiology. Neuroscience 153 (4 ):881–900, PMID: , 10.1016/j.neuroscience.2008.02.077.18440718
36. Bodin T, Björk J, Ardö J, Albin M. 2015. Annoyance, sleep and concentration problems due to combined traffic noise and the benefit of quiet side. Int J Environ Res Public Health 12 (2 ):1612–1628, PMID: , 10.3390/ijerph120201612.25642690
37. Langenbach M, Olderog M, Michel O, Albus C, Köhle K. 2005. Psychosocial and personality predictors of tinnitus-related distress. Gen Hosp Psychiatry 27 (1 ):73–77, PMID: , 10.1016/j.genhosppsych.2004.08.008.15694221
38. Wallhäusser-Franke E, Schredl M, Delb W. 2013. Tinnitus and insomnia: is hyperarousal the common denominator? Sleep Med Rev 17 (1 ):65–74, PMID: , 10.1016/j.smrv.2012.04.003.22750224
39. Probst T, Pryss RC, Langguth B, Rauschecker JP, Schobel J, Reichert M, et al. 2017. Does tinnitus depend on time-of-day? An ecological momentary assessment study with the “TrackYourTinnitus” application. Front Aging Neurosci 9 :253–259, PMID: , 10.3389/fnagi.2017.00253.28824415
40. Sindhusake D, Golding M, Newall P, Rubin G, Jakobsen K, Mitchell P. 2003. Risk Factors for tinnitus in a population of older adults: the blue mountains hearing study. Ear Hear 24 (6 ):501–507, PMID: , 10.1097/01.AUD.0000100204.08771.3D.14663349
41. Eggermont JJ, Roberts LE. 2004. The neuroscience of tinnitus. Trends Neurosci 27 (11 ):676–682, PMID: , 10.1016/j.tins.2004.08.010.15474168
42. Coles RRA. 1995. Classification of causes, mechanisms of patient disturbance, and associated counseling. In: Mechanisms of Tinnitus. Vernon JA, Moller AR, eds. Needham Heights, MA: Allyn & Bacon, 11–19.
43. Shapiro SB, Noij KS, Naples JG, Samy RN. 2021. Hearing loss and tinnitus. Med Clin North Am 105 (5 ):799–811, PMID: , 10.1016/j.mcna.2021.05.003.34391534
44. König O, Schaette R, Kempter R, Gross M. 2006. Course of hearing loss and occurrence of tinnitus. Hear Res 221 (1–2 ):59–64, PMID: , 10.1016/j.heares.2006.07.007.16962270
45. Vernon JA, Meikle MB. 2003. Tinnitus: clinical measurement. Otolaryngol Clin North Am 36 (2 ):293–305, vi, PMID: , 10.1016/s0030-6665(02)00162-7.12856298
46. Hobson J, Chisholm EJ, el Refaie A. 2012. Sound therapy (masking) in the management of tinnitus in adults. Cochrane Database Syst Rev 11 (11 ):CD006371–22, PMID: , 10.1002/14651858.CD006371.pub3.23152235
47. Bertakis KD, Azari R, Helms LJ, Callahan EJ, Robbins JA. 2000. Gender differences in the utilization of health care services. J Fam Pract 49 (2 ):147–152, PMID: .10718692
48. Wiltshire JC, Roberts V, Brown R, Sarto GE. 2009. The effects of socioeconomic status on participation in care among middle-aged and older adults. J Aging Health 21 (2 ):314–335, PMID: , 10.1177/0898264308329000.19091692
49. Michaud DS, Marro L, McNamee JP. 2021. Self-reported occupational noise exposure and cardiovascular disease in Canada: results from the Canadian Health Measures Survey. J Acoust Soc Am 150 (2 ):990–1000, PMID: , 10.1121/10.0005588.34470300
50. Suadicani P, Hein HO, Gyntelberg F. 2012. Occupational noise exposure, social class, and risk of ischemic heart disease and all-cause mortality – a 16-year follow-up in the Copenhagen Male Study. Scand J Work Environ Health 38 (1 ):19–26, PMID: , 10.5271/sjweh.3200.21986898
51. Niemann U, Boecking B, Brueggemann P, Mazurek B, Spiliopoulou M. 2020. Gender-specific differences in patients with chronic tinnitus—baseline characteristics and treatment effects. Front Neurosci 14 :1–11, PMID: , 10.3389/fnins.2020.00487.32038151
52. Miedema HME, Vos H. 1998. Exposure-response relationships for transportation noise. J Acoust Soc Am 104 (6 ):3432–3445, PMID: , 10.1121/1.423927.9857505
|
PMC009xxxxxx/PMC9894152.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36729393
EHP12555
10.1289/EHP12555
Invited Perspective
Invited Perspective: Metals and Menarche
https://orcid.org/0000-0002-4313-3298
Pollack Anna Z. 1
Marroquin Joanna M. 1
1 Department of Global and Community Health, College of Public Health, George Mason University, Fairfax, Virginia, USA
Address correspondence to Anna Z. Pollack, 4400 University Dr., MS5B7, Fairfax, VA 22030 USA. Email: [email protected]
2 2 2023
2 2023
131 2 02130107 12 2022
03 1 2023
04 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11121
==== Body
pmcMenarche marks the start of the reproductive life span and has implications for chronic disease and mortality. Earlier age at menarche is a recognized risk factor for breast cancer.1,2 Mortality and cardiovascular disease have also been associated with age at menarche.3 Age at menarche has decreased globally over the last several decades4–9; these temporal changes underscore the likely influence of environmental factors, such as endocrine disrupting chemicals.10 Metals may also act as endocrine disruptors.11 In this issue of Environmental Health Perspectives, Malin Igra et al. have advanced the understanding of how three nonessential metals—cadmium, lead, and arsenic—potentially affect menarche in a longitudinal cohort with the measurement of those metals during fetal development and at 5 and 10 years of age.12
The main study finding of later age at menarche being associated with maternal blood arsenic during pregnancy is notable.12 Evidence of arsenic’s influence on menarche and pubertal markers is lacking in prospective epidemiologic studies, apart from an earlier finding of delayed menarche with increased in utero exposure to arsenic reported from this mother–child cohort in rural Bangladesh,13 underscoring the need for further prospective research on this toxic metal with widespread exposure. A body of ecologic and experimental evidence suggests that arsenic exposure may delay menarche.14–19 This evidence supports the findings by Malin Igra et al. of prenatal arsenic exposure with later age at menarche.12 Of note, urinary arsenic in childhood was not associated with age at menarche. This suggests that arsenic exposure in utero, a critical period of development, but not later life levels, can influence timing of menarche.
Cadmium, too, was associated with later age at menarche but only for childhood exposures. This is in line with a few studies on this topic20,21 although the evidence is mixed.22 The variety of methodological approaches makes comparisons of these studies challenging. Longitudinal studies with multiple individual exposure biomarkers of metals are needed to confirm these findings by Malin Igra et al.12
Other studies have reported that blood iron is inversely associated with cadmium exposure.23–25 This may help explain why cadmium exposure later in childhood, but not in utero, was associated with later menarche given that in utero cadmium exposure was attenuated by iron supplementation.23–25 There is limited evidence of lead’s effect on menarche. Although Malin Igra et al. reported no association between lead and menarche, the point estimates consistently indicated earlier menarche during childhood.12 Evidence for an association is mixed.26–28 Again, the lack of consistency of these findings and the various approaches to lead measurement—in bone, blood, and urine—underscore the need for further studies on this topic with consistent exposure measurement approaches.
The study by Malin Igra et al.12 advances the field in several important ways. The longitudinal design had multiple exposure measurements (one prenatal and twice in childhood), a key strength. This captures exposure at critical stages of development in the life course. Of note, the childhood biomarkers of different urinary metals reflect exposure over different time periods, depending on the metal. Urinary cadmium reflects long-term exposure, on the order of years.29 In contrast, urinary lead and arsenic reflect short-term exposure, on the order of days or weeks.30
Although this study is an important contribution, some important gaps should be addressed in future work in metals and pubertal development. Because of the time span of follow-up in this cohort, participants were <14 years of age. However, clinically delayed menarche is considered the absence of menses after 15 years of age31; therefore, additional follow-up is warranted. This study also had some variability around the preparation of maternal biospecimens. The method for maternal erythrocyte preparation differed for the 34.8% of the samples that had acid digestion instead of alkali dilution prior to inductively coupled plasma mass spectrometry (ICP-MS).12 Values for samples prepared using acid digestion were consistently lower than for samples prepared via alkali dilution, although samples prepared by these two methods were highly correlated in another cohort.32 In addition, correlated exposures that differ from the true exposure of interest introduce a source of measurement error from the true exposure.33,34 The correction approach applied here is helpful. However, an additional approach that excludes the one-third of samples prepared via acid digestion method would have better evaluated how this correction may have influenced the findings. Finally, additional metals exposure following menarche may come from the use of menstrual products, an important and understudied exposure route for menstruators.35,36
The analysis by Malin Igra et al. evaluated how specific time points in development and metal concentrations differed by age. There was moderate correlation between urine concentration of cadmium at 5 and 10 years of age (rho=0.40) but weaker correlation with maternal erythrocyte cadmium (rho=0.10). Urine concentrations of lead at 5 years of age was approximately double the concentrations at 10 years of age.12 Future studies should evaluate the trajectory of exposure over time using longitudinal methods that appropriately address correlated measures over time. One recent study implemented a longitudinal approach incorporating multiple exposure measures to metals in relation to menarche and pubertal measures.37
Measurement of lead using blood or bone also should be a priority in children because these tissues better reflect childhood lead exposure compared with urine.38 Consideration of linear associations or splines for nonlinear approaches should be considered in future studies. In addition, exposures do not occur in isolation, so future studies should evaluate exposure mixtures.39–41 Considering the role that societal factors contribute along with environmental exposures is key.42 In particular, social norms can influence exposure to beauty products, which contain metals and other chemicals.43–45 Leveraging pregnancy cohort studies with follow-up of children provides an efficient approach to address this question, and incorporating measures of pubertal development and timing of menarche should be a priority.
Menarche is an important event in the reproductive life span that has implications for chronic disease development. There is a growing body of studies that show nonessential metals affect menarche and pubertal development. The study by Malin Igra et al.12 highlights the importance of timing of exposure assessment and underscores that there are critical periods of exposure that can affect later life health and development.
==== Refs
References
1. Fuhrman BJ, Moore SC, Byrne C, Makhoul I, Kitahara CM, Berrington de González A, et al. 2021. Association of the age at menarche with site-specific cancer risks in pooled data from nine cohorts. Cancer Res 81 (8 ):2246–2255, PMID: , 10.1158/0008-5472.CAN-19-3093.33820799
2. Goldberg M, D’Aloisio AA, O’Brien KM, Zhao S, Sandler DP. 2020. Pubertal timing and breast cancer risk in the Sister Study cohort. Breast Cancer Res 22 (1 ):112, PMID: , 10.1186/s13058-020-01326-2.33109223
3. Lakshman R, Forouhi NG, Sharp SJ, Luben R, Bingham SA, Khaw KT, et al. 2009. Early age at menarche associated with cardiovascular disease and mortality. J Clin Endocrinol Metab 94 (12 ):4953–4960, PMID: , 10.1210/jc.2009-1789.19880785
4. Asrullah M, L’Hoir M, Feskens EJM, Melse-Boonstra A. 2022. Trend in age at menarche and its association with body weight, body mass index and non-communicable disease prevalence in Indonesia: evidence from the Indonesian Family Life Survey (IFLS). BMC Public Health 22 (1 ):628, PMID: , 10.1186/s12889-022-12995-3.35361192
5. Euling SY, Selevan SG, Pescovitz OH, Skakkebaek NE. 2008. Role of environmental factors in the timing of puberty. Pediatrics 121 (suppl 3 ):S167–S171, PMID: , 10.1542/peds.2007-1813C.18245510
6. Flash-Luzzatti S, Weil C, Shalev V, Oron T, Chodick G. 2014. Long-term secular trends in the age at menarche in Israel: a systematic literature review and pooled analysis. Horm Res Paediatr 81 (4 ):266–271, PMID: , 10.1159/000357444.24504328
7. Kim H, Choe SA, Lee SJ, Sung J. 2021. Causal relationship between the timing of menarche and young adult body mass index with consideration to a trend of consistently decreasing age at menarche. PLoS One 16 (2 ):e0247757, PMID: , 10.1371/journal.pone.0247757.33635908
8. Liu W, Yan X, Li C, Shu Q, Chen M, Cai L, et al. 2021. A secular trend in age at menarche in Yunnan Province, China: a multiethnic population study of 1,275,000 women. BMC Public Health 21 (1 ):1890, PMID: , 10.1186/s12889-021-11951-x.34666747
9. Marván ML, Castillo-López RL, Del-Callejo-Canal DD, Canal-Martínez ME, Núñez-de la Mora A. 2020. Secular trends in age at menarche in 20th century Mexico: differences by ethnicity, area of residency, and socioeconomic status. Am J Hum Biol 32 (6 ):e23404, PMID: , 10.1002/ajhb.23404.32052905
10. Mouritsen A, Aksglaede L, Sørensen K, Mogensen SS, Leffers H, Main KM, et al. 2010. Hypothesis: exposure to endocrine-disrupting chemicals may interfere with timing of puberty. Int J Androl 33 (2 ):346–359, PMID: , 10.1111/j.1365-2605.2010.01051.x.20487042
11. Iavicoli I, Fontana L, Bergamaschi A. 2009. The effects of metals as endocrine disruptors. J Toxicol Environ Health B Crit Rev 12 (3 ):206–223, PMID: , 10.1080/10937400902902062.19466673
12. Malin Igra A, Rahman A, Johansson ALV, Pervin J, Svefors P, El Arifeen S, et al. 2023. Early life environmental exposure to cadmium, lead, and arsenic and age at menarche: a longitudinal mother–child cohort study in Bangladesh. Environ Health Perspect 131 (2 ):027003, 10.1289/EHP11121.36729392
13. Rahman A, Kippler M, Pervin J, Tarafder C, Lucy IJ, Svefors P, et al. 2021. A cohort study of the association between prenatal arsenic exposure and age at menarche in a rural area, Bangladesh. Environ Int 154 :106562, PMID: , 10.1016/j.envint.2021.106562.33866057
14. Kumar S, Mukherjee TK, Guptasarma P. 2016. Arsenic and 17-β-estradiol bind to each other and neutralize each other’s signaling effects. Biochem Biophys Res Commun 477 (4 ):575–580, PMID: , 10.1016/j.bbrc.2016.06.087.27346132
15. Parodi DA, Greenfield M, Evans C, Chichura A, Alpaugh A, Williams J, et al. 2015. Alteration of mammary gland development and gene expression by in utero exposure to arsenic. Reprod Toxicol 54 :66–75, PMID: , 10.1016/j.reprotox.2014.12.011.25543096
16. Reilly MP, Saca JC, Hamilton A, Solano RF, Rivera JR, Whitehouse-Innis W, et al. 2014. Prepubertal exposure to arsenic(III) suppresses circulating insulin-like growth factor-1 (IGF-1) delaying sexual maturation in female rats. Reprod Toxicol 44 :41–49, PMID: , 10.1016/j.reprotox.2013.09.005.24090629
17. Sen J, Chaudhuri ABD. 2007. Effect of arsenic on the onset of menarcheal age. Bull Environ Contam Toxicol 79 (3 ):293–296, PMID: , 10.1007/s00128-007-9206-7.17668141
18. Sengupta M. 2004. Does arsenic consumption influence the age at menarche of woman. Indian Pediatr 41 (9 ):960–961, PMID: .15475645
19. Stoica A, Pentecost E, Martin MB. 2000. Effects of arsenite on estrogen receptor-α expression and activity in MCF-7 breast cancer cells. Endocrinology 141 (10 ):3595–3602, PMID: , 10.1210/endo.141.10.7704.11014213
20. Chen X, Zhu G, Jin T. 2017. Effects of cadmium exposure on age of menarche and menopause. Toxics 6 (1 ):6, PMID: , 10.3390/toxics6010006.29280952
21. Reynolds P, Canchola AJ, Duffy CN, Hurley S, Neuhausen SL, Horn-Ross PL, et al. 2020. Urinary cadmium and timing of menarche and pubertal development in girls. Environ Res 183 :109224, PMID: , 10.1016/j.envres.2020.109224.32085996
22. Choi HS. 2020. Relationships of lead, mercury and cadmium levels with the timing of menarche among Korean girls. Child Health Nurs Res 26 (1 ):98–106, PMID: , 10.4094/chnr.2020.26.1.98.35004455
23. Berglund M, Akesson A, Nermell B, Vahter M. 1994. Intestinal absorption of dietary cadmium in women depends on body iron stores and fiber intake. Environ Health Perspect 102 (12 ):1058–1066, PMID: , 10.1289/ehp.941021058.7713018
24. Ekström EC, Lindström E, Raqib R, El Arifeen S, Basu S, Brismar K, et al. 2016. Effects of prenatal micronutrient and early food supplementation on metabolic status of the offspring at 4.5 years of age. The MINIMat randomized trial in rural Bangladesh. Int J Epidemiol 45 (5 ):1656–1667, PMID: , 10.1093/ije/dyw199.27694568
25. Kippler M, Goessler W, Nermell B, Ekström EC, Lönnerdal B, El Arifeen S, et al. 2009. Factors influencing intestinal cadmium uptake in pregnant Bangladeshi women—a prospective cohort study. Environ Res 109 (7 ):914–921, PMID: , 10.1016/j.envres.2009.07.006.19646688
26. dos Santos NR, Rodrigues JLG, Bandeira MJ, Anjos ALDS, Araújo CFDS, Adan LFF, et al. 2022. Manganese and lead exposure and early puberty onset in children living near a ferromanganese alloy plant. Int J Environ Res Public Health 19 (12 ):7158, PMID: , 10.3390/ijerph19127158.35742410
27. Jansen EC, Zhou L, Song PXK, Sánchez BN, Mercado A, Hu H, et al. 2018. Prenatal lead exposure in relation to age at menarche: results from a longitudinal study in Mexico City. J Dev Orig Health Dis 9 (4 ):467–472, PMID: , 10.1017/S2040174418000223.29706142
28. Liu Y, Téllez-Rojo MM, Sánchez BN, Zhang Z, Afeiche MC, Mercado-García A, et al. 2019. Early lead exposure and pubertal development in a Mexico City population. Environ Int 125 :445–451, PMID: , 10.1016/j.envint.2019.02.021.30763831
29. ATSDR (Agency for Toxic Substances and Disease Registry). 2012. Toxicological Profile for Cadmium. CAS#: 7440-43-9. Atlanta, GA: ATSDR. https://wwwn.cdc.gov/TSP/ToxProfiles/ToxProfiles.aspx?id=48&tid=15 [accessed 22 November 2022].
30. Hu H, Shih R, Rothenberg S, Schwartz BS. 2007. The epidemiology of lead toxicity in adults: measuring dose and consideration of other methodologic issues. Environ Health Perspect 115 (3 ):455–462, PMID: , 10.1289/ehp.9783.17431499
31. Bakhtiani P, Geffner M. 2022. Delayed puberty. Pediatr Rev 43 (8 ):426–435, PMID: , 10.1542/pir.2020-005291.35909138
32. Lu Y, Kippler M, Harari F, Grandér M, Palm B, Nordqvist H, et al. 2015. Alkali dilution of blood samples for high throughput ICP-MS analysis—comparison with acid digestion. Clin Biochem 48 (3 ):140–147, PMID: , 10.1016/j.clinbiochem.2014.12.003.25498303
33. Schisterman EF, Little RJ. 2010. Opening the black box of biomarker measurement error. Epidemiology 21 (4 ):S1–S3, PMID: , 10.1097/EDE.0b013e3181dda514.20539119
34. Wacholder S. 1995. When measurement errors correlate with truth: surprising effects of nondifferential misclassification. Epidemiology 6 (2 ):157–161, PMID: , 10.1097/00001648-199503000-00012.7742402
35. Singh J, Mumford SL, Pollack AZ, Schisterman EF, Weisskopf MG, Navas-Acien A, et al. 2019. Tampon use, environmental chemicals and oxidative stress in the BioCycle study. Environ Health 18 (1 ):11, PMID: , 10.1186/s12940-019-0452-z.30744632
36. Upson K, Shearston JA, Kioumourtzoglou MA. 2022. Menstrual products as a source of environmental chemical exposure: a review from the epidemiologic perspective. Curr Environ Health Rep 9 (1 ):38–52, PMID: , 10.1007/s40572-022-00331-1.35302185
37. Ashrap P, Sánchez BN, Téllez-Rojo MM, Basu N, Tamayo-Ortiz M, Peterson KE, et al. 2019. In utero and peripubertal metals exposure in relation to reproductive hormones and sexual maturation and progression among girls in Mexico City. Environ Res 177 :108630, PMID: , 10.1016/j.envres.2019.108630.31421446
38. Sommar JN, Hedmer M, Lundh T, Nilsson L, Skerfving S, Bergdahl IA. 2014. Investigation of lead concentrations in whole blood, plasma and urine as biomarkers for biological monitoring of lead exposure. J Expo Sci Environ Epidemiol 24 (1 ):51–57, PMID: , 10.1038/jes.2013.4.23443239
39. Gibson EA, Goldsmith J, Kioumourtzoglou MA. 2019. Complex mixtures, complex analyses: an emphasis on interpretable results. Curr Environ Health Rep 6 (2 ):53–61, PMID: , 10.1007/s40572-019-00229-5.31069725
40. Joubert BR, Kioumourtzoglou MA, Chamberlain T, Chen HY, Gennings C, Turyk ME, et al. 2022. Powering Research through Innovative methods for Mixtures in Epidemiology (PRIME) program: novel and expanded statistical methods. Int J Environ Res Public Health 19 (3 ):1378, PMID: , 10.3390/ijerph19031378.35162394
41. Taylor KW, Joubert BR, Braun JM, Dilworth C, Gennings C, Hauser R, et al. 2016. Statistical approaches for assessing health effects of environmental chemical mixtures in epidemiology: lessons from an innovative workshop. Environ Health Perspect 124 (12 ):A227–A229, PMID: , 10.1289/EHP547.27905274
42. Ford CL, Airhihenbuwa CO. 2010. The public health critical race methodology: praxis for antiracism research. Soc Sci Med 71 (8 ):1390–1398, PMID: , 10.1016/j.socscimed.2010.07.030.20822840
43. Borowska S, Brzóska MM. 2015. Metals in cosmetics: implications for human health. J Appl Toxicol 35 (6 ):551–572, PMID: , 10.1002/jat.3129.25809475
44. Gaston SA, James-Todd T, Harmon Q, Taylor KW, Baird D, Jackson CL. 2020. Chemical/straightening and other hair product usage during childhood, adolescence, and adulthood among African-American women: potential implications for health. J Expo Sci Environ Epidemiol 30 (1 ):86–96, PMID: , 10.1038/s41370-019-0186-6.31641276
45. Liu S, Hammond SK, Rojas-Cheatham A. 2013. Concentrations and potential health risks of metals in lip products. Environ Health Perspect 121 (6 ):705–710, PMID: , 10.1289/ehp.1205518.23674482
|
PMC009xxxxxx/PMC9894154.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36729392
EHP11121
10.1289/EHP11121
Research
Early Life Environmental Exposure to Cadmium, Lead, and Arsenic and Age at Menarche: A Longitudinal Mother–Child Cohort Study in Bangladesh
Malin Igra Annachiara 1
Rahman Anisur 2
Johansson Anna L.V. 3
Pervin Jesmin 2
Svefors Pernilla 4
Arifeen Shams El 2
Vahter Marie 1
Persson Lars-Åke 4 5
Kippler Maria 1
1 Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden
2 Maternal and Child Health Division, International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh
3 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden
4 Department of Women’s and Children’s Health, Uppsala University, Uppsala, Sweden
5 London School of Hygiene and Tropical Medicine, London, UK
Address correspondence to Maria Kippler, Institute of Environmental Medicine, Karolinska Institutet, Box 210, SE-171 77, Stockholm, Sweden. Telephone: 46 7030303131. Email: [email protected]
2 2 2023
2 2023
131 2 02700317 2 2022
07 12 2022
03 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Several metals act as endocrine disruptors, but there are few large longitudinal studies about associations with puberty onset.
Objectives:
We evaluated whether early life cadmium, lead, and arsenic exposure was associated with timing of menarche.
Methods:
In a mother–child cohort in rural Bangladesh (n=935), the exposure was assessed by concentrations in maternal erythrocytes in early pregnancy and in girls’ urine at 5 and 10 years of age using inductively coupled plasma mass spectrometry. The girls were interviewed twice, at average ages 13.3 [standard deviation (SD)=0.43] and 13.8 (SD=0.43) y, and the date of menarche, if present, was recorded. Associations were assessed using Kaplan–Meier analysis and multivariable-adjusted Cox regression.
Results:
In total, 77% of the girls (n=717) had reached menarche by the second follow-up. The median age of menarche among all girls was 13.0 y (25th–75th percentiles: 12.4–13.7 y). At 10 years of age, median urinary cadmium was 0.25μg/L (5th–95th percentiles: 0.087–0.72μg/L), lead 1.6μg/L (0.70–4.2μg/L), and arsenic 54μg/L (19–395μg/L). Given the same age, girls in the highest quartile of urinary cadmium at 5 and 10 years of age had a lower rate of menarche than girls in the lowest quartile, with an adjusted hazard ratio of (HR) 0.80 (95% CI: 0.62, 1.01) at 5 years of age, and 0.77 (95% CI: 0.60, 0.98) at 10 years of age. This implies that girls in the highest cadmium exposure quartile during childhood had a higher age at menarche. Comparing girls in the highest to the lowest quartile of urinary lead at 10 years of age, the former had a higher rate of menarche [adjusted HR = 1.23 (95% CI: 0.97, 1.56)], implying lower age at menarche, whereas there was no association with urinary lead at 5 years of age. Girls born to mothers in the highest quartile of erythrocyte arsenic during pregnancy were less likely to have attained menarche than girls born to mothers in the lowest quartile [adjusted HR= 0.79 (95% CI: 0.62, 0.99)]. No association was found with girls’ urinary arsenic exposure.
Discussion:
Long-term childhood cadmium exposure was associated with later menarche, whereas the associations with child lead exposure were inconclusive. Maternal exposure to arsenic, but not cadmium or lead, was associated with later menarche. https://doi.org/10.1289/EHP11121
Supplemental Material is available online (https://doi.org/10.1289/EHP11121).
The authors have no conflicts of interest to declare.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Puberty is the transitional phase between childhood and reproductive maturity. Besides being associated with fertility, puberty is also associated with development of secondary sexual characteristics, rapid growth, and behavioral changes.1 Altered timing of puberty onset has been linked to various health conditions later in life.2–5 Earlier menarche, the first menstruation, has been associated with increased risk of metabolic syndrome,2 cardiovascular disease,3 and cancer4 later in life, whereas later menarche has been associated with lower peak bone mass and increased risk of osteoporosis.5 Puberty starts with the activation of the hypothalamic–pituitary–gonadal (HPG) axis and the secretion of gonadotropin-releasing hormone (GnRH), triggering a hormonal cascade.1 The timing of puberty depends on a variety of factors, such as genetics (which account for about half of the variability),6 epigenetic regulation,7 nutrition, growth trajectories, and exposure to endocrine disrupting chemicals.8,9
Emerging evidence indicates that metals, especially lead, commonly present in food, dust, drinking water, and various household items,10 may alter pubertal development. Studies in Mexico have found that elevated maternal lead exposure during pregnancy, assessed via concentrations in blood11 and bone,12 was associated with later menarche in their daughters. In cross-sectional studies from the United States and South Africa, girls’ blood lead concentrations were associated with later breast and pubic hair development or menarche13–15 and with changes in hormonal markers consistent with later puberty.16 Interestingly, one of these studies reported that a combination of elevated exposure to both lead and cadmium, another metal with endocrine disrupting properties17 and a common dietary pollutant,18 was associated with lower inhibin B in girls 6–11 years of age.16 Inhibin B is a protein involved in the feedback regulation of the HPG axis, and lower concentrations are indicative of pubertal delay.19 In addition, maternal gestational exposure to cadmium was associated with slower breast development in Mexican girls, and the girls’ peripubertal urinary cadmium concentration was associated with later menarche.20 Another study of girls in the United States reported that childhood exposure to cadmium was associated with later menarche and pubic hair growth.21 In contrast, an ecological Chinese study reported that women in two heavily cadmium-polluted areas had undergone menarche earlier than women in a control area (∼100–300 girls in each area).22 Thus, there is a need for large prospective studies on the association between cadmium and puberty.
The aim of this study was to assess whether exposure to cadmium and lead during pregnancy and childhood was associated with age at menarche in a large longitudinal birth cohort in rural Bangladesh with documented low-to-moderate exposure to cadmium and lead.23,24 In addition, this is a region where drinking water frequently contains elevated arsenic concentrations and we have previously reported that consumption of such water by mothers during pregnancy was associated with later menarche in this cohort.25 Therefore, a secondary aim of this study was to investigate whether prenatal and childhood exposure to arsenic, assessed through individual exposure markers, was associated with age at menarche, and whether it modified the possible associations of cadmium and lead exposure.
Methods
Study Population
This study was conducted in Matlab, a region in Bangladesh located ∼50km southeast of the capital, Dhaka. Participants were children born to women in the Maternal and Infant Nutrition Interventions trial in Matlab (MINIMat, isrctn.org identifier: ISRCTN16581394). The MINIMat trial was a factorial randomized trial with the overarching objective to evaluate whether a combination of multiple micronutrients and food supplementations would improve pregnancy outcomes.26 Pregnant women were randomly assigned to one of three different supplementations from gestational week (GW) 14; 30mg iron and 400μg folate, 60mg iron and 400μg folate, or a capsule of 15 recommended micronutrients that included 30mg iron and 400μg folate, in combination with food supplementation, which was either provided early (at around GW9) or at the usual timing (at around GW20).26 From November 2001 to October 2003, community health research workers recruited 4,436 women during early pregnancy, and these pregnancies resulted in 3,625 live-born infants, of which 3,560 were singleton births.
Since birth, extensive follow-ups of the children have been conducted at ∼4.5–5, 9–10, and at 12–15 years of age.27 Of the 3,560 children, 2,307 participated (1,175 girls and 1,132 boys) in the puberty assessment, consisting of two follow-up visits, within the age interval of 12–15 y.25,28 Reasons for loss to follow-up were outmigration, refusal to participate, and death (Figure S1). In children born in the MINIMat trial between October 2002 and November 2003, the exposure to metals has previously been assessed for the mothers during pregnancy and the children at 5 and 10 years of age (n=1,530 at the 10-y follow-up) and thereafter related to outcomes such as anthropometry and neurodevelopment.29–33 Furthermore, the exposure to metals has also been assessed at 4.5 and 9 years of age in a smaller subsample of the MINIMat children (born at the hospital in Matlab between June 2003 and June 2004; n=551 at the 9-y follow-up) to evaluate the impact on immune function.34,35 The girls and their mothers in this rural setting are primarily exposed to cadmium through their diet,36 which is largely based on rice, known to easily take up cadmium from the soil.18 Lead exposure occurs mainly through food and drinking water, dust, housing materials, and various utensils.24 Elevated exposure to inorganic arsenic through drinking water and to some extent also via rice is prevalent.37,38 In the present study, we included all girls who participated in the puberty assessment, had data on menarche, had complete covariate data for the primary analysis and for whom metal exposure had been previously assessed either for their mothers during pregnancy (n=771) or for themselves at 5 (n=750) or at 10 years of age (n=745), resulting in a total of 935 girls (Figure S1).
Written and oral informed consent was obtained from the mothers at enrollment and from the girls, as well as the mother or other guardian, at the time of the puberty follow-ups. The study was conducted in concordance with the Helsinki Declaration, and approved by the Ethical Review Committee at the International Center for Diarrheal Disease Research, Bangladesh (ICDDR,B) in Bangladesh and the ethical review board in Sweden.
Exposure Assessment
Exposure to cadmium, lead, and arsenic was assessed by the concentrations in the mothers’ blood (erythrocyte fraction) during early pregnancy (GW14) and in the girls’ urine at 5 and 10 years of age. Both blood and spot urine samples were collected at the health care facilities in Matlab. Erythrocytes were separated from plasma by centrifugation within a couple of hours. The collection and handling of blood samples and spot urine samples has been described in detail.31,33,39
Metal concentrations in erythrocytes reflect the exposure over the last few months, given that the erythrocytes’ lifespan is about 3–4 months.40 Cadmium concentration in urine is a measure of long-term exposure, because cadmium accumulates in the renal cortex with a half-life of decades.41 Urinary lead and arsenic, on the other hand, are short-term exposure biomarkers because lead and arsenic are excreted in urine within a few days of exposure.24 No blood samples were available from the children at these time points. However, because arsenic is present in the daily consumed water and food, the urinary concentrations reflect ongoing exposure quite well.38 Arsenic in erythrocytes and urine was measured as total arsenic. We have previously reported that total arsenic in urine and the sum of arsenic metabolites, reflecting exposure to inorganic arsenic specifically, were in good agreement in this cohort, both in the mothers during pregnancy (linear regression coefficient β=0.93, p<0.001, and R2=0.96)42 and in the previously mentioned smaller subsample of children that participated in the follow-up at 9 years of age (Spearman’s correlation rho=0.98, p<0.001).43 In addition, there was a strong correlation between arsenic in erythrocytes and urine in the mothers during pregnancy (linear regression coefficient β=0.83, p<0.001, and R2=0.83)42 and in the children at 9 years of age (Spearman’s correlation rho=0.79, p<0.001).43
The concentrations of cadmium (m/z 111), lead (m/z 208), and arsenic (m/z 75) were measured using inductively coupled plasma mass spectrometry (ICPMS; Agilent 7500ce or 7700ce; Agilent Technologies) at our laboratory at Karolinska Institutet, Stockholm, Sweden. Before analysis, the erythrocyte samples were diluted 1:25 in an alkali solution [2% (wt:vol) 1-butanol, 0.05% (wt:vol) ethylenediaminetetraacetic acid, 0.05% (wt:vol) Triton X-100, 1% (wt:vol) ammonium hydroxide, and 20μg/L internal standard; Sigma-Aldrich]. Then they were vortex mixed, sonicated for 5 min, and centrifuged at 179×g for 2 min (MSE centrifuge, Super Minor; MSE Ltd.).44 Urine samples were diluted 1:10 with 1% nitric acid (Scharlau, Scharlab, Sentmenat, Spain or Ultrapure Normatom; VWR Chemicals). The limit of detection was 0.006μg/L, 0.015μg/L, and 0.011μg/L for erythrocyte cadmium, lead, and arsenic, respectively, and 0.005μg/L, 0.007μg/L, and 0.023μg/L for urinary cadmium, lead, and arsenic, respectively. No samples were found to have metal concentrations below each respective limit of detection. The details of the quality control have been previously published,30,34 and in general there was a good agreement between the expected values and our measurements.
A minor fraction of the maternal erythrocyte samples (n=269 of total 771) was analyzed using acid digestion prior to the ICPMS analyses39 instead of the alkali dilution described above. The results from the two methods were previously found to be strongly correlated when analyzing samples from another mother–child cohort in the Argentinean Andes (R2>0.96 for cadmium, lead, and arsenic).44 However, because the cadmium, lead, and arsenic concentrations were found to be consistently lower using the alkali method (by 9%, 10%, and 5% respectively),44 the concentrations obtained with the acidic method were multiplied by 0.91 for cadmium, 0.90 for lead, and 0.95 for arsenic. To compensate for urine dilution, metal concentrations in urine were adjusted to the mean specific gravity (mean 1.012 in the children), which was measured with a digital refractometer (EUROMEX RD712 Clinical Refractometer; EUROMEX Holland).45 Adjustment for specific gravity instead of creatinine has been shown to be more suitable, particularly in growing adolescents.46
Outcome Assessment
The assessment of puberty was conducted at health care facilities run by ICCDR,B in Matlab. Information about menstruation, breast, and pubic hair development was collected on two separate occasions, spaced 6 months apart, to optimize the validity of reported menarche.25,28 Female nurses interviewed the girls about if they had had their first menstruation and its date. Mothers were asked to assist if needed, and a calendar with local events was used. Age at menarche was calculated using the date of birth and the recalled date of first menstruation. This study used menarche data obtained from the second puberty follow-up visit. Data from the first puberty follow-up visit were used for girls who did not participate in the second follow-up (n=43). If two different dates of menarche were reported at the first and second puberty follow-up, we used data from the first, given that it was closer to the event. For girls who did not remember the month, but only the year of menarche (n=3), the month June was imputed, and for girls who remembered the year and month but not the exact date (n=362), the 15th of that month was imputed.
The girls also self-assessed their breast and pubic hair development according to Tanner, guided by pictures.47 Available information on Tanner stages was primarily used to evaluate consistency with potential associations of the metals and age at menarche.
Covariates
Information concerning maternal and household characteristics was obtained either from the Health and Demographic Surveillance System (HDSS), which has been ongoing in Matlab since 1966 and is maintained by ICDDR,B, Dhaka, Bangladesh, or from the clinic visits and questionnaires administered during the MINIMat trial. The HDSS is continuously updated with data collected by community health research workers who visit the families on a monthly basis, and from here we obtained data to generate a household socioeconomic asset score. The household socioeconomic asset score was thereafter generated through principal component analysis using extensive information on household ownership of various assets, housing structure, and dwelling characteristics.48 This asset score was categorized into tertiles. From the MINIMat trial we obtained data on maternal smoking during pregnancy (yes/no), maternal age, maternal education, and maternal anthropometry. In accordance with the social norms of the area, none of the women smoked. Maternal weight (in kilograms) was measured at the clinic visit at around GW8 with an electronic scale (Uniscale; SECA) with a precision of 0.1kg, and maternal height (in centimeters) with a stadiometer to the nearest 0.1cm. Maternal body mass index (BMI) was calculated as weight/(height in meters).2 The children’s weight and length or height were recorded at birth,49 during the follow-up survey at 4.5 years of age, and two times during the puberty follow-up in the age interval of 12–15 y28; they were thereafter converted into weight- and height-for-age Z-scores.50 Child stunting was defined as lower than (stunted) or above (not stunted) −2 Z-scores. The micronutrient supplementation during pregnancy had three categories (30mg iron and 400μg folate, 60mg iron and 400μg folate, or 15 recommended micronutrients), and the food supplementation during pregnancy had two categories (early or usual timing).
Statistical Analyses
All analyses were performed using Stata (version 16; StataCorp), and the statistical significance level was 0.05. To assess whether there were differences in characteristics between the mother–daughter pairs included (n=935) and not included (owing to missing data on exposure, menarche, or covariates; n=240) in this study, we compared these two groups using Mann-Whitney U-tests for continuous variables and Pearson’s chi-square tests for categorical variables. We used Spearman’s rank correlation to assess correlations between continuous variables.
Associations of gestational and childhood exposure to cadmium, lead, and arsenic with age at menarche were evaluated using so-called survival analysis. We calculated person-time for each girl from date of birth until date of menarche, or censored at second puberty assessment, if available, or first puberty assessment if the second was not available. Age was used as the timescale in all survival analyses. The Kaplan–Meier method was used to calculate the median age at menarche by the exposure biomarkers (maternal erythrocyte concentrations of cadmium, lead, and arsenic and girls’ urinary concentrations of cadmium, lead, and arsenic) categorized into quartiles. Thereafter, we conducted Cox proportional hazards regression models to evaluate associations of the exposures (categorized into quartiles) with timing of menarche. The assumption of proportional hazards was evaluated by visually assessing the hazards plot and by testing that the Schoenfeld residuals were independent of time. The effect estimates were presented as hazard ratios (HRs) with 95% confidence intervals (CIs). Model 1 presents the crude association of exposure and age at menarche. Model 2 was adjusted for household socioeconomic asset score (in tertiles), maternal BMI at GW8 (in kilograms per meter squared; continuously), and maternal education (years; continuous variable). We selected these three covariates because they were identified as the minimal sufficient adjustment using a directed acyclic graph (DAG; R package dagitty; version 3.0).51 The DAG indicated no need for adjustment for maternal age (Figure S2). Household socioeconomic asset score and maternal education were correlated (rho=0.54, p<0.001), but they were both included in the adjustment given that they do not reflect the same aspects. Moreover, these three variables were found to differ between girls who reached menarche earlier or later in this cohort.28 Body composition or anthropometry at peripubertal age were not included in adjustments because puberty itself is a driver of growth.52 Finally, we conducted a combined exposure model (model 3) including the time-specific exposure of cadmium, lead, and arsenic simultaneously.
In sensitivity analysis, we additionally adjusted for stunting at 4.5 years of age because childhood (1, 2, and 4.5 years of age) stunting has been associated with later menarche in this cohort28 and exposure to cadmium and arsenic during the mother’s pregnancy and during childhood (at 5 and 10 years of age) has been associated with decreased growth in these children from birth to 10 years of age.23,29,30,49 Moreover, lead exposure has been associated with impaired growth in another Bangladeshi cohort.53 Stunting was used as a binary variable, and it was categorized as having a height-for-age Z-score (calculated with the measured height at 4.5 years of age through the child growth international reference values developed by the World Health Organization54) lower than −2 (stunted) or above (not stunted). We also tested adjusting for the randomized micronutrient and food supplementation groups during pregnancy (three and two categories, respectively).
We used the Kruskal-Wallis H-test to assess differences in cadmium, lead, and arsenic exposure (as continuous variables) between the girls at different Tanner developmental stages. Ordered logistic regression models were used to calculate the odds ratios (ORs) representing the relative odds of reaching a more advanced Tanner stage at the second puberty follow-up. The models included the biomarkers of cadmium, lead, and arsenic exposure (categorized into quartiles using the lowest quartile as the reference category) during the mother’s pregnancy, at 5 or at 10 years of age, and they were adjusted for household socioeconomic asset score (in tertiles), maternal BMI at GW8 (in kilograms per meter squared; continuously), and maternal education (years; continuous variable).
For the primary survival analysis, we conducted complete-subject analysis and included only mother–daughter pairs with complete covariate data. In the sensitivity analyses of Cox regression data, we again performed the main analyses (models 1–3), as well as models 4 and 5, including the mother–daughter pairs who had available data on stunting at 4.5 years of age, which was the limiting factor. For the analysis of associations between metal exposure and Tanner stages, we included the mother–daughter pairs who had complete covariate data and had participated in the second puberty follow-up.
Results
Participant Selection and Participant Characteristics
Among all girls who participated in at least one of the puberty follow-ups (n=1,175), the girls who were included in the present study (n=935) and those who were not (owing to lacking data on menarche or any metal exposure data either during their mother’s pregnancy or at 5 or 10 years of age; n=240) were comparable with respect to all measured background characteristics, such as age, weight, and height, as well as household socioeconomic asset score, and maternal BMI and education (Table S1). At enrollment in the MINIMat study, the pregnant women were all nonsmokers, had an average age of 26 y (min–max: 14–50 y) and 22% were considered underweight (BMI<18 kg/m2) (Table 1). The mothers had on average attended 4.5 y of schooling (min–max: 0–16 y), and around one-third was illiterate, having never attended school.
Table 1 Characteristics of the girls and their mothers during early pregnancy, from a Bangladeshi mother–child cohort recruited between 2001 and 2003, who were included in the present study (n=935).
Characteristics n or n (%) Missing data (n) Mean±SD or median (5th–95th percentiles)a
Mothers or householdb
Socioeconomic asset score (min–max) 935 0 —
Low (–5.93 to –0.81) 349 (37) — —
Medium (–0.80 to 1.46) 317 (34) — —
High (1.47 to 3.98) 269 (29) — —
Age (y) 935 0 26±6.1
BMI at GW8 (kg/m2) 935 0 20±2.6
Education (y) 935 0 4.4±3.9
Parity (no. of children) 935 0 1 (0–4)
Micronutrient supplementation 935 0
30mg iron and 400μg folate 313 (33) — —
60mg iron and 400μg folate 313 (33) — —
15 recommended micronutrients 309 (33) — —
Food supplementation 935 0 —
Early timing (at around GW9) 478 (51) — —
Usual timing (at around GW20) 457 (49) — —
Girls
Age
1st follow-up (y) 935 0 13.3±0.43
2nd follow-up (y)c 892 0 13.8±0.43
Weight
1st follow-up (kg)
932 3 38.4±8.3
2nd follow-up (kg)c 891 1 40.6±8.1
Height
1st follow-up (cm) 933 2 148±6.3
2nd follow-up (cm)c 891 1 150±5.8
Breast development (Tanner stage)c 892 0 —
1 0 (0) — —
2 126 (14) — —
3 617 (69) — —
4 149 (17) — —
5 0 (0) — —
Pubic hair development (Tanner stage)c 892 0 —
1 3 (0.34) — —
2 413 (46) — —
3 456 (51) — —
4 20 (2.2) — —
5 0 (0) — —
Menarche by 1st follow-up 935 0 —
Yes 572 (61) — —
No 363 (39) — —
Menarche by 2nd follow-up 935 0 —
Yes 717 (77) — —
No 218 (23) — —
Age at menarche (y)d 717 0 12.66±0.79
Birth weight (g) 935 0 2,647±381
Stunting at 4.5 years of age 890 45 —
Yes 315 (35) — —
No 575 (65) — —
Note: —, not applicable; BMI, body mass index; GW, gestational week; max, maximum; min, minimum; SD, standard deviation.
a Median (5th–95th percentiles) reported for parity, mean±SD reported for all other variables.
b At enrollment.
c At second puberty follow-up visit. Participants at second follow-up visit n=892 girls.
d Menarche data available for all girls in this study (n=935); 218 girls who had not reached menarche were censored.
The mean weight of girls at the first and second puberty assessment was 38.4kg (min–max: 19.0–91.6kg) and 40.6kg (min–max: 22.7–93.0kg), respectively, and their mean height was 148cm (min–max: 121–167cm) and 150cm (min–max: 122–167cm) at the first and the second follow-up. During the first puberty assessment, girls were on average 13.3 years of age [standard deviation (SD)=0.43, min–max: 12.4–14.2 y] and 61% reported to have had menarche. At the second assessment the mean age was 13.8 y (SD=0.43, min–max: 12.9–14.7 y) and the percentage who had reached menarche increased to 77%. The median age at menarche was 13.0 y (25th–75th percentiles: 12.4–13.7 y), and the mean recall time between menarche and data collection was 0.8 y (min–max: 0.0–3.6 y). Table S2 summarizes the frequency distribution of the reported age at menarche of the 717 girls who reached this milestone during this study, and the 218 girls who did not. None of the girls had reached menarche before their most recent metal exposure assessment at 10 years of age (min–max: 8.7–10.0 y), and the girl with the youngest age at menarche was 9.9 years old.
At the second puberty follow-up, more than two-thirds of the girls assessed themselves as belonging to Tanner stage 3 of breast development, and the other girls were quite evenly split between stage 2 and stage 4 (Table 1). Regarding pubic hair development, almost all girls were either stage 2 or 3, with only a few participants assessing themselves as stage 1 (n=3) or 4 (n=20) (Table 1).
The concentrations of the exposure biomarkers in this study are presented in Table 2. Median maternal erythrocyte concentrations at GW14 were 0.92μg/kg (5th–95th percentiles: 0.27–2.3μg/kg) for cadmium, 72μg/kg (5th–95th percentiles: 22–150μg/kg) for lead, and 4.8μg/kg (5th–95th percentiles: 1.2–22μg/kg) for arsenic. Maternal erythrocyte cadmium and lead concentrations were moderately correlated (rho=0.38, p<0.001), whereas maternal arsenic was not correlated with either maternal cadmium or lead. The girls’ median urinary cadmium concentrations at 5 and 10 years of age were 0.23μg/L (5th–95th percentiles: 0.083–0.73μg/L) and 0.25μg/L (5th–95th percentiles: 0.087–0.72μg/L), respectively, and the exposure biomarkers at the two time points were moderately correlated (rho=0.40, p<0.001). The median urinary lead concentration was 3.4μg/L (5th–95th percentiles: 1.4–9.8μg/L) at 5 years of age and 1.6μg/L (5th–95th percentiles: 0.70–4.2μg/L) at 10 years of age, and the correlation between them was weaker than for cadmium (rho=0.24, p<0.001). The girls’ urinary lead concentrations at 5 and 10 years of age were weakly correlated with the maternal erythrocyte lead concentrations (rho=0.082, p<0.050). The median total urinary arsenic concentrations at 5 and 10 years of age were 56μg/L (5th–95th percentiles: 17–353μg/L) and 54μg/L (5th–95th percentiles: 19–395μg/L), respectively, and the arsenic urinary concentrations at the two time points were strongly correlated (rho=0.57, p<0.001). The girls’ urinary arsenic concentrations were moderately correlated with the maternal erythrocyte arsenic concentrations (rho ∼ 0.45) (Table S3).
Table 2 Concentrations of cadmium, lead, and arsenic in maternal erythrocytes during pregnancy and child urine at 5 and 10 years of age and range of concentration in each quartile of exposure.
Exposure biomarker n Min Max Mean±SD Median Percentile
25th 75th
Erythrocyte Cd GW14 (μg/kg) 771 0.076 4.9 1.1±0.72 0.92 0.61 1.4
Q1 193 0.076 0.61 — — — —
Q2 193 0.61 0.92 — — — —
Q3 193 0.93 1.4 — — — —
Q4 192 1.4 4.9 — — — —
Urinary Cd at 5 years of age (μg/L)a 750 0.20 3.8 0.31±0.32 0.23 0.16 0.36
Q1 187 0.02 0.16 — — — —
Q2 188 0.16 0.23 — — — —
Q3 188 0.23 0.36 — — — —
Q4 187 0.36 3.8 — — — —
Urinary Cd at 10 years of age (μg/L)a 745 0.016 2.9 32±0.25 0.25 0.17 0.39
Q1 187 0.016 0.17 — — — —
Q2 186 0.17 0.25 — — — —
Q3 186 0.26 0.39 — — — —
Q4 186 0.39 2.9 — — — —
Erythrocyte Pb GW14 (μg/kg) 771 7.0 607 78±44 72 50 96
Q1 193 7 50 — — — —
Q2 193 50 72 — — — —
Q3 193 72 97 — — — —
Q4 192 97 607 — — — —
Urinary Pb at 5 years of age (μg/L)a 750 0.30 44 4.3±3.4 3.4 2.4 5.2
Q1 187 0.3 2.4 — — — —
Q2 188 2.4 3.4 — — — —
Q3 188 3.4 5.2 — — — —
Q4 187 5.3 44 — — — —
Urinary Pb at 10 years of age (μg/L)a 745 0.059 12 1.9±1.2 1.6 1.2 2.2
Q1 187 0.059 1.2 — — — —
Q2 186 1.2 1.6 — — — —
Q3 186 1.6 2.2 — — — —
Q4 186 2.2 12 — — — —
Erythrocyte As GW14 (μg/kg) 771 0.15 62 7.8±7.6 4.8 2.4 11
Q1 193 0.15 2.4 — — — —
Q2 193 2.4 4.8 — — — —
Q3 193 4.8 11 — — — —
Q4 192 11 62 — — — —
Urinary As at 5 years of age (μg/L)a 750 6 1,152 104±129 56 32 124
Q1 188 6 32 — — — —
Q2 187 32 56 — — — —
Q3 188 56 124 — — — —
Q4 187 124 1,152 — — — —
Urinary As at 10 years of age (μg/L)a 745 8.9 847 106±126 54 33 121
Q1 187 8.9 33 — — — —
Q2 186 33 54 — — — —
Q3 186 54 121 — — — —
Q4 186 121 847 — — — —
Note: —, not shown; As, arsenic; Cd, cadmium; max, maximum; min, minimum; Pb, lead; Q, quartile; SD, standard deviation.
a Adjusted for urinary specific gravity.
Associations of Exposure to Cadmium, Lead, and Arsenic and Age at Menarche
Tables 3–5 show the median age at menarche for the quartiles of cadmium, lead, and arsenic exposure measured in maternal erythrocytes at GW14 and in the girls’ urine at 5 and 10 years of age (unadjusted). A weak trend was observed between increasing concentrations of mothers’ erythrocyte cadmium and higher median age at menarche (1.8 months difference between highest and lowest quartile). However, the trends between the girls’ urinary cadmium at 5 years of age (2.9 months difference), and especially that at 10 years of age, with increasing age at menarche were more marked. Girls in the highest quartile of urinary cadmium at 10 years of age (>0.39μg/L, median=0.51μg/L, 5th–95th percentiles: 0.40–1.1μg/L) reached menarche 3.8 months later than girls in the lowest quartile (<0.17μg/L; Table 3). Median age at menarche varied less consistently across quartiles of urinary lead (Table 4), but girls in the highest quartiles of urinary lead at 5 and 10 years of age reached menarche 2.2 and 3.0 months earlier, respectively, than girls in the lowest quartile (Table 4). We observed a trend of increasing age at menarche with increasing maternal erythrocyte arsenic concentrations, but not with increasing childhood urinary arsenic concentrations. Girls born to mothers in the highest quartile of arsenic exposure reached menarche 5.5 months later than those born to mothers in the lowest quartile (Table 5). Figures S3–S5 show the cumulative incidence of menarche per quartile of cadmium, lead, and arsenic exposure at the three exposure time points.
Table 3 Cox regression models of girls’ early life cadmium exposure (maternal erythrocyte concentrations at gestational week 14, urinary concentrations at 5 and 10 years of age), and age at menarche.
Exposure biomarker (median) n Time at risk (person-years) Median age at menarche (y)a Model 1 Model 2 Model 3
HR (95% CI) p-Value HR (95% CI) p-Value HR (95% CI) p-Value
Erythrocyte Cd GW14 (μg/kg) 771 — — — — — — — —
Q1 (0.44) 193 2,472 12.93 1.00 — 1.00 — 1.00 —
Q2 (0.77) 193 2,490 12.97 0.91 (0.73, 1.13) 0.39 0.94 (0.75, 1.18) 0.59 0.99 (0.78, 1.25) 0.91
Q3 (1.1) 193 2,499 13.11 0.86 (0.68, 1.07) 0.17 0.90 (0.72, 1.13) 0.36 0.94 (0.74, 1.19) 0.61
Q4 (1.8) 192 2,490 13.08 0.84 (0.67, 1.05) 0.13 0.91 (0.72, 1.14) 0.40 0.97 (0.76, 1.24) 0.83
Urinary Cd at 5 years of age (μg/L) 750 — — — — — — — —
Q1 (0.12) 187 2,391 12.84 1.00 — 1.00 — 1.00 —
Q2 (0.19) 188 2,417 13.01 0.86 (0.68, 1.08) 0.20 0.88 (0.70, 1.10) 0.26 0.86 (0.68, 1.09) 0.22
Q3 (0.27) 188 2,415 13.11 0.87 (0.69, 1.10) 0.24 0.91 (0.72, 1.14) 0.41 0.89 (0.71, 1.13) 0.34
Q4 (0.50) 187 2,421 13.08 0.76 (0.60, 0.95) 0.018 0.81 (0.64, 1.02) 0.077 0.80 (0.62, 1.01) 0.065
Urinary Cd at 10 years of age (μg/L) 745 — — — — — — — —
Q1 (0.12) 187 2,406 12.95 1.00 — 1.00 — 1.00 —
Q2 (0.21) 186 2,375 12.89 1.10 (0.88, 1.38) 0.42 1.11 (0.88, 1.40) 0.36 1.10 (0.87, 1.38) 0.43
Q3 (0.30) 186 2,383 12.89 1.07 (0.85, 1.35) 0.55 1.16 (0.92, 1.47) 0.21 1.15 (0.91, 1.45) 0.25
Q4 (0.51) 186 2,420 13.27 0.73 (0.57, 0.93) 0.010 0.77 (0.61, 0.99) 0.039 0.77 (0.60, 0.98) 0.035
Note: —, not applicable; Cd, cadmium; CI, confidence interval; GW, gestational week; HR, hazard ratio; Q, quartile.
a Unadjusted, calculated from Kaplan–Meier.
Model 1: adjusted for age of the child.
Model 2: additionally adjusted for household socioeconomic asset score at enrollment (tertiles), maternal body mass index at GW8 (kg/m2), and maternal education at enrollment (y).
Model 3: additionally adjusted for lead and arsenic exposure in quartiles at the corresponding time points (maternal erythrocyte concentrations at GW14 and girls’ urinary concentrations at 5 and 10 years of age).
Table 4 Cox regression models of girls’ early life lead exposure (maternal erythrocyte concentrations at gestational week 14, urinary concentrations at 5 and 10 years of age), and age at menarche.
Exposure biomarker (median) n Time at risk (person-years) Median age at menarche (y)a Model 1 Model 2 Model 3
HR (95% CI) p-Value HR (95% CI) p-Value HR (95% CI) p-Value
Erythrocyte Pb GW14 (μg/kg) 771 — — — — — — — —
Q1 (36) 193 2,468 12.82 1.00 — 1.00 — 1.00 —
Q2 (61) 193 2,506 13.18 0.78 (0.62, 0.98) 0.030 0.78 (0.62, 0.98) 0.032 0.81 (0.64, 1.03) 0.086
Q3 (82) 193 2,501 13.06 0.87 (0.70, 1.09) 0.22 0.86 (0.69, 1.07) 0.18 0.89 (0.70, 1.12) 0.32
Q4 (118) 192 2,476 12.99 0.90 (0.72, 1.13) 0.37 0.85 (0.68, 1.06) 0.14 0.86 (0.68, 1.09) 0.20
Urinary Pb at 5 years of age (μg/L) 750 — — — — — — — —
Q1 (1.8) 187 2,423 13.11 1.00 — 1.00 — 1.00 —
Q2 (2.9) 188 2,409 13.05 1.09 (0.85, 1.38) 0.50 1.08 (0.85, 1.37) 0.53 1.09 (0.86, 1.39) 0.49
Q3 (4.3) 188 2,409 12.99 1.15 (0.91, 1.46) 0.25 1.14 (0.90, 1.44) 0.29 1.17 (0.92, 1.49) 0.19
Q4 (7.1) 187 2,403 12.93 1.13 (0.89, 1.42) 0.33 1.09 (0.86, 1.38) 0.49 1.15 (0.90, 1.47) 0.25
Urinary Pb at 10 years of age (μg/L) 745 — — — — — — — —
Q1 (0.89) 187 2,426 13.12 1.00 — 1.00 — 1.00 —
Q2 (1.4) 186 2,385 12.96 1.26 (0.99, 1.59) 0.054 1.27 (1.00, 1.60) 0.049 1.24 (0.98, 1.57) 0.076
Q3 (1.9) 186 2,394 13.14 1.04 (0.82, 1.32) 0.75 0.99 (0.78, 1.27) 0.97 0.99 (0.78, 1.27) 0.98
Q4 (2.8) 186 2,379 12.87 1.28 (1.01, 1.62) 0.039 1.21 (0.95, 1.53) 0.12 1.23 (0.97, 1.56) 0.094
Note: —, not applicable; CI, confidence interval; GW, gestational week; HR, hazard ratio; Pb, lead; Q, quartile.
aUnadjusted, calculated from Kaplan–Meier.
Model 1: adjusted for age of the child.
Model 2: additionally adjusted for household socioeconomic asset score at enrollment (tertiles), maternal body mass index at GW8 (kg/m2), and maternal education at enrollment (y).
Model 3: additionally adjusted for cadmium and arsenic exposure in quartiles at the corresponding time points (maternal erythrocyte concentrations at GW14 and girls’ urinary concentrations at 5 and 10 years of age).
Table 5 Cox regression models of girls’ early life arsenic exposure (maternal erythrocyte concentrations at gestational week 14, urinary concentrations at 5 and 10 years of age), and age at menarche.
Exposure biomarker (median) n Time at risk (person-years) Median age at menarche (y)a Model 1 Model 2 Model 3
HR (95% CI) p-Value HR (95% CI) p-Value HR (95% CI) p-Value
Erythrocyte As GW14 (μg/kg) 771 — — — — — — — —
Q1 (1.6) 193 2,464 12.82 1.00 — 1.00 — 1.00 —
Q2 (3.4) 193 2,498 13.06 0.89 (0.71, 1.11) 0.29 0.92 (0.74, 1.15) 0.46 0.93 (0.74, 1.17) 0.55
Q3 (7.6) 193 2,481 12.91 0.93 (0.74, 1.16) 0.51 0.95 (0.76, 1.19) 0.66 0.95 (0.76, 1.19) 0.66
Q4 (16) 192 2,508 13.28 0.73 (0.58, 0.91) 0.006 0.78 (0.62, 0.97) 0.029 0.79 (0.62, 0.99) 0.043
Urinary As at 5 years of age (μg/L) 750 — — — — — — — —
Q1 (22) 188 2,399 12.85 1.00 — 1.00 — 1.00 —
Q2 (0.42) 187 2,405 13.01 0.91 (0.72, 1.14) 0.41 0.89 (0.70, 1.12) 0.33 0.89 (0.70, 1.13) 0.33
Q3 (77) 188 2,420 13.02 0.85 (0.67, 1.07) 0.17 0.85 (0.67, 1.08) 0.18 0.87 (0.68, 1.10) 0.23
Q4 (212) 187 2,420 13.05 0.86 (0.69, 1.09) 0.21 0.89 (0.71, 1.12) 0.33 0.89 (0.70, 1.12) 0.31
Urinary As at 10 years of age (μg/L) 745 — — — — — — — —
Q1 (25) 187 2,399 12.97 1.00 — 1.00 — 1.00 —
Q2 (42) 186 2,391 13.02 1.02 (0.81, 1.29) 0.87 1.13 (0.89, 1.43) 0.31 1.14 (0.90, 1.44) 0.29
Q3 (73) 186 2,388 12.99 1.04 (0.82, 1.32) 0.73 1.12 (0.89, 1.42) 0.34 1.16 (0.91, 1.48) 0.22
Q4 (237) 186 2,407 13.02 0.93 (0.74, 1.18) 0.55 1.04 (0.82, 1.32) 0.76 1.07 (0.84, 1.36) 0.58
Note: —, not applicable; As, arsenic; CI, confidence interval; GW, gestational week; HR, hazard ratio; Q, quartile.
aUnadjusted, calculated from Kaplan–Meier.
Model 1: adjusted for age of the child.
Model 2: additionally adjusted for household socioeconomic asset score at enrollment (tertiles), maternal body mass index at GW8 (kg/m2), and maternal education at enrollment (y).
Model 3: additionally adjusted for cadmium and lead exposure in quartiles at the corresponding time points (maternal erythrocyte concentrations at GW14 and girls’ urinary concentrations at 5 and 10 years).
In the Cox proportional hazard models (Tables 3–5), we did not find any consistent associations between maternal erythrocyte cadmium or lead concentration during pregnancy and the daughters’ age at menarche. Instead, we found that girls born to mothers belonging to the highest quartile of erythrocyte arsenic had a 21% (HR= 0.79; 95% CI: 0.62, 0.99) lower rate of menarche than girls born to mothers in the lowest quartile of exposure. The HRs did not vary markedly with adjustments (Table 5). Regarding the relationship between metal exposure during childhood and age at menarche, we observed that girls in the highest quartile of urinary cadmium at 5 and 10 years of age had 20% (HR= 0.80; 95% CI: 0.62, 1.01) and 23% (HR= 0.77; 95% CI: 0.60, 0.98) lower rate of menarche at a given age than girls in the lowest quartile of exposure. The associations were very consistent regardless of adjustment (Table 3). For lead, we observed that the girls in the highest quartile of urinary lead at 10 years of age, but not at 5 years of age, obtained menarche earlier than those in the lowest quartile (Table 4), and after adjustments the HR remained similar but with a CI including 1.00 (HR= 1.23; 95% CI: 0.97, 1.56). No associations were found between urinary arsenic concentrations during childhood and age at menarche (Table 5).
In sensitivity analysis, the association between the girls’ urinary cadmium concentrations at 5 years of age and later menarche was weakened following adjustment for stunting at 4.5 years of age (HR= 0.84; 95% CI: 0.66, 1.07), whereas that with urinary cadmium at 10 years of age remained the same (HR= 0.78; 95% CI: 0.61, 0.99) (Table S4, model 4). Adjustment for stunting at 4.5 years of age also weakened the association between urinary lead at 10 years of age and earlier menarche (Table S5, model 4). Instead, the association between maternal erythrocyte arsenic and later menarche was not affected by adjusting for stunting at 4.5 years of age (Table S6, model 4). Following additional adjustment for micronutrient and food supplementation groups during pregnancy, all the HRs remained unchanged (Tables S4–S6, model 5).
Associations of Exposure to Cadmium, Lead, and Arsenic and Tanner Developmental Stages
There was a statistically significant difference in urinary cadmium concentrations at 10 years of age between the girls at different stages of both breast and pubic hair development (Table S7). Girls assessing themselves to more advanced breast and pubic hair developmental Tanner stages had lower concentrations of urinary cadmium at 10 years of age. We also observed a significant difference in maternal erythrocyte lead concentrations by stages of breast development and in urinary lead concentrations at 5 years of age by stages of pubic hair development. Girls assessing themselves to more advanced breast developmental Tanner stages were born to mothers with higher erythrocyte lead concentrations during pregnancy, and girls assessing themselves to more advanced pubic hair developmental Tanner stages had higher concentrations of urinary lead at 5 years of age. No statistically significant difference in concentrations of urinary lead at 10 years of age could be observed between girls at different stages of either breast or pubic hair development. There was a statistically significant difference in concentrations of maternal erythrocyte arsenic between girls at different stages of breast development, as well as in concentrations of urinary arsenic at 10 years of age by stages of pubic hair development. The erythrocyte arsenic concentrations during pregnancy were lower in mothers of girls assessing themselves to more advanced breast developmental Tanner stages, and the urinary arsenic concentrations at 10 years of age were lower in girls with more advanced pubic hair development.
In ordered logistic regression models, neither gestational cadmium exposure, nor early childhood exposure (until 5 years of age) was associated with Tanner stages (Table S8). Girls belonging to the highest quartile of cadmium exposure at 10 years of age had lower odds of reaching an advanced stage of breast development (OR= 0.63; 95% CI: 0.40, 0.99) than the girls in the lowest quartile. Although maternal erythrocyte lead concentrations were not associated with Tanner stages in the daughters, the girls in the third and fourth quartiles of urinary lead concentrations at 5 years of age had higher odds of being at a more advanced stage of pubic hair development than the girls in the first quartile (Table S9). Moreover, girls belonging to the highest quartile of urinary lead exposure at 10 years of age had higher odds of reaching a more advanced breast development stage (OR= 1.62; 95% CI: 1.03, 2.54) than girls in the lowest quartile of exposure. We found no statistically significant associations between maternal or childhood arsenic exposure and Tanner stages in ordered logistic regression models (Table S10). However, there was a statistically significant trend between erythrocyte arsenic concentrations and more advanced breast development.
Discussion
The first menstruation does not only change girls’ lives during adolescence, but its timing also has the potential of influencing their health for decades to come.2–5 In this large prospective cohort study of 935 girls, long-term elevated exposure to cadmium during childhood, determined as concentrations in urine at 5 and 10 years of age, was associated with a delay of menarche of ∼3–4 months. There was a suggested association between ongoing lead exposure at 10 years of age, but not at 5 years of age or prenatally, and earlier menarche. Elevated maternal exposure to arsenic in early pregnancy was associated with later menarche in the daughters.
The finding of later menarche in girls with the highest long-term cadmium exposure during childhood (median urinary cadmium median∼0.5μg/L in the highest quartile at both 5 and 10 years of age), which still implies low-level exposure, is consistent with the observation that urinary cadmium at 10 years of age was associated with later breast development at the second puberty follow-up. It is also in agreement with the results of two previous longitudinal studies. The first study20 included 132 Mexican girls, who were 8–13 years of age at baseline and 14–18 years of age at the follow-up, and whose urinary cadmium concentration at baseline (median=0.1μg/L) was approximately one-half of that in the present study (median=0.23μg/L at 5 years of age and 0.25μg/L at 10 years of age). Similar to our findings, they found that the girls’ peripubertal urinary cadmium concentration was associated with later menarche, whereas maternal urinary cadmium during pregnancy showed no association.20 The other study was conducted in the United States and included 211 girls, 11–13 years of age at baseline, who were followed for up to 2 y.21 The reported urinary cadmium concentrations were similar (mean 0.26μg/L) to those we observed, and elevated concentrations were associated with later menarche. However, while we compensated for the variation in urine dilution by adjusting for specific gravity, this U.S. study adjusted either the measured urinary concentrations or the Cox regression models for urinary creatinine. Creatinine adjustment may not be optimal for growing children and teenagers,46 given that creatinine is associated with muscle mass, which is likely to increase during the peripubertal growth spurt.55 Therefore, creatinine adjustment can lead to an underestimation of the exposure in more-developed children. Indeed, they found that the creatinine-adjusted urinary cadmium concentrations decreased with increasing age.21 Because urinary cadmium reflects the accumulation in the renal cortex,41 it usually increases with age, as found in the present study. The two previous studies also adjusted the models for the girls’ body fat mass21 or the BMI Z-score around puberty20 although an increment in fat mass, like growth, is part of normal pubertal development.52 We adjusted for stunting at 4.5 years of age in our sensitivity analyses to evaluate whether the association under study was mediated through a cadmium-related decrease in growth during early childhood.23,29,30 Still, this adjustment did not alter the results. This suggests that cadmium may affect puberty onset through other modes of action, possibly through endocrine disruption, as discussed further below.
Animal studies on cadmium and reproductive maturity have yielded inconsistent results, probably due to differences in doses and mode of administration.17,56–59 However, some of the experimental studies that administered cadmium orally have also reported later reproductive maturity in rats.56–58 Later vaginal opening and an extended estrous cycle were observed in young females exposed after weaning58 and in female offspring following exposure of the pregnant dams.56,57 In contrast to the latter experimental studies, we did not observe any association between maternal erythrocyte cadmium during pregnancy and age at menarche. This could partly be due to the fact that cadmium accumulates in the human placenta with only little passing over to the fetus,60 whereas it has been observed to readily pass through the rodent placenta, although it cannot be excluded that the difference is dose dependent.61
Cadmium has been reported to have estrogen-mimicking properties both in vitro and in vivo,17,62–64 and could, therefore, be expected to result in earlier menarche. However, studies concerning the relationship between cadmium exposure and the hormones regulating puberty and menarche onset are scarce. The Mexican study mentioned above reported an association between peripubertal urinary cadmium and later menarche but did not find any associations between cadmium exposure and concentrations of pubertal hormones, including estradiol, although there was a tendency of inverse association with inhibin B, suggestive of pubertal delay.20 A National Health and Nutrition Examination Survey (NHANES) III study also reported that girls with the combination of urinary cadmium and blood lead above the median values had lower inhibin B.16 However, the girls included in that study were very young (6–11 years of age, n=260) and a third of them had cadmium concentrations below the limit of detection. Some animal studies have found a decrease in serum testosterone, estradiol, and progesterone following cadmium exposure, as well as increased lipid peroxidation and decreased antioxidant enzymes in the ovaries.57,58,65 Others have reported an activation of steroidogenesis and increased serum estradiol and progesterone concentrations in the female offspring of cadmium-treated pregnant dams.66 These conflicting results emphasize the need for further research on cadmium exposure and hormonal levels before and during reproductive maturation in large epidemiological studies with girls in the right developmental window.
Exposure to lead prenatally and during childhood has repeatedly been reported to be associated with later menarche,11,12,14,15 whereas a large (n=918) longitudinal study in the UK found no association between prenatal lead exposure and timing of menarche.67 Instead, we found indications that the girls in the highest quartile of lead exposure at 10 years of age reached menarche earlier than those in the lowest quartile. This was consistent with the association of more advanced breast development in the same girls. We observed an association between urinary lead at 5 years of age, but not at 10 years of age, and more advanced pubic hair development. However, because Tanner stages were self-assessed and less reliable than data on menarche, these findings should be interpreted with caution. Furthermore, we cannot exclude that the indicated association of urinary lead at 10 years of age and menarche might be due to reverse causality. During rapid pubertal growth, which may even precede menarche, the increased bone tissue turnover may release lead, which accumulates in bone,10 thus leading to increased urinary lead concentrations. The lack of association between age at menarche and urinary lead at 5 years of age, that is, before the peripubertal growth spurt occurs, is consistent with this hypothesis. Furthermore, urinary lead, which was used to assess the girls’ childhood exposure, is not an optimal biomarker for lead exposure. It is a short-time marker and can, therefore, be susceptible to day-to-day variation.68 In addition, the urinary lead concentrations at 5 years of age were about twice those at 10 years of age, indicating changes in childhood exposure over time.
Our result of an association between elevated exposure to inorganic arsenic in early pregnancy and later menarche in the daughters is in accordance with the previous study in this cohort, showing an association between elevated arsenic concentrations in drinking water consumed in pregnancy and later menarche in the daughters.25 In both studies, the association of later menarche was observed only in girls born to mothers belonging to the highest quartile of arsenic exposure (median erythrocyte arsenic =16μg/kg in this study). Besides the advantage that arsenic exposure was measured through biomarkers in the present study, it was also measured both during the mother’s pregnancy and twice during childhood. Interestingly, we did not observe any association with the girls’ own arsenic exposure in childhood. Although urinary arsenic is a short-term exposure biomarker, we found a fairly strong correlation between the concentrations at 5 and 10 years of age (median concentrations =56μg/L and 54μg/L, rho=0.57), indicating quite stable exposure over time on a group level, as shown previously.38 However, at an individual level, the exposure may fluctuate because the children use different water sources at home and at school. It appears difficult to mitigate the elevated arsenic exposure through drinking water and there is additional exposure through rice, the main staple food.38
It can be speculated that the influence of arsenic exposure on menarche may be due to an epigenetic effect by arsenic during fetal life, previously documented in this cohort.69,70 Animal studies have reported contradictory results. Studies in rats exposed to arsenic pre- and postnatally have reported later puberty and lower estrogen levels,71–73 whereas studies of prenatal exposure to arsenic in mice have indicated an earlier puberty onset.74,75 Given the conflicting evidence, and the many millions of people exposed to arsenic worldwide, further research on arsenic-related effects on timing of puberty onset is warranted.
The strengths of this study include the large number of participating girls compared with prior studies on menarche and the longitudinal design with individual exposure assessment of all metals at two time points in childhood (each with narrow age span) and in the mothers during pregnancy. Furthermore, the timing of the two puberty follow-ups, one at 13.3 years of age and the other at 13.8 years of age, resulted in short periods between menarche and the data collection, on average only 0.8 y, minimizing recall bias. Menarche is a relatively late marker of sexual maturation, but an acknowledged and reliable indicator of the onset of puberty. It is a significant milestone for girls and highly correlated with age at thelarche (the time of first breast development), the first sign of female puberty.52 Another strength is that the ranges of exposure to cadmium, lead, and arsenic were quite wide. The 5th–95th percentiles of urinary cadmium at 10 years of age were 0.09–0.72μg/L, thus encompassing concentrations reported in children and pregnant women living in widely different settings around the world.20,21,76,77 Cadmium exposure occurs predominantly through food, especially cereals and root vegetables because the plants easily take up cadmium from the soil,18 and tobacco smoking.18 Therefore, the observed associations regarding cadmium may be particularly relevant for the many millions of girls who consume a rice-based diet.
This study also has several limitations. Foremost, there were no measures of concentrations of lead in blood of the children. Another limitation is that data on age at menarche of the mothers was unavailable. Maternal menarche would have been an important factor to consider, given that it has been reported that genetics account for around half of the variability in age at menarche.6 Moreover, although we knew that all the mothers were nonsmokers, we did not have data on other family members’ smoking habits. Such exposure could result in unmeasured confounding, given that prenatal and childhood exposure to passive smoke has previously been associated with an earlier onset of puberty.78,79 Although tobacco smoking is a source of cadmium and lead, exposure to passive smoking is not associated with higher cadmium exposure in children.80 However, it may contribute to children’s exposure to lead.81 The role of air pollution due to indoor cooking, which is frequent in the study area, in puberty development is still largely unknown. Another limitation is that we were not able to adjust for other environmental pollutants such as polycyclic aromatic hydrocarbons (PAHs) and dichlorodiphenyltrichloroethane (DDT), which have previously been associated with altered pubertal timing,82–84 and to which the participants may have been exposed to via indoor air pollution or food. We also did not have data on when the spot urine samples were collected during the day, although the possible exposure misclassification would bias our results toward the null hypothesis.
In conclusion, we found that exposure to cadmium during childhood, but not gestational exposure, was associated with later menarche. We could not ascertain an association between maternal or childhood exposure to lead and age at menarche. In addition, elevated arsenic exposure during pregnancy was associated with later menarche in the daughters. More research is needed to assess whether the found shift in menarche might play a role in the health effects of cadmium and arsenic exposure later in life.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank all the pregnant women and their daughters who participated in this study, as well as the health care workers and all other staff in Bangladesh involved in data collection during all these years. We also thank the personnel involved in the metal analysis at Karolinska Institutet, Sweden.
This work was funded by the Swedish Research Council [grants 2018-04303 (to M.K.), 2017-01172 (to M.V.), 2015-03206 (to M.V.), 2015-03655 (to M.K.), and 2013-2269 (to M.V.)], the Swedish International Development Cooperation Agency [Sida; grants 2002-067, 2003-201A, and 2015-03206 (all to M.V.)], Karolinska Institutet and the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B). The Maternal and Infant Nutrition Interventions trial in Matlab (MINIMat), a supplementation trial during pregnancy, was funded by the Sida, Uppsala University, United Nations Children’s Fund, UK Medical Research Council, Swedish Research Council, UK Department for International Development, ICDDR,B, Global Health Research Fund Japan, Child Health and Nutrition Research Initiative, and U.S. Agency for International Development.
==== Refs
References
1. Pinilla L, Aguilar E, Dieguez C, Millar RP, Tena-Sempere M. 2012. Kisspeptins and reproduction: physiological roles and regulatory mechanisms. Physiol Rev 92 (3 ):1235–1316, PMID: , 10.1152/physrev.00037.2010.22811428
2. Kim Y, Je Y. 2019. Early menarche and risk of metabolic syndrome: a systematic review and meta-analysis. J Womens Health (Larchmt) 28 (1 ):77–86, PMID: , 10.1089/jwh.2018.6998.30285527
3. Lee JJ, Cook-Wiens G, Johnson BD, Braunstein GD, Berga SL, Stanczyk FZ, et al. 2019. Age at menarche and risk of cardiovascular disease outcomes: findings From the National Heart Lung and Blood Institute-sponsored Women’s Ischemia Syndrome Evaluation. J Am Heart Assoc 8 (12 ):e012406, PMID: , 10.1161/JAHA.119.012406.31165670
4. Werneck AO, Coelho-e-Silva MJ, Padilha CS, Ronque ERV, Cyrino ES, Szwarcwald CL, et al. 2018. Age at menarche and cancer risk at adulthood. Ann Hum Biol 45 (4 ):369–372, PMID: , 10.1080/03014460.2018.1470670.29807473
5. Bonjour JP, Chevalley T. 2014. Pubertal timing, bone acquisition, and risk of fracture throughout life. Endocr Rev 35 (5 ):820–847, PMID: , 10.1210/er.2014-1007.25153348
6. Towne B, Czerwinski SA, Demerath EW, Blangero J, Roche AF, Siervogel RM. 2005. Heritability of age at menarche in girls from the Fels Longitudinal Study. Am J Phys Anthropol 128 (1 ):210–219, PMID: , 10.1002/ajpa.20106.15779076
7. Toro CA, Aylwin CF, Lomniczi A. 2018. Hypothalamic epigenetics driving female puberty. J Neuroendocrinol 30 (7 ):e12589, PMID: , 10.1111/jne.12589.29520866
8. Schoeters G, Den Hond E, Dhooge W, van Larebeke N, Leijs M. 2008. Endocrine disruptors and abnormalities of pubertal development. Basic Clin Pharmacol Toxicol 102 (2 ):168–175, PMID: , 10.1111/j.1742-7843.2007.00180.x.18226071
9. Lopez-Rodriguez D, Franssen D, Heger S, Parent AS. 2021. Endocrine-disrupting chemicals and their effects on puberty. Best Pract Res Clin Endocrinol Metab 35 (5 ):101579, PMID: , 10.1016/j.beem.2021.101579.34563408
10. EFSA (European Food Safety Authority). 2013. Scientific opinion on lead in food. EFSA J 8 (4 ):1570, 10.2903/j.efsa.2010.1570.
11. Jansen EC, Zhou L, Song PXK, Sánchez BN, Mercado A, Hu H, et al. 2018. Prenatal lead exposure in relation to age at menarche: results from a longitudinal study in Mexico City. J Dev Orig Health Dis 9 (4 ):467–472, PMID: , 10.1017/S2040174418000223.29706142
12. Liu Y, Téllez-Rojo MM, Sánchez BN, Zhang Z, Afeiche MC, Mercado-García A, et al. 2019. Early lead exposure and pubertal development in a Mexico City population. Environ Int 125 :445–451, PMID: , 10.1016/j.envint.2019.02.021.30763831
13. Nkomo P, Richter LM, Kagura J, Mathee A, Naicker N, Norris SA. 2018. Environmental lead exposure and pubertal trajectory classes in South African adolescent males and females. Sci Total Environ 628–629 :1437–1445, PMID: , 10.1016/j.scitotenv.2018.02.150.30045563
14. Wu T, Buck GM, Mendola P. 2003. Blood lead levels and sexual maturation in U.S. girls: the Third National Health and Nutrition Examination Survey, 1988-1994. Environ Health Perspect 111 (5 ):737–741, PMID: , 10.1289/ehp.6008.12727603
15. Selevan SG, Rice DC, Hogan KA, Euling SY, Pfahles-Hutchens A, Bethel J. 2003. Blood lead concentration and delayed puberty in girls. N Engl J Med 348 (16 ):1527–1536, PMID: , 10.1056/NEJMoa020880.12700372
16. Gollenberg AL, Hediger ML, Lee PA, Himes JH, Louis GMB. 2010. Association between lead and cadmium and reproductive hormones in peripubertal U.S. girls. Environ Health Perspect 118 (12 ):1782–1787, PMID: , 10.1289/ehp.1001943.20675266
17. Johnson MD, Kenney N, Stoica A, Hilakivi-Clarke L, Singh B, Chepko G, et al. 2003. Cadmium mimics the in vivo effects of estrogen in the uterus and mammary gland. Nat Med 9 (8 ):1081–1084, PMID: , 10.1038/nm902.12858169
18. EFSA. 2009. Scientific opinion of the Panel on Contaminants in the Food Chain on a request from the European Commission on cadmium in food. EFSA J 7 (3 ):980, 10.2903/j.efsa.2009.980.
19. Sims EK, Addo OY, Gollenberg AL, Himes JH, Hediger ML, Lee PA. 2012. Inhibin B and luteinizing hormone levels in girls aged 6–11 years from NHANES III, 1988–1994. Clin Endocrinol (Oxf) 77 (4 ):555–563, PMID: , 10.1111/j.1365-2265.2012.04393.x.22443272
20. Ashrap P, Sánchez BN, Téllez-Rojo MM, Basu N, Tamayo-Ortiz M, Peterson KE, et al. 2019. In utero and peripubertal metals exposure in relation to reproductive hormones and sexual maturation and progression among girls in Mexico City. Environ Res 177 :108630, PMID: , 10.1016/j.envres.2019.108630.31421446
21. Reynolds P, Canchola AJ, Duffy CN, Hurley S, Neuhausen SL, Horn-Ross PL, et al. 2020. Urinary cadmium and timing of menarche and pubertal development in girls. Environ Res 183 :109224, PMID: , 10.1016/j.envres.2020.109224.32085996
22. Chen X, Zhu G, Jin T. 2017. Effects of cadmium exposure on age of menarche and menopause. Toxics 6 (1 ):6, PMID: , 10.3390/toxics6010006.29280952
23. Kippler M, Tofail F, Gardner R, Rahman A, Hamadani JD, Bottai M, et al. 2012. Maternal cadmium exposure during pregnancy and size at birth: a prospective cohort study. Environ Health Perspect 120 (2 ):284–289, PMID: , 10.1289/ehp.1103711.21862444
24. Bergkvist C, Kippler M, Hamadani JD, Grandér M, Tofail F, Berglund M, et al. 2010. Assessment of early-life lead exposure in rural Bangladesh. Environ Res 110 (7 ):718–724, PMID: , 10.1016/j.envres.2010.07.004.20656285
25. Rahman A, Kippler M, Pervin J, Tarafder C, Lucy IJ, Svefors P, et al. 2021. A cohort study of the association between prenatal arsenic exposure and age at menarche in a rural area, Bangladesh. Environ Int 154 :106562, PMID: , 10.1016/j.envint.2021.106562.33866057
26. Persson LÅ, Arifeen S, Ekström EC, Rasmussen KM, Frongillo EA, Yunus M, et al. 2012. Effects of prenatal micronutrient and early food supplementation on maternal hemoglobin, birth weight, and infant mortality among children in Bangladesh: the MINIMat randomized trial. JAMA 307 (19 ):2050–2059, PMID: , 10.1001/jama.2012.4061.22665104
27. Arifeen SE, Ekström EC, Frongillo EA, Hamadani J, Khan AI, Naved RT, et al. 2018. Cohort profile: the Maternal and Infant Nutrition Interventions in Matlab (MINIMat) cohort in Bangladesh. Int J Epidemiol 47 (6 ):1737–1738e, PMID: , 10.1093/ije/dyy102.29868907
28. Svefors P, Pervin J, Islam Khan A, Rahman A, Ekström EC, El Arifeen S, et al. 2020. Stunting, recovery from stunting and puberty development in the MINIMat cohort, Bangladesh. Acta Paediatr 109 (1 ):122–133, PMID: , 10.1111/apa.14929.31283046
29. Gardner RM, Kippler M, Tofail F, Bottai M, Hamadani J, Grandér M, et al. 2013. Environmental exposure to metals and children’s growth to age 5 years: a prospective cohort study. Am J Epidemiol 177 (12 ):1356–1367, PMID: , 10.1093/aje/kws437.23676282
30. Malin Igra A, Warnqvist A, Rahman SM, Ekström EC, Rahman A, Vahter M, et al. 2021. Environmental metal exposure and growth to 10 years of age in a longitudinal mother–child cohort in rural Bangladesh. Environ Int 156 :106738, PMID: , 10.1016/j.envint.2021.106738.34246127
31. Gustin K, Tofail F, Vahter M, Kippler M. 2018. Cadmium exposure and cognitive abilities and behavior at 10 years of age: a prospective cohort study. Environ Int 113 :259–268, PMID: , 10.1016/j.envint.2018.02.020.29459184
32. Vahter M, Skröder H, Rahman SM, Levi M, Derakhshani Hamadani J, Kippler M. 2020. Prenatal and childhood arsenic exposure through drinking water and food and cognitive abilities at 10 years of age: a prospective cohort study. Environ Int 139 :105723, PMID: , 10.1016/j.envint.2020.105723.32298878
33. Skröder H, Kippler M, Tofail F, Vahter M. 2017. Early-life selenium status and cognitive function at 5 and 10 years of age in Bangladeshi children. Environ Health Perspect 125 (11 ):117003, PMID: , 10.1289/EHP1691.29116931
34. Malin Igra A, Vahter M, Raqib R, Kippler M. 2019. Early-life cadmium exposure and bone-related biomarkers: a longitudinal study in children. Environ Health Perspect 127 (3 ):37003, PMID: , 10.1289/EHP3655.30848671
35. Raqib R, Ahmed S, Ahsan KB, Kippler M, Akhtar E, Roy AK, et al. 2017. Humoral immunity in arsenic-exposed children in rural Bangladesh: total immunoglobulins and vaccine-specific antibodies. Environ Health Perspect 125 (6 ):067006, PMID: , 10.1289/EHP318.28657894
36. Kippler M, Nermell B, Hamadani J, Tofail F, Moore S, Vahter M. 2010. Burden of cadmium in early childhood: longitudinal assessment of urinary cadmium in rural Bangladesh. Toxicol Lett 198 (1 ):20–25, PMID: , 10.1016/j.toxlet.2010.04.029.20466048
37. Vahter ME, Li L, Nermell B, Rahman A, El Arifeen S, Rahman M, et al. 2006. Arsenic exposure in pregnancy: a population-based study in Matlab, Bangladesh. J Health Popul Nutr 24 (2 ):236–245, PMID: .17195565
38. Kippler M, Skröder H, Rahman SM, Tofail F, Vahter M. 2016. Elevated childhood exposure to arsenic despite reduced drinking water concentrations—a longitudinal cohort study in rural Bangladesh. Environ Int 86 :119–125, PMID: , 10.1016/j.envint.2015.10.017.26580026
39. Kippler M, Goessler W, Nermell B, Ekström EC, Lönnerdal B, El Arifeen S, et al. 2009. Factors influencing intestinal cadmium uptake in pregnant Bangladeshi women—a prospective cohort study. Environ Res 109 (7 ):914–921, PMID: , 10.1016/j.envres.2009.07.006.19646688
40. Nève J. 1995. Human selenium supplementation as assessed by changes in blood selenium concentration and glutathione peroxidase activity. J Trace Elem Med Biol 9 (2 ):65–73, PMID: , 10.1016/S0946-672X(11)80013-1.8825978
41. Akerstrom M, Barregard L, Lundh T, Sallsten G. 2013. The relationship between cadmium in kidney and cadmium in urine and blood in an environmentally exposed population. Toxicol Appl Pharmacol 268 (3 ):286–293, PMID: , 10.1016/j.taap.2013.02.009.23454399
42. Gardner RM, Nermell B, Kippler M, Grandér M, Li L, Ekström EC, et al. 2011. Arsenic methylation efficiency increases during the first trimester of pregnancy independent of folate status. Reprod Toxicol 31 (2 ):210–218, PMID: , 10.1016/j.reprotox.2010.11.002.21078382
43. Skröder Löveborn H, Kippler M, Lu Y, Ahmed S, Kuehnelt D, Raqib R, et al. 2016. Arsenic metabolism in children differs from that in adults. Toxicol Sci 152 (1 ):29–39, PMID: , 10.1093/toxsci/kfw060.27056082
44. Lu Y, Kippler M, Harari F, Grandér M, Palm B, Nordqvist H, et al. 2015. Alkali dilution of blood samples for high throughput ICP-MS analysis—comparison with acid digestion. Clin Biochem 48 (3 ):140–147, PMID: , 10.1016/j.clinbiochem.2014.12.003.25498303
45. Nermell B, Lindberg AL, Rahman M, Berglund M, Persson LA, El Arifeen S, et al. 2008. Urinary arsenic concentration adjustment factors and malnutrition. Environ Res 106 (2 ):212–218, PMID: , 10.1016/j.envres.2007.08.005.17900556
46. De Craemer S, Croes K, van Larebeke N, De Henauw S, Schoeters G, Govarts E, et al. 2017. Metals, hormones and sexual maturation in Flemish adolescents in three cross-sectional studies (2002–2015). Environ Int 102 :190–199, PMID: , 10.1016/j.envint.2017.02.014.28318602
47. Marshall WA, Tanner JM. 1969. Variations in pattern of pubertal changes in girls. Arch Dis Child 44 (235 ):291–303, PMID: , 10.1136/adc.44.235.291.5785179
48. Gwatkin DR, Rustein S, Johnson K, Pande RP, Wagstaff A. 2000. Socio-Economic Differences in Health, Nutrition, and Population in Bangladesh. Washington, DC: World Bank.
49. Rahman A, Vahter M, Smith AH, Nermell B, Yunus M, El Arifeen S, et al. 2009. Arsenic exposure during pregnancy and size at birth: a prospective cohort study in Bangladesh. Am J Epidemiol 169 (3 ):304–312, PMID: , 10.1093/aje/kwn332.19037006
50. Svefors P, Rahman A, Ekström EC, Khan AI, Lindström E, Persson LÅ, et al. 2016. Stunted at 10 years. Linear growth trajectories and stunting from birth to pre-adolescence in a rural Bangladeshi cohort. PLoS One 11 (3 ):e0149700, PMID: , 10.1371/journal.pone.0149700.26934484
51. Textor J, van der Zander B, Gilthorpe MS, Liskiewicz M, Ellison GT. 2016. Robust causal inference using directed acyclic graphs: the R package ‘dagitty.’ Int J Epidemiol 45 (6 ):1887–1894, PMID: , 10.1093/ije/dyw341.28089956
52. Tanner JM. 1962. Growth at Adolescence. 2nd ed. Oxford, UK: Blackwell Scientific.
53. Gleason KM, Valeri L, Shankar AH, Hasan MOSI, Quamruzzaman Q, Rodrigues EG, et al. 2016. Stunting is associated with blood lead concentration among Bangladeshi children aged 2-3 years. Environ Health 15 (1 ):103, PMID: , 10.1186/s12940-016-0190-4.27809911
54. WHO (World Health Organization). 2006. WHO Child Growth Standards. Geneva, Switzerland: WHO. https://www.who.int/tools/child-growth-standards [accessed 18 February 2018].
55. Fuse Y, Ito Y, Shishiba Y, Irie M. 2022. Current iodine status in Japan: a cross-sectional nationwide survey of schoolchildren, 2014–2019. J Clin Endocrinol Metab 107 (5 ):e2065–e2079, PMID: , 10.1210/clinem/dgab919.34935958
56. Salvatori F, Talassi CB, Salzgeber SA, Spinosa HS, Bernardi MM. 2004. Embryotoxic and long-term effects of cadmium exposure during embryogenesis in rats. Neurotoxicol Teratol 26 (5 ):673–680, PMID: , 10.1016/j.ntt.2004.05.001.15315816
57. Samuel JB, Stanley JA, Princess RA, Shanthi P, Sebastian MS. 2011. Gestational cadmium exposure-induced ovotoxicity delays puberty through oxidative stress and impaired steroid hormone levels. J Med Toxicol 7 (3 ):195–204, PMID: , 10.1007/s13181-011-0143-9.21373971
58. Saedi S, Shirazi MRJ, Zamiri MJ, Totonchi M, Dadpasand M, Sedaghati F. 2020. Impaired follicular development and endocrine disorders in female rats by prepubertal exposure to toxic doses of cadmium. Toxicol Ind Health 36 (2 ):63–75, PMID: , 10.1177/0748233720912060.32279652
59. Höfer N, Diel P, Wittsiepe J, Wilhelm M, Degen GH. 2009. Dose- and route-dependent hormonal activity of the metalloestrogen cadmium in the rat uterus. Toxicol Lett 191 (2–3 ):123–231, PMID: , 10.1016/j.toxlet.2009.08.014.19703529
60. Osman K, Akesson A, Berglund M, Bremme K, Schütz A, Ask K, et al. 2000. Toxic and essential elements in placentas of Swedish women. Clin Biochem 33 (2 ):131–138, PMID: , 10.1016/S0009-9120(00)00052-7.10751591
61. Whelton BD, Toomey JM, Bhattacharyya MH. 1993. Cadmium-109 metabolism in mice. IV. Diet versus maternal stores as a source of cadmium transfer to mouse fetuses and pups during gestation and lactation. J Toxicol Environ Health 40 (4 ):531–546, PMID: , 10.1080/15287399309531817.8277517
62. Garcia-Morales P, Saceda M, Kenney N, Kim N, Salomon DS, Gottardis MM, et al. 1994. Effect of cadmium on estrogen receptor levels and estrogen-induced responses in human breast cancer cells. J Biol Chem 269 (24 ):16896–16901, PMID: , 10.1016/S0021-9258(19)89474-7.8207012
63. Stoica A, Katzenellenbogen BS, Martin MB. 2000. Activation of estrogen receptor-α by the heavy metal cadmium. Mol Endocrinol 14 (4 ):545–553, PMID: , 10.1210/mend.14.4.0441.10770491
64. Ali I, Penttinen-Damdimopoulou PE, Mäkelä SI, Berglund M, Stenius U, Akesson A, et al. 2010. Estrogen-like effects of cadmium in vivo do not appear to be mediated via the classical estrogen receptor transcriptional pathway. Environ Health Perspect 118 (10 ):1389–1394, PMID: , 10.1289/ehp.1001967.20525538
65. Zhang W, Pang F, Huang Y, Yan P, Lin W. 2008. Cadmium exerts toxic effects on ovarian steroid hormone release in rats. Toxicol Lett 182 (1–3 ):18–23, PMID: , 10.1016/j.toxlet.2008.07.016.18708132
66. Li Z, Li T, Leng Y, Chen S, Liu Q, Feng J, et al. 2018. Hormonal changes and folliculogenesis in female offspring of rats exposed to cadmium during gestation and lactation. Environ Pollut 238 :336–347, PMID: , 10.1016/j.envpol.2018.03.023.29574358
67. Maisonet M, Jaakkola JJ, Taylor CM, Marcus M, Golding J. 2014. Prenatal lead exposure and puberty timing in girls. Epidemiology 25 (1 ):153–155, PMID: , 10.1097/EDE.0000000000000021.24296930
68. Sommar JN, Hedmer M, Lundh T, Nilsson L, Skerfving S, Bergdahl IA. 2014. Investigation of lead concentrations in whole blood, plasma and urine as biomarkers for biological monitoring of lead exposure. J Expo Sci Environ Epidemiol 24 (1 ):51–57, PMID: , 10.1038/jes.2013.4.23443239
69. Broberg K, Ahmed S, Engström K, Hossain MB, Jurkovic Mlakar S, Bottai M, et al. 2014. Arsenic exposure in early pregnancy alters genome-wide DNA methylation in cord blood, particularly in boys. J Dev Orig Health Dis 5 (4 ):288–298, PMID: , 10.1017/S2040174414000221.24965135
70. Gliga AR, Engström K, Kippler M, Skröder H, Ahmed S, Vahter M, et al. 2018. Prenatal arsenic exposure is associated with increased plasma IGFBP3 concentrations in 9-year-old children partly via changes in DNA methylation. Arch Toxicol 92 (8 ):2487–2500, PMID: , 10.1007/s00204-018-2239-3.29947889
71. Chattopadhyay S, Ghosh S, Chaki S, Debnath J, Ghosh D. 1999. Effect of sodium arsenite on plasma levels of gonadotrophins and ovarian steroidogenesis in mature albino rats: duration-dependent response. J Toxicol Sci 24 (5 ):425–431, PMID: , 10.2131/jts.24.5_425.10656165
72. Dávila-Esqueda ME, Jiménez-Capdeville ME, Delgado JM, De la Cruz E, Aradillas-García C, Jiménez-Suárez V, et al. 2012. Effects of arsenic exposure during the pre- and postnatal development on the puberty of female offspring. Exp Toxicol Pathol 64 (1–2 ):25–30, PMID: , 10.1016/j.etp.2010.06.001.20580540
73. Reilly MP, Saca JC, Hamilton A, Solano RF, Rivera JR, Whitehouse-Innis W, et al. 2014. Prepubertal exposure to arsenic(III) suppresses circulating insulin-like growth factor-1 (IGF-1) delaying sexual maturation in female rats. Reprod Toxicol 44 :41–49, PMID: , 10.1016/j.reprotox.2013.09.005.24090629
74. Rodriguez KF, Ungewitter EK, Crespo-Mejias Y, Liu C, Nicol B, Kissling GE, et al. 2016. Effects of in utero exposure to arsenic during the second half of gestation on reproductive end points and metabolic parameters in female CD-1 mice. Environ Health Perspect 124 (3 ):336–343, PMID: , 10.1289/ehp.1509703.26295903
75. Li X, Sun Z, Manthari RK, Li M, Guo Q, Wang J. 2018. Effect of gestational exposure to arsenic on puberty in offspring female mice. Chemosphere 202 :119–126, PMID: , 10.1016/j.chemosphere.2018.03.095.29567609
76. Gustin K, Barman M, Stråvik M, Levi M, Englund-Ögge L, Murray F, et al. 2020. Low-level maternal exposure to cadmium, lead, and mercury and birth outcomes in a Swedish prospective birth-cohort. Environ Pollut 265 (pt B ):114986, PMID: , 10.1016/j.envpol.2020.114986.32585550
77. Watanabe T, Nakatsuka H, Shimbo S, Yaginuma-Sakurai K, Ikeda M. 2013. High cadmium and low lead exposure of children in Japan. Int Arch Occup Environ Health 86 (8 ):865–873, PMID: , 10.1007/s00420-012-0821-1.23079791
78. Chen Y, Liu Q, Li W, Deng X, Yang B, Huang X. 2018. Association of prenatal and childhood environment smoking exposure with puberty timing: a systematic review and meta-analysis. Environ Health Prev Med 23 (1 ):33, PMID: , 10.1186/s12199-018-0722-3.30021511
79. Windham GC, Lum R, Voss R, Wolff M, Pinney SM, Teteilbaum SL, et al. 2017. Age at pubertal onset in girls and tobacco smoke exposure during pre- and postnatal susceptibility windows. Epidemiology 28 (5 ):719–727, PMID: , 10.1097/EDE.0000000000000704.28661938
80. Willers S, Attewell R, Bensryd I, Schutz A, Skarping G, Vahter M. 1992. Exposure to environmental tobacco smoke in the household and urinary cotinine excretion, heavy metals retention, and lung function. Arch Environ Health 47 (5 ):357–363, PMID: , 10.1080/00039896.1992.9938375.1444598
81. Apostolou A, Garcia-Esquinas E, Fadrowski JJ, McLain P, Weaver VM, Navas-Acien A. 2012. Secondhand tobacco smoke: a source of lead exposure in US children and adolescents. Am J Public Health 102 (4 ):714–722, PMID: , 10.2105/AJPH.2011.300161.21852639
82. Kehm RD, Oskar S, Tehranifar P, Zeinomar N, Rundle AG, Herbstman JB, et al. 2021. Associations of prenatal exposure to polycyclic aromatic hydrocarbons with pubertal timing and body composition in adolescent girls: implications for breast cancer risk. Environ Res 196 :110369, PMID: , 10.1016/j.envres.2020.110369.33131678
83. Dobraca D, Laurent CA, Greenspan LC, Hiatt RA, Sjödin A, Kushi LH, et al. 2020. Urinary polycyclic aromatic hydrocarbons in relation to anthropometric measures and pubertal development in a cohort of Northern California girls. Environ Epidemiol 4 (4 ):e0102, PMID: , 10.1097/EE9.0000000000000102.32832841
84. Ouyang F, Perry MJ, Venners SA, Chen C, Wang B, Yang F, et al. 2005. Serum DDT, age at menarche, and abnormal menstrual cycle length. Occup Environ Med 62 (12 ):878–884, PMID: , 10.1136/oem.2005.020248.16299097
|
PMC009xxxxxx/PMC9895304.txt
|
==== Front
Workplace Health Saf
Workplace Health Saf
WHS
spwhs
Workplace Health & Safety
2165-0799
2165-0969
SAGE Publications Sage CA: Los Angeles, CA
36708021
10.1177/21650799221147814
10.1177_21650799221147814
Research Articles
Exploring University and Healthcare Workers’ Physical Activity, Diet, and Well-Being During the COVID-19 Pandemic
https://orcid.org/0000-0001-6948-0127
Gilbert Amanda MSW, MPH 1
Eyler Amy PhD, CHES 1
Cesarone Gabriella BA 1
https://orcid.org/0000-0002-3576-5906
Harris Jenine PhD 1
https://orcid.org/0000-0002-7586-8037
Hayibor Lisa MD, MPH 1
Evanoff Bradley MD, MPH 12
1 Washington University
2 University of Iowa
Amanda Gilbert, MSW, MPH, Prevention Research Center, Brown School, Washington University, One Brookings Drive MSC 1196-257-220, St. Louis, MO 63130, USA; email: [email protected].
27 1 2023
8 2023
27 1 2023
71 8 384394
© 2023 The Author(s)
2023
American Association of Occupational Health Nurses
This article is made available via the PMC Open Access Subset for unrestricted re-use and analyses in any form or by any means with acknowledgement of the original source. These permissions are granted for the duration of the COVID-19 pandemic or until permissions are revoked in writing. Upon expiration of these permissions, PMC is granted a perpetual license to make this article available via PMC and Europe PMC, consistent with existing copyright protections.
Background:
The COVID-19 pandemic affected well-being and health behaviors, especially among healthcare workers and employees in other fields. This is of public health concern because health behaviors and well-being influence long-term negative health outcomes. The purpose of this study was to explore health behaviors and well-being among university and medical center staff during COVID-19.
Methods:
EMPOWER (Employee Well-being during Epidemic Response) was a three-wave observational study (wave 1: 1,994; wave 2: 1,426; wave 3: 1,363) measuring health behaviors and well-being of university and medical center staff. Surveys were disseminated online to all employees between April and September 2020. Descriptive statistics explored trends across waves for health behaviors (physical activity [PA], diet), and well-being (mental well-being [MWB], depression, anxiety, and stress). Logistic regressions explored associations between health behaviors and well-being factors adjusting for demographics and clinical role. Interactions explored moderation by clinical role.
Results:
Most participants reported same/healthier changes in PA (54–65%) and diet (57–73%) and decreased MWB across waves (62%–69%). Nonclinical workers were less likely than clinical workers to experience worse MWB and moderate/severe anxiety and stress (odds ratios [ORs] ranged from 0.38 to 0.58 across waves and well-being outcomes). Participants who maintained/increased PA and diet were less likely to experience worse well-being (ORs ranged from 0.44 to 0.69 across waves and well-being outcomes). Interactions by clinical role were not significant.
Conclusion/Application to Practice:
Maintaining/increasing health behaviors during COVID-19 may be protective of mental health/well-being in some healthcare workers. These findings support health promotion efforts focused on maintaining or improving diet and PA.
workplace
clinical
fitness
nutrition
population health
National Heart, Lung, and Blood Institute https://doi.org/10.13039/100000050 T32 HL130357 Centers for Disease Control and Prevention https://doi.org/10.13039/100000030 U19OH008868 Centers for Disease Control and Prevention https://doi.org/10.13039/100000030 U48DP006395 National Center for Advancing Translational Sciences https://doi.org/10.13039/100006108 UL1TR002345 typesetterts1
==== Body
pmcBackground
From the start of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus (COVID-19) pandemic in December of 2019 through July of 2021, around 35 million cases and over 600,000 deaths were confirmed in the United States (The Centers for Disease Control and Prevention, 2020). This high level of COVID-19 illness and death along with public health measures such as stay-at-home orders, mask mandates, and closures of nonessential businesses to prevent the spread of COVID-19, have had significant impacts on daily life and well-being.
Throughout the pandemic, public health restrictions, such as stay-at-home orders, led to limited in-person contact and disruption to daily work, causing increased feelings of loneliness, anxiety, and stress (Brooks et al., 2020; Jewell et al., 2020; Mukhtar, 2020; Newby et al., 2020; Tull et al., 2020). One study by Daly et al. found the percentage of U.S. adults with depression significantly increased from around 9% prior to the pandemic to a little over 14% during the pandemic in the spring of 2020 (Daly et al., 2021). COVID-19 illnesses and deaths have also contributed to psychological and emotional distress, particularly among healthcare workers who have been at an increased risk of exposure to COVID-19 and often been responsible for treating severe COVID-19 cases (Hennein & Lowe, 2020; Sharma et al., 2021; Thomaier et al., 2020). These disruptions to daily life have also led to changes in health behaviors such as physical activity and diet. Limited access to exercise facilities, changes in daily routine, and increased work and parenting responsibilities have decreased levels of physical activity, while increased time at home has changed eating behaviors, resulting in higher calorie consumption, fewer fruits and vegetables, and more snacking (Barkley et al., 2020; Dwyer et al., 2020; Farah et al., 2021; Martinez-Ferran et al., 2020; Tison et al., 2020).
The impact of the COVID-19 pandemic on well-being and health behaviors is of public health concern. Decreased physical activity, poor nutrition, and poor mental health and well-being are all influential risk factors for chronic diseases such as obesity, cardiovascular disease, and diabetes, many types of cancer, and even for severe COVID-19 (Butler & Barrientos, 2020; Na & Oliynyk, 2011; Prince et al., 2007; Salleh, 2008; Sallis et al., 2021; U.S. Department of Health and Human Services, 2018; U.S. Department of Health and Human Services and U.S. Department of Agriculture [USDA], 2015). In addition, we know a complex relationship exists between health behaviors and mental health, wherein negative changes in one may negatively impact the other and fuel a cycle of worsening health, potentially leading to even worse health outcomes (Prince et al., 2007; Rossa-Roccor et al., 2021; Salleh, 2008).
Given this complex relationship and potential health outcomes, it is important to build on our still-limited knowledge of how the COVID-19 pandemic has influenced well-being and health behaviors. This is the case for workers both within healthcare and employed in other fields because changes to daily routine and work have been driving some of the changes in well-being and health behaviors (Brooks et al., 2020; Tull et al., 2020). We used data from the EMPOWER (Employee Well-being during Epidemic Response) study to (a) describe changes in health behaviors (physical activity and diet) and well-being outcomes (mental well-being, depression, anxiety, and stress) across three waves of data collection during the first 4 months of the COVID-19 pandemic among staff at a university and associated medical center and (b) to explore the associations between health behavior changes and well-being outcomes and whether these relationships varied by clinical role.
Methods
Study Design and Sample
The EMPOWER study was an observational study conducted at a large university and affiliated medical center from April 17, 2020 to September 8, 2020. Three waves of surveys were conducted, wave 1 (April 17, 2020—May 1, 2020), wave 2 (May 28, 2020–June 23, 2020), and wave 3 (July 20, 2020–September 8, 2020), throughout the study period (Evanoff et al., 2020). Online surveys were sent via email to all benefits-eligible employees at the university and medical school to measure mental health and well-being of faculty and staff during the COVID-19 pandemic. Employees received emails with a link to the anonymous and voluntary survey, which could be completed in 10 minutes. All participants provided written informed consent. Topics in the survey covered demographics, well-being, and health behaviors. The study was approved by the Washington University in St. Louis institutional review board.
Measures
The survey measured demographics by asking participants to describe their race (White, Black, Asian, two or more races, other), their gender (male and female), Hispanic ethnicity (yes/no), age (less than 30 years, 30–39 years, 40–49 years, 50–59 years, 60 years or older), annual household income (less than $40,000, $40,000–$80,000, More than $80,000), clinical role (onsite clinical, onsite nonclinical, working at home), and job type (nursing, clinical support/records, clinical research, basic research, accounting/finance, student/academic services, security, animal care, facilities/transportation/health and safety, administrative support, management/administration, human resources, information technology, library services, public affairs/marketing).
The survey measured health behaviors (physical activity and diet) by asking participants to answer the questions, “Compared to a month ago, my eating habits are . . .” and “Compared to a month ago my exercise habits are . . .” with one of five response options provided (much healthier, somewhat healthier, about the same, somewhat less healthy, and much less healthy). Responses were then dichotomized into same or healthier (much healthier, somewhat healthier, and about the same) and less healthy (somewhat less healthy and much less healthy).
Stress, anxiety, and depression were measured using the Depression, Anxiety, and Stress Scale-21 Items (DASS-21) (S. H. Lovibond & P. F. Lovibond, 1995). The DASS-21 is a validated measure composed of three self-reported scales measuring the presence of symptoms relating to depression, anxiety, and stress. Each scale is composed of 7 items. Participants are asked to indicate how much each item applies to them over the past week. Response options include (0—does not apply to me at all, 1—applied to me to some degree or some of the time, 2—applied to me to a considerable degree or a good part of the time, and 3—applied to me much or most of the time). Scores for each scale are calculated and recommended cut points used to rate normal, mild, moderate, severe, and extremely severe levels of depression, anxiety, and stress (Brumby et al., 2011; John & Julie, 2003). The cut points vary for depression, anxiety, and stress. For depression (normal 0–9, mild 10–13, moderate 14–20, severe 21–27, extremely severe 28+), anxiety (normal 0–7, mild 8–9, moderate 10–14, severe 15–19, extremely severe 20+), and stress (normal 0–14, mild 15–18, moderate 19–25, severe 26–33, extremely severe 37+). Responses were then dichotomized into none/mild (normal and mild) and moderate/severe (moderate, severe, and extremely severe).
Since well-being is not solely the absence of clinical symptoms, a measure of mental well-being was also included in the analysis to assess quality of life relating to psychological, emotional, and cognitive functioning (Huppert, 2014; Linton et al., 2016; Ryan & Deci, 2001). To measure mental well-being, participants were asked to report on changes in mental well-being from before the COVID-19 pandemic to the present time during the COVID-19 pandemic. The survey question, “To what extent have COVID-19 work/life changes impacted your mental well-being?” included response options ranged on a 4-point scale (much worse, somewhat worse, about the same, much better/somewhat better). Responses were then dichotomized into decreased (much worse and somewhat worse) and same or improved (about the same and much better/somewhat better).
Statistical Analysis
All statistical analyses were conducted using SAS version 9.5 (SAS Institute Inc). To compare results across the three waves of data collection, data were only included in this analysis for participants who completed at least two of the three waves. Analysis treated each wave as a cross-sectional sample of the population. Using chi-square tests, we found no significant demographic differences between the samples for each of the three waves. We computed descriptive statistics and chi-square analyses to examine the frequencies of demographics (age, race, ethnicity, gender, clinical role, and income) health behaviors (physical activity and diet) and well-being outcomes (mental well-being, depression, anxiety, and stress) and to explore associations between health behaviors and well-being outcomes. We also ran logistic regression models assessing the relationship between health behaviors and well-being outcomes. In the logistic regression models, the dependent variables were the well-being outcomes of interest (mental well-being, depression, anxiety, and stress) and the independent variables of interest were health behaviors (physical activity and diet). We adjusted the models with the covariates of age, race, income, gender, ethnicity, and clinical role. We then ran the adjusted logistic regressions using interaction terms to explore how the relationship between health behaviors and well-being outcomes might vary by clinical role. Assumptions of sample size were met for chi-square analyses and of multicollinearity for logistic regressions.
Results
The analytic sample consisted of participants who completed at least two of the three waves of the EMPOWER study (Wave 1: 1994, Wave 2: 1426 and Wave 3: 1363; Table 1). Around 85% of the sample were female. Most of the participants were White (90%) and made more than $80,000 (58%–61%). Many of the participants were between the ages of 30 and 59 years old (77%). Regarding clinical role, the majority (75%–82%) worked at home, while 11% to 14% were onsite clinical workers.
Table 1. Demographic and Occupational Characteristics of Participants in Waves 1, 2, and 3 of the EMPOWER Study
Demographics, n (%) Wave 1 (n = 1,994) Wave 2 (n = 1,426) Wave 3 (n = 1,363)
Female 1682 (84.35) 1191 (83.52) 1163 (85.08)
Race
White 1788 (89.67) 1299 (91.09) 1227 (89.76)
Black 104 (5.22) 60 (4.21) 66 (4.83)
Asian 40 (2.01) 26 (1.82) 32 (2.34)
Two or more races 41 (2.06) 25 (1.75) 30 (2.19)
Other 21 (1.05) 16 (1.12) 12 (0.88)
Hispanic/Latino 50 (2.51) 34 (2.38) 43 (3.15)
Age
Less than 30 years 239 (11.99) 160 (11.22) 163 (11.92)
30–39 years 551 (27.63) 388 (27.21) 380 (27.80)
40–49 years 517 (25.93) 383 (26.86) 367 (26.85)
50–59 years 450 (22.57) 343 (24.05) 315 (23.04)
60 years or older 237 (11.89) 152 (10.66) 142 (10.39)
Income
$40,000 or less 187 (9.38) 106 (7.43) 105 (7.68)
$40,00–$80,000 735 (32.20) 452 (31.70) 438 (32.04)
More than $80,000 1165 (58.43) 868 (60.87) 824 (60.28)
Clinical
Onsite clinical 226 (11.33) 191 (13.39) 100 (11.04)
Onsite nonclinical 142 (7.12) 162 (11.36) 91 (10.04)
Working at home 1626 (81.54) 1073 (75.25) 715 (78.92)
Job type
Nursing 165 (8.27) 120 (8.42) 96 (7.02)
Clinical support/records 116 (5.82) 81 (5.68) 92 (6.73)
Clinical research 301 (15.10) 233 (16.34) 221 (16.17)
Basic research 189 (9.48) 128 (8.98) 137 (10.02)
Accounting/finance 198 (9.93) 153 (10.73) 142 (10.39)
Student/academic services 124 (6.22) 81 (5.68) 80 (5.85)
Security 8 (0.40) 6 (0,42) 3 (0.22)
Animal care 14 (0.70) 11 (0.77) 8 (0.59)
Facilities/transportation/health and safety 56 (2.81) 38 (2.66) 34 (2.49)
Administrative support 329 (16.50) 236 (16.55) 232 (16.97)
Management/administration 167 (8.38) 110 (7.71) 103 (7.53)
Human resources 34 (1.71) 29 (2.03) 30 (2.19)
Information technology 147 (7.37) 100 (7.01) 92 (6.80)
Library services 35 (1.76) 24 (1.68) 20 (1.46)
Public affairs/marketing 111 (5.57) 76 (5.33) 76 (5.56)
Over half of participants reported the same or healthier physical activity (54%) and diet (57%) (Table 2). Across the waves, the proportion of participants reporting the same or healthier physical activity and diet than the month before increased, from 54% in wave 1 to 65% in wave 3 for physical activity and 57% to 73% from wave 1 to wave 3 for diet. The proportion of participants reporting less healthy physical activity and diet decreased from 46% in wave 1 to 35% in wave 3 for physical activity and 41% in wave 1 to 27% in wave 3 for diet.
Table 2. Frequencies of Health Behaviors and Well-Being of Participants in Waves 1, 2, and 3 of the EMPOWER Study
Health Behaviors and Well-being Wave 1 (n = 1994) Wave 2 (n = 1426) Wave 3 (n = 1363)
Health behaviors, n (%)
Physical activity
Same/Healthier 1070 (53.66) 875 (61.36) 883 (64.59)
Less healthy 924 (46.34) 551 (38.64) 484 (35.41)
Diet
Same/healthier 1189 (56.63) 969 (67.95) 997 (72.93)
Less healthy 805 (40.37) 457 (32.05) 370 (27.07)
Well-being, n (%)
Mental well-being
Not decreased 629 (31.54) 537 (37.66) 527 (38.55)
Decreased 1365 (68.46) 889 (62.34) 840 (61.54)
Depression
None/mild 1680 (84.25) 1206 (84.57) 1,139 (83.32)
Moderate-severe 314 (15.75) 220 (15.43) 228 (16.68)
Anxiety
None/mild 1736 (87.06) 1241 (87.03) 1175 (85.95)
Moderate-severe 258 (12.94) 185 (12.97) 192 (14.05)
Stress
None/mild 1735 (87.01) 1235 (86.61) 1184 (86.61)
Moderate-severe 259 (12.99) 191 (13.39) 183 (13.39)
More than half of participants reported decreased mental well-being across the three waves (62%–68%), while most participants reported none/mild depression (83%–85%), none/mild anxiety (86%–87%), and none/mild stress (87%). Overall, well-being outcomes stayed somewhat stable across the three waves. The proportion of participants reporting a decrease in mental well-being started at 68% in wave 1, dropped to 62% in wave 2, and then to 62% in wave 3. The proportion of participants reporting moderate to severe depression, stayed constant at around 15% to 17% across the three waves, as did participants reporting moderate to severe stress (13%), and moderate to severe anxiety (13%–14%).
To understand the influence of clinical role on well-being outcomes, we also examined the clinical role as a factor in multivariate analyses (Table 3). We observed that onsite nonclinical staff and those working at home were less likely to experience worse mental well-being across most waves, compared to onsite clinical staff. There were no significant associations between clinical role and depression regardless of wave. Only at wave 2 were onsite nonclinical staff and those working at home less likely to experience moderate/severe stress, compared to onsite clinical staff. Onsite nonclinical staff were less likely at wave 2 to experience moderate/severe anxiety compared to onsite clinical staff, while those working at home were less likely to experience moderate/severe anxiety compared to onsite clinical staff at all waves. There was no effect modification of clinical role on the relationship between health behaviors and well-being outcomes.
Table 3. Multivariate Analysis of Clinical Role as Correlates of Well-Being (Mental Well-Being, Depression, Anxiety, and Stress) for Participants in Waves 1, 2, and 3 of the EMPOWER Study
Well-being a Onsite nonclinical Working at home
OR b 95% CI OR b 95% CI
Decreased mental well-being wave 1 c 0.63 [0.38, 1.03] 0.64 [0.45, 0.90]
Decreased mental well-being wave 2 d 0.37 [0.22, 0.60] 0.42 [0.29, 0.62]
Decreased mental well-being wave 3 e 0.49 [0.31, 0.77] 0.38 [0.26, 0.55]
Moderate/severe depression wave 1 0.93 [0.51, 1.69] 0.91 [0.62, 1.33]
Moderate/severe depression wave 2 0.80 [0.43, 1.48] 0.95 [0.62, 1.47]
Moderate/severe depression wave 3 1.11 [0.65, 1.88] 1.02 [0.67, 1.57]
Moderate/severe anxiety wave 1 0.81 [0.44, 1.50] 0.57 [0.39, 0.84]
Moderate/severe anxiety wave 2 0.33 [0.17, 0.65] 0.45 [0.30, 0.68]
Moderate/severe anxiety wave 3 0.89 [0.53, 1.51] 0.58 [0.38, 0.88]
Moderate/severe stress wave 1 1.07 [0.57, 2.01] 0.93 [0.62, 1.40]
Moderate/severe stress wave 2 0.47 [0.25, 0.89] 0.56 [0.37, 0.85]
Moderate/severe stress wave 3 0.82 [0.46, 1.48] 0.85 [0.54, 1.33]
a Reference groups (maintained/improved mental well-being, none/mild depression, none/mild anxiety, none/mild stress, onsite clinical). bOdds ratio adjusted for physical activity, diet, income, age, race, gender, ethnicity. cWave 1 (n = 1994). dWave 2 (n = 1426). eWave 3 (n = 1363).
Significant associations were observed between reported physical activity and all well-being outcomes of mental well-being, depression, anxiety, and stress at each of the three waves. After adjusting for diet, age, race, income, gender, ethnicity, and clinical role, same or healthier physical activity was no longer significantly associated with moderate/severe depression at waves 2 and 3 or stress at waves 2 and 3 (Table 4). Only at wave 1 were participants whose physical activity remained the same or was healthier compared to a month prior, 32% less likely to report moderate/severe depression (odds ratio [OR] = 0.68, 95% confidence interval [CI]: 0.52–0.89) and 39% less likely to report moderate/severe stress (OR = 0.61, 95% CI: 0.45–0.81) than participants whose physical activity was less healthy. Physical activity was associated across all three waves with both mental well-being and anxiety. Participants whose physical activity remained the same or was healthier compared to a month prior were 24% to 48% less likely at waves 1 (OR = 0.76, 95% CI: 0.62–0.94), 2 (OR = 0.52, 95% CI: 0.40–0.68), and 3 (OR = 0.55, 95% CI: 0.41–0.73) to experience decreased mental well-being than participants whose physical activity was less healthy. Similar results were found for anxiety, although with decreasing odds across each wave. Participants whose physical activity remained the same or was healthier compared to a month prior were 56% less likely at wave 1 (OR = 0.44, 95% CI: 0.32–0.59), 52% less likely at wave 2, (OR = 0.48, 95% CI: 0.33–0.70), and 56% less likely at wave 3 (OR = 0.54, 95% CI: 0.37–0.79) to report moderate/severe anxiety than participants whose physical activity was less healthy.
Table 4. Analysis of Health Behaviors (Physical Activity and Diet) as Correlates of Well-Being (Mental Well-Being, Depression, Anxiety, and Stress) for Participants in Waves 1, 2, and 3 of the EMPOWER Study
Well-being a Same/healthier physical activity Same/healthier diet
OR b 95% CI OR b 95% CI
Decreased mental well-being wave 1 c 0.76 [0.62, 0.94] 0.45 [0.36, 0.57]
Decreased mental well-being wave 2 d 0.52 [0.40, 0.68] 0.47 [0.35, 0.62]
Decreased mental well-being wave 3 e 0.55 [0.41, 0.73] 0.61 [0.45, 0.83]
Moderate/severe depression wave 1 0.68 [0.52, 0.89] 0.60 [0.46, 0.78]
Moderate/severe depression wave 2 0.80 [0.57, 1.13] 0.52 [0.37, 0.74]
Moderate/severe depression wave 3 0.80 [0.57, 1.13] 0.65 [0.45, 0.94]
Moderate/severe anxiety wave 1 0.44 [0.32, 0.59] 0.69 [0.51, 0.92]
Moderate/severe anxiety wave 2 0.48 [0.33, 0.70] 0.63 [0.44, 0.91]
Moderate/severe anxiety wave 3 0.54 [0.37, 0.79] 0.73 [0.50, 1.08]
Moderate/severe stress wave 1 0.61 [0.45, 0.81] 0.63 [0.47, 0.84]
Moderate/severe stress wave 2 0.74 [0.51, 1.06] 0.62 [0.43, 0.89]
Moderate/severe stress Wave 3 0.76 [0.52, 1.11] 0.53 [0.36, 0.78]
a Reference groups (maintained/improved mental well-being, none/mild depression, none/mild anxiety, none/mild stress, less healthy physical activity, less healthy diet). bOdds ratio adjusted for income, age, race, gender, ethnicity, and clinical role. cWave 1 (n = 1994). dWave 2 (n = 1426). eWave 3 (n = 1363).
Diet was significantly correlated with well-being outcomes of mental well-being, depression, anxiety, and stress at all three waves. After adjusting for physical activity, age, race, income, gender, ethnicity, and clinical role, diet was no longer significantly correlated with anxiety at wave 3. However, at waves 1 and 2 participants whose diet was the same or healthier than a month prior, were 31% (OR = 0.69, 95% CI: 0.51–0.92) and 37% (OR = 0.63, 95% CI: 0.44–0.91) less likely to report moderate/severe anxiety than participants whose diets were less healthy. Mental well-being was associated with diet across all three waves, with decreased odds at each wave. Participants whose reported diet was the same or healthier compared to a month prior were 55% (OR = 0.45, 95% CI: 0.36–0.57), 53% (OR = 0.47, 95% CI: 0.35–0.62), and 39% (OR = 0.61, 95% CI: 0.45–0.83) less likely at waves 1, 2, and 3, respectively to experience decreased mental well-being than participants whose diet was less healthy. We found participants were 40% less likely to report moderate/severe depression at wave 1 (OR = 0.60, 95% CI: 0.46–0.78), 48% less likely to report moderate/severe depression at wave 2 (OR = 0.52, 95% CI: 0.37–0.74), and 35% less likely to report moderate/severe depression at wave 3 (OR = 0.65, 95% CI: 0.45–0.94) if their diet stayed the same or was healthier compared to participants whose diet was less healthy. Finally, we found participants were 37% to 38% less likely at waves 1 (OR = 0.63, 95% CI: 0.47–0.84) and 2 (OR = 0.62, 95% CI: 0.43–0.89), and 47% less likely at wave 3 (OR = 0.53, 95% CI: 0.36–0.78) to report moderate/severe stress if their diet stayed the same or was healthier compared to participants whose diet was less healthy than a month prior.
Discussion
This study aimed to describe changes in health behaviors and well-being during the COVID-19 pandemic among staff at a university and associated medical center. We found most participants reported maintained or improved health behaviors, and while most reported none or mild depression, anxiety, and stress, and maintained or improved mental well-being, the prevalence of worse well-being outcomes in our sample were still higher than national averages and similar to other studies conducted at the time. Well-being outcomes varied by clinical role with higher rates of worse mental well-being and moderate to severe stress and anxiety for onsite clinical workers; however, we did not find that clinical role moderated the relationship between health behaviors and well-being outcomes. Importantly, maintained or improved health behaviors were associated with better well-being outcomes.
We observed a high prevalence of moderate to severe depression and stress and decreased mental well-being in our sample, along with differences in well-being outcomes by clinical role. In general, most participants reported none or mild levels of depression, anxiety, and stress; however, the proportion of participants reporting moderate to severe depression (16%–17%) were higher than the national average of 7%, and similar to other studies assessing mental health during the COVID-19 pandemic (Jewell et al., 2020; National Institute of Mental Health, 2017a, 2017b; Newby et al., 2020). Around 13% to 14% of participants reported moderate to severe stress, which falls in the range of other studies using the DASS-21, which had findings ranging from 9% to 28% of participants experiencing moderate to severe stress during the COVID-19 pandemic (Chew et al., 2020; Mazza et al., 2020; Stanton et al., 2020; Wang et al., 2020). While we did not find anxiety to be reported at the same level as national averages, 13% to 14% compared to 20%, these findings are generally in line with previous research which also found high levels of depression and stress among U.S. adults and healthcare workers during the COVID-19 pandemic (Brooks et al., 2020; Evanoff et al., 2020; Jewell et al., 2020; National Institute of Mental Health, 2017a; Newby et al., 2020; Tull et al., 2020).
We observed differences in well-being outcomes by clinical role. Compared to onsite clinical staff, onsite nonclinical staff and those working at home were less likely to experience worse mental well-being, anxiety, and stress at most waves. Staff working from home were less likely to experience moderate to severe anxiety compared to onsite clinical workers. Even though most participants reported none or mild depression and stress, depression was still higher than national averages and moderate to severe stress was in line with other research conducted during the COVID-19 pandemic. While anxiety was not higher overall, it was higher for the group of onsite clinical workers. This suggests mental health may be a concern for university and medical staff during the pandemic and that particular attention should be paid to onsite clinical staff who may experience worse anxiety. Ongoing monitoring of mental health in this population will be critical, especially as the pandemic is ongoing, and understanding mental health and well-being trends in this population can inform public health interventions and policies (Bianchi et al., 2015; Bubonya et al., 2017; Prince et al., 2007).
Another important finding from our study was that health behaviors may act as a protective factor for mental health and well-being. Participants whose physical activity and diet stayed the same or increased were significantly less likely to experience worse mental well-being and moderate to severe levels of depression, anxiety, and stress. Our findings around the association between health behaviors and well-being are in line with previous research and lend support to the importance of maintaining or increasing health behaviors during the COVID-19 pandemic (Ball & Lee, 2000; Cerin et al., 2009; Ingram et al., 2020). Given the sample of university and medical staff, one such avenue would be to focus on robust workplace wellness programs that can help workers maintain and improve health behaviors, such as diet and physical activity. Employee wellness programs have been found to improve worker health behavior, promoting well-being and worker satisfaction (Goetzel et al., 2014; Kaspin et al., 2013). As such, access to these programs may not only have long-term health benefits but also aid in making sure employees maintain and improve health behaviors and well-being during times of crisis, such as during the COVID-19 pandemic.
Finally, we found a high proportion of participants reporting maintained or improved health behaviors. These findings are not in line with most of the previous research showing the COVID-19 pandemic led to decreases in physical activity and worse dietary behaviors (Ammar et al., 2020; Dwyer et al., 2020; Gallo et al., 2020; López-Bueno et al., 2020; Robinson et al., 2021; Tison et al., 2020). One potential explanation for our findings is that health behavior change during the pandemic may be influenced by prepandemic diet and physical activity. One study by Barkley et al. found similar changes of increased physical activity in the university setting for participants who had low or moderate rates of physical activity prior to the pandemic (Barkley et al., 2020). We did not collect data on baseline levels of physical activity and diet; however, it is possible variation in prepandemic health behaviors could be driving some of our findings. The health behavior findings may also be due to the measure we used for physical activity and diet change. Since we asked about changes in overall diet and physical activity, we may miss nuances in health behavior change that could affect our results. For example, we do not have information on increases in healthy foods versus decreases in unhealthy foods, snacking, or changes to home cooking and eating out. In addition, we do not have information about changes to the frequency, intensity, duration, or location for physical activity. Future research should explore these more nuanced aspects of health behavior change to better understand if and how the COVID-19 pandemic influences healthy lifestyles in this population.
Our study has a few limitations which are important to note. While we described self-reported changes to health behavior and mental health and well-being outcomes across three waves of data, the cross-sectional nature of this study means we cannot report on causation. This is important to note because the relationship between health behaviors and mental health and well-being are likely bi-directional. In terms of measurement, we asked participants to report on whether physical activity and diet changed through a scale of much healthier to much less healthy. As “healthy” may be interpreted differently by respondents, there is the potential for self-report bias. This is also a limitation for our measurement of Mental well-being, as this term can be perceived differently by different respondents. We also did not ask participants what their health behaviors were like prior to the pandemic, limiting our ability to assess how prepandemic health behaviors and well-being related to changes during the pandemic. In addition, each survey asked participants to report on how their diet or physical activity changed compared to 1 month prior, so changes made and sustained early in the pandemic may not be reflected as changes in later survey waves. As, health behaviors were measured by self-report, there may also be recall bias. In terms of the sample, many participants were working from home, and data on worker roles were collected at a high-level preventing comparison across job types for this analysis. Although unable to explore comparisons across job type, clinical role was adjusted for in the analyses. We also report on the association between clinical role and well-being outcomes from our adjusted models to better capture how clinical role and working from home may relate to well-being. While we adjusted our models for factors that may influence well-being outcomes, we did not account for the potential role of employee wellness programs. This was beyond the scope of our analysis, which focused on describing health behaviors and well-being during the COVID-19 pandemic and the relationship between health behaviors and well-being among staff at a university and medical center. In addition, our sample, although large, is limited to one employer, making it difficult to generalize our findings beyond our study population. Despite these limitations, this study collected data in multiple waves, beginning early in the pandemic from a large group of employees in clinical, academic, and staff positions. This information provides insight into how the pandemic might impact well-being and health behaviors and informs strategies for improvement.
Conclusion
The main finding of our study was the importance of maintaining or increasing health behaviors during the COVID-19 pandemic to mitigate negative influence on mental health and well-being. Participants who maintained or increased healthy physical activity and diet were less likely to experience worse mental health and well-being. This finding is especially salient considering the high prevalence of poor mental health outcomes and decreased mental well-being within the study sample, and specifically among clinical workers. Our findings of the observed relationship between mental health and healthy diet and physical activity during the pandemic support employer-based and public health efforts to promote wellness. Future research should continue to monitor health behaviors and mental health and well-being among workers and the impact of public health and policy intervention.
Implications for Occupational Health Practice
This study provides insight into health behaviors and well-being for clinical and nonclinical staff at a university and medical center during the first few months of the COVID-19 pandemic. Results show the mental health experiences and health behaviors of workers in medical and academic settings over the course of the first few months of the pandemic. Our findings show a relationship between health behaviors and mental health during this time. These findings suggest that healthy eating and physical activity behaviors may have mitigated the negative influence of the COVID-19 pandemic on mental health and well-being in this population. Given these findings, it is important to consider the need for workplace health promotion efforts focused on maintaining or improving healthy eating and physical activity during the pandemic and beyond.
Applying Research to Occupational Health Practice
This study explored trends in health behaviors (physical activity, diet) and well-being among university and medical center staff during the beginning few months of the COVID-19 pandemic. Using three waves of observational data from the Employee Well-being during Epidemic Response study, we found most university and medical center staff experienced worsening mental health and well-being. We also found significant associations between health behaviors and mental health and well-being across each wave. Our findings suggest the need for workplace health promotion programs to support health behaviors among employees to improve mental health and well-being. Such health promotion programs may help employees maintain or improve health behaviors and mitigate negative mental health and well-being impacts due to the COVID-19 pandemic and beyond.
Author Contributions: A.G. conducted data analysis and was a major contributor in writing this manuscript. A.E. provided guidance on data analysis and was a major contributor in writing this manuscript. G.C. assisted in data analysis and contributed to writing this manuscript. J.H. provided guidance around data analysis and reviewed and provided feedback and edits to this manuscript. L.H. conducted data management and provided feedback and edits to this manuscript. B.E. designed the EMPOWER study, provided guidance on data analysis, and provided feedback and edits to this manuscript. All authors read and approved the final manuscript.
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding: The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Healthier Workforce Center of the Midwest grant no. U19OH008868 from the Centers for Disease Control and Prevention (CDC), by the Washington University Institute of Clinical and Translational Sciences grant no. UL1TR002345 from the National Center for Advancing Translational Sciences (NCATS) of the National Institutes of Health (NIH), by the Centers for Disease Control and Prevention (grant no. U48DP006395), and by the National Heart, Lung, and Blood Institute (NHLBI), National Institutes of Health (grant no. T32 HL130357). The content is solely the responsibility of the authors and does not necessarily represent the official views of the CDC, NCATS, or NIH.
ORCID iDs: Amanda Gilbert https://orcid.org/0000-0001-6948-0127
Jenine Harris https://orcid.org/0000-0002-3576-5906
Lisa Hayibor https://orcid.org/0000-0002-7586-8037
==== Refs
References
Ammar A. Brach M. Trabelsi K. Chtourou H. Boukhris O. Masmoudi L. Bouaziz B. Bentlage E. How D. Ahmed M. Müller P. Müller N. Aloui A. Hammouda O. Paineiras-Domingos L. L. Braakman-Jansen A. Wrede C. Bastoni S. Pernambuco C. S. . . .Hoekelmann A. (2020). Effects of COVID-19 home confinement on eating behaviour and physical activity: Results of the ECLB-COVID19 International Online Survey. Nutrients, 12 (6 ), 1583.32481594
Ball K. Lee C. (2000). Relationships between psychological stress, coping and disordered eating: A review. Psychology and Health, 14 (6 ), 1007–1035. 10.1080/08870440008407364 22175259
Barkley J. E. Lepp A. Glickman E. Farnell G. Beiting J. Wiet R. Dowdell B. (2020). The acute effects of the COVID-19 pandemic on physical activity and sedentary behavior in university students and employees. International Journal of Exercise Science, 13 (5 ), 1326–1339. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7523895/ 33042377
Bianchi R. Schonfeld I. S. Laurent E. (2015). Burnout-depression overlap: A review. Clinical Psychology Review, 36 , 28–41. 10.1016/j.cpr.2015.01.004 25638755
Brooks S. K. Webster R. K. Smith L. E. Woodland L. Wessely S. Greenberg N. Rubin G. J. (2020). The psychological impact of quarantine and how to reduce it: Rapid review of the evidence. The Lancet, 395 (10227 ), 912–920. 10.1016/S0140-6736(20)30460-8
Brumby S. Chandrasekara A. McCoombe S. Torres S. Kremer P. Lewandowski P. (2011). Reducing psychological distress and obesity in Australian farmers by promoting physical activity. BMC Public Health, 11 , 362. 10.1186/1471-2458-11-362 21600058
Bubonya M. Cobb-Clark D. A. Wooden M. (2017). Mental health and productivity at work: Does what you do matter? Labour Economics, 46 (September ), 150–165. 10.1016/j.labeco.2017.05.001
Butler M. J. Barrientos R. M. (2020). The impact of nutrition on COVID-19 susceptibility and long-term consequences. Brain, Behavior, and Immunity, 87 , 53–54.32311498
The Centers for Disease Control and Prevention. (2020). CDC COVID data tracker. https://covid.cdc.gov/covid-data-tracker/?CDC_AA_refVal=https%3A%2F%2Fwww.cdc.gov%2Fcoronavirus%2F2019-ncov%2Fcases-updates%2Fcases-in-us.html#cases_totalcases
Cerin E. Leslie E. Sugiyama T. Owen N. (2009). Associations of multiple physical activity domains with mental well-being. Mental Health and Physical Activity, 2 (2 ), 55–64. 10.1016/j.mhpa.2009.09.004
Chew N. W. S. Lee G. K. H. Tan B. Y. Q. Jing M. Goh Y. Ngiam N. J. H. Yeo L. L. L. Ahmad A. Ahmed Khan F. Napolean Shanmugam G. Sharma A. K. Komalkumar R. N. Meenakshi P. V. Shah K. Patel B. Chan B. P. L. Sunny S. Chandra B. Ong J. J. Y. . . .Sharma V. K. (2020). A multinational, multicentre study on the psychological outcomes and associated physical symptoms amongst healthcare workers during COVID-19 outbreak Nicholas. Brain, Behavior, and Immunity, 88 (January ), 559–565. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7172854/ 32330593
Daly M. Sutin A. R. Robinson E. (2021). Depression reported by US adults in 2017–2018 and March and April 2020. Journal of Affective Disorders, 278 , 131–135.32956962
Dwyer M. J. Pasini M. De Dominicis S. Righi E. (2020). Physical activity: Benefits and challenges during the COVID-19 pandemic. Scandinavian Journal of Medicine and Science in Sports, 30 (7 ), 1291–1294. 10.1111/sms.13710 32542719
Evanoff B. A. Strickland J. R. Dale A. M. Hayibor L. Page E. Duncan J. G. Kannampallil T. Gray D. L. (2020). Work-related and personal factors associated with mental well-being during the COVID-19 response: Survey of healthcare and other workers. Journal of Medical Internet Research, 22 (8 ), e21366. 10.2196/21366 32763891
Farah B. Q. do Prado W. L. Malik N. Lofrano-Prado M. C. de Melo P. H. Botero J. P. Cucato G. G. Correia M. A. Ritti-Dias R. M. (2021). Barriers to physical activity during the COVID-19 pandemic in adults: A cross-sectional study. Sport Sciences for Health, 17 (2 ), 441–447. 10.1007/s11332-020-00724-5 33815618
Gallo L. A. Gallo T. F. Young S. L. Moritz K. M. Akison L. K. (2020). The impact of isolation measures due to COVID-19 Australian University students. Nutrients, 12 (6 ), 1865.32585830
Goetzel R. Z. Henke R. M. Tabrizi M. Pelletier K. R. Loeppke R. Ballard D. W. Grossmeier J. Anderson D. R. Yach D. Kelly R. K. McCalister T. Serxner S. Selecky C. Shallenberger L. G. Fries J. F. Baase C. Isaac F. Crighton K. A. Wald P. . . .Metz R. D. (2014). Do workplace health promotion (wellness) programs work? Journal of Occupational and Environmental Medicine, 56 (9 ), 927–934. 10.1097/JOM.0000000000000276 25153303
Hennein R. Lowe S. (2020). A hybrid inductive-abductive analysis of health workers’ experiences and wellbeing during the COVID-19 pandemic in the United States. PLOS ONE, 15 (10 ), Article e0240646. 10.1371/journal.pone.0240646
Huppert F. A. (2014). The state of wellbeing science concepts, measures, interventions, and policies. In Huppert F. A. Cooper C. L. (Eds.), Interventions and policies to enhance wellbeing: Wellbeing: A complete reference guide (Vol. VI , pp. 2–50). John Wiley & Sons. 10.1002/9781118539415
Ingram J. Maciejewski G. Hand C. J. (2020). Changes in diet, sleep, and physical activity are associated with differences in negative mood during COVID-19 lockdown. Frontiers in Psychology, 11 (May ), 588604. 10.3389/fpsyg.2020.588604 32982903
Jewell J. S. Farewell C. V. Welton-Mitchell C. Lee-Winn A. Walls J. Leiferman J. A. (2020). Mental health during the COVID-19 pandemic in the United States: Online survey. JMIR Formative Research, 4 (10 ), e22043. 10.2196/22043 33006939
John R. C. Julie D. H. (2003). The Depression Anxiety Stress Scales (DASS): Normative data and latent structure in a large non-clinical sample. British Journal of Clinical Psychology, 42 (2 ), 111–131. http://ejournals.ebsco.com/direct.asp?ArticleID=46B8A63971E13A62629D 12828802
Kaspin L. C. Gorman K. M. Miller R. M. (2013). Systematic review of employer-sponsored wellness strategies and their economic and health-related outcomes. Population Health Management, 16 (1 ), 14–21. 10.1089/pop.2012.0006 23113636
Linton M. J. Dieppe P. Medina-Lara A. (2016). Review of 99 self-report measures for assessing well-being in adults: Exploring dimensions of well-being and developments over time. BMJ Open, 6 (7 ), 1–16. 10.1136/bmjopen-2015-010641
López-Bueno R. Calatayud J. Casaña J. Casajús J. A. Smith L. Tully M. A. Andersen L. L. López-Sánchez G. F. (2020). COVID-19 confinement and health risk behaviors in Spain. Frontiers in Psychology, 11 (June ), 1426 . 10.3389/fpsyg.2020.01426
Lovibond S. H. Lovibond P. F. (1995). Manual for the depression anxiety & stress scales. Psychology Foundation.
Martinez-Ferran M. de la Guía-Galipienso F. Sanchis-Gomar F. Pareja-Galeano H. (2020). Metabolic impacts of confinement during the COVID-19 pandemic due to modified diet and physical activity habits. Nutrients, 12 (6 ), 1549.32466598
Mazza C. Ricci E. Biondi S. Colasanti M. Ferracuti S. Napoli C. Roma P. (2020). A nationwide survey of psychological distress among Italian people during the covid-19 pandemic: Immediate psychological responses and associated factors. International Journal of Environmental Research and Public Health, 17 (9 ), 1–14. 10.3390/ijerph17093165
Mukhtar S. (2020). Psychological health during the coronavirus disease 2019 pandemic outbreak. International Journal of Social Psychiatry, 66 (5 ), 512–516. 10.1177/0020764020925835 32434402
Na H.-K. Oliynyk S. (2011). Effects of physical activity on cancer prevention. Annals of the New York Academy of Sciences, 1229 , 176–183. 10.1111/j.1749-6632.2011.06105.x 21793853
National Institute of Mental Health. (2017a). Prevalence of anxiety disorder among adults. https://www.nimh.nih.gov/health/statistics/any-anxiety-disorder
National Institute of Mental Health. (2017b). Prevalence of major depressive episode among adults. https://www.nimh.nih.gov/health/statistics/major-depression
Newby J. M. O’Moore K. Tang S. Christensen H. Faasse K. (2020). Acute mental health responses during the COVID-19 pandemic in Australia. PLOS ONE, 15 (7 ), Article e0236562. 10.1371/journal.pone.0236562
Prince M. Patel V. Saxena S. Maj M. Maselko J. Phillips M. R. Rahman A. (2007). No health without mental health. Lancet, 370 (9590 ), 859–877. 10.1016/S0140-6736(07)61238-0 17804063
Robinson E. Boyland E. Chisholm A. Harrold J. Maloney N. G. Marty L. Mead B. R. Noonan R. Hardman C. A. (2021). Obesity, eating behavior and physical activity during COVID-19 lockdown: A study of UK adults. Appetite, 156 (August ), 104853. 10.1016/j.appet.2020.104853 33038479
Rossa-Roccor V. Richardson C. G. Murphy R. A. Gadermann A. M. (2021). The association between diet and mental health and wellbeing in young adults within a biopsychosocial framework. PLOS ONE, 16 (6 ), Article e0252358. 10.1371/journal.pone.0252358
Ryan R. M. Deci E. L. (2001). On happiness and human potentials: A review of research on hedonic and Eudaimonic well-being. Annual Review of Psychology, 52 , 141–166. 10.1146/annurev.psych.52.1.141
Salleh M. R. (2008). Life event, stress and illness. Malaysian Journal of Medical Sciences, 15 (4 ), 9–18.
Sallis R. Young D. R. Tartof S. Y. Sallis J. F. Sall J. Li Q. Smith G. N. Cohen D. A. (2021). Physical inactivity is associated with a higher risk for severe COVID-19 outcomes: A study in 48 440 adult patients. British Journal of Sports Medicine, 55 (19 ), 1099–1105. 10.1136/bjsports-2021-104080 33849909
Sharma M. Creutzfeldt C. J. Lewis A. Patel P. V. Hartog C. Jannotta G. E. Blissitt P. Kross E. K. Kassebaum N. Greer D. M. Curtis J. R. Wahlster S. (2021). Health-care professionals’ perceptions of critical care resource availability and factors associated with mental well-being during coronavirus disease 2019 (COVID-19): Results from a US Survey. Clinical Infectious Diseases: An Official Publication of the Infectious Diseases Society of America, 72 (10 ), e566–e576. 10.1093/cid/ciaa1311 32877508
Stanton R. To Q. G. Khalesi S. Williams S. L. Alley S. J. Thwaite T. L. Fenning A. S. Vandelanotte C. (2020). Depression, anxiety and stress during COVID-19: Associations with changes in physical activity, sleep, tobacco and alcohol use in Australian adults. International Journal of Environmental Research and Public Health, 17 (11 ), 4065. 10.3390/ijerph17114065 32517294
Thomaier L. Teoh D. Jewett P. Beckwith H. Parsons H. Yuan J. Blaes A. H. Lou E. Hui J. Y. C. Vogel R. I. (2020). Emotional health concerns of oncology physicians in the United States: Fallout during the COVID-19 pandemic. PLOS ONE, 15 (11 ), Article e0242767. 10.1371/journal.pone.0242767
Tison G. Avram R. Kuhar P. Abreau S. Marcus G. Pletcher M. Olgin J. (2020). Worldwide effect of COVID-19 on physical activity: A descriptive study. Annals of Internal Medicine, 173 (9 ), 767–770. 10.7326/M20-2665 32598162
Tull M. T. Edmonds K. A. Scamaldo K. M. Richmond J. R. Rose J. P. Gratz K. L. (2020). Psychological outcomes associated with stay-at-home orders and the perceived impact of COVID-19 on daily life. Psychiatry Research, 289 (April ), 113098. 10.1016/j.psychres.2020.113098 32434092
U.S. Department of Health and Human Services. (2018). 2018 physical activity guidelines advisory committee. Physical Activity Guidelines Advisory Committee Scientific Report. U.S. Department of Health and Human Services. 10.1111/j.1753-4887.2008.00136.x
U.S. Department of Health and Human Services and U.S. Department of Agriculture. (2015). 2015-2020 dietary guidelines for Americans. http://health.gov/dietaryguidelines/2015/guidelines/
Wang C. Pan R. Wan X. Tan Y. Xu L. Ho C. S. Ho R. C. (2020). Immediate psychological responses and associated factors during the initial stage of the 2019 coronavirus disease (COVID-19) epidemic among the general population in China. International Journal of Environmental Research and Public Health, 17 (5 ), 1–25.
|
PMC009xxxxxx/PMC9901500.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36745128
EHP12546
10.1289/EHP12546
Science Selection
Unfulfilled Promise: Pollinator Declines, Crop Deficits, and Diet-Associated Disease
Nicole Wendee
6 2 2023
2 2023
131 2 02400106 12 2022
22 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
A collage of four images of pollinators on flowers. On top left, a rufous hummingbird. On top right, a western honey bee. At bottom left, a grey-headed flying fox. At bottom right, a swallowtail butterfly.
==== Body
pmcAlthough difficult to track, populations of wild pollinators across the globe appear to have declined for the past several decades.1,2 Animal pollinators (e.g., birds, bees, wasps, small mammals) ensure the yield and quality of approximately one-third of the world’s agricultural crops, including vegetables, fruits, nuts, and seeds.3 A shortage of these foods in the diet influences human health in myriad ways, ranging from undernutrition to development of chronic diseases.4,5 In a new study in Environmental Health Perspectives,6 a team of scientists estimated effects of pollinator deficits on five diet-associated disease end points: stroke, cancer, type 2 diabetes, coronary heart disease, and all-cause mortality associated with changes in weight.
Most plants are pollinated by self-pollination, the wind, or animals such as birds, insects, bats, and other small mammals. Animals pollinate one-third of agricultural crops worldwide. Images, clockwise from top left: rufous hummingbird, Tom Koerner/U.S. Fish and Wildlife Service, under CC BY 2.0 license; western honey bee, Kirsten Strough/U.S. Department of Agriculture; swallowtail butterfly, Preston Keres/U.S. Department of Agriculture; grey-headed flying fox, Andrew Mercer, under CC BY-SA 4.0.
A collage of four images of pollinators on flowers. On top left, a rufous hummingbird. On top right, a western honey bee. At bottom left, a grey-headed flying fox. At bottom right, a swallowtail butterfly.
The authors started with a model of worldwide attainable yields for 63 important pollinator-dependent crops. They combined this model with a globally derived data set that quantified the proportion of crop yield gaps (difference between current and attainable yields) attributable to insufficient pollination. Then, for each country where those crops are grown, they estimated how much additional food would have been produced if pollination were sufficient. The team then used an agricultural–economic model to examine who would have consumed that food, and a comparative risk assessment framework to estimate the impact of altered consumption patterns on human disease and mortality.
“Linking all these different fields—pollination ecology, agriculture, food and diets, global economics, and human health—required building a multidisciplinary team with expertise in each area and linking their models together to understand how they interact,” says lead author Matthew Smith, a research scientist at the Harvard T.H. Chan School of Public Health.
The researchers estimated that insufficient pollination results in a global loss of 4.7% of fruit, 3.2% of vegetables, and 4.7% of nuts produced every year. Next, using the International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT),7 a multimarket model of the global food system, they estimated how these lost foods might have been either traded globally or eaten locally. Combining these calculated dietary changes with relative risks that link dietary factors to health outcomes, they estimated that 427,000 deaths—mainly from stroke, coronary heart disease, and cancer—would have been avoided had there been no pollinator declines.
“That is on par with other major global risks like interpersonal violence, substance use disorders, or prostate cancer,” says Smith. “It’s a sobering reminder that a failure to protect nature has concrete and tangible consequences for human lives.”
The research team estimated that food production losses were greatest in lower-income countries, with an estimated 26% of vegetable production and 8% of nut production lost. When the overall loss in production was translated to human consumption worldwide, estimates included 2%–5% lower consumption of fruits, 3% lower consumption of vegetables, and 4%–12% lower consumption of nuts, depending on income level and region. North America saw the greatest estimated reductions in consumption across all food categories, including: 5.7% for fruit, 4.4% for vegetables, and 11.3% for nuts.
Interestingly, the modeling suggested that effects of pollinator declines on human health would be less pronounced in lower-income countries. Smith explains, “The health effects of lost pollination are concentrated in areas where rates of chronic disease are [already] higher but where populations often cannot pay for expensive fruits, vegetables, and nuts,” namely, Russia, Eastern Europe, China, India, and parts of Southeast Asia and North Africa. In other words, surplus food is routinely exported to higher-income countries, so consumption of those crops in lower-income countries would not be expected to change with fluctuations in yield. However, the results also showed significant economic losses for lower-income countries: Case studies on Honduras, Nepal, and Nigeria estimated 12%–31% of agricultural value was lost due to insufficient pollination.
“Most large-scale studies into the impacts of pollination have focused on fairly extreme versions of what we stand to lose,” says Tom Breeze, a postdoctoral scholar in sustainable land management at the University of Reading in England, who was not associated with the study. For example, he points to a previous paper by Smith et al. that modeled consumption changes resulting from a more extreme loss of 50%–100% of pollinators.6 “This new study, by including trade and yield deficit projections, shows not only what we stand to lose in a more realistic way, but also the scale of what we may already have lost,” he says.
Breeze adds that although the study is very strong and the team used the best data available, they were limited by gaps in information data from across the world. For example, he says, “We simply don’t have good [enough] pollination service monitoring to be able to accurately estimate any yield gaps at a national, never mind international, scale.” Breeze adds that although the IMPACT model is global, it is very much grounded in western market economy thinking, so it does not factor in relevant activity in lower-income countries, such as bartering and trade in locally valuable crops produced on a small scale.
“Protecting pollinators is more than a crucially important goal for biodiversity; doing so also [protects] human health,” says Smith. “The policy prescriptions are straightforward: Set aside nearby natural or semi-natural habitat, provide adequate food by planting a diversity of flowers on or near farms, and halt the use of harmful pesticides like neonicotinoids.”
Wendee Nicole, a San Diego–based writer, contributes regularly to Environmental Health Perspectives.
==== Refs
References
1. Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services. 2016. The Assessment Report on Pollinators, Pollination and Food Production: summary for Policymakers. http://digitallibrary.un.org/record/1664349 [accessed 23 January 2023].
2. Jacquemin F, Violle C, Munoz F, Mahy G, Rasmont P, Roberts SPM, et al. 2020. Loss of pollinator specialization revealed by historical opportunistic data: insights from network-based analysis. PLoS One 5 (7 ):e0235890, PMID: , 10.1371/journal.pone.0235890.32658919
3. Klein AM, Vaissière BE, Cane JH, Steffan-Dewenter I, Cunningham SA, Kremen C, et al. 2007. Importance of pollinators in changing landscapes for world crops. Proc Biol Sci 274 (1608 ):303–313, PMID: , 10.1098/rspb.2006.3721.17164193
4. Chaplin-Kramer R, Dombeck E, Gerber J, Knuth KA, Mueller ND, Mueller M, et al. 2014. Global malnutrition overlaps with pollinator-dependent micronutrient production. Proc Biol Sci 281 (1794 ):20141799, PMID: , 10.1098/rspb.2014.1799.25232140
5. Liu RH. 2013. Dietary bioactive compounds and their health implications. J Food Sci 78 (suppl 1 ):A18–A25, PMID: , 10.1111/1750-3841.12101.23789932
6. Smith MR, Mueller ND, Springmann M, Sulser TB, Garibaldi LA, Gerber J, et al. 2022. Pollinator deficits, food consumption, and consequences for human health: a modeling study. Environ Health Perspect 130 (12 ):127003, PMID: , 10.1289/EHP10947.36515549
7. Robinson S, Mason-d’Croz D, Islam S, Sulser TB, Robertson RD, Zhu T, et al. 2015. The International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT): Model Description for Version 3. IFPRI Discussion Paper 01483. http://ebrary.ifpri.org/cdm/ref/collection/p15738coll2/id/129825 [accessed 23 January 2023].
|
PMC009xxxxxx/PMC9904409.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36749608
EHP11171
10.1289/EHP11171
Research
Associations of Prenatal Chemical and Nonchemical Stressors with Early-Adulthood Anxiety and Depressive Symptoms
https://orcid.org/0000-0003-2147-0397
Rokoff Lisa B. 1 2
Coull Brent A. 1 3
Bosquet Enlow Michelle 4 5
Korrick Susan A. 1 6
1 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
2 Population Health Sciences Program, Harvard University, Cambridge, Massachusetts, USA
3 Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
4 Department of Psychiatry and Behavioral Sciences, Boston Children’s Hospital, Boston, Massachusetts, USA
5 Department of Psychiatry, Harvard Medical School, Boston, Massachusetts, USA
6 Channing Division of Network Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts, USA
Address correspondence to Lisa B. Rokoff, Department of Environmental Health, Harvard T.H. Chan School of Public Health, 665 Huntington Ave., SPH-1 Room 1406, Boston, MA 02115. Email: [email protected]
7 2 2023
2 2023
131 2 02700425 2 2022
01 12 2022
09 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Prenatal exposure to environmental chemicals may increase risk of childhood internalizing problems, but few studies have explored the potential for longer-term consequences of such exposures.
Objective:
We evaluated associations between prenatal organochlorine and metal levels and early adulthood internalizing symptoms, considering whether sociodemographic/nonchemical stressors modified these associations.
Methods:
Participants were 209 young adults, born (1993–1998) to mothers residing in or near New Bedford, Massachusetts. As part of the early-adult assessment, self-reported anxiety (7-item Generalized Anxiety Disorder scale) and depressive (8-item Patient Health Questionnaire) symptoms (≥10: elevated symptoms) were ascertained. We previously analyzed levels of cord serum organochlorines [hexachlorobenzene, dichlorodiphenyldichloroethylene (p,p′-DDE), polychlorinated biphenyls (ΣPCB4: sum of congeners 118, 138, 153, 180)] and whole blood lead shortly after participants’ birth, and levels of cord whole blood manganese from archived samples at the time of the adolescent study visit. We used modified Poisson regression models and quantile g-computation, adjusting for sociodemographics, and explored whether biological sex, race/ethnicity (proxy for unmeasured consequences of racism), prenatal social disadvantage (assessed when participants were neonates), and quality of the home environment (assessed during adolescence) modified these associations.
Results:
Participants were (mean±standard deviation) 22.1±1.5 y old, 76% Non-Hispanic White, and 67% female. Prenatal hexachlorobenzene, p,p′-DDE, and lead exposures were moderately associated with increased risk of elevated anxiety symptoms. There were strata-specific associations for prenatal social disadvantage and quality of home environment such that adverse associations of p,p′-DDE and lead and the overall mixture with anxiety and depressive symptoms were largely only evident in those with lower nonchemical stress [e.g., risk ratio and 95% confidence interval (CI) per doubling p,p′-DDE for anxiety: 1.54 (95% CI: 1.20, 1.99) in high-quality home environments and 0.77 (95% CI: 0.51, 1.16) in low-quality home environments]. Associations between prenatal hexachlorobenzene and p,p′-DDE and anxiety symptoms were stronger for underrepresented racial/ethnic group participants vs. Non-Hispanic Whites. We found minimal evidence for sex-specific effects, and no consistent associations with manganese or ΣPCB4.
Discussion:
Prenatal organochlorine pesticides and lead exposure possibly increases risk of internalizing problems, particular anxiety symptoms, in young adults. Varying risk was observed by sociodemographic/nonchemical stressor strata, demonstrating the importance of considering interactions between chemical and other stressors. https://doi.org/10.1289/EHP11171
Supplemental Material is available online (https://doi.org/10.1289/EHP11171).
B.A.C. receives research funding from Apple, Inc. The other authors (L.B.R., M.B.E., and S.A.K) declare they have no financial or other conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
As the prevalence of anxiety disorders and depression in young adults continues to rise,1,2 a better understanding of the contributing role of early-life stressors is critical to identifying susceptible individuals and reducing physical, mental, and societal burden caused by these internalizing disorders.3 Certain risk factors for anxiety and depression, such as female sex,3,4 lower socioeconomic status,3,4 and a poorer-quality home and family environment,4,5 have been well characterized, but there is limited research on the extent to which prenatal exposures to environmental neurotoxicant chemicals influence risk in adulthood.
Prenatal exposures to environmental chemicals, including organochlorines and metals, have been implicated as detrimental to neurobehavioral development. However, most studies evaluated these exposures in relation to externalizing (e.g., hyperactivity, inattention) rather than internalizing (e.g., anxiety, depression) symptoms and focused on outcomes in childhood prior to major core risk periods for anxiety and depression.6,7 Prenatal organochlorine and metal exposures may play a role in the development of anxiety and depression. These chemicals, which readily cross the placenta and blood–brain barrier, are hypothesized to alter the function of neurotransmitters implicated in emotion regulation, such as dopamine, gamma-aminobutyric acid (GABA), and glutamate.8–14 Insults to these neurobiological pathways may program an individual for dysregulation of emotional and behavioral function across the life span.
Previous studies of internalizing symptoms in childhood, including our prior work in the New Bedford Cohort (NBC),15 have provided limited evidence of associations (i.e., largely null results) with prenatal exposure to organochlorines, including hexachlorobenzene (HCB),16–19 dichlorodiphenyldichloroethylene (p,p′-DDE),16–20 and polychlorinated biphenyls (PCBs),16–21 and mixed evidence of associations (i.e., positive and null results) with prenatal exposure to lead (Pb)16,21–24 and manganese (Mn).24,25 In the NBC, we found an adverse association between prenatal Pb and self-reported anxiety-related symptoms in adolescence but not with parent-reported symptoms in mid-childhood,15 highlighting that neurobehavioral effects related to early chemical exposures may not be apparent until later in development, during or after core risk periods for onset of internalizing disorders,3,26 and when more accurate self-report of symptoms27,28 can be ascertained. Additionally, most of this literature did not explore whether coexposures to other risk factors for neurobehavioral disorders (e.g., sociodemographic, nonchemical, and other chemical stressors) modify associations.
In the current study, we examined prospective associations between prenatal organochlorine and metal levels and early-adulthood anxiety and depressive symptoms. Because sociodemographic and nonchemical risk factors, such as sex, race/ethnicity (proxy for unmeasured consequences of racism), prenatal social disadvantage, and quality of the childhood home environment, may modify the impact of chemical exposures on neurobehavior,29–31 we explored possible interactions with these stressors. We hypothesized that greater prenatal chemical levels would be associated with higher risk of elevated symptoms, with stronger adverse effects among those with sociodemographic-related or nonchemical stress risk factors (i.e., biological female sex, racial discrimination and/or structural racism, and social/material deprivation reflected in prenatal social disadvantage or a poorer-quality home environment).
Methods
Study Population: New Bedford Cohort (NBC)
The NBC is a longitudinal birth cohort of mother–child pairs who were recruited after delivery at St. Luke’s Hospital (New Bedford, Massachusetts) from 1993 to 1998.32,33 During pregnancy, NBC mothers resided in one of four towns (New Bedford, Acushnet, Fairhaven, Dartmouth) adjacent to the New Bedford Harbor Superfund site, a waste site highly contaminated with PCBs and heavy metals between 1940 and the late 1970s. Eligible women recruited into the study were ≥18 y old and English- and/or Portuguese-speaking and had delivered vaginally (postsurgical narcotic analgesia restricted consent after cesarean delivery). Infants were excluded if they were not available for study-required neonatal examination. The NBC ultimately consented and enrolled 788 mother–child pairs. Child participants have undergone periodic follow-up assessments from birth through early adulthood.15,32–35
Between 2017 and 2019, a series of web-based neurobehavioral assessments were completed by NBC participants at the early-adult visit (age range: 19–25 y). Specifically, eight distinct neurobehavioral or questionnaire assessments were emailed, one at a time, typically with a few months between mailings. One of these emails included both the anxiety and depressive symptom assessments; that information was used in the present study along with data from a general questionnaire sent separately. Eligibility criteria included intact cognition (i.e., no history of central nervous system cancer or posttraumatic cognitive impairment that precluded testing), availability of at least one biomarker of prenatal chemical exposure, and trackable contact information. Among the 788 child participants enrolled at birth, 565 were eligible for young-adult follow-up and had an email address to receive study questionnaires. Of these, 325 consented to participate and completed at least one early-adult visit study questionnaire, with 240 specifically completing an anxiety and/or depressive symptom evaluation (mean age: 22.1 y). A subset of participants also completed in-person testing to validate the web-based neurobehavioral approach (89 participants did both in-person and web-based anxiety and depressive symptom assessments).
The study research protocol was approved by the human subjects committee of Brigham and Women’s Hospital (Boston, Massachusetts). Web-based written consent was obtained from all participants prior to early-adult study evaluations. Before other study visit evaluations, we obtained written consent from all participating parents and assent from all adolescent participants.
Chemical Exposure Assessment
Umbilical cord blood samples for organochlorine analyses (BD Vacutainer® red top tubes without anticoagulant) and metal analyses (BD Vacutainer® lavender top tubes with EDTA anticoagulant) were collected at birth. For organochlorines, blood samples were centrifuged, and the serum fraction was removed and transferred into solvent-rinsed glass vials and stored at −20°C until analysis. Organochlorine analyses were performed within approximately 2 to 12 months of sample collection at the Harvard T.H. Chan School of Public Health Organic Chemistry Laboratory (Boston, Massachusetts). Using high-resolution gas chromatography with electron capture detection, with primary and confirmatory capillary columns, cord serum samples were analyzed for HCB, p,p′-DDE, and 51 PCB congeners (see Korrick et al.33 for complete list of congeners and their concentrations). In our analysis, we used the sum of the four most prevalent PCBs in the general population during the 1990s and early 2000s (ΣPCB4; congeners 153, 118, 138, 180),36–38 because these were measured with the least error and are often used when exploring congener-specific effects. The limits of detection (LOD) for HCB, p,p′-DDE, and most individual PCB congeners were 0.01, 0.07, and 0.01 ng/g serum, respectively. For the organochlorine analyses, within-batch coefficients of variation (CV) ranged from 3% to 7% and between-batch CV from 20% to 39%.32,33,39 CVs reflected replicate analyses of cord serum quality assurance (QA)/quality control (QC) samples.
Cord whole blood samples for metal analyses were stored at 4°C and then digested prior to analysis at the Harvard T.H. Chan School of Public Health Trace Metals Laboratory (Boston, Massachusetts). We selected Pb and Mn as exposures for this analysis because a) there is animal and epidemiological evidence of associations between prenatal exposure to these metals and internalizing symptoms or other adverse neurobehavioral outcomes; and b) these metals were the only ones measured in the same biomarker as organochlorines in the NBC, and there was minimal missingness for cord blood samples in comparison with other neurotoxic metals (e.g., arsenic, cadmium, mercury) measured in maternal peripartum hair or toenails. Pb analyses were performed on average about 6 months after sample collection, and analyses of archived cord blood digests for Mn were performed as part of the 15-y assessments. Chemical analyses were conducted using isotope dilution (ID) inductively coupled plasma–mass spectrometry (ICP-MS) for assessment of concentrations of Pb (Sciex ELAN 5000; Perkin Elmer) and external calibration on a dynamic reaction cell (DRC) ICP–mass spectrometer (DRC-ICP-MS; ELAN 6100, Perkin Elmer) for assessment of concentrations of Mn. Metal concentrations were reported as the mean of five replicate measurements. The LOD for Pb and Mn was 0.02μg/dL whole blood. Laboratory recovery rates for QC standards and spiked samples ranged from 90% to 110%. The relative standard deviation (SD) percentage of replicate sample analyses was <5%.40
In the current analyses, machine concentration readings <LOD for organochlorines and metals were used to optimize statistical power and avoid biased exposure estimates due to censoring at the detection limit.41 For this analytic cohort, machine concentration readings were above zero for all included organochlorine and metals.
In addition to evaluating Pb levels in cord blood, for a subset of the NBC where comprehensive pediatric medical record review for routine Pb exposure screening data was possible (n=576), infant and childhood blood Pb levels were abstracted. Peak early childhood Pb, an available postnatal exposure index in the NBC that has been previously linked to Pb-associated cognitive deficits,42,43 was estimated as the maximum value between ages 12 to 36 months.
Outcome Assessments
On web-based and/or in-person assessments at the early-adult visit (age range: 19–25 y), participants completed the 7-item Generalized Anxiety Disorder scale (GAD-7)44 and 8-item Patient Health Questionnaire (PHQ-8).45 The GAD-7 and PHQ-8 are questionnaires that assess severity of anxiety and depressive symptoms, respectively, in the prior 2 wk, using a 4-point Likert scale (0–3). These questionnaires are used as diagnostic screening tools for generalized anxiety disorder and depression, dichotomized at a score of ≥10 to indicate moderate levels of symptoms and a clinical cut point for further evaluation.44,45
In analyses, we included GAD-7 and PHQ-8 scores that were fully complete or only lacked a response to one question. In a few cases where one response was not answered [n=3 (1%) for GAD-7 and PHQ-8, each], we prorated the score and rounded it to the nearest whole number (e.g., multiplying an incomplete GAD-7 score by seven-sixths). Web-based questionnaire scores were used in cases where participants also completed an in-person study visit, but in some instances participants only filled out these assessments in-person, and in those cases, we used the available scores from the in-person assessment (n=31 for GAD-7, n=32 for PHQ-8; 13%). Although web-based and in-person visits were typically completed 4 months apart [median (range): 4.0 (0.8–19) months], retest scores were positively and highly correlated, with Spearman’s rank correlations (rs) of 0.71 for GAD-7 and 0.81 for PHQ-8 scores. Reliability among retest scores remained high regardless of timing between questionnaires (e.g., rs for GAD-7 was 0.70 when retested in a time span of <4 months, and 0.71 for ≥4 months). Of the 240 participants who completed either questionnaire at least once, 239 participants had available GAD-7 and PHQ-8 scores, because two individuals completed only one of the assessments.
On an early-adult visit general questionnaire, participants reported whether, during the past 10 y, they had ever had a diagnosis of an anxiety disorder or depression by a psychologist, therapist, school professional, or other doctor. At this time, study participants also reported medications taken on a regular basis, including psychotropic drugs frequently used to treat anxiety or mood disorders (see Table S1).
Covariate Assessment
Just after the child’s birth, a trained study nurse abstracted information on the medical course of each NBC mother’s pregnancy and delivery, and on infant’s race/ethnicity, biological sex, and birth outcomes from hospital records. As previously described,34 an obstetrical risk score was derived to summarize adverse conditions prior to or during pregnancy, labor and delivery, and the neonatal period.46
During home visits occurring approximately 2 wk after birth, information on parental and household characteristics, including maternal age, marital status, diet, smoking during pregnancy, maternal and paternal education, and annual household income were ascertained via questionnaires completed by study mothers. Maternal marital status and smoking status, parental education, and annual household income questions were asked of study mothers at the 6-month, 8-y, and/or 15-y assessments to update information. Generally, child race/ethnicity was ascertained from infant medical record review data, but where those data were not available, race/ethnicity information collected from study mothers at 6-month or 8-y assessments was used. The child’s race/ethnicity was determined using questionnaire categories that included White; Hispanic; Black or African American; Asian American; Native American; and Other, with a free-text option for specifying. Because a significant portion of “Other” responses included “Cape Verdean,” it was added as its own category.
Modifying methods used in prior studies that examined childhood social disadvantage,47,48 we constructed a measure of prenatal social disadvantage specific to the NBC, based on five sociodemographic factors at birth. This index summarized data on maternal age (<20, ≥20 y); maternal and paternal education (≤high school graduate, post–high school education); maternal marital status (not married, married); and annual household income (<USD $20,000, ≥$20,000), scored as 1 or 0, for a total possible range of 0–5. In cases where data on one of these covariates were missing in our analytic cohort (n=3; 1%), we weighted the composite score to five responses and rounded to the nearest whole number (i.e., multiplied by five-fourths); if >1 covariate was missing data, then the prenatal social disadvantage index (PNSDI) score was considered missing. For analyses, we derived a dichotomized variable in which ≥3 was considered high disadvantage,15,40,49,50 because this cut point was sensitive to greater risk of exposure to other prenatal and childhood environmental and social stressors in the NBC.
The 8-y and 15-y study assessments included a home visit in which the Home Observation for Measurement of the Environment (HOME) was used to assess and summarize the quality of the home environment and parenting skills. Consisting of direct questions as well as observed items, the HOME is a validated tool that measures the physical environment, as well as the amount and quality of support and stimulation available to a child at home, with a higher score indicating a more enriching environment.51,52 We prioritized HOME scores (total score: 0–60) from the 15-y assessment, using the previous (8-y) scores if available in cases (n=16; 8%) when 15-y scores were not. For analyses, we derived a dichotomized variable in which less than the sample median HOME score value of 45 was considered a low score that indicated a poorer-quality home environment.
Statistical Analysis
For all analyses, we log2-transformed biomarkers of chemical exposure levels. Using prior literature and with consideration of avoiding overfitting regression models,53 we used directed acyclic graphs (Figure S1) to select a priori potential confounders, predictors of internalizing symptoms, and effect modifiers in regression models.54 The covariates included in all models were parental and household characteristics [PNSDI score (<3, ≥3), maternal smoking during pregnancy (no, yes), HOME score (≥median of 45, <45)] and participant characteristics [biological sex (female, male), age at early-adult assessment (continuous; years), race/ethnicity (Non-Hispanic White, underrepresented racial/ethnic group)]. Due to limited sample size in racial/ethnic identities besides Non-Hispanic White, we categorized Hispanic, Black or African American, Asian American, Native American, Cape Verdean, and Other racial/ethnic groups together for analysis, because we hypothesize these groups experience a greater, although varying, degree of discrimination and racism.
Because elevated GAD-7 and PHQ-8 scores were not a rare outcome and because we were interested in relative risk, we used modified Poisson regression with a robust error variance approach55 to estimate associations between prenatal chemical concentrations and risk of elevated symptoms. We first examined potential nonlinear associations between prenatal biomarker concentrations and the log risk ratio (RR) of outcomes using covariate-adjusted generalized additive models and, in most cases, found linear relationships. Some of the modeled splines indicated possible quadratic associations; consequently, we compared goodness of fit, using the quasi-likelihood under the independence model information criterion (QICu), between models including linear terms vs. linear and quadratic terms for exposures. In these cases, QICu values were better for the linear term model or not appreciably different (within 1 point of each other) between the two models (Figure S2); thus, for the sake of parsimony, all exposures were modeled as linear terms, with effect estimates representing risk per doubling cord blood biomarker concentration. We evaluated the presence of associations based on both magnitude and precision of effect estimates.
Based on prior literature, we identified sociodemographic and nonchemical stressors (hereafter referred to as other stressors), including participant biological sex, race/ethnicity (surrogate for experience with racial discrimination and/or structural racism), PNSDI, and adolescent HOME score; we explored whether these other stressors modified the association of prenatal chemical concentrations with internalizing symptoms. We did so by a) including product interaction terms between chemical concentrations and the other stressors in models (our primary approach) and b) running main effects models in data sets stratified by each dichotomized other stressor. These two approaches were used due to possible concerns of varying confounder relationships in exposure–outcome relationships by strata, in which the use of an interaction term would be inappropriate and bias results.
In main analyses, we fit single-exposure covariate-adjusted models for the overall analytic cohort and then models that additionally included product interaction terms between the exposure and one of the other stressors. Thus, for each exposure and outcome, we ran five separate models (i.e., main effects model; ones additionally including an exposure–sex, exposure–race/ethnicity, exposure–PNSDI, or exposure–HOME product interaction term). With respect to an explicit statistical significance level, we used α=0.05, but we also highlighted findings that were suggestive but did not meet this threshold. For possible heterogeneity in strata-specific effects, we assessed the magnitude and precision of associations within strata and interaction term p-values.
Additionally, to evaluate the effect of the five-chemical mixture on risk of elevated internalizing symptoms, we used quantile-based g-computation.56 Briefly, this approach specifies an index formed as a weighted combination of quantile scores for each pollutant in the model and applies g-computation to estimate the joint association of a quantile increase of the total mixture. Based on our assessment of generalized additive models, we assumed linearity of associations. We fit logistic regression models with no bootstraps to obtain weights for each exposure (i.e., strength/direction of the association between each exposure and outcome). Then, running 10,000 bootstrap samples, we estimated RRs (95% “pointwise” bounds) for the overall mixture effect. We ran quantile g-computation in the overall analytic cohort and in data sets stratified by the other stressors to obtain effect estimates. Then, in the overall analytic cohort, we fit models with a product term between the chemical mixture and each of the other stressors to test statistical interactions.
We investigated concordance between elevated GAD-7 and PHQ-8 symptoms in the overall analytic cohort. Furthermore, in the subset of participants who completed medical history questionnaires at the early-adult assessment (n=167), we first determined concordance between a) elevated GAD-7 symptoms and self-report of clinical diagnosis of an anxiety disorder and b) elevated PHQ-8 symptoms and self-report of a clinical diagnosis of depression. Then, we performed secondary single-chemical analyses considering self-report of an anxiety disorder and of depression in the prior 10 y as outcomes of interest. As with the GAD-7 and PHQ-8, we explored overall and strata-specific associations.
We performed sensitivity analyses in which we examined potential selection bias, residual confounding, or influence of outliers for the single-chemical main analysis of GAD-7 and PHQ-8 outcomes. First, we used inverse probability of censoring weighting (IPCW) to quantify potential selection bias due to cohort attrition and exclusion of participants without complete case information.57,58 We explored predictors of dropout (i.e., missing exposure, outcome, and/or covariate data) and then calculated stabilized truncated weights to include in the analysis. To evaluate possible coexposure confounding, we a) mutually adjusted for all chemical exposures in one model (i.e., by including the five separate chemical variables) and b) further adjusted models of organochlorine exposures for maternal seafood (continuous; servings per day) and local produce consumption (yes/no) during pregnancy, because these dietary exposures may act as negative confounders (i.e., consumption is associated with higher organochlorine exposures, and with positive neurobehavioral effects due to coexposure to essential nutrients).59 We also fit models that excluded participants on regular psychotropic medications, because these individuals’ symptoms may be altered by medication use. When considering cord blood Pb as the exposure of interest, we included peak early childhood Pb as a covariate to investigate the effect of Pb in these two separate exposure windows. Last, in generalized additive models, we observed that the HCB minimum value was an influential outlier, and so we ran all single-chemical HCB models excluding that observation.
Among 240 participants who completed the GAD-7 and/or PHQ-8 assessments in early adulthood, 234 had cord serum organochlorine measures, and of those, 204 had complete covariate data. For analyses of prenatal Pb and Mn, 228 and 220 participants had cord blood biomarker measures, respectively. Of those with these prenatal metal biomarkers, 198 and 193 participants had measures of cord blood Pb and Mn, respectively, as well as complete data on covariates. Data from 188 participants were used in five-chemical Poisson regression and quantile g-computation analyses. A total of 209 participants were included in at least one complete case analysis (see flow chart: Figure S3).
Results
Study Population Characteristics
Of those 209 participants included in the analytic data set, a large proportion of their parents [49% (n=101) of mothers and 63% (n=131) of fathers] had an education level of high school graduate or lower at the time of their child’s birth. One-quarter of mothers reported smoking during pregnancy (n=53, 25%), and 33% (n=68) were categorized with a high PNSDI (more disadvantaged). At the time of GAD-7 and PHQ-8 assessment, the mean±SD age of study participants was 22.1±1.5 y. Approximately two-thirds of early-adult study participants were female (n=139, 67%), and most identified as Non-Hispanic White (n=158, 76%) (Table 1). Of the 51 participants from underrepresented groups, most participants identified as Cape Verdean (n=20, 10%) or Hispanic (n=18, 9%); the others identified as Black or African American, Native American, Asian American, or another race/ethnicity. Among those categorized as belonging to a higher risk sociodemographic or psychosocial stressor group, concordances between high PNSDI, low HOME score, and the underrepresented racial/ethnic group were moderate to strong, with the greatest overlap between those with a high PNSDI also having a low HOME score (Table S2).
Table 1 Characteristics of 209 young adult New Bedford Cohort participants included in at least one complete case analysis and their associated RRs (95% CI) [RR (95% CI)] for elevated anxiety symptoms on the GAD-7 scale and elevated depressive symptoms on the PHQ-8.
Mean±SD or n (%) RR (95% CI) of elevated GAD-7 anxiety symptoms RR (95% CI) of elevated PHQ-8 depressive symptoms
Parental and household characteristics
PNSDI score (0–5 scale)
High (≥3) 68 (33) 2.33 (1.34, 4.02) 2.33 (1.46, 3.72)
Low (<3) 141 (67) Ref Ref
Maternal age at child’s birthb
<20 yb 24 (11) 2.01 (1.06, 3.82) 1.65 (0.92, 2.95)
≥20 y 185 (89) Ref Ref
Maternal educationa,b
High school graduate or lessb 101 (49) 2.82 (1.49, 5.34) 1.78 (1.08, 2.94)
Junior college, college, and/or postgraduate 107 (51) Ref Ref
Paternal educationa,b
High school graduate or lessb 131 (63) 1.97 (0.99, 3.91) 1.65 (0.94, 2.91)
Junior college, college, and/or postgraduate 76 (37) Ref Ref
Annual household income at birthb (USD)
<$20,000b 64 (31) 1.88 (1.09, 3.26) 2.01 (1.27, 3.20)
≥$20,000 145 (69) Ref Ref
Maternal marital status at birthb
Not marriedb 68 (33) 2.33 (1.34, 4.02) 2.16 (1.35, 3.44)
Married 141 (67) Ref Ref
Maternal prenatal smoking
Yes 53 (25) 1.57 (0.89, 2.78) 1.47 (0.90, 2.41)
No 156 (75) Ref Ref
HOME score (0–60 scale)c
Low [<median (45)] 93 (44) 1.89 (1.07, 3.35) 2.29 (1.38, 3.79)
High [≥median (45)] 116 (56) Ref Ref
Participant characteristics
Age at GAD-7 and PHQ-8 assessment (per year) 22.1±1.5 1.04 (0.87, 1.24) 1.11 (0.95, 1.29)
Sex
Female 139 (67) 1.52 (0.79, 2.93) 1.47 (0.84, 2.58)
Male 70 (33) Ref Ref
Race/ethnicity
Underrepresented groupd 51 (24) 1.90 (1.09, 3.30) 1.55 (0.95, 2.53)
Non-Hispanic White 158 (76) Ref Ref
Early adult anxiety and depressive symptom characteristics
GAD-7 anxiety symptoms score (0–21 scale)a 5.6±4.1 — —
PHQ-8 depressive symptoms score (0–24 scale) 6.4±5.7 — —
GAD-7 symptomsa
Elevated (≥10) 40 (19) — —
Not elevated (<10) 168 (81) — —
PHQ-8 symptoms
Elevated (≥10) 51 (24) — —
Not elevated (<10) 158 (76) — —
Psychotropic medication usea
Yes 15 (9) 3.29 (1.81, 5.98) 2.88 (1.72, 4.84)
No 149 (91) Ref Ref
Diagnosis of an anxiety disordera
Yes 43 (25) 4.58 (2.50, 8.40) 2.73 (1.66, 4.49)
No 129 (75) Ref Ref
Diagnosis of depressiona
Yes 36 (21) 5.77 (3.19, 10.4) 3.78 (2.34, 6.11)
No 136 (79) Ref Ref
Note: —, no data; CI, confidence interval; GAD-7, Generalized Anxiety Disorder scale; HOME: Home Observation for Measurement of the Environment; PHQ-8, Patient Health Questionnaire; PNSDI, prenatal social disadvantage index; Ref, reference; SD, standard deviation.
a For participants included in any of the complete case analyses, the following covariates have missing data (n missing): 1 with maternal education level, 2 for paternal education level; 1 with GAD-7 score; 45 with psychotropic medication use; 37 for diagnosis of an anxiety disorder/depression.
b These parental and household characteristics contribute toward the total score for the PNSDI.
c For 16 participants who were missing the HOME score at the 15-y assessment, the HOME score from the 8-y assessment was used.
d The underrepresented group includes those categorized as Cape Verdean, Hispanic, Black or African American, Native American, Asian American, or another race/ethnicity.
Participants who completed the GAD-7 and PHQ-8 assessments at the early-adult visit were more likely to be female and Non-Hispanic White and have low PNSDI and high HOME scores than those initially enrolled in the cohort who did not participate in these assessments (Table S3).
Organochlorine and Metal Exposures
The median biomarker levels of organochlorines in cord serum were 0.02, 0.33, and 0.20 ng/g for HCB, p,p′-DDE, and ΣPCB4, respectively. Median peak early childhood Pb concentration was higher than cord blood Pb (5.8 vs. 1.1μg/dL), and median cord blood Mn was 4.0μg/dL (Table 2). The cord serum organochlorines were positively and moderately correlated, with rs of 0.36 for HCB and ΣPCB4, 0.38 HCB and p,p′-DDE, and 0.61 p,p′-DDE and ΣPCB4. The cord blood metals, Pb and Mn, were positively correlated (rs=0.18). Across the two chemical classes, concentrations were not strongly correlated (rs: −0.03 to 0.14). Peak early childhood Pb was positively and moderately correlated with cord blood Pb (rs=0.39) (Figure S4).
Table 2 Distributions of cord blood biomarkers of organochlorine and metal exposures among 209 New Bedford Cohort participants included in at least one complete case analysis.
Mean±SD GM±GSD Minimum Percentiles Maximum
25th 50th 75th
Organochlorines (ng/g)a,b,c
HCB 0.03±0.02 0.02±1.88 0.001 0.02 0.02 0.03 0.13
p,p′-DDE 0.48±0.77 0.33±2.09 0.03 0.21 0.33 0.46 9.74
ΣPCB4d 0.28±0.40 0.20±2.23 0.01 0.12 0.20 0.31 4.41
Metals (μg/dL)a,b,c
Pb
Cord blood 1.43±1.12 1.17±1.87 0.07 0.77 1.14 1.66 9.39
Peak childhoode 6.56±3.79 5.60±1.79 1.00 4.00 5.82 8.66 22.8
Mn 4.30±1.67 4.05±1.40 1.92 3.19 4.01 4.91 14.6
Note: GM, geometric mean; GSD, geometric standard deviation; HCB, hexachlorobenzene; LOD, limit of detection; Mn, manganese; Pb, lead; PCB, polychlorinated biphenyl; p,p′-DDE, dichlorodiphenyldichloroethylene; SD, standard deviation; ΣPCB4, sum of four polychlorinated biphenyls congeners.
a Organochlorines were measured in cord serum; prenatal Pb and Mn were measured in cord whole blood; peak childhood Pb was measured in whole blood.
b For participants included in any of the complete case analyses, the organochlorine and metal exposures have missing data (n missing): 5 with HCB, p,p′-DDE, and ΣPCB4; 11 with cord blood Pb; 16 with peak early childhood Pb and Mn.
c The LOD for the cord blood biomarkers were: 0.01 ng/g for HCB, 0.07 ng/g for p,p′-DDE, 0.01 ng/g for individual PCB congeners, and 0.02μg/dL for Pb and Mn. In cases where concentrations were <LOD but >0, we used machine-read values in analyses.
d Sum of four PCB congeners (118, 138, 153, 180).
e Peak early childhood blood lead levels from ages 12 to 36 months.
Average concentrations of chemical biomarker levels were similar across those included and not included in any complete case analysis, except for cord blood and peak early childhood Pb, which were lower among those in complete case analyses (Table S3).
Symptoms and Diagnoses of Anxiety and Depression
In the analytic cohort, 19% (n=40) of participants had elevated GAD-7 anxiety symptoms, and 24% (n=51) had elevated PHQ-8 depressive symptoms (Table 1; Table S4). The strongest risk factors for elevated symptoms included high PNSDI, low HOME score, underrepresented racial/ethnic group, psychotropic medication use, and prior diagnosis of an anxiety disorder and depression (Table 1). Elevated anxiety and depressive symptoms were often comorbid, although there were more cases of elevated depressive symptoms in the absence of elevated anxiety symptoms in comparison with to anxiety without depressive symptoms (Table S5). Among participants who completed the GAD-7 and/or PHQ-8 and self-reported on medical diagnoses in the past 10 y, there was moderate concordance between elevated recent symptoms and prior clinical diagnosis. For example, 61% (n=20) of participants with elevated recent anxiety symptoms also self-reported a diagnosis of an anxiety disorder, and 50% (n=21) of participants with elevated recent depressive symptoms also self-reported a diagnosis of depression (Table S5).
Single-Chemical Associations with Elevated Anxiety and Depressive Symptoms
Overall (i.e., main effects models), a doubling of prenatal Pb level was associated with 1.41 (95% CI: 1.03, 1.94) times the risk of elevated anxiety symptoms. There was also a positive association between Pb and elevated depressive symptoms, but the RR was weaker than for anxiety, and CIs included the null [1.18 (95% CI: 0.91, 1.53)] (Figure 1; Table S6). Although CIs included the null, prenatal HCB and p,p′-DDE levels were also associated with higher risk of elevated anxiety, but not depressive, symptoms (Figure 2; Table S6). Prenatal ΣPCB4 and Mn levels were not associated with risk of elevated symptoms (Figures 1 and 2; Table S6). In comparison with effect estimates in unadjusted models (Table S7), covariate-adjusted RRs were typically attenuated for prenatal Pb and Mn biomarkers and slightly strengthened for HCB, p,p′-DDE, and ΣPCB4 biomarkers (Table S6).
Figure 1. Overall and sociodemographic/nonchemical stressor strata-specific RRs (95% CI) [RR (95% CI)] for early adulthood elevated anxiety (GAD-7) and depressive symptoms (PHQ-8) associated with a doubling of cord whole blood Pb and Mn (micrograms per deciliter). Note: Single-exposure modified Poisson regression models adjusted for parental and household characteristics (PNSDI score, maternal smoking during pregnancy) and participant characteristics [age at assessment, race/ethnicity (URG, NHW), sex, HOME score]. For each outcome and chemical, RRs were estimated from five models: one with no interaction term and four including a product interaction term between the chemical and each of the four specified other stressors. See Table S6 for corresponding numeric data. CI, confidence interval; GAD-7, Generalized Anxiety Disorder scale; HOME, Home Observation for Measurement of the Environment; Mn, manganese; NHW, Non-Hispanic White; Pb, lead; PHQ-8, Patient Health Questionnaire; PNSDI, prenatal social disadvantage index score; RR, risk ratio; URG, underrepresented group.
Figure 1 is a set of ten error bar graphs. On the top, the five error bar graphs are titled cord blood lead, plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 0 to 4 in unit increments (y-axis) across total cohort; sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Generalized Anxiety Disorder scale 7 (anxiety) and Patient Health Questionnaire- 8 (depression). At the bottom, the five error bar graphs are titled cord blood manganese, plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 0 to 4 in unit increments (y-axis) across total cohort; sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Generalized Anxiety Disorder scale 7 (anxiety) and Patient Health Questionnaire- 8 (depression).
Figure 2. Overall and sociodemographic/nonchemical stressor strata-specific RRs (95% CI) [RR (95% CI)] for early adulthood elevated anxiety (GAD-7) and depressive symptoms (PHQ-8) associated with a doubling of cord serum HCB, p,p′-DDE, and ΣPCB4 (congeners: 118, 138, 153, 180) (nanograms per gram). Note: Single-exposure modified Poisson regression models adjusted for parental and household characteristics (PNSDI score, maternal smoking during pregnancy) and participant characteristics [age at assessment, race/ethnicity (URG; NHW), sex, HOME score]. For each outcome and chemical, RRs were estimated from five models: one with no interaction term and four including a product interaction term between the chemical and each of the four specified other stressors; See Table S6 for corresponding numeric data. CI, confidence interval; GAD-7, Generalized Anxiety Disorder scale; HCB, hexachlorobenzene; HOME, Home Observation for Measurement of the Environment; NHW, Non-Hispanic White; PHQ-8, Patient Health Questionnaire; PNSDI, prenatal social disadvantage index; p,p′-DDE, dichlorobiphenyldichloroethylene; RR, risk ratio; ΣPCB4, sum of four polychlorinated biphenyls congeners; URG, underrepresented group.
Figure 2 is a set of fifteen error bar graphs. On the top, the five error bar graphs are titled cord serum hexachlorobenzene, plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 1 to 3 in unit increments (y-axis) across total cohort; sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Generalized Anxiety Disorder scale 7 (anxiety) and Patient Health Questionnaire- 8 (depressive). In the middle, the five error bar graphs are titled cord serum p,p′-dichlorobiphenyldichloroethylene, plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 1 to 3 in unit increments (y-axis) across total cohort; sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Generalized Anxiety Disorder scale 7 (anxiety) and Patient Health Questionnaire- 8 (depressive). At the bottom, the five error bar graphs are titled cord serum sum of four polychlorinated biphenyls congeners, plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 1 to 3 in unit increments (y-axis) across total cohort; sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Generalized Anxiety Disorder scale 7 (anxiety) and Patient Health Questionnaire- 8 (depressive).
In analyses exploring whether other stressors modified associations between prenatal chemical exposures and risk of elevated symptoms, we observed statistically significant interactions (p<0.05) for PNSDI and HOME score, but not sex or race/ethnicity, with at least one chemical (Figures 1 and 2; see Table S6 for interaction p-values). For prenatal Pb, p,p′-DDE, and ΣPCB4, a doubling in chemical concentration was associated with higher risk of elevated depressive symptoms for those with a low PNSDI but associated with no or lower risk of symptoms for those with a high PNSDI (more disadvantaged) [e.g., RR per doubling Pb: 1.99 (95% CI: 1.34, 2.98) for low vs. 0.85 (95% CI: 0.58, 1.25) for high PNSDI]. Although CIs included the null, a doubling of prenatal levels of p,p′-DDE and ΣPCB4 were also associated with higher risk of elevated anxiety symptoms only for those with a low PNSDI [RR=1.35 (95% CI: 0.98, 1.86) for p,p′-DDE, RR=1.33 (95% CI: 0.92, 1.91) for ΣPCB4]. Similarly, for those with a high HOME score (better home environment), a doubling of prenatal p,p′-DDE was associated with higher risk of elevated anxiety symptoms [RR=1.54 (95% CI: 1.20, 1.99)], and prenatal Pb was associated with higher risk of depressive symptoms [RR=1.85 (95% CI: 1.22, 2.81)]. The pattern of adverse risk of depressive symptoms only for those with a low PSNDI score or high HOME score (lower nonchemical stress groups) was also observed for prenatal HCB exposure, but CIs included the null and interactions were not statistically significant at the p=0.05 level [e.g., RR per doubling prenatal HCB: 1.29 (95% CI: 0.89, 1.88) in low PNSDI strata] (Figures 1 and 2; Table S6).
Additionally, prenatal HCB and p,p′-DDE levels were associated with higher risk of elevated anxiety symptoms among underrepresented racial/ethnic group participants, and associations were null for Non-Hispanic White participants [e.g., RR=1.86 (95% CI: 1.18, 2.95) for the underrepresented racial/ethnic group vs. 1.07 (95% CI: 0.66, 1.72) for Non-Hispanic White per doubling HCB] (Figure 2; Table S6). Although there was no statistically significant evidence of sex-specific associations with any chemical exposure, the adverse association between prenatal Pb and anxiety symptoms was stronger for males [RR per doubling Pb: 2.22 (95% CI: 1.18, 4.16) vs. 1.24 (95% CI: 0.86, 1.78) for females] (Figure 1; Table S6). Magnitude and directionality of effect estimates did not appreciably change when estimating strata-specific effects using interaction terms vs. stratifying the data set by the modifier (Table S8).
Mixture Associations with Anxiety and Depressive Symptoms
In the overall analytic cohort, quantile g-computation analysis showed that a quartile increase in the chemical mixture was moderately, albeit imprecisely with CIs that included the null, associated with an increased risk of elevated anxiety symptoms [RR=1.32 (95% CI: 0.71, 2.44)]. The overall mixture was not associated with risk of elevated depressive symptoms [RR=1.07 (95% CI: 0.67, 1.69)] (Table 3). The chemicals did not have coefficients in the same direction, with HCB, p,p′-DDE, and Pb having positive scaled effect weights (i.e., contributing to higher risk of elevated symptoms), and ΣPCB4 and Mn having negative scaled effect weights (i.e., lower risk of elevated symptoms) (Figure S5). In models stratified by the four other stressors, RRs between the overall mixture and elevated symptoms somewhat varied by strata, but most estimates were imprecise, and CIs overlapped for strata-specific estimates. The strongest example of a stressor modifying the association between the mixture and an outcome was in a model stratified by PNSDI score. A quartile increase in all chemicals was associated with increased risk of depressive symptoms in the low PNSDI group (lower nonchemical stress) and with lower risk of depressive symptoms in the high PNSDI group (higher nonchemical stress); specifically, with 2.04 (95% CI: 1.01, 4.12) times the risk of elevated depressive symptoms for those with a low PNSDI and 0.64 (95% CI: 0.32, 1.26) times the risk for those with a high PNSDI. This was the only stratified quantile g-computation model in which the interaction between the mixture and another stressor was statistically significant at the p=0.05 level (Table 3).
Table 3 Adjusted overall and sociodemographic/nonchemical stressor strata-specific RRs (95% CI) [RR (95% CI)] for early adulthood elevated anxiety (GAD-7) and depressive symptoms (PHQ-8) associated with a quartile increase in all five chemicals [cord blood HCB, p,p′-DDE, ΣPCB4 (congeners: 118, 138, 153, 180), lead, and manganese] in quantile g-computation models.
Anxiety (GAD-7) Depressive (PHQ-8)
n RR (95% CI)a,b pinteraction c n RR (95% CI)a,b pinteraction c
Overall 187 1.32 (0.71, 2.44) — 188 1.07 (0.67, 1.69) —
Sex-stratified — — 1.00 — — 0.85
Female 125 1.39 (0.68, 2.83) — 126 1.27 (0.73, 2.22) —
Male 62 1.46 (0.10, 20.6) — 62 0.86 (0.25, 2.98) —
Race/ethnicity-stratified — — 0.36 — — 0.72
Underrepresented group 47 1.68 (0.72, 3.94) — 48 0.87 (0.39, 1.95) —
Non-Hispanic White 140 0.99 (0.36, 2.72) — 140 1.05 (0.55, 1.99) —
PNSDI-stratified — — 0.73 — — 0.04
High (≥3) 64 1.19 (0.39, 3.59) — 65 0.64 (0.32, 1.26) —
Low (<3) 123 1.66 (0.45, 6.06) — 123 2.04 (1.01, 4.12) —
HOME-stratified — — 0.78 — — 0.30
Low (<45) 85 1.22 (0.49, 3.02) — 86 0.85 (0.49, 1.50) —
High (≥45) 102 1.71 (0.35, 8.35) — 102 1.86 (0.72, 4.76) —
Note: —, no data; CI, confidence interval; GAD-7, Generalized Anxiety Disorder scale; HCB, hexachlorobenzene; HOME, Home Observation for Measurement of the Environment; PHQ-8, Patient Health Questionnaire; PNSDI, prenatal social disadvantage index; p,p′-DDE, dichlorodiphenyldichloroethylene; RR, risk ratio; ΣPCB4, sum of four polychlorinated biphenyls congeners.
a Adjusted for parental and household characteristics (PNSDI, maternal smoking during pregnancy, HOME score) and participant characteristics (sex, race/ethnicity, age at assessment).
b Effect estimates obtained from quantile g-computation models run in data sets stratified by the four other stressors.
c Interaction term p-values obtained from quantile g-computation models run in overall analytic data set and fit with a product term between the mixture and each of the four other stressors.
Single-Chemical Associations with Physician Diagnosis of Anxiety Disorders or Depression
Prenatal p,p′-DDE and Pb levels were moderately associated with higher risk of diagnosis of an anxiety disorder or depression in the prior ten years, but CIs included the null [e.g., RR=1.20 (95% CI: 0.90, 1.61)] for diagnosis of an anxiety disorder per doubling Pb]. Strata-specific associations by HOME score were observed with a doubling of prenatal p,p′-DDE and ΣPCB4, with higher risk of diagnoses for those with a high HOME score and lower risk for those with a low HOME score (poorer-quality home environment) [e.g., RR for depression diagnosis: 1.51 (95% CI: 1.08, 2.11) for high HOME score vs. 0.67 (95% CI: 0.45, 1.00) for low HOME score]. A similar strata-specific pattern was present for ΣPCB4 and PNSDI with diagnosis of depression [RR=1.26 (95% CI: 0.89, 1.77) for low PNSDI vs. 0.70 (95% CI: 0.46, 1.05) for high PNSDI (more disadvantaged) per doubling ΣPCB4]. Additionally, a doubling of prenatal p,p′-DDE was associated with higher risk of anxiety diagnosis among the underrepresented racial/ethnic group participants [RR=1.27 (95% CI: 1.01, 1.60)], but not for Non-Hispanic White participants [RR=0.99 (95% CI: 0.64, 1.53)]. Prenatal HCB and Mn levels were not associated with self-report of physician-diagnosed anxiety or depression (Table S9).
Sensitivity Analyses
Findings were largely unchanged when using IPCW vs. complete case analysis, although the association between doubling of prenatal HCB and elevated anxiety symptoms in underrepresented racial/ethnic group participants was strengthened with IPCW [RR=2.18 (95% CI: 1.49, 3.18)] (Figure 3; Table S10). When excluding the minimum HCB outlier, the association with depressive symptoms did not change, but prenatal HCB was more strongly associated with elevated anxiety symptoms [i.e., overall RR=1.50 (95% CI: 1.09, 2.07) per doubling HCB] (Table S11). We saw no meaningful changes to the magnitude of associations when further adjusting for coexposure by other chemicals (Figure 3; Table S12) or when additionally adjusting for potential dietary confounders (Figure 3; Table S13). Effect estimates also did not appreciably differ when comparing results from the subset that had information available on medication use (Table S14) to the further restricted subset of those not taking regular psychotropic medications (Table S15). However, the overall RR of prenatal Pb and elevated anxiety symptoms was attenuated in the 155 participants with data available on medication use (Table S14) and with exclusion of those on medications (Table S15) vs. the 197 participants in the complete case subset (Table S6).
Figure 3. Overall RRs (95% CI) [RR (95% CI)] for early adulthood elevated anxiety (GAD-7) and depressive symptoms (PHQ-8) associated with a doubling of cord serum HCB, p,p’-DDE, ΣPCB4 (congeners: 118, 138, 153, 180) (nanograms per gram), and cord blood Pb and Mn (micrograms per deciliter), in the main models vs. sensitivity models using a) IPCW, b) adjusting for all five chemical coexposures, and c) adjusting for dietary factors associated with organochlorine exposure. Note: Modified Poisson regression models adjusted for parental and household characteristics (PNSDI score, maternal smoking during pregnancy) and participant characteristics (age at assessment, race/ethnicity, sex, HOME score). See Table S6 (main), Table S9 (IPCW), Table S11 (five exposures), and Table S12 (diet adjustment) for corresponding numeric data of overall and stratified models. CI, confidence interval; GAD-7, Generalized Anxiety Disorder scale; HCB, hexachlorobenzene; HOME, Home Observation for Measurement of the Environment; IPCW, inverse probability of censoring weighted; Mn, manganese; Pb, lead; PHQ-8, Patient Health Questionnaire; PNSDI, prenatal social disadvantage index; p,p′-DDE, dichlorobiphenyldichloroethylene; RR risk ratio; ΣPCB4, sum of four polychlorinated biphenyls congeners.
Figure 3 is a set of ten error bar graphs. On the top, the five error bar graphs are titled hexachlorobenzene, p,p′-dichlorobiphenyldichloroethylene, sum of four polychlorinated biphenyls congeners, lead, and manganese under Generalized Anxiety Disorder scale 7 (anxiety), plotting Main, inverse probability of censoring weighted, five-exposures, and diet adjustment (y-axis) across relative risks (95 percent confidence interval) per doubling chemical level, ranging from 0.5 to 2.5 in increments of 0.5 (x-axis) for Main, inverse probability of censoring weighted, five-exposures, and diet adjustment. At the bottom, the five error bar graphs are titled hexachlorobenzene, p,p′-dichlorobiphenyldichloroethylene, sum of four polychlorinated biphenyls congeners, lead, and manganese under Patient Health Questionnaire- 8 (depressive), plotting Main, inverse probability of censoring weighted, five-exposures, and diet adjustment (y-axis) across relative risks (95 percent confidence interval) per doubling chemical level, ranging from 0.5 to 2.5 in increments of 0.5 (x-axis) for Main, inverse probability of censoring weighted, five-exposures, and diet adjustment.
In regression models including both prenatal and peak early childhood blood Pb, cord blood Pb was associated with higher risk of elevated anxiety and depressive symptoms, whereas peak childhood Pb was not associated with elevated symptoms. In analyses assessing strata-specific estimates, associations with peak childhood Pb remained largely null, except for associations with higher risk of elevated anxiety symptoms among the underrepresented racial/ethnic group participants and with lower risk of depressive symptoms for male, Non-Hispanic White, low PNSDI, and high HOME score participants (Figure 4; Table S16).
Figure 4. Overall and sociodemographic/nonchemical stressor strata-specific RRs (95% CI) [RR (95% CI)] for early adulthood moderately elevated anxiety (GAD-7) and depressive symptoms (PHQ-8) associated with a doubling of prenatal and postnatal (i.e., peak early childhood) Pb (micrograms per deciliter) (n=197 for GAD-7; n=198 for PHQ-8). Note: Modified Poisson regression models, including both prenatal and postnatal lead, were adjusted for parental and household characteristics (PNSDI score, maternal smoking during pregnancy) and participant characteristics [age at assessment, race/ethnicity (URG, NHW), sex, HOME score]. See Table S16 for corresponding numeric data. CI, confidence interval; GAD-7, Generalized Anxiety Disorder scale; HOME, Home Observation for Measurement of the Environment; Mn, manganese; NHW, Non-Hispanic White; Pb, lead; PHQ-8, Patient Health Questionnaire; PNSDI, prenatal social disadvantage index; RR, risk ratio; URG, underrepresented group.
Figure 4 is a set of ten error bar graphs. On the top, the five error bar graphs are titled Generalized Anxiety Disorder scale 7 (anxiety), plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 0 to 4 in unit increments (y-axis) across sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Prenatal and Postnatal. At the bottom, the five error bar graphs are titled Patient Health Questionnaire- 8 (depressive), plotting relative risks (95 percent confidence interval) per doubling chemical level, ranging from 0 to 4 in unit increments (y-axis) across sex, including female and male; Race or ethnicity, including Underrepresented group and Non-Hispanic White; prenatal social disadvantage index score, including high (greater than or equal to 3) and Low (less than 3); and Home Observation for Measurement of the Environment, including low (less than 45) and high (greater than or equal to 45) (x-axis) for Prenatal and Postnatal.
Discussion
In our study of NBC participants followed from birth through early adulthood, we found evidence of associations between greater prenatal exposure to HCB, p,p′-DDE, and Pb and increased risk of elevated anxiety symptoms at approximately age 22 y. These adverse associations were strongest with prenatal Pb exposure, and more modest with the organochlorine exposures. In addition, higher prenatal organochlorines and Pb levels, as well as the overall effect of the mixture, were associated with increased risk of elevated depressive symptoms for those categorized as exposed to lower levels of nonchemical stress (i.e., low PNSDI, high HOME score). These findings suggest a possible “signal-to-noise ratio” scenario60 in which strong psychosocial risk factors for internalizing disorders obscure modest additional risk from prenatal chemical exposures. The strata-specific associations among those with a high PNSDI (more disadvantaged) or low HOME score (poorer-quality home environment) were mostly null, but in some analyses, prenatal p,p′-DDE and ΣPCB4 appeared to be paradoxically associated with lower risk of elevated symptoms in these other stressor groups. In the total analytic cohort, the overall mixture was not associated with risk of depressive symptoms, but there was evidence of a relatively strong in magnitude, but imprecise, association with anxiety symptoms. We observed an interaction between the overall mixture and PNSDI score, in which a quartile increase in the mixture was associated with greater risk of depressive symptoms for the low PNSDI group (lower nonchemical stress group). Other than that, the overall effect of the mixture appeared to slightly vary in the presence other stressors, but strata-specific CIs were wide and overlapped. In general, prenatal ΣPCB4 and Mn were not adversely associated with risk of early adulthood anxiety and depressive symptoms. We did not observe strong evidence of sex-specific effects of these prenatal chemicals on risk for these symptoms in early adulthood.
Consistent with associations of prenatal Pb and self-report anxiety symptoms in adolescence in the NBC,15 higher prenatal Pb levels were associated with increased risk of elevated anxiety symptoms in early adulthood. Prior studies of prenatal Pb exposure and internalizing symptoms have mainly assessed outcomes in early or mid-childhood,16,21–24 prior to core risk periods for the onset of anxiety and depressive disorders,3,26 and largely have not observed associations. Our study findings suggested that in utero exposure to Pb may be one factor that increases anxiety vulnerability, possibly via the hypothesized mechanisms of disruption of dopaminergic and GABAergic systems9; notably, these effects may not manifest until later in development during or after critical risk periods for anxiety. Similar findings have been reported in rodent models, with low-level exposure to Pb during pregnancy and lactation shown to have little impact on weaned rats but to be associated with heightened anxiety behaviors in adult rats.14
Research on postnatal Pb exposure, either prospectively examining childhood exposure61 or cross-sectionally assessing Pb levels in adulthood,62 has demonstrated increased risk for internalizing symptoms in early adulthood. Despite the facts that, in the NBC, average peak early childhood blood Pb levels were a) higher than those measured in cord blood and b) higher than the CDC’s recently updated (in 2021) and previous blood reference levels of 3.5 and 5μg/dL, respectively,63 we observed that prenatal Pb was more consistently associated with risk of elevated symptoms than early-childhood blood Pb. Although the literature on early Pb exposure and child neurodevelopment supports the postnatal period as a critical time,64 our study indicated that prenatal exposure may also influence risk for developing internalizing problems, particularly anxiety symptoms.
Although most prior literature, including in the NBC, has reported no associations between prenatal organochlorine levels and subsequent internalizing symptoms,15–21 we found evidence of prenatal HCB and p,p′-DDE as risk factors for elevated internalizing symptoms in early adulthood, in most cases with stronger associations in certain levels of the sociodemographic or nonchemical stressors. The adverse associations of prenatal HCB and p,p′-DDE levels and risk of elevated anxiety symptoms was stronger for young adult participants who identified as underrepresented racial/ethnic group participants. This finding aligns with the hypothesis that individuals experiencing stress due to racial discrimination and/or structural racism may have increased susceptibility to the effects of environmental exposures.65,66 It is important to note that our study did not have the power to look at more distinct categories of race/ethnicity. Experiences with racial discrimination and structural racism may differ vastly for those who identify as Cape Verdean, Hispanic, Black or African American, Native American, Asian American, or in other underrepresented racial/ethnic groups. Our study used a composite definition of race/ethnicity and thus was unable to parse out possible strata-specific effects by more meaningful constructs. Thus, to better understand possible synergistic effects between the stressors of chemicals and racism on subsequent mental health, future studies should be conducted in racially diverse cohorts that measure participants’ personal experiences with discrimination and other manifestations of structural racism using validated questionnaires (e.g., Everyday Discrimination Scale67).
Aside from potential differential susceptibility to p,p′-DDE-associated anxiety by race/ethnicity, associations of prenatal p,p′-DDE exposure with higher risk of internalizing symptoms were otherwise only apparent for those in lower stressor groups as indicated by PNSDI and HOME score ratings. This pattern of findings suggests a possible signal-to-noise ratio problem60 in which the modest effects of p,p′-DDE are only detectable among those experiencing fewer psychosocial stressors. Strata-specific patterns may be more similar for PNSDI and HOME score strata in comparison with race/ethnicity strata, because a large portion of participants with a low HOME score (poorer-quality home environment) were also categorized with high PNSDI (more disadvantaged).
In comparison with most other studies of prenatal organochlorine exposures and internalizing symptoms, our study is unique in that we evaluated anxiety and depressive symptoms in early adulthood. In the NBC, a study in which prenatal concentrations of organochlorines were relatively low in comparison with other high-risk exposure cohorts,33,68 prenatal HCB, p,p′-DDE, and ΣPCB4 were not associated with mid-childhood and adolescent internalizing symptoms, even when considering sex-specific effects.15 Similarly, other studies exploring prenatal HCB16,18,19 and p,p′-DDE16,18–20 exposures and symptoms in childhood observed no adverse associations. One study with follow-up through early adulthood found no associations of these organochlorines with diagnosis of depression.17 When we considered self-report of physician diagnosed anxiety or depression as outcomes, we also observed no associations with prenatal HCB and mainly strata-specific adverse associations with p,p′-DDE and ΣPCB4. Our findings a) highlight the importance of considering coexposures to chemical and other stressors because adverse associations with HCB and p,p′-DDE appeared to be strata-specific and b) suggest that modest effects of prenatal chemical exposures may be more likely to be observed with self-report of symptoms rather than physician diagnoses. As is the case for metals such as Pb, organochlorines are also hypothesized to dysregulate dopamine-mediated, glutamate, and GABAergic functions8,10 and thereby could contribute to later susceptibility to anxiety or depressive symptoms, which may not become apparent until adulthood.
With prenatal HCB, p,p′-DDE, and Pb exposures, we found more consistent evidence of associations with anxiety in comparison with depressive symptoms. Although anxiety and depressive symptoms are often comorbid,69 as was the case in our study, and several of the hypothesized biological pathways by which prenatal exposures may alter neurobiological functions would be expected to increase vulnerability to both types of symptoms, it is plausible that neural pathways associated specifically with an anxiety phenotype are more susceptible to early-life chemical insults. For example, anxiety and depression phenotypes have been proposed to align with different constructs within the systems of negative valence (anxiety with “acute threat” or “potential threat” and depression with “loss”), and these constructs are associated with differing neural and physiological pathways.70,71 Thus, certain chemical exposures may dysregulate pathways more relevant to anxiety vulnerability (e.g., bed nucleus of the stria terminalis). Research is needed to explicate the potential neural mechanisms by which specific chemical exposures exert their effects.
We did not observe consistent associations between prenatal ΣPCB4 exposure and internalizing symptoms in early adulthood. In fact, in some cases, cord blood levels of ΣPCB4, as well as p,p′-DDE, were associated with lower risk of elevated symptoms for those categorized as having a high PNSDI and low HOME score (i.e., hypothesized higher nonchemical stressor groups). These apparent protective associations, in the opposite direction than hypothesized, could be due to chance findings in an analytic cohort with a moderate sample size or differential sources/levels of confounding in nonchemical stressor strata. In addition to fitting models with interaction terms, we stratified the data set by PNSDI and HOME score groups to explore whether this strata-specific pattern was due to differing associations of confounders with exposure–outcome relationships for high vs. low groups. However, this did not appear to be the case, because RRs remained similar when estimating effect estimates both ways. Alternatively, a small study of mother–child pairs found that greater maternal stress and adversity during pregnancy was associated with decreased cord blood DNA methylation of the oxytocin receptor gene.72 This decrease may lead to increased expression and easier activation of the oxytocin receptor gene in childhood, in turn mitigating the adverse impacts of stress and exerting a protective effect toward chemical exposures. Further exploration into this potential and other mechanisms is needed to better understand whether higher nonchemical maternal stress could induce certain protective adaptations.
Prenatal Mn levels were not associated with internalizing symptoms here, in contrast to adverse associations observed for girls in mid-childhood and adolescence in the NBC.15 When analogous outcomes were evaluated in the NBC at earlier ages, symptoms were mainly assessed using continuous measures (i.e., Conners’ Rating Scale,73 Behavior Assessment System for Children, Second Edition74) whereas the GAD-7 and PHQ-8 are binary screening tools. Thus, this attenuation of the sex-specific association by early adulthood could be due to reduced power to detect moderate differences with a dichotomized outcome and smaller sample size. It is also possible that prenatal Mn exposure does not influence anxiety and depressive symptoms evident in early adulthood. In other cohorts with follow-up through mid-childhood,24,25 there is limited, mixed evidence of associations between prenatal Mn and subsequent internalizing symptoms. Because Mn is an essential metal, prenatal exposure via diet is fundamental for growth and neurodevelopment, unlike prenatal exposure to Pb. However, deficient or excess levels Mn can cause neurotoxicity.75 Thus, if NBC participants were largely exposed to nutritionally optimal Mn levels while in utero, it is plausible that Mn would have no impact on anxiety and depressive symptoms in early adulthood.
Although this study is novel by virtue of its prospective longitudinal assessment of the effects of prenatal exposures on internalizing symptom risk in early adulthood, our findings may have been influenced by selection bias due to loss to follow-up of 20 y or longer, residual and unmeasured confounding, a limited sample size, and type II error. We aimed to address possible selection bias due to cohort attrition with use of IPCW and residual confounding bias by adjusting for prenatal seafood consumption, local produce consumption, and chemical coexposures. Although these sensitivity analyses showed that results were robust, IPCW model misspecification or dietary measurement error still may result in remaining bias. Our study also had a modest overall sample size, with relatively small numbers of participants in some of the other stressor subgroups. Consequently, we were only able to evaluate participants from underrepresented racial/ethnic groups in a composite group, and strata-specific findings may not be generalizable. Moreover, our analyses may not have been sufficiently powered to detect modest effect sizes or interactive effects. External validity of our results may have also been impacted by the timing of biomarker collection (1993–1998) and changes in exposure distributions of organochlorines and metals over time. Additionally, we did not correct for multiple comparisons in our models. Therefore, some of the observed associations, including both those that supported and contradicted our a priori hypotheses, may be the result of type II error (i.e., false positives). We elected to prioritize risk of false positives over false negatives; thus, replication of our findings, particularly those examining interactive effects between prenatal chemical exposures and other stressors, is needed in other cohort studies. Furthermore, missing data on medical history at the early-adult visit (i.e., different analytic cohorts for sensitivity analyses) limited our ability to determine whether treatment for a psychiatric illness contributed to outcome measurement error. Last, we were unable to assess other critical windows of exposure vulnerability, because all chemicals, except peak early childhood Pb, were only measured in cord blood. Although this limits our ability to understand specific exposure windows when the developing brain may be most susceptible to chemical impacts, postnatal exposure levels would not confound the observed associations, unless they were highly correlated with prenatal ones; thus, their absence does not detract from the validity of these findings.
That said, elevated anxiety and depressive symptoms were prevalent among the young adult NBC participants, and we observed associations between HCB, p,p′-DDE, and Pb and internalizing symptoms and differing associations in strata of other stressor groups. In addition, our study had several strengths and addressed gaps in the extant literature. In limited prior studies, there has been evaluation of coexposures to prenatal organochlorines or metals with mid-childhood internalizing symptoms,18,19,24 single-chemical associations with early adulthood outcomes,17 or sex-specific effects.25 However, to our knowledge, the present study is distinct as a singular study that examined prenatal organochlorines and metal exposures as individual components and as a mixture, assessed internalizing symptoms in early adulthood using both self-report questionnaires and physician-diagnosed disorders, and considered interactions between chemical exposures and several sociodemographic and nonchemical stressors. We used quantile g-computation, in addition to standard parametric regression models, to better understand how simultaneous exposure to these two classes of chemicals may influence risk of elevated symptoms. Additionally, we examined the outcome in early adulthood, a core risk period for anxiety and depressive disorders. By including assessments a) not dependent on access and use of medical care and b) not subject to variability in clinical practice (i.e., standardized questionnaires), we minimized outcome misclassification that can occur when only considering physician diagnosis of internalizing disorders. The use of self-report questionnaires also better captures subclinical disease and thus may have greater sensitivity for detecting more modest associations. In fact, we observed stronger magnitudes of associations for GAD-7 anxiety symptoms than for diagnosis of an anxiety disorder with HCB, p,p′-DDE, and Pb. Finally, the NBC is socioeconomically diverse, which not only improves the generalizability of the findings, but also allowed us to consider a number of other stressors as potential modifiers of the associations between chemical exposures and internalizing symptoms.
In summary, our findings indicated adverse associations of prenatal HCB, p,p′-DDE, and Pb exposure with risk of early adulthood internalizing symptoms, particularly anxiety symptoms. Prevalence of anxiety in young adults has been rapidly increasing in recent years1; thus, understanding how early-life chemical exposures, in addition to more well-studied psychosocial stressors, influence susceptibility to internalizing disorders is critical. Because anxiety is a risk factor for poor work and social and psychiatric outcomes over the life course,76 it is important to consider that reductions to chemical exposures during pregnancy, even if associations are modest, may improve the overall population’s well-being.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
Support for this research was provided by grants P42ES005947, R01ES014864, P30ES000002, and R21ES024513 from the National Institute of Environmental Health Sciences (NIEHS/NIH). L.B.R. was additionally supported by the CDC/NIOSH Harvard Education and Research Center (T42OH008416).
==== Refs
References
1. Goodwin RD, Weinberger AH, Kim JH, Wu M, Galea S. 2020. Trends in anxiety among adults in the United States, 2008–2018: rapid increases among young adults. J Psychiatr Res 130 :441–446, PMID: , 10.1016/j.jpsychires.2020.08.014.32905958
2. Mojtabai R, Olfson M, Han B. 2016. National Trends in the prevalence and treatment of depression in adolescents and young adults. Pediatrics 138 (6 ):e20161878, PMID: , 10.1542/peds.2016-1878.27940701
3. Koenen KC, Rudenstine S, Susser E, Galea S. 2013. A Life Course Approach to Mental Disorders. Oxford, UK: Oxford University Press.
4. Blanco C, Rubio J, Wall M, Wang S, Jiu CJ, Kendler KS. 2014. Risk factors for anxiety disorders: common and specific effects in a national sample. Depress Anxiety 31 (9 ):756–764, PMID: , 10.1002/da.22247.24577934
5. Sander JB, McCarty CA. 2005. Youth depression in the family context: familial risk factors and models of treatment. Clin Child Fam Psychol Rev 8 (3 ):203–219, PMID: , 10.1007/s10567-005-6666-3.16151618
6. Bellinger DC. 2013. Prenatal exposures to environmental chemicals and children’s neurodevelopment: an update. Saf Health Work 4 (1 ):1–11, PMID: , 10.5491/SHAW.2013.4.1.1.23515885
7. Korrick SA, Sagiv SK. 2008. Polychlorinated biphenyls, organochlorine pesticides and neurodevelopment. Curr Opin Pediatr 20 (2 ):198–204, PMID: , 10.1097/MOP.0b013e3282f6a4e9.18332718
8. Bell MR. 2014. Endocrine-disrupting actions of PCBs on brain development and social and reproductive behaviors. Curr Opin Pharmacol 19 :134–144, PMID: , 10.1016/j.coph.2014.09.020.25310366
9. Mason LH, Harp JP, Han DY. 2014. Pb neurotoxicity: neuropsychological effects of lead toxicity. Biomed Res Int 2014 :1–8, PMID: , 10.1155/2014/840547.24516855
10. Addae C, Cheng H, Martinez-Ceballos E. 2013. Effect of the environmental pollutant hexachlorobenzene (HCB) on the neuronal differentiation of mouse embryonic stem cells. Int J Environ Res Public Health 10 (10 ):5244–5256, PMID: , 10.3390/ijerph10105244.24157519
11. Betharia S, Maher TJ. 2012. Neurobehavioral effects of lead and manganese individually and in combination in developmentally exposed rats. Neurotoxicology 33 (5 ):1117–1127, PMID: , 10.1016/j.neuro.2012.06.002.22732189
12. van den Bosch M, Meyer-Lindenberg A. 2019. Environmental exposures and depression: biological mechanisms and epidemiological evidence. Annu Rev Public Health 40 :239–259, PMID: , 10.1146/annurev-publhealth-040218-044106.30633709
13. de Souza Lisboa SF, Gonçalves G, Komatsu F, Queiroz CA, Almeida AA, Moreira EG. 2005. Developmental lead exposure induces depressive-like behavior in female rats. Drug Chem Toxicol 28 (1 ):67–77, PMID: , 10.1081/dct-39696.15720036
14. Moreira EG, Vassilieff I, Vassilieff VS. 2001. Developmental lead exposure: behavioral alterations in the short and long term. Neurotoxicol Teratol 23 (5 ):489–495, PMID: , 10.1016/S0892-0362(01)00159-3.11711252
15. Rokoff LB, Shoaff JR, Coull BA, Enlow MB, Bellinger DC, Korrick SA. 2022. Prenatal exposure to a mixture of organochlorines and metals and internalizing symptoms in childhood and adolescence. Environ Res 208 :112701, 10.1016/j.envres.2022.112701.35016863
16. Sioen I, Den Hond E, Nelen V, Van de Mieroop E, Croes K, Van Larebeke N, et al. 2013. Prenatal exposure to environmental contaminants and behavioural problems at age 7–8 years. Environ Int 59 :225–231, PMID: , 10.1016/j.envint.2013.06.014.23845936
17. Strøm M, Hansen S, Olsen SF, Haug LS, Rantakokko P, Kiviranta H, et al. 2014. Persistent organic pollutants measured in maternal serum and offspring neurodevelopmental outcomes–a prospective study with long-term follow-up. Environ Int 68 :41–48, PMID: , 10.1016/j.envint.2014.03.002.24704638
18. Jedynak P, Maitre L, Guxens M, Gützkow KB, Julvez J, López-Vicente M, et al. 2021. Prenatal exposure to a wide range of environmental chemicals and child behaviour between 3 and 7 years of age – an exposome-based approach in 5 European cohorts. Sci Total Environ 763 :144115, PMID: , 10.1016/j.scitotenv.2020.144115.33422710
19. Maitre L, Julvez J, López-Vicente M, Warembourg C, Tamayo-Uria I, Philippat C, et al. 2021. Early-life environmental exposure determinants of child behavior in Europe: a longitudinal, population-based study. Environ Int 153 :106523, PMID: , 10.1016/j.envint.2021.106523.33773142
20. Rosenquist AH, Høyer BB, Julvez J, Sunyer J, Pedersen HS, Lenters V, et al. 2017. Prenatal and postnatal PCB-153 and p,p’-DDE exposures and behavior scores at 5–9 years of age among children in Greenland and Ukraine. Environ Health Perspect 125 (10 ):107002, PMID: , 10.1289/EHP553.28974479
21. Plusquellec P, Muckle G, Dewailly E, Ayotte P, Bégin G, Desrosiers C, et al. 2010. The relation of environmental contaminants exposure to behavioral indicators in Inuit preschoolers in arctic Quebec. Neurotoxicology 31 (1 ):17–25, PMID: , 10.1016/j.neuro.2009.10.008.19854214
22. Bellinger D, Leviton A, Allred E, Rabinowitz M. 1994. Pre- and postnatal lead exposure and behavior problems in school-aged children. Environ Res 66 (1 ):12–30, PMID: , 10.1006/enrs.1994.1041.8013435
23. Arbuckle TE, Davis K, Boylan K, Fisher M, Fu J. 2016. Bisphenol A, phthalates and lead and learning and behavioral problems in Canadian children 6–11 years of age: CHMS 2007–2009. Neurotoxicology 54 :89–98, PMID: , 10.1016/j.neuro.2016.03.014.27021348
24. Horton MK, Hsu L, Claus Henn B, Margolis A, Austin C, Svensson K, et al. 2018. Dentine biomarkers of prenatal and early childhood exposure to manganese, zinc and lead and childhood behavior. Environ Int 121 (pt 1 ):148–158, PMID: , 10.1016/j.envint.2018.08.045.30205321
25. Mora AM, Arora M, Harley KG, Kogut K, Parra K, Hernández-Bonilla D, et al. 2015. Prenatal and postnatal manganese teeth levels and neurodevelopment at 7, 9, and 10.5 years in the CHAMACOS cohort. Environ Int 84 :39–54, PMID: , 10.1016/j.envint.2015.07.009.26209874
26. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE. 2005. Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the National Comorbidity Survey Replication. Arch Gen Psychiatry 62 (6 ):593–602, PMID: , 10.1001/archpsyc.62.6.593.15939837
27. Sourander A, Helstelä L, Helenius H. 1999. Parent-adolescent agreement on emotional and behavioral problems. Soc Psychiatry Psychiatr Epidemiol 34 (12 ):657–663, PMID: , 10.1007/s001270050189.10703276
28. van der Ende J, Verhulst FC, Tiemeier H. 2012. Agreement of informants on emotional and behavioral problems from childhood to adulthood. Psychol Assess 24 (2 ):293–300, PMID: , 10.1037/a0025500.21928910
29. Appleton AA, Holdsworth EA, Kubzansky LD. 2016. A Systematic review of the interplay between social determinants and environmental exposures for early-life outcomes. Curr Environ Health Rep 3 (3 ):287–301, PMID: , 10.1007/s40572-016-0099-7.27344145
30. Barrett ES, Padula AM. 2019. Joint impact of synthetic chemical and non-chemical stressors on children’s health. Curr Environ Health Rep 6 (4 ):225–235, PMID: , 10.1007/s40572-019-00252-6.31637664
31. Cowell WJ, Wright RJ. 2017. Sex-specific effects of combined exposure to chemical and non-chemical stressors on neuroendocrine development: a review of recent findings and putative mechanisms. Curr Envir Health Rpt 4 (4 ):415–425, 10.1007/s40572-017-0165-9.
32. Choi AL, Levy JI, Dockery DW, Ryan LM, Tolbert PE, Altshul LM, et al. 2006. Does living near a superfund site contribute to higher polychlorinated biphenyl (PCB) exposure? Environ Health Perspect 114 (7 ):1092–1098, PMID: , 10.1289/ehp.8827.16835064
33. Korrick SA, Altshul LM, Tolbert PE, Burse VW, Needham LL, Monson RR. 2000. Measurement of PCBs, DDE, and hexachlorobenzene in cord blood from infants born in towns adjacent to a PCB-contaminated waste site. J Expo Anal Environ Epidemiol 10 (6 pt 2 ):743–754, PMID: , 10.1038/sj.jea.7500120.11138666
34. Sagiv SK, Nugent JK, Brazelton TB, Choi AL, Tolbert PE, Altshul LM, et al. 2008. Prenatal organochlorine exposure and measures of behavior in infancy using the Neonatal Behavioral Assessment Scale (NBAS). Environ Health Perspect 116 (5 ):666–673, PMID: , 10.1289/ehp.10553.18470320
35. Sagiv SK, Thurston SW, Bellinger DC, Altshul LM, Korrick SA. 2012. Neuropsychological measures of attention and impulse control among 8-year-old children exposed prenatally to organochlorines. Environ Health Perspect 120 (6 ):904–909, PMID: , 10.1289/ehp.1104372.22357172
36. Hansen LG. 1998. Stepping backward to improve assessment of PCB congener toxicities. Environ Health Perspect 106 (suppl 1 ):171–189, PMID: , 10.1289/ehp.98106s1171.9539012
37. Schantz SL, Widholm JJ, Rice DC. 2003. Effects of PCB exposure on neuropsychological function in children. Environ Health Perspect 111 (3 ):357–576, PMID: , 10.1289/ehp.5461.12611666
38. U.S. Environmental Protection Agency. 2013. America’s Children and the Environment. 3rd ed. https://www.epa.gov/system/files/documents/2022-10/ace3_2013.pdf [accessed 16 November 2022].
39. Sagiv SK, Thurston SW, Bellinger DC, Tolbert PE, Altshul LM, Korrick SA. 2010. Prenatal organochlorine exposure and behaviors associated with attention deficit hyperactivity disorder in school-aged children. Am J Epidemiol 171 (5 ):593–601, PMID: , 10.1093/aje/kwp427.20106937
40. Oppenheimer AV, Bellinger DC, Coull BA, Weisskopf MG, Zemplenyi M, Korrick SA. 2021. Prenatal exposure to chemical mixtures and inhibition among adolescents. Toxics 9 (11 ):311, PMID: , 10.3390/toxics9110311.34822702
41. Kim R, Aro A, Rotnitzky A, Amarasiriwardena C, Hu H. 1995. K x-ray fluorescence measurements of bone lead concentration: the analysis of low-level data. Phys Med Biol 40 (9 ):1475–1485, PMID: , 10.1088/0031-9155/40/9/007.8532760
42. Canfield RL, Henderson CR Jr, Cory-Slechta DA, Cox C, Jusko TA, Lanphear BP. 2003. Intellectual impairment in children with blood lead concentrations below 10 μg per deciliter. N Engl J Med 348 (16 ):1517–1526, PMID: , 10.1056/NEJMoa022848.12700371
43. Lanphear BP, Hornung R, Khoury J, Yolton K, Baghurst P, Bellinger DC, et al. 2005. Low-level environmental lead exposure and children’s intellectual function: an international pooled analysis. Environ Health Perspect 113 (7 ):894–899, PMID: , 10.1289/ehp.7688.16002379
44. Spitzer RL, Kroenke K, Williams JB, Löwe B. 2006. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med 166 (10 ):1092–1097, PMID: , 10.1001/archinte.166.10.1092.16717171
45. Kroenke K, Strine TW, Spitzer RL, Williams JB, Berry JT, Mokdad AH. 2009. The PHQ-8 as a measure of current depression in the general population. J Affect Disord 114 (1–3 ):163–173, PMID: , 10.1016/j.jad.2008.06.026.18752852
46. Hobel CJ, Hyvarinen MA, Okada DM, Oh W. 1973. Prenatal and intrapartum high-risk screening. I. Prediction of the high-risk neonate. Am J Obstet Gynecol 117 (1 ):1–9, PMID: , 10.1016/0002-9378(73)90720-5.4722373
47. Non AL, Rewak M, Kawachi I, Gilman SE, Loucks EB, Appleton AA, et al. 2014. Childhood social disadvantage, cardiometabolic risk, and chronic disease in adulthood. Am J Epidemiol 180 (3 ):263–271, PMID: , 10.1093/aje/kwu127.24970845
48. Gilman SE, Hornig M, Ghassabian A, Hahn J, Cherkerzian S, Albert PS, et al. 2017. Socioeconomic disadvantage, gestational immune activity, and neurodevelopment in early childhood. Proc Natl Acad Sci USA 114 (26 ):6728–6733, PMID: , 10.1073/pnas.1617698114.28607066
49. Oppenheimer AV, Bellinger DC, Coull BA, Weisskopf MG, Korrick SA. 2022. Prenatal exposure to chemical mixtures and working memory among adolescents. Environ Res 205 :112436, PMID: , 10.1016/j.envres.2021.112436.34843721
50. Oppenheimer AV, Bellinger DC, Coull BA, Weisskopf MG, Korrick SA. 2021. Prenatal exposure to chemical mixtures and cognitive flexibility among adolescents. Toxics 9 (12 ):329, PMID: , 10.3390/toxics9120329.34941764
51. Caldwell BM, Bradley RH. 2003. HOME Inventory Administration Manual: Comprehensive Edition. Little Rock, AR. University of Arkansas for Medical Sciences and University of Arkansas at Little Rock.
52. Bradley RH, Corwyn RF, Caldwell BM, Whiteside-Mansell L, Wasserman GA, Mink IT. 2000. Measuring the home environments of children in early adolescence. J Res Adolesc 10 (3 ):247–288, 10.1207/SJRA1003_1.
53. Hosmer DW, Lemeshow S, Sturdivant RX. 2013. Applied Logistic Regression. 3rd ed. Hoboken, NJ: John Wiley & Sons.
54. Greenland S, Pearl J, Robins JM. 1999. Causal diagrams for epidemiologic research. Epidemiology 10 (1 ):37–48, PMID: .9888278
55. Zou G. 2004. A modified Poisson regression approach to prospective studies with binary data. Am J Epidemiol 159 (7 ):702–706, PMID: , 10.1093/aje/kwh090.15033648
56. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2020. A Quantile-Based g-Computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 128 (4 ):47004, PMID: , 10.1289/EHP5838.32255670
57. Cole SR, Hernán MA. 2008. Constructing inverse probability weights for marginal structural models. Am J Epidemiol 168 (6 ):656–664, PMID: , 10.1093/aje/kwn164.18682488
58. Hernán MA, Hernández-Díaz S, Robins JM. 2004. A structural approach to selection bias. Epidemiology 15 (5 ):615–625, PMID: , 10.1097/01.ede.0000135174.63482.43.15308962
59. Choi AL, Cordier S, Weihe P, Grandjean P. 2008. Negative confounding in the evaluation of toxicity: the case of methylmercury in fish and seafood. Crit Rev Toxicol 38 (10 ):877–893, PMID: , 10.1080/10408440802273164.19012089
60. Vieira VM, Levy JI, Fabian MP, Korrick S. 2021. Assessing the relation of chemical and non-chemical stressors with risk-taking related behavior and adaptive individual attributes among adolescents living near the New Bedford Harbor Superfund site. Environ Int 146 :106199, PMID: , 10.1016/j.envint.2020.106199.33126063
61. Reuben A, Schaefer JD, Moffitt TE, Broadbent J, Harrington H, Houts RM, et al. 2019. Association of childhood lead exposure with adult personality traits and lifelong mental health. JAMA Psychiatry 76 (4 ):418–425, PMID: , 10.1001/jamapsychiatry.2018.4192.30673063
62. Bouchard MF, Bellinger DC, Weuve J, Matthews-Bellinger J, Gilman SE, Wright RO, et al. 2009. Blood lead levels and major depressive disorder, panic disorder, and generalized anxiety disorder in US young adults. Arch Gen Psychiatry 66 (12 ):1313–1319, PMID: , 10.1001/archgenpsychiatry.2009.164.19996036
63. U.S. Centers for Disease Control and Prevention. 2022. Childhood Lead Poisoning Prevention: Blood Lead Reference Value. https://www.cdc.gov/nceh/lead/data/blood-lead-reference-value.htm [accessed 18 November 2022].
64. Bellinger DC. 2008. Very low lead exposures and children’s neurodevelopment. Curr Opin Pediatr 20 (2 ):172–177, PMID: , 10.1097/MOP.0b013e3282f4f97b.18332714
65. Hicken MT, Gee GC, Connell C, Snow RC, Morenoff J, Hu H. 2013. Black-white blood pressure disparities: depressive symptoms and differential vulnerability to blood lead. Environ Health Perspect 121 (2 ):205–209, PMID: , 10.1289/ehp.1104517.23127977
66. Yitshak-Sade M, Lane KJ, Fabian MP, Kloog I, Hart JE, Davis B, et al. 2020. Race or racial segregation? Modification of the PM2.5 and cardiovascular mortality association. PLoS One 15 (7 ):e0236479, PMID: , 10.1371/journal.pone.0236479.32716950
67. Williams DR, Yan Y, Jackson JS, Anderson NB. 1997. Racial differences in physical and mental health: socio-economic status, stress and discrimination. J Health Psychol 2 (3 ):335–351, PMID: , 10.1177/135910539700200305.22013026
68. Longnecker MP, Wolff MS, Gladen BC, Brock JW, Grandjean P, Jacobson JL, et al. 2003. Comparison of polychlorinated biphenyl levels across studies of human neurodevelopment. Environ Health Perspect 111 (1 ):65–70, PMID: , 10.1289/ehp.5463.12515680
69. Pollack MH. 2005. Comorbid anxiety and depression. J Clin Psychiatry 66 (suppl 8 ):22–29, PMID: .16336033
70. Ahmed AT, Frye MA, Rush AJ, Biernacka JM, Craighead WE, McDonald WM, et al. 2018. Mapping depression rating scale phenotypes onto research domain criteria (RDoC) to inform biological research in mood disorders. J Affect Disord 238 :1–7, PMID: , 10.1016/j.jad.2018.05.005.29807322
71. National Institute of Mental Health. Definitions of the RDoC Domains and Constructs. https://www.nimh.nih.gov/research/research-funded-by-nimh/rdoc/definitions-of-the-rdoc-domains-and-constructs [accessed 29 August 2021].
72. Unternaehrer E, Bolten M, Nast I, Staehli S, Meyer AH, Dempster E, et al. 2016. Maternal adversities during pregnancy and cord blood oxytocin receptor (OXTR) DNA methylation. Soc Cogn Affect Neurosci 11 (9 ):1460–1470, PMID: , 10.1093/scan/nsw051.27107296
73. Conners CK. 1997. Conners’ Rating Scales-Revised Technical Manual. Multi-Health Systems. North Tonawanda, NY.
74. Reynolds CR, Kamphaus RW. 2004. Behavior Assessment System for Children (BASC-2). 2nd ed. Circle Pines, MN: American Guidance Service.
75. Balachandran RC, Mukhopadhyay S, McBride D, Veevers J, Harrison FE, Aschner M, et al. 2020. Brain manganese and the balance between essential roles and neurotoxicity. J Biol Chem 295 (19 ):6312–6329, PMID: , 10.1074/jbc.REV119.009453.32188696
76. Wittchen HU. 2002. Generalized anxiety disorder: prevalence, burden, and cost to society. Depress Anxiety 16 (4 ):162–171, PMID: , 10.1002/da.10065.12497648
|
PMC009xxxxxx/PMC9924307.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36779965
EHP11745
10.1289/EHP11745
Research Letter
Large-Scale Analysis of the Association between Air Pollutants and Leucocyte Telomere Length in the UK Biobank
Bountziouka Vasiliki 1 2
Hansell Anna L. 3 4
Nelson Christopher P. 1 2
Codd Veryan 1 2
https://orcid.org/0000-0002-3286-8133
Samani Nilesh J. 1 2
1 Department of Cardiovascular Sciences, University of Leicester, Leicester, UK
2 National Institute for Health Research (NIHR) Leicester Biomedical Research Centre, Glenfield Hospital, Leicester, UK
3 Centre for Environmental Health and Sustainability, University of Leicester, Leicester, UK
4 NIHR Health Protection Research Unit in Environmental Exposures and Health, University of Leicester, Leicester, UK
Address correspondence to Nilesh J. Samani, Department of Cardiovascular Sciences, University of Leicester, British Heart Foundation Cardiovascular Research Centre, Glenfield Hospital, Groby Rd., Leicester, UK. Telephone: 44 116 2044758. Email: [email protected]
13 2 2023
2 2023
131 2 02770121 6 2022
06 12 2022
21 12 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Shorter telomere length (TL) and air pollution are both associated with higher risk of aging-related diseases.1,2 Oxidative stress may mediate the adverse effects of air pollution and also accelerate telomere attrition. Therefore, some of the adverse effects of air pollution could be mediated through an effect on TL. A few small studies, summarized in a recent systematic review and meta-analysis,3 suggested a possible association of exposure to particulate matter (PM) with an aerodynamic diameter of ≤2.5μm (PM2.5) with shorter leukocyte TL (LTL), but overall findings were inconclusive.4 Here, we have leveraged our recent measurement of LTL in over 474,000 participants in the UK Biobank (UKB)5 and available individual-level estimates of exposures to particulates [PM2.5; PM2.5–10, PM with an aerodynamic diameter range of from 2.5 to 10μm (referred to as PMcoarse); PM10, PM with an aerodynamic diameter of ≤10μm; PM2.5absorbance, a proxy for elemental carbon] and gaseous [nitrogen oxides (NOx) and nitrogen dioxide (NO2)] air pollutants in UKB participants4 to undertake a large-scale study of the associations of air pollutants with LTL.
Methods
Recruitment and phenotyping of UKB participants have been described in detail previously.6 LTL was measured using a quantitative polymerase chain reaction method and expressed as a ratio (the T/S ratio) between the telomere repeat copy number (T) and that of a single copy gene (S). Details of the LTL measurements (UKB field code 22192) and their quality control are reported in full elsewhere.5 Air pollution estimates for the year 2010 were modeled for each address using land use regression models developed and validated as part of the European Study of Cohorts for Air Pollution Effects (ESCAPE).7,8 The data are available in the following UKB field codes: 24006 (PM2.5), 24007 (PMabsorbance), 24008 (PMcoarse), 24005 (PM10), 24004 (NOx), and 24003 (NO2).
From the participants with LTL measurements in the UKB (n=474,074), we excluded participants without information on ethnicity, or white blood cell (WBC) count (n=14,836) and randomly excluded one of any genetically related (Kinship K>0.088) participant pairs (n=36,441). Of the remaining 422,797 participants, complete data on air pollution markers, socioeconomic status (SES) indicators and smoking variables were available in 299,786 participants (the available data). Those with complete data tended to be younger, male, of White ethnic background, and have longer LTL than those without. To account for any selective bias, we undertook analyses in imputed data generated through multiple imputation by chained equations (MICE)9 with 10 imputed data sets (n=422,927; the imputed data).
Correlation coefficients between air pollutants and with SES were assessed with Pearson’s correlation coefficient (r). The association between each individual pollutant with LTL was investigated using linear regression models in R (version 4.2.1; R Development Core Team), adjusted for age, sex, ethnic background [defined by the UKB as Asian (including Asian or Asian British, Indian, Pakistani, Bangladeshi, and any other Asian background), Black (including Black or Black British, Caribbean, African, and any other Black background), Chinese, Mixed (including White and Black African, White and Black Caribbean, White and Asian, and any other Mixed background), Other, and White (including British, Irish, and any other White background)], and WBC (all determinants of LTL6; i.e., the base model). Nonlinearity was assessed via the inclusion of a quadratic term for the pollutant.
The fully adjusted model was additionally adjusted for the area-level Townsend deprivation index (UKB field code 189), annual gross family income (738), ever smoked (20160) and passive smoking (1269 and 1279), using UKB-defined category levels, and for highest educational level (6138), which was recategorized into None (no qualification), Compulsory (O-levels/CSE/GCSE), Advanced (A-levels/nonvocational/other professional), and Degree (university/college degree). To avoid the potential for collider bias between LTL and air pollution with SES, we further examined the association between individual air pollutants and LTL within substrata of the Townsend index. The UKB has ethical approval from the North West Centre for Research Ethics Committee (application 11/NW/0382), which covers the UK. The UKB obtained informed consent from all participants. The generation and analysis of the data presented in this paper was approved by the UK Biobank access committee under UK Biobank application no. 6007.
Results
Participants were 40–69 years of age at recruitment [mean±standard deviation (SD)=56.3±8.0y] with more women (52.3%) and predominantly of White ethnic background (95.2%). Mean±SD exposures for NO2, NOx, PM10, PM2.5, and PMcoarse were 26.3 ± 7.6, 43.4 ± 15.5, 16.2 ± 1.9, 10.0 ± 1.1, and 6.4 ± 0.90 μg/m3, respectively, and 1.2 ± 0.27 ×10−5/m for PM2.5 absorbance. There was a strong correlation (r=0.92) between NO2 and NOx levels. The correlations between the gaseous pollutants and the particulate pollutants ranged from r=0.24 between NO2 and PMcoarse to r=0.87 between NO2 and PM2.5. Among the particulate pollutants, the correlations ranged from r=0.22 between PMcoarse and PM2.5 to r=0.82 between PMcoarse and PM10 (all p<1×10−300).
Accounting for all major determinants of LTL, there was a nominal (p=0.03) inverse association of PM2.5 with LTL (Table 1, base models). However, allowing for multiple testing (Bonferroni p=0.008), no pollutant showed a significant association with LTL in the available data. The findings were unchanged after adjustment for SES indicators and smoking (Table 1, adjusted models). In the imputed data, there was a small but significant inverse association of PM2.5 and NO2 with LTL in the base models that became nonsignificant when adjusted for SES and smoking (Table 2). There was no strong evidence of nonlinear associations between the air pollutants and LTL or evidence of a threshold effect.
Table 1 Association of individual air pollution markers with leukocyte telomere length using the complete available data (n=299,786).
Exposure response estimate per 1 SD Base model Fully adjusted model
Beta (95% CI) p-Value Beta (95% CI) p-Value pInteraction
NO2, per 7.6 μg/m3 −0.001 (−0.004, 0.003) 0.71 −0.004 (−0.011, 0.002) 0.18 8.25×10−4
NOx, per 15.5 μg/m3 −0.003 (−0.006, 0.001) 0.11 −0.005 (−0.011, 0.002) 0.15 0.004
PM10, per 1.9 μg/m3 −0.001 (−0.005, 0.002) 0.47 −0.002 (−0.008, 0.003) 0.41 0.11
PM2.5, per 1.1 μg/m3 −0.004 (−0.007, 0.000) 0.03 −0.004 (−0.011, 0.003) 0.27 0.19
PM2.5absorbance, per 0.27×10−5/m 0.003 (−0.001, 0.006) 0.15 −0.007 (−0.014, 0.000) 0.05 1.33×10−6
PMcoarse, per 0.9 μg/m3 −0.002 (−0.005, 0.002) 0.31 −0.004 (−0.009, 0.002) 0.20 0.03
Note: The base linear regression model was adjusted for age, sex, ethnicity, and white blood cell count. The fully adjusted model was further adjusted for markers of socioeconomic status (SES) and smoking. SES includes quintiles of the 2011 Townsend deprivation index, gross annual family income, and higher educational qualifications. Smoking factors include ever and passive smoking. pInteraction gives the global p-value for interaction, obtained through a Wald-test, where terms of pollution markers with SES markers were tested. In the presence of a pollution marker×Townsend interaction index, the interactions with income and education were not significant and removed from the models. CI, confidence interval; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5μm; PM10, particulate matter with an aerodynamic diameter of ≤10μm; PM2.5absorbance, a proxy for elemental carbon; PMcoarse, particulate matter with an aerodynamic diameter range of from 2.5 to 10μm; SD, standard deviation.
Table 2 Association of individual air pollution markers with leukocyte telomere length using the imputed data (n=422,797).
Exposure response estimate per 1 SD Base model Fully adjusted model
Beta (95% CI) p-Value Beta (95% CI) p-Value pInteraction
NO2, per 7.6 μg/m3 −0.003 (−0.006, 0.000) 0.02 −0.002 (−0.008, 0.004) 0.46 0.0105
NOx, per 15.5 μg/m3 −0.005 (−0.008, −0.002) 3.74×10−4 −0.003 (−0.009, 0.003) 0.30 0.04
PM10, per 1.9 μg/m3 −0.002 (−0.005, 0.002) 0.32 −0.001 (−0.006, 0.004) 0.65 0.04
PM2.5, per 1.1 μg/m3 −0.006 (−0.009, −0.003) 5.62×10−5 −0.001 (−0.007, 0.005) 0.70 0.47
PM2.5absorbance, per 0.27×10−5/m 0.000 (−0.003, 0.003) 0.86 −0.006 (−0.013, 0.000) 0.04 1.38×10−6
PMcoarse, per 0.9 μg/m3 −0.002 (−0.005, 0.001) 0.27 −0.004 (−0.009, 0.001) 0.11 0.02
Note: The base linear regression model was adjusted for age, sex, ethnicity, and white blood cell count. The fully adjusted model was further adjusted for markers of socioeconomic status (SES) and smoking. SES includes quintiles of the 2011 Townsend deprivation index, gross annual family income, and higher educational qualifications. Smoking factors include ever and passive smoking. pInteraction gives the global p-value for interaction, obtained through a Wald-test, where terms of pollution markers with SES markers were tested. In the presence of a pollution marker×Townsend index interaction, the interactions with income and education were not significant and removed from the models. Imputed data were generated through multiple imputation by chained equations (MICE)9 with 10 imputed data sets (n=422,927). CI, confidence interval; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5μm; PM10, particulate matter with an aerodynamic diameter of ≤10 μm; PM2.5absorbance, a proxy for elemental carbon; PMcoarse, particulate matter with an aerodynamic diameter range of from 2.5 to 10μm; SD, standard deviation.
As expected, there were highly significant correlations (all p<1×10−300) between air pollution markers and the Townsend index (NO2: 0.496; NOx: 0.436; PM10: 0.219; PM2.5: 0.448; PM2.5absorbance: 0.406; and PMcoarse: 0.111). In analyses stratified by level of deprivation, higher PM2.5absorbance, NOx, and NO2 concentrations were positively associated with LTL in the most deprived areas in the base model (Figure 1). However, further adjustment for additional SES factors and smoking attenuated the observed associations.
Figure 1. Association between leukocyte telomere length and air pollutants, stratified by quintiles of deprivation. Data are split into quintiles using the Townsend deprivation index, with the first quintile representing the least deprived and the fifth quintile, the most deprived. Beta estimates and 95% confidence intervals (CIs) were obtained from models run within strata. The base model was adjusted for age, sex, ethnicity, and WBC count. The fully adjusted model was further adjusted for markers of socioeconomic status (income and education) and smoking (both ever and passive smokers). Note: NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5 μm; PM10, particulate matter with an aerodynamic diameter of ≤10μm; PM2.5absorbance, a proxy for elemental carbon; PMcoarse, particulate matter with an aerodynamic diameter range of from 2.5 to 10μm; WBC, white blood cell.
Figure 1 is a set of twelve graphs. On the left, the six graphs are titled Base model, plotting Beta coefficient (95 percent confidence interval), ranging from negative 0.02 to 0.04 in increments of 0.02 (left y-axis) and particulate matter coarse, particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen oxides, and nitrogen dioxide (right y-axis) across Quintiles of Townsend index, ranging as first (low deprivation), second, third, fourth, and fifth (high deprivation) (x-axis). On the right, the six graphs are titled Fully adjusted model, plotting Beta coefficient (95 percent confidence interval), ranging from negative 0.02 to 0.04 in increments of 0.02 (left y-axis) and particulate matter coarse, particulate matter begin subscript 2.5 end subscript absorbance, particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, nitrogen oxides, and nitrogen dioxide (right y-axis) across Quintiles of Townsend index, ranging as first (low deprivation), second, third, fourth, and fifth (high deprivation) (x-axis).
Discussion
In this large-scale analysis in a contemporary cohort, we found no evidence for a significant association between several air pollutants and LTL, with similar findings in both available and imputed data. The findings are surprising given prior, albeit weak, evidence suggesting that higher air pollution levels are associated with shorter LTL.3 Furthermore, in the same data set we have been able to replicate associations of several traits with LTL and identify novel associations.5,10 Similarly, the ESCAPE air pollution models have been extensively used, including in >60 studies in the UKB, demonstrating adverse associations with air pollution.4
Given that air pollution exposure is strongly linked with SES, as we also show here, we further examined the association between air pollution markers and LTL in groups stratified by the Townsend index, but this did not alter the findings.
Although our findings suggest that it is unlikely that the adverse health effects of several major air pollutants are substantially mediated through accelerated attrition of TL, several limitations of our study need to be considered. First, the analyses were cross-sectional owing to the availability of only single time point measurements for both LTL and air pollution. Second, we have analyzed only a single cohort, albeit large, of predominantly White ethnicity, living in a high-income country with lower ambient air pollution. Therefore, our findings may not be generalizable to other settings. Furthermore, for the particulate pollutants, because only particle concentrations are available, we are unable to evaluate any association between LTL and particle composition or toxicity. Finally, our study does not exclude the possibility that exposure to other pollutants, including those from indoor exposures, affect TL.
Acknowledgments
V.C., A.L.H., and N.J.S. conceived the project; V.C., C.P.N., and N.J.S. secured funding for the LTL measurements and oversaw the generation and curation of the LTL measurements; V.B. and C.P.N. developed the analysis plan; V.B. performed the analysis; V.B., A.L.H., and N.J.S. drafted the manuscript; and all authors revised the manuscript and approved the submitted version.
This work was funded by the UK Medical Research Council (MRC), Biotechnology and Biological Sciences Research Council and British Heart Foundation (BHF) through MRC grant MR/M012816/1 (to N.J.S., V.C., and C.P.N.). V.B., V.C., C.P.N., and N.J.S. are supported by the National Institute for Health Research (NIHR) Leicester Cardiovascular Biomedical Research Centre (BRC–1215–20010). C.P.N. is funded by the BHF (SP/16/4/32697). A.L.H. acknowledges funding from the NIHR Health Protection Research Unit in Environmental Exposures and Health at the University of Leicester, a partnership between the UK Health Security Agency, the Health and Safety Executive, and the University of Leicester. The views expressed are those of the author(s) and not necessarily those of the National Health Service, the NIHR, the Department of Health and Social Care, or the UK Health Security Agency.
All data used in this study, including telomere length measurements, are available through application to the UK Biobank. Further information on registration to access the data can be found at http://www.ukbiobank.ac.uk/register-apply/.
==== Refs
References
1. Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K, et al. 2017. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 389 (10082 ):1907–1918, PMID: , 10.1016/S0140-6736(17)30505-6.28408086
2. Codd V, Wang Q, Allara E, Musicha C, Kaptoge S, Stoma S, et al. 2021. Polygenic basis and biomedical consequences of telomere length variation. Nat Genet 53 (10 ):1425–1433, PMID: , 10.1038/s41588-021-00944-6.34611362
3. Miri M, Nazarzadeh M, Alahabadi A, Ehrampoush MH, Rad A, Lotfi MH, et al. 2019. Air pollution and telomere length in adults: a systematic review and meta-analysis of observational studies. Environ Pollut 244 :636–647, PMID: , 10.1016/j.envpol.2018.09.130.30384069
4. Doiron D, de Hoogh K, Probst-Hensch N, Fortier I, Cai Y, De Matteis S, et al. 2019. Air pollution, lung function and COPD: results from the population-based UK Biobank study. Eur Respir J 54 (1 ):1802140, PMID: , 10.1183/13993003.02140-2018.31285306
5. Codd V, Denniff M, Swinfield C, Warner SC, Papakonstantinou M, Sheth S, et al. 2022. Measurement and initial characterization of leucocyte telomere length in 474,074 participants in UK Biobank. Nat Aging 2 (2 ):170–179, 10.1038/s43587-021-00166-9.37117760
6. Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. 2018. The UK Biobank resource with deep phenotyping and genomic data. Nature 562 (7726 ):203–209, PMID: , 10.1038/s41586-018-0579-z.30305743
7. Beelen R, Hoek G, Vienneau D, Eeftens M, Dimakopoulou K, Pedeli X, et al. 2013. Development of NO2 and NOX land use regression models for estimating air pollution exposure in 36 study areas in Europe—the ESCAPE project. Atmos Environ 72 :10–23, 10.1016/j.atmosenv.2013.02.037.
8. Eeftens M, Beelen R, de Hoogh K, Bellander T, Cesaroni G, Cirach M, et al. 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PMcoarse in 20 European study areas; results of the ESCAPE project. Environ Sci Technol 46 (20 ):11195–11205, PMID: , 10.1021/es301948k.22963366
9. White IR, Royston P, Wood AM. 2011. Multiple imputation using chained equations: issues and guidance for practice. Stat Med 30 (4 ):377–399, PMID: , 10.1002/sim.4067.21225900
10. Bountziouka V, Musicha C, Allara E, Kaptoge S, Wang Q, Angelantonio ED, et al. 2022. Modifiable traits, healthy behaviours, and leucocyte telomere length: a population–based study in UK Biobank. Lancet Healthy Longev 3 (5 ):e321–e331, PMID: , 10.1016/S2666-7568(22)00072-1.35685390
|
PMC009xxxxxx/PMC9940783.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12800
10.1289/EHP12800
Response to Letter
Response to “Comment on ‘Impacts of Sugarcane Fires on Air Quality and Public Health in South Florida’”
Holmes Christopher D. 1
https://orcid.org/0000-0001-7291-7887
Nowell Holly K. 1
1 Department of Earth, Ocean and Atmospheric Science, Florida State University, Tallahassee, Florida
Address correspondence to Christopher D. Holmes. Email: [email protected]
20 2 2023
2 2023
131 2 02800226 1 2023
26 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
All authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP12236
==== Body
pmcShapero et al.1 claim the conclusions of our study2 are undermined by “erroneous assumptions and misapplied technical approaches.” However, their letter ignores most of the evidence that we provided in our article, incorrectly describes the methods we used, and fails to identify any errors in our work.
We quantified the contribution of preharvest sugarcane burning to air concentrations of fine particulate matter [PM ≤2.5μm in aerodynamic diameter (PM2.5)] using multiple independent data sets and lines of evidence. We showed that ground-based PM2.5 monitors and satellite-derived surface PM2.5 observations both recorded higher average PM2.5 concentrations in Florida’s sugarcane-growing region during harvest burning season than the rest of the year, a pattern not seen elsewhere on the Florida peninsula. This pattern matched the magnitude and spatial extent of PM2.5 expected from sugarcane fires, which we simulated in a state-of-the-art atmospheric dispersion model using emissions derived from sugarcane burn authorization records. The mean diurnal cycle of PM2.5 in Belle Glade, Florida, a city surrounded by sugarcane fields, also featured a peak shortly after sugarcane fires began in the morning; this peak did not appear in non-harvest months.
The consistency and corroboration between these independent sources provided a confident estimate of PM2.5 concentrations caused by sugarcane fires. Our results also cohere with numerous past studies showing that PM2.5 from many sources causes premature mortality3,4 and that PM2.5 from agricultural fires specifically is associated with around 600 premature deaths per year in the United States.5
Much of the letter by Shapero et al.1 concerns a statistical significance test (p value) at one surface monitor site, but their critique does not apply to the methods we used. As we wrote in our paper, PM2.5 surface observations were averaged over harvest and non-harvest seasons (6 months each) before performing the significance test. The statements by Shapero et al.1 about clustered standard errors on subseasonal timescales are, therefore, irrelevant and misrepresent our paper.
Shapero et al. are incorrect in saying that our analysis neglected “meteorological conditions” or the “temporal specifics of actual harvest activities.”1 In reality, we estimated the contribution of sugarcane fires to PM2.5 using an atmospheric dispersion model driven by high-resolution meteorological data from the National Oceanic and Atmospheric Administration, and we accounted for the date and location of every sugarcane fire, as well as the times of day when harvest fires occurred. Shapero et al. also say that our analysis contained large uncertainties and high biases,1 but all the sources of uncertainty they listed (e.g., fuel loading, emission factors, plume rise, secondary aerosol) were explicitly accounted for in the confidence intervals reported in our article.
Our study focused on the years 2009–2018 because the satellite-derived PM2.5 data,6 which informed the health impacts assessment, were not available for later years at the time of our analysis. If we look at the PM2.5 monitor in Belle Glade (Figure 1), as Shapero et al.1 suggest, we see that the mean PM2.5 concentrations were consistently higher during harvest seasons than non-harvest seasons for 2009–2017, except for a couple years when large wildfires burned nearby during summer. In more recent years, harvest season PM2.5 concentrations have not been as elevated, coinciding with new restrictions on sugarcane burning implemented in 2019.7 The change in harvest season mean PM2.5 after 2019 is therefore consistent with sugarcane fires contributing to PM2.5 during the years of our study. Future work should examine whether the recent PM2.5 changes in Belle Glade are regionally representative or limited to the vicinity of the monitor, and whether the changes persist in future harvest seasons.
Figure 1. Change in PM2.5 in Belle Glade, Florida, during harvest season (October through March) compared with the preceding (left arrow) and following (right arrow) non-harvest seasons (April through September). Positive values indicate mean PM2.5 concentrations were higher during harvest season. Nearby wildfires in the summers of 2011 and 2017 reduced the harvest season PM2.5 change in those years. Note: PM2.5 is particulate matter ≤2.5μm in aerodynamic diameter.
Figure 1 is a graph, plotting harvest season particulate matter begin subscript 2.5 end subscript change percentage, ranging from negative 20 to 50 in increments of 10 (y-axis) across harvest year, ranging as 2010 to 2011, 2012 to 2013, 2014 to 2015, 2016 to 2017, 2018 to 2019, and 2020 to 2021 (x-axis) for mean particulate matter begin subscript 2.5 end subscript elevated during most harvest seasons 2009 to 2017 and new restrictions on sugarcane burning.
In summary, we maintain that the letter by Shapero et al.1 contains errors, incorrectly describes our analysis, and does not identify any source of uncertainty that was not already accounted for in our article. We stand behind our analysis and the central findings of our article.
==== Refs
References
1. Shapero A, Keck S, Goswami E, Love AH. 2023. Comment on “Impacts of sugarcane fires on air quality and public health in South Florida.” Environ Health Perspect 131 (2 ):028001, 10.1289/EHP12236.36802828
2. Nowell HK, Wirks C, Val Martin M, van Donkelaar A, Martin RV, Uejio CK, et al. 2022. Impacts of sugarcane fires on air quality and public health in South Florida. Environ Health Perspect 130 (8 ):87004, PMID: , 10.1289/EHP9957.35929976
3. U.S. EPA (U.S. Environmental Protection Agency). 2019. Integrated Science Assessment for Particulate Matter. EPA/600/R-19/188. Washington, DC: U.S. EPA.
4. U.S. EPA. 2022. Supplement to the 2019 Integrated Science Assessment for Particulate Matter (Final). EPA/600/R-22/028. Washington, DC: U.S. EPA.
5. McDuffie EE, Martin RV, Spadaro JV, Burnett R, Smith SJ, O’Rourke P, et al. 2021. Source sector and fuel contributions to ambient PM2.5 and attributable mortality across multiple spatial scales. Nature Comm 12 (1 ):3594, PMID: , 10.1038/s41467-021-23853-y.34127654
6. van Donkelaar A, Martin RV, Li C, Burnett RT. 2019. Regional estimates of chemical composition of fine particulate matter using a combined geoscience–statistical method with information from satellites, models, and monitors. Environ Sci Technol 53 (5 ):2595–2611, PMID: , 10.1021/acs.est.8b06392.30698001
7. Florida Forest Service. 2019. Commissioner Nikki Fried announces major changes to prescribed burning. Press release 1 October 2019. https://web.archive.org/web/20220705030307/https://www.fdacs.gov/News-Events/Press-Releases/2019-Press-Releases/Commissioner-Nikki-Fried-Announces-Major-Changes-to-Prescribed-Burning [accessed 15 January 2023].
|
PMC009xxxxxx/PMC9940784.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12236
10.1289/EHP12236
Letter to the Editor
Comment on “Impacts of Sugarcane Fires on Air Quality and Public Health in South Florida”
Shapero Andrew 1
Keck Stella 1
Goswami Emily 2
https://orcid.org/0000-0002-9943-5463
Love Adam H. 2
1 Roux, Inc., Burlington, Massachusetts, USA
2 Roux, Inc., Oakland, California, USA
Address correspondence to Adam H. Love, 555 12th St., Suite 250, Oakland, CA 94607 USA. Email: [email protected]
20 2 2023
2 2023
131 2 02800104 10 2022
26 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Roux, Inc. has Florida Sugar Cane League as a client and has been compensated for the time required to review and evaluate the original journal article and provide this technical critique.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP9957
==== Body
pmcNowell et al.1 evaluated potential community impacts from sugarcane harvesting related to fine particulate matter [PM ≤2.5μm in aerodynamic diameter (PM2.5)] exposure. However, in an attempt to fill numerous measurement gaps, they made erroneous assumptions and misapplied technical approaches, undermining their conclusions.
The assumption of a seasonal increase in community PM2.5 air concentration from sugarcane harvesting is supported by neither the data the authors relied on nor their analysis of those data. They reported a confidence interval in PM2.5 increase that was not statistically significant, yet they used the assumed increase to justify calculating sugarcane harvest emissions.
We also disagree with the authors’ use of the t-test to compare the harvest and non-harvest seasons. At a minimum, they should have conducted a regression that accounted for yearly fixed effects and clustered standard errors because the data they used were likely not independent—a requirement for conducting a t-test. Similarly, by pooling each year’s harvest season data as a single measure and non-harvest season data as a separate single measure, the authors failed to evaluate whether there were consistent yearly PM2.5 differences between each harvest and non-harvest season. In addition, although the authors’ publication was submitted in 2022, the data set they relied upon spanned only 2009 to 2018. No explanation was provided for why they used an incomplete data set for their modeled calculation of observed PM2.5 differences over time.
Finally, the use of months to categorize the data into harvest vs. non-harvest seasons is not appropriate for causal inference. This is because the temporal specifics of actual harvest activities are not included in their analysis, and their analysis does not account for other confounding seasonal factors that may affect PM2.5 concentrations (e.g., meteorological conditions, other PM2.5 sources).
The authors’ calculation of sugarcane harvest emissions demonstrates numerous instances of model inputs with errors, large uncertainties, and high bias that cannot justify the analysis precision reported and the certainty with which they stated their conclusion. These model issues occurred at nearly every step in the authors’ methodology: in the fire area and location2 and the fuel loading factor they drew from Pouliot et al.,3 in the PM2.5 emission factors they drew from McCarty,4 in the HYSPLIT model plume rise,5,6 and in the secondary particle formation they drew from Yokelson et al.7 In addition, by adjusting the satellite PM2.5 data to identify apparent increases in PM2.5 concentrations in the sugar-growing region, the authors increased uncertainty and bias through their choice of a limited sample of ground-based air monitors and the temporal mismatch between the monthly satellite data set and daily ground-based data set. Finally, although the authors concede that burning activity occurs discontinuously during the harvest season, they applied their differential PM2.5 exposure amounts for community exposure throughout the entire year.
Ultimately, correcting any of these highlighted deficiencies would have supported the null hypothesis. The corrected analysis would thus demonstrate there is no scientific basis to assert community health impacts associated with an increase in PM2.5 from sugarcane harvesting.
==== Refs
References
1. Nowell HK, Wirks C, Val Martin M, van Donkelaar A, Martin RV, Uejio CK, et al. 2022. Impacts of sugarcane fires on air quality and public health in south Florida. Environ Health Perspect 130 (8 ):87004, PMID: , 10.1289/EHP9957.35929976
2. Nowell HK, Holmes CD, Robertson K, Teske C, Hiers JK. 2018. A new picture of fire extent, variability, and drought interaction in prescribed fire landscapes: insights from Florida government records. Geophys Res Lett 45 (15 ):7874–7884, PMID: , 10.1029/2018GL078679.31031448
3. Pouliot G, Rao V, McCarty JL, Soja A. 2017. Development of the crop residue and rangeland burning in the 2014 National Emissions Inventory using information from multiple sources. J Air Waste Manag Assoc 67 (5 ):613–622, PMID: , 10.1080/10962247.2016.1268982.27964698
4. McCarty JL. 2011. Remote sensing-based estimates of annual and seasonal emissions from crop residue burning in the contiguous United States. J Air Waste Manag Assoc 61 (1 ):22–34, PMID: , 10.3155/1047-3289.61.1.22.21305885
5. Draxler RR, Hess GD. 1998. An overview of the HYSPLIT_4 modelling system of trajectories, dispersion, and deposition. Aust Meteorol Mag 47 :295–308.
6. Stein AF, Rolph GD, Draxler RR, Stunder B, Ruminski M. 2009. Verification of the NOAA Smoke Forecasting System: model sensitivity to the injection height. Weather Forecast 24 (2 ):379–394, 10.1175/2008WAF2222166.1.
7. Yokelson RJ, Crounse JD, DeCarlo PF, Karl T, Urbanski S, Atlas E, et al. 2009. Emissions from biomass burning in the Yucatan. Atmos Chem Phys 9 (15 ):5785–5812, 10.5194/acp-9-5785-2009.
|
PMC009xxxxxx/PMC9942819.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36802347
EHP11347
10.1289/EHP11347
Research
Effects of Sociodemographic Characteristics, Comorbidity, and Coexposures on the Association between Air Pollution and Type 2 Diabetes: A Nationwide Cohort Study
https://orcid.org/0000-0002-7302-4789
Sørensen Mette 1 2
Poulsen Aslak Harbo 1
Hvidtfeldt Ulla Arthur 1
Christensen Jesper H. 3
Brandt Jørgen 3 4
Frohn Lise Marie 3 4
Ketzel Matthias 3 5
Andersen Christopher 3
Valencia Victor H. 3
Lassen Christina Funch 6
Raaschou-Nielsen Ole 1 3
1 Work, Environment and Cancer, Danish Cancer Society Research Center, Copenhagen, Denmark
2 Department of Natural Science and Environment, Roskilde University, Roskilde, Denmark
3 Department of Environmental Science, Aarhus University, Roskilde, Denmark
4 iClimate – Interdisciplinary Centre for Climate Change, Aarhus University, Roskilde, Denmark
5 Global Centre for Clean Air Research (GCARE), Department of Civil and Environmental Engineering, University of Surrey, Guildford, UK
6 Centre of Social Medicine, University Hospital Bispebjerg-Frederiksberg, Frederiksberg, Denmark
Address correspondence to Mette Sørensen, Work, Environment and Cancer, Danish Cancer Society Research Center, Strandboulevarden 49, 2100 Copenhagen Ø, Denmark. Telephone: +45 3525 7626. Email: [email protected]
21 2 2023
2 2023
131 2 02700804 4 2022
05 1 2023
17 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Exposure to air pollution has been associated with a higher risk of type 2 diabetes (T2D), but studies investigating whether deprived groups are more susceptible to the harmful effects of air pollution are inconsistent.
Objectives:
We aimed to investigate whether the association between air pollution and T2D differed according to sociodemographic characteristics, comorbidity, and coexposures.
Methods:
We estimated residential exposure to PM2.5, ultrafine particles (UFP), elemental carbon, and NO2 for all persons living in Denmark in the period 2005–2017. In total, 1.8 million persons 50–80 y of age were included for main analyses of whom 113,985 developed T2D during follow-up. We conducted additional analyses on 1.3 million persons age 35–50 y. Using Cox proportional hazards model (relative risk) and Aalens additive hazard model (absolute risk), we calculated associations between 5-y time-weighted running means of air pollution and T2D in strata of sociodemographic variables, comorbidity, population density, road traffic noise, and green space proximity.
Results:
Air pollution was associated with T2D, especially among people age 50–80 y, with hazard ratios of 1.17 [95% confidence interval (CI): 1.13, 1.21] per 5 μg/m3 PM2.5 and 1.16 (95% CI: 1.13, 1.19) per 10,000 UFP/cm3. In the age 50–80 y population, we found higher associations between air pollution and T2D among men in comparison with women, people with lower education vs. individuals with high education, people with medium income vs. those with low or high income, people cohabiting vs. those living alone, and people with comorbidities vs. those without comorbidities. We observed no marked changes according to occupation, population density, road noise, or surrounding greenness. In the age 35–50 y population, similar tendencies were observed, except in relation to sex and occupation, where we observed associations with air pollution only among women and blue-collar workers.
Discussion:
We found stronger associations between air pollution and T2D among people with existing comorbidities and weaker associations among people with high socioeconomic status in comparison with those with lower socioeconomic status. https://doi.org/10.1289/EHP11347
Supplemental Material is available online (https://doi.org/10.1289/EHP11347).
All authors declare no competing interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
The prevalence of type 2 diabetes (T2D) has increased markedly in countries of all income levels, from 108 million in 1980 to 422 million in 2014.1 The main risk factor for T2D is an unhealthy lifestyle, particularly obesity and physical inactivity,1 but a number of studies have also linked T2D with exposure to ambient air pollution.2 Proposed mechanistic pathways include air pollution–induced oxidative stress and systemic inflammation, which are both involved in the pathogenesis of T2D.3,4 Furthermore, epidemiological studies have found air pollution to be associated with early markers of T2D, including decreased glucose tolerance and insulin insensitivity.5,6 A recent meta-analysis of air pollution and incident T2D found risk estimates of 1.10 [95% confidence interval (CI): 1.04, 1.16] per 10 μg/m3 particulate matter (PM) with a diameter <2.5μm (PM2.5) and 1.02 (95% CI: 0.99, 1.05) per 10 μg/m3 nitrogen oxide (NO2).2 Ultrafine particles (UFP; <0.1μm in diameter) are potentially more harmful than larger particles,7,8 and the two studies on UFP and diabetes found long-term exposure to UFP associated with increased risk of diabetes.9,10
Previous studies have found socioeconomic inequalities according to air pollution exposure, mainly showing higher exposure among people with low socioeconomic status (SES),11 although this finding varies across regions/countries if, e.g., living centrally is highly attractive.12,13 It is unclear whether the harmfulness of air pollution differs in relation to the development of T2D across different socioeconomic groups, because the studies investigating effect modification by sex are inconsistent.14–16 Systemic inflammation is thought to be a main biological pathway underlying an effect of air pollution on T2D.17 People with low SES are more likely to have an unhealthy lifestyle, a higher body mass index (BMI), and are at higher risk of, e.g., COPD and cardiovascular disease, which are all habits and conditions characterized by chronic inflammation.18,19 It is therefore possible that people with low SES are more susceptible to the harmful effects of air pollution on risk of T2D.14
Associations between air pollution and T2D may differ between men and women due to socially derived exposure differences according to gender (e.g., in some countries women spend more time at home than men); to physiological differences related to sex (e.g., differences in hormones, lung size, and deposition of particles); or to a combination of these.20 The studies investigating these associations and exposure differences are inconsistent, because some studies find the highest risk estimates among men,16,21 some studies find them among women,14,15 and some studies report no difference in risk according to sex.9,22
Road traffic noise has been found to increase the risk of T2D.23–25 Noise is believed harmful through some of the same biological mechanisms as air pollution, including systemic inflammation and oxidative stress.26 A recent study found that road traffic noise was associated with higher risk of T2D among people exposed to high levels of air pollution in comparison with people exposed to low levels of air pollution, suggesting that high exposure to one of these two traffic pollutants can increase the susceptibility to the other.24 Surrounding greenness has been found inversely associated with T2D, potentially by promoting physical activity, decreasing psychological stress, or as a result of lower air pollution and traffic noise levels in such areas.27,28 If surrounding greenness results in a healthier lifestyle, it might protect against the harmful effects of air pollution, but this possible mitigation has not been investigated in relation to T2D.
We aimed to investigate whether the association between long-term exposure to air pollution (PM2.5, elemental carbon (EC), UFP, and NO2) and risk of T2D differed according to sociodemographic characteristics, financial stress, comorbidity, population density, road traffic noise, and green space, based on the entire Danish population.
Methods
Study Population
All persons living in Denmark can be followed across all health and administrative registers based on a unique identification number.29 Using the Danish Civil Registration System, which contains continuously updated information on exact addresses,29 we identified address histories for all inhabitants born after 1 January 1921 and living in Denmark from 1979 onward (after 1979, address information is virtually complete). We censored people at the date of missing address information (>14 consecutive days), emigration, death, or 31 December 31 2017. Based on this population, we defined a study base with baseline at 1 January 2005 or age 35 y, whichever came last, such that a person who was below 35 y of age in 2005 was included into the cohort at the time the person turned 35 y (N=2,757,813).
Outcome
We identified incident diabetes cases based on the National Patient Registry30 and the National Prescription Registry,31 using an algorithm developed by the Danish Health Data Agency for the purpose of monitoring diabetes prevalence and incidence in Denmark.32 This algorithm has been used in various register-based studies based on the Danish population.33,34
The Prescription Registry holds information on all dispensed drugs. We defined T2D cases as persons with two contacts with a pharmacy [Anatomical Therapeutic Chemical system (ATC) codes A10B (blood glucose–lowering drugs, excluding insulins), though excluding A10BJ02 (liraglutide: only Saxenda®), as well as A10AE54 (insulin glargine and lixisenatide) and A10AE56 (insulin degludec and liraglutide)] and/or T2D-related hospital contacts (International Classification of Diseases (ICD) 8 code 250 or ICD10 code E11). We defined a person as case from the second register record. A diagnosis of type 1 diabetes [ICD-8 code 249 or ICD-10 code E10 and/or at least one dispensed prescription with ATC A10A (insulins and analogs), excluding A10AE54 (insulin glargine and lixisenatide) and A10AE56 (insulin degludec and liraglutide)] resulted in censoring (exclusion if before baseline). All persons with a diagnosis of T2D before baseline (identified as described above for incident cases) were excluded.
Estimation of Air Pollution Exposure
We modeled air pollution concentrations of PM2.5, EC, NO2, and UFP outside the front door of all addresses in Denmark (identified from the Building and Housing Registry) using the Danish Eulerian Hemispheric Model (DEHM) DEHM/ Urban Background Model (UBM) /AirGIS modeling system.35 This modeling system calculates air pollution contributions from a) the regional background, modeled using the DEHM36; b) the local background, modeled using the UBM37 covering Denmark in a 1×1km grid; and c) traffic in the address street (modeled for streets with >500 vehicles per day), modeled using the Operational Street Pollution Model (OSPM®), which takes into account emission factors, traffic composition and intensity, meteorology, and street and building configurations.35,38 We recently implemented modeling of particle number concentration, as an indicator for UFP (in this paper denoted as UFP), into the DEHM/UBM/AirGIS modeling system. In brief, the regional scale model, DEHM, was extended with the M7 aerosol dynamics module39 to account for number concentrations of particles with a diameter <1μm.40 We furthermore developed models for estimating particle number concentrations at the local scale (UBM) and street scale (OSPM).41 A validation of the model results with long-term UFP measurements in Denmark showed correlations of 0.86, 0.87, and 0.95 between measured and predicted annual averages at, respectively, the regional, urban, and street scale.41 For PM2.5, EC, and NO2, correlation coefficients between measured and modeled air pollution (using DEHM/UBM/AirGIS) across various measurement periods and locations have been found to be, respectively, 0.67–0.85, 0.77–0.79, and 0.60–0.80.42,43 Using the DEHM/UBM/AirGIS system, we estimated hourly address-specific concentrations (the modeling system operates at all scales in a 1-h time resolution) of the four air pollutants from 2000 through 2017, which we summarized into monthly averages for each address. We attached the monthly exposures to person-specific address histories and calculated person-specific time-weighted 5-y running means for the four exposures.
Sociodemographic Variables
All SES variables in the present study were collected from the nationwide registers that, based on yearly input from relevant authorities (e.g., the Danish tax authorities for income and all educational institutions for education), accumulate this information. From the registries at Statistics Denmark, we obtained information on a number of individual- and area-level SES variables, selected based on availability and findings of previous papers showing associations between the SES variables and the outcome of interest (T2D) as well as exposure to air pollution (see Directed Acyclic Graph in Figure S1 generated in DAGitty, version 3.044).45 More specifically, we obtained yearly individual-level information from 2005 to 2017 on highest attained education categorized as short (mandatory), medium (secondary/vocational), and long (e.g., university, nursing, and teaching) education, disposable individual income (calculated as calendar year and sex-specific quintiles based on the income distribution in the Danish population), occupational status (blue-collar, white-collar, unemployed/retired), cohabiting status [“live alone,” corresponding to divorced/widowed/never-married persons who do not share address with others (except their children) and “cohabiting,” corresponding to married people as well as people sharing address with one or more persons (except children)] and country of birth (Denmark, other). We also obtained yearly information on three neighborhood-level SES indicators: proportion of inhabitants in each parish with only basic education, with a non-Western background (corresponding to being born in a non-Western country), and with a criminal record. Furthermore, we obtained yearly information on population density within each parish (<100, 100 to<2,000, and ≥2,000 persons per square kilometer). In 2017, there were 2,160 Danish parishes with a median of 1,032 inhabitants and a mean size of 16 km2.
We excluded all persons missing information on one or more of the SES variables described above from the study population.
Financial Stress and Comorbidity
We used the registers of Statistics Denmark to identify people experiencing one or more “financial stress event(s)” defined as family income below the Danish relative poverty limit (time-dependent), personal income drop of 50% or more between 2 consecutive years, family income drop of 50% or more between 2 consecutive years, and/or loss of job. Based on this approach, we created a time-dependent dichotomous variable of one or more financial stressful event(s) in the prior 5 y (yes/no).
Using the National Patient Registry,30 we calculated a Charlson Comorbidity Index for all cohort members, which is a standard method of categorizing comorbidities of patients based on ICD codes.46 The index was calculated as a time-dependent variable, summing up a score based on diseases during 5 previous years, calculated with a 1-y lag period (0–1 y; to ensure that the diagnosis of T2D did not impact the index). In analyses, we categorized the comorbidity index score into 0, 1, or ≥2.
Road Traffic Noise
We modeled road traffic noise at all residential addresses at the most exposed facade using the Nordic prediction method47 for the years 2000, 2005, 2010, and 2015 as previously described.48 Input variables included address-specific geocodes; height, road type, light/heavy vehicle distributions, travel speed, and annual average daily traffic for all Danish road links38; and screening effects from buildings, terrain, and noise barriers. We calculated noise as the equivalent A-weighted sound pressure level for day (0700–1900 hours), evening (1900–2200 hours) and night (2200–0700 hours) and aggregated it as Lden. We used linear interpolation between the 5-y exposure calculations to quantify exposure for all years in the period 2000–2017.
Green Space
We used BASEMAP02, which classifies land use in a high-resolution map of Denmark, to calculate area proportions of 36 land-use classes within a 1,000 meter radius around all addresses.49,50 Green space of high quality was defined as forest, recreational areas, and wet/dry open nature areas.
Statistical Analyses
Correlations between air pollutants were calculated as Spearman’s correlation coefficients. We calculated associations between 5-y exposure to air pollution and risk of T2D using two different models: Cox proportional hazard model and Aalen additive hazard model. Based on the Cox model we calculated hazard ratio (HR; relative risk estimate) and based on the Aalen model we calculated the rate difference per 100,000 person-years (absolute risk estimate). In both models, we included age as the underlying time scale (continuous), and air pollution was modeled as 5-y time-weighted running means. In brief, this modeling was done by calculating mean exposure for the 5 y before the T2D diagnosis for all cases, taking all present and historical addresses in this period into account (including exposure before baseline when relevant) and subsequently for each case, and then comparing this exposure with the 5-y exposure for all noncases at the exact same age as the case at the time of diagnosis.
In initial analyses based on the whole study population of people above 35 y of age (using the Aalen model), we observed that all four air pollutants were associated with higher risk of T2D, mainly among people between 50 and 80 y of age (Figure 1). We therefore restricted all main analyses to include only the 1,843,597 persons within this age group. Start of follow-up in all analyses was age 50 y or year 2005 (whichever came last), and people were censored at type 1 diabetes or T2D diagnosis, age 80 y, death, missing address, emigration, or end of follow-up (31 December 2017), whichever came first. Furthermore, we conducted additional analyses using the Cox model on the population age 35–50 y, including 1,300,108 persons, with start of follow-up at age 35 y or year 2005 and censoring at type 1 diabetes or T2D diagnosis, age 50 y, death, missing address, emigration, or end of follow-up (31 December 2017).
Figure 1. Associations between 5-y exposure to air pollution (PM2.5, ultrafine particles, elemental carbon, and NO2) and risk of type 2 diabetes according to age, expressed as cumulative coefficients (middle curve) with 95% confidence intervals (upper and lower curve).
Figures 1A to 1D are line graphs titled particulate matter begin subscript 2.5 end subscript, Ultrafine particles, Elmental carbon, and Nitrogen Dioxide, plotting Cumulative coefficients, ranging from 0.00 to 0.06 in increments of 0.01; 0.00 to 0.05 in increments of 0.01; 0.00 to 0.05 in increments of 0.01; and 0.000 to 0.030 in increments of 0.010 (y-axis) across Age (years), ranging from 40 to 80 in increments of 10 (x-axis), respectively.
We calculated risk estimates for the association between air pollution and T2D adjusted for sex, calendar year (2-y categories), educational level, individual income, cohabiting status, country of birth, and occupation, as well as area-level proportion of inhabitants with only basic education, of non-Western background, and with a criminal record. Estimates were calculated per 5 μg/m3 PM2.5, 10 μg/m3 NO2, 1 μg/m3 EC, and 10,000 particles/cm3 for UFP. To examine the shape of the exposure–response relationship, we also analyzed associations between the four air pollutants and T2D among people age 50–80 y based on the Cox model in the following categories: <10th (reference group), 10th to<25th, 25th to<50th, 50th to<75th, 75th to<90th, 90th to<95th, and ≥95th percentiles.
For all four air pollutants, we investigated associations between air pollution and T2D in strata of sociodemographic variables (sex, education, income, occupation, and cohabiting status), financial stress (yes, no), comorbidity (Charlson Comorbidity Index; 0, 1, ≥2), population density (<100, 100to<2,000, ≥2,000 persons/km2), road traffic noise (<55, 55to<60, ≥60 dB), and green space within 1,000m (<9.8%, 9.8% to<17.6%, ≥17.6%).
Descriptive analyses and Cox proportional hazards model analyses were done in SAS 9.4 (SAS Institute Inc.) and Aalen additive hazard model analyses were performed in R (version 3.6.3; R Development Core Team).
Results
From the study base of 2,757,813 people, we excluded 13,535 persons with type 1 diabetes and 88,934 with T2D before baseline. Also, we excluded 23,856 persons missing information on one or more potential confounders. Of the remaining 2,631,488 persons, for the main analyses we excluded 787,891 persons who were below age 50 y at end of follow-up or above 80 y of age at start of follow-up. This approach yielded a study population of 1,843,597 persons with a median follow-up of 9.5 y during which 113,985 developed T2D. Also, we conducted additional analyses on a population of persons between 35–50 y of age, consisting of 1,300,108 persons, of whom 19,662 developed T2D.
We found that people exposed to UFP above the median were more likely to be women, live alone, have high income, have longer education, be of non-Danish origin, be retired/unemployed, and live in neighborhoods with a higher proportion of people with a non-Western background and a criminal record in comparison with people exposed to UFP below the median (Table 1; Table S1). T2D cases were more likely to be men, have a low SES, with low education and income, and working in blue-collar jobs, as well as have comorbidities in comparison with noncases (Table S2). The distributions of 5-y exposure to PM2.5, UFP, EC, and NO2 at baseline are shown in Figure S2, Table 1, and Table S3. UFP, EC, and NO2 were found skewed to the right. The four air pollutants were correlated with RSpearman coefficients between 0.75 and 0.94 (Table 2).
Table 1 Baseline sociodemographic characteristics and exposures among the Danish study population of people age 50–80 y in the period from 2005–2017 according to baseline 5-y exposure to UFP below and above the median.
Baseline Characteristics Cohort (N=1,843,597) UFP<11,064 particles/cm3 (n=921,797) UFP≥11,064 particles/cm3 (n=921,800)
Individual level
Men (%) 47.7 49.4 46.4
Age [y (mean±SD)] 58.9±9.1 57.6±8.9 60.3±9.1
Cohabiting status (%)
Cohabiting 74.9 79.0 70.9
Living alone 25.1 21.0 29.1
Individual income (%)
Low (quintile 1) 23.8 24.0 23.6
Low-medium (quintile 2) 20.5 20.8 20.3
Medium (quintile 3) 17.0 18.3 15.6
Medium-high (quintile 4) 17.8 18.6 16.9
High (quintile 5) 21.0 18.4 23.5
Highest attained education (%)
Mandatory education 34.8 36.8 32.7
Secondary or vocational education 46.1 46.5 45.7
Medium or long education 19.1 16.7 21.6
Country of birth (%)
Danish 98.1 99.0 97.3
Other 1.9 1.0 2.7
Occupational status (%)
Blue-collar 31.4 35.5 27.2
White-collar level 26.6 27.1 26.0
Retired or unemployed 42.1 37.4 46.8
Financial stress (%)
Yes 17.8 18.9 16.7
No 82.2 81.1 83.4
Charlson Comorbidity Index (%)
0 86.9 88.6 85.3
1 7.4 6.6 8.3
≥2 5.7 4.8 6.5
Address level
Road traffic noise [5-y (%)]
<55 dB 50.6 57.6 43.6
55 to<60 dB 21.6 20.0 23.1
≥60 dB 27.9 22.4 33.4
High-quality green space in 1,000 m (%)
<9.8 % 34.0 38.2 29.8
9.8 to<17.6 % 33.4 30.0 36.8
≥17.6 % 32.6 31.8 33.5
Air pollution [5-y (mean±SD)]
PM2.5 (μg/m3) 10.9±1.3 10.1±1.1 11.6±1.0
UFP (particles/cm3) 11,578±3,231 9,075±1,316 14,082±2,571
EC (μg/m3) 0.70±0.29 0.54±0.10 0.87±0.31
NO2 (μg/m3) 16.5±5.9 12.5±2.5 20.5±5.6
Area level
Area-level SES (mean±SD)
% with only basic education 10.3±3.4 11.2±3.3 9.5±3.3
% non-Western background 5.2±6.0 3.3±3.9 7.1±7.0
% with criminal record 0.49±0.31 0.40±0.25 0.57±0.34
Population density
<100/km2 26.7 47.4 6.0
100 to<2,000/km2 55.2 48.6 61.8
≥2,000/km2 18.2 4.1 32.2
Note: EC, elemental carbon; SD, standard deviation; SES, socioeconomic status; UFP, ultrafine particles.
Table 2 Spearman correlations between 5-y exposure to PM2.5, ultrafine particles, elemental carbon, and NO2 for the main study population (ages 50–80 y) in 2005 (N=1,252,432).
PM2.5 Ultrafine particles Elemental carbon NO2
PM2.5 1 0.75 0.75 0.79
Ultrafine particles 0.75 1 0.90 0.92
Elemental carbon 0.75 0.90 1 0.94
NO2 0.79 0.92 0.94 1
In the main population of people age 50–80 y, exposure contrasts of 5 μg/m3 PM2.5, 10,000 UFP per cm3, 1 μg/m3 EC, and 10 μg/m3 NO2 were associated with a higher risk of incident T2D, with overall HRs 95% CI of 1.17 (95% CI: 1.13, 1.21), 1.16 (95% CI: 1.13, 1.19), 1.10 (95% CI: 1.08, 1.12) and 1.10 (95% CI: 1.08, 1.11), respectively, and overall rate differences of 123 (95% CI: 96, 150), 124 (95% CI: 104, 144), 102 (95% CI: 80, 124), and 73 (95% CI: 62, 85), respectively. Inspection of the exposure–response relationships between the four air pollutants and T2D indicated a linear relationship for EC and NO2, whereas for PM2.5 and UFP there were some indications of a leveling off at high exposures (Figure S3; Table S4). For all four air pollutants, the association with T2D was stronger among men in comparison with women and among people living with a partner in comparison with people living alone (Table 3). Similar trends were observed for relative (HR) and absolute (rate difference) risk estimates. We found lower risk estimates among people with high education in comparison with low/medium education. When comparing people with low vs. medium education, risk estimates were highest among people with medium education. For income, the risk estimates were generally highest in the medium category; one exception was PM2.5, where similar size estimates were observed for people with medium and high income. No marked differences were observed between people working in white-collar vs. blue-collar occupations.
Table 3 Association between air pollution and incidence of type 2 diabetes by sociodemographic factors.
n cases PM2.5 (per 5 μg/m3) UFP (per 10,000 particles/cm3) Elemental carbon (per 1 μg/m3) NO2 (per 10 μg/m3)
Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b
All 113,985 1.17 (1.13, 1.21) 123 (96, 150) 1.16 (1.13, 1.19) 124 (104, 144) 1.10 (1.08, 1.12) 102 (80, 124) 1.10 (1.08, 1.11) 73 (62, 85)
Sex
Men 65,245 1.23 (1.18, 1.28) 198 (165, 232) 1.22 (1.19, 1.26) 210 (183, 279) 1.12 (1.10, 1.15) 183 (147, 220) 1.13 (1.11, 1.15) 119 (104, 134)
Women 48,740 1.09 (1.04, 1.14) 51 (23, 79) 1.07 (1.04, 1.11) 47 (25, 69) 1.03 (1.00, 1.07) 26 (2, 49) 1.06 (1.04, 1.08) 37 (26, 49)
Education
Short 49,162 1.12 (1.07, 1.17) 96 (58, 133) 1.12 (1.08, 1.16) 121 (90, 152) 1.07 (1.04, 1.10) 101 (66, 136) 1.08 (1.06, 1.10) 75 (57, 92)
Medium 50,992 1.24 (1.19, 1.30) 169 (138, 199) 1.23 (1.19, 1.27) 172 (147, 196) 1.12 (1.10, 1.15) 147 (113, 181) 1.14 (1.12, 1.16) 103 (90, 117)
Long 13,831 1.10 (1.02, 1.19) 48 (12, 84) 1.05 (0.99, 1.11) 16 (−12, 44) 0.98 (0.92, 1.06) −25 (−58, 8) 1.04 (1.01, 1.07) 14 (−1, 29)
Disposable income
Quintile 1 (low) 41,633 1.12 (1.07, 1.17) 71 (30, 112) 1.11 (1.07, 1.15) 89 (56, 122) 1.05 (1.02, 1.09) 59 (23, 96) 1.07 (1.05, 1.09) 56 (38, 75)
Quintile 2–4 60,530 1.20 (1.15, 1.25) 140 (111, 170) 1.20 (1.16, 1.23) 147 (122, 171) 1.13 (1.11, 1.16) 141 (109, 173) 1.13 (1.11, 1.15) 90 (77, 103)
Quintile 5 (high) 11,822 1.23 (1.14, 1.33) 141 (103, 179) 1.15 (1.08, 1.22) 93 (63, 122) 1.06 (0.99, 1.13) 54 (18, 88) 1.08 (1.04, 1.11) 51 (35, 67)
Occupation
White-collar 14,107 1.27 (1.18, 1.36) 194 (157, 231) 1.19 (1.12, 1.25) 138 (109, 167) 1.15 (1.09, 1.21) 141 (106, 175) 1.12 (1.09, 1.15) 78 (62, 94)
Blue-collar 21,423 1.20 (1.13, 1.27) 199 (161, 236) 1.26 (1.21, 1.32) 209 (177, 240) 1.15 (1.11, 1.19) 206 (171, 242) 1.16 (1.14, 1.19) 122 (105, 139)
Cohabiting status
Cohabiting 78,945 1.22 (1.17, 1.26) 141 (113, 168) 1.20 (1.17, 1.24) 145 (123, 167) 1.11 (1.09, 1.14) 126 (100, 152) 1.14 (1.13, 1.16) 101 (89, 113)
Living alone 35,040 1.08 (1.03, 1.13) 67 (29, 105) 1.07 (1.04, 1.11) 70 (40, 101) 1.03 (0.99, 1.07) 45 (6, 84) 1.04 (1.02, 1.06) 27 (12, 43)
Note: CI, confidence interval; HR, hazard ratio; py, person-years: .
a Analyses were adjusted for age, sex, calendar year, education, cohabiting status, personal income, country of birth, and area-level percentage of population with only basic education, with non-Western background, and with criminal record.
b Risk estimates calculated in Aalen additive hazard model is given as a rate difference per 100,000 py.
People with any financial stress during the last 5 y, such as job loss and an income below the poverty limit, were found to have lower air pollution-T2D risk estimates than people without such events (Table 4). We found stronger associations between air pollution and T2D among people with a comorbidity or comorbidities in comparison with no comorbidity; the association was strongest for people with a comorbidity score of ≥2. We found no consistent indications of effect modification by population density, road traffic noise, or surrounding green space across the four air pollutants.
Table 4 Association between air pollution and incidence of type 2 diabetes by financial stress, comorbidity, population density, road traffic noise and surrounding green space.
N cases PM2.5 (per 5 μg/m3) UFP (per 10,000 particles/cm3) Elemental carbon (per 1 μg/m3) NO2 (per 10 μg/m3)
Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b Cox model
HR (95% CI)a Aalen model
Rate (per 100,000 py) difference (95% CI)a,b
All 113,985 1.17 (1.13, 1.21) 123 (96, 150) 1.16 (1.13, 1.19) 124 (104, 144) 1.10 (1.08, 1.12) 102 (80, 124) 1.10 (1.08, 1.11) 73 (62, 85)
Financial stressc
Yes 19,290 1.05 (0.98, 1.11) 52 (6, 98) 1.08 (1.03, 1.13) 77 (39, 115) 1.03 (0.97, 1.08) 41 (−3, 85) 1.05 (1.02, 1.08) 40 (20, 61)
No 94,695 1.20 (1.15, 1.24) 133 (106, 159) 1.18 (1.15, 1.21) 131 (110, 152) 1.11 (1.08, 1.13) 114 (89, 139) 1.11 (1.09, 1.13) 81 (70, 92)
Charlson Comorbidity Indexd
0 77,432 1.04 (1.00, 1.08) 34 (8, 60) 1.09 (1.06, 1.12) 63 (44, 83) 1.07 (1.05, 1.10) 62 (40, 85) 1.08 (1.06, 1.09) 50 (40, 61)
1 18,592 1.28 (1.20, 1.36) 308 (230, 386) 1.19 (1.14, 1.25) 250 (184, 315) 1.08 (1.03, 1.13) 133 (57, 208) 1.07 (1.04, 1.10) 74 (38, 1,109)
≥2 17,961 1.40 (1.31, 1.49) 441 (360, 522) 1.26 (1.20, 1.32) 338 (270, 405) 1.12 (1.08, 1.16) 255 (166, 344) 1.13 (1.10, 1.16) 155 (117, 192)
Population density
<100/km2 29,689 1.09 (1.04, 1.16) 74 (35, 113) 1.12 (1.06, 1.18) 103 (56, 150) 1.05 (1.02, 1.09) 56 (12, 101) 1.13 (1.09, 1.18) 97 (65, 129)
100 to<2,000/km2 63,665 1.16 (1.11, 1.21) 109 (78, 141) 1.15 (1.11, 1.18) 111 (85, 136) 1.10 (1.07, 1.14) 98 (67, 129) 1.13 (1.11, 1.15) 94 (78, 109)
≥2,000/km2 20,631 1.21 (1.14, 1.28) 144 (101, 188) 1.17 (1.12, 1.23) 127 (93, 162) 1.16 (1.10, 1.22) 122 (82, 162) 1.06 (1.04, 1.09) 50 (34, 66)
Road traffic noise
<55 dB 55,998 1.12 (1.07, 1.18) 87 (55, 119) 1.14 (1.10, 1.18) 103 (77, 130) 1.11 (1.06, 1.16) 93 (54, 132) 1.13 (1.10, 1.16) 89 (69, 108)
55 to<60 dB 24,918 1.18 (1.11, 1.25) 125 (82, 169) 1.16 (1.11, 1.21) 121 (86, 156) 1.05 (1.01, 1.09) 56 (21, 92) 1.12 (1.08, 1.15) 85 (62, 107)
≥60 dB 33,069 1.15 (1.10, 1.21) 118 (83, 153) 1.15 (1.11, 1.19) 123 (95, 150) 1.09 (1.07, 1.12) 100 (68, 131) 1.08 (1.06, 1.10) 61 (48, 74)
High-quality green space within 1,000 me
<9.8 % 38,931 1.14 (1.08, 1.20) 104 (69, 139) 1.13 (1.09, 1.17) 106 (78, 134) 1.14 (1.09, 1.20) 111 (74, 148) 1.08 (1.06, 1.10) 59 (44, 74)
9.8 to<17.6 % 38,953 1.16 (1.11, 1.22) 118 (82, 154) 1.14 (1.10, 1.18) 110 (81, 138) 1.16 (1.11, 1.21) 126 (89, 163) 1.09 (1.07, 1.11) 68 (53, 83)
≥17.6 % 36,101 1.18 (1.12, 1.24) 121 (87, 156) 1.19 (1.15, 1.24) 148 (119, 177) 1.08 (1.05, 1.10) 80 (50, 110) 1.13 (1.10, 1.15) 93 (76, 109)
Note: CI, confidence interval; HR, hazard ratio; py, person-years; UFP, ultrafine particles: .
a Analyses were adjusted for age, sex, calendar year, education, cohabiting status, personal income, country of birth, and area-level percentage of population with only basic education, with non-Western background, and with criminal record.
b Risk estimates calculated in Aalen additive hazard model is given as a rate difference per 100,000 py.
c Financial stress was defined as ≥1 of the following events during the last 5 years: family income below Danish relative poverty limit, personal income drop of ≥50% between two consecutive years, family income drop of ≥50% between two consecutive years and/or loss of job.
d Comorbidity defined as a Charlson Comorbidity Index of 0, 1, or ≥2 during the last 1–6 y with a lag-period of 1 y.
e Green space defined as the proportion of recreational areas, forest and wet/dry open nature areas within 1,000m of the residence.
In the population of people age 35–50 y, exposure contrasts of 5 μg/m3 PM2.5, 10,000 UFP per cm3, 1 μg/m3 EC, and 10 μg/m3 NO2 were associated with overall HRs of 1.05 (95% CI: 0.96, 1.14), 1.05 (95% CI: 0.99, 1.12), 1.05 (95% CI: 0.98, 1.13) and 1.02 (95% CI: 0.99, 1.06), respectively. We observed effect modification trends for the age 35–50 y population to be similar to those for those age 50–80 y [except for sex, where we found stronger association among women in comparison with that of men, and occupation, where we found stronger associations among blue-collar workers in comparison with those of white-collar workers (Tables S5–S6)].
Discussion
In a nationwide study of Denmark, we found that air pollution was associated with higher risk of T2D among people age 50–80 y in comparison with people age 35–50 y. In the population of people age 50–80 y, we found higher risk estimates among men in comparison with women and a pattern of higher risk estimates among people with low or medium education (highest for medium education) in comparison with high education, among people with medium income vs. low or high income, among people living with a partner vs. living alone, among people with comorbidities vs. without comorbidities, and among people without financial stress vs. people with financial stress. No marked changes in risk estimates were observed according to occupation, population density, road traffic noise, and surrounding green space. We observed similar tendencies among people 35–50 y of age, except in relation to sex and occupation, where we observed associations with air pollution only among women and blue-collar workers. The results showed similar tendencies for relative and absolute risk estimates.
For all four exposures, we observed weaker associations with risk of T2D among people age 35–50 y in comparison with people age 50–80 y. A potential explanation is that a diagnosis of T2D at a young age may have a stronger genetic component than diabetes later in life, and therefore environmental pollutants like air pollution may play a minor role in the development of diabetes in this age group. Also, there could be a higher degree of outcome misclassification among people below 50 y of age, e.g., the general practitioner (GP) may be less likely to test for T2D in younger patients, because the disease is less frequent in this age group.
The previous studies investigating associations between air pollution and metabolic syndrome/T2D according to sex are inconsistent, with some studies reporting the highest risk estimates for men16,21 and others for women,14,15 whereas some studies report no difference.9,22 We found the association between all four air pollutants and T2D to be stronger among men, especially for the absolute risk estimates, where the rate differences were 3–7 times higher in comparison with women. It is unclear whether these differences are caused by socially derived differences in exposure according to gender, by physiological differences related to sex, or a combination of these.20 Previous studies observing stronger associations between air pollution and T2D among women in comparison with men, suggested that it may be due to less exposure misclassification in women, because they may spend more time at home (where exposure is modeled).14 However, in Denmark both parents usually work when bringing up their children, and thus fewer differences according to time spend at home are expected in our population. Therefore societal differences across countries may partly explain differences in results across studies. The physiological differences between men and women that could lead to different risk in association with air pollution are numerous, such as differences in lung size, in deposition of particles, and in inflammatory responses.20,51 Also, men have a higher incidence of T2D and comorbidities, such as cardiovascular disease, than women, partly due to a protective effect of estrogen.52,53 It is therefore possible that men due to inherent sex-related physiological differences are more susceptible to the hazardous effects of air pollution. An interesting finding was that in the subpopulation of people age 35–50 y, air pollution was associated with higher risk of T2D only among women, whereas no associations were observed among men. These opposite findings in different age groups may partly explain inconsistencies in previous studies with regard to effect modification by sex.
We observed that air pollution was associated with a lower risk of T2D in people with long education in comparison with people with short or medium education, which was most pronounced for the absolute risk estimates. Only a few studies have investigated associations between air pollution and T2D in different strata of education, with one study reporting highest risk estimates among people with high education16 and two studies observing slightly lower risk estimates among the highly educated.14,15 Two of these studies investigated only two levels of education, an approach that may be too crude to capture the potentially complex relationship between SES, air pollution, and T2D. Having a short or medium education is associated with a lifestyle that is less healthy than that of people with a long education, e.g., physical inactivity, smoking, and high BMI.54 These are all risk factors for T2D, which are believed harmful through some of the same mechanistic pathways as air pollution, including oxidative stress and systemic inflammation.18,19 It is possible that people with an unhealthy lifestyle are more susceptible to the harmful effects of air pollution because their systems are already challenged, which could explain the lower risk found among the highly educated in our study.
We found that air pollution was associated with lower risk of T2D among people with short education in comparison with people with a medium education. A similar pattern was observed for income. Although it seems counterintuitive that the people with the lowest education and income are less susceptible to the harmful effects of air pollution than groups with higher SES, outcome misclassification may be part of the explanation: At least 24% of all T2D cases in Denmark are estimated to be undiagnosed.55 It is well known that, even in countries like Denmark with free health care for all residents, people with low SES are less likely to visit a GP for regular examinations or act on mild symptoms, such as frequent urination, weight loss, and fatigue, which are early symptoms of T2D.56–58 Therefore, they will in general be diagnosed later than people with higher SES. If such delay of a diagnosis last for several years, the 5-y exposure time window preceding the diagnosis (or part of it), which we applied in the present study, will also cover exposure after the person should have been censored, and thus result in exposure misclassification, which potentially could drive the risk estimates toward the null. In support, we also observed lower risk estimates between air pollution and T2D among people experiencing one or more “financial stress events,” which is a group of people who potentially have reduced “resources” to act on mild symptoms, as well as among people living alone, and therefore with no spouse to encourage seeking health care.59
Previous studies investigating comorbidity as a potential modifier of the association between air pollution and T2D are inconsistent, with some studies reporting stronger associations among people without COPD,9 myocardial infarction,9,14,15 and/or hypertension,9,14 whereas others report stronger associations in people with COPD14,15 and/or hypertension.15 We found that air pollution was associated with a substantially higher risk of T2D in people with comorbidities (assessed by the Charlson Comorbidity Index) in comparison with people without comorbidity. Furthermore, risk estimates were higher among people with a score of ≥2 compared to a score of 1. Part of the explanation is probably that people with, e.g., cardiovascular disease or COPD are automatically tested for T2D when hospitalized. There will, therefore, be fewer persons with undiagnosed T2D among people with comorbidities than among people without comorbidities and thus lower risk of outcome misclassification in this group. However, the difference in risk estimates for PM2.5 and UFP exposure among people with and without comorbidities is substantial, and it seems unlikely that outcome misclassification is the only explanation. Another explanation might be that comorbidities like cardiovascular disease, COPD, and asthma are characterized by systemic inflammation and/or oxidative stress, e.g., as a result of an unhealthy lifestyle and high BMI, which could make people with these diseases more vulnerable to the harmful effect of air pollution, because air pollution is believed to be harmful through the same biological pathways.17
We have previously found noise to be associated with T2D, with a HR of 1.03 (95% CI: 1.02–1.03) per 10 dB higher road traffic noise based on the entire Danish population (>35 y).24 However, although previous studies have suggested that noise can be hazardous through similar mechanisms as air pollution,26 our study does not suggest that noise exposure can modify the association between air pollution and T2D. Also, we found no marked differences in risk estimates according to the level of green space around the home address, suggesting that although previous studies have found greenness to be associated with a lower risk of T2D,27 it does not protect against the harmful effects of air pollution on risk for T2D.
A strength of the present study is the nationwide design, which minimized the risk of selection bias, and a large number of T2D cases identified using high-quality hospital and prescription registries.30,31 Furthermore, we obtained information on individual- and area-level SES covariates, comorbidity, financial stress, and residential address history from 2000–2017, using high-quality nationwide registries, and estimated air pollution and road traffic noise using validated models with high spatial resolution and high-quality input data.35
Limitations include the large proportion of people with undiagnosed T2D not captured in the present study. A validation study estimated that approximately 24% in Denmark had undiagnosed T2D.55 However, that study applied a different identification of T2D, including data from the Health Services Register (diabetic foot therapy) and two clinical databases in addition to the Patient and Prescription registries used in the present study. Although results are thus not directly comparable, it is unlikely that the percentage of undiagnosed T2D cases in the present study deviates substantially from the results obtained by Jørgensen et al.55 Outcome misclassification may be differential because people with low SES more often live with undiagnosed T2D (as described above). Another limitation is the lack of information on lifestyle covariates, especially adiposity. We, however, adjusted for various socioeconomic variables, which are associated with lifestyle. Furthermore, we recently conducted a study on long-term exposure to NO2 and PM2.5 and risk of T2D using a questionnaire-based cohort of 250,000 participants randomly selected across Denmark and with information on lifestyle habits.13 We found that after adjusting for various register-based individual- and area-level covariates (similar to the present study), further adjustment for lifestyle, resulted in only small changes in HRs, e.g., for PM2.5 the HR was 1.27 before and 1.24 after lifestyle adjustment, including smoking status, BMI and physical activity and intake of fruit, vegetables, and red meat. Limitations also include the lack of information on nonresidential exposure to air pollution, e.g., during work. We expect such misclassification to be mainly unrelated to T2D and draw the estimates toward the null. Another limitation is that the indicators of green space, financial stress, and comorbidities used in the present study are based on objective register-based data that may not capture all relevant aspects of these factors. Last, although our study was based on the entire Danish population and we thus believe that our results can be generalized to other Western populations, differences in, e.g., genetics, air pollution sources, and concentrations of air pollution have to be considered when generalizing the results.
In conclusion, we found men and individuals with preexisting comorbidities to be highly susceptible to the harmful effect of air pollution in relation to T2D, whereas people with high SES were less susceptible than people with lower SES. These findings suggest that the health burden of air pollution is not evenly distributed.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by the Health Effects Institute (HEI) (Assistance Award No. R-82811201). HEI is an organization jointly funded by the U.S. Environmental Protection Agency (U.S. EPA) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI or its sponsors, nor do they necessarily reflect the views and policies of the U.S. EPA or motor vehicle and engine manufacturers.
The study funder was not involved in the design of the study; the collection, analysis, and interpretation of the data; and writing the paper, and the study funder did not impose any restrictions regarding the publication of the paper.
The data that support the findings of this study are available from Statistics Denmark (and only at a secure server at Statistics Denmark). However, restrictions apply to the availability of these data, which were used under license for the current study, and therefore are not publicly available. Access to data requires permission from Statistics Denmark and the Danish Cancer Society.
The present study is strictly register based, with no contact with the participants. According to Danish legislation, no ethical permission or informed consent is needed for strictly register-based studies.
M.S., A.H.P., and O.R.N. contributed to the study concept and design. M.S., A.H.P., U.A.H., J.H.C., J.B., L.M.F., M.K., C.A., and V.H.V. obtained, generated and/or cleaned data important for the analyses. M.S. did the statistical analyses and drafted the paper, and all authors contributed to a critical revision of the manuscript and the final approval of the version to be published.
==== Refs
References
1. World Health Organization. 2016. Global Report on Diabetes. Geneva Switzerland: World Health Organization.
2. Liu F, Chen G, Huo W, Wang C, Liu S, Li N, et al. 2019. Associations between long-term exposure to ambient air pollution and risk of type 2 diabetes mellitus: a systematic review and meta-analysis. Environ Pollut 252 (pt B ):1235–1245, PMID: , 10.1016/j.envpol.2019.06.033.31252121
3. Lucht S, Hennig F, Moebus S, Führer-Sakel D, Herder C, Jöckel KH, et al. 2019. Air pollution and diabetes-related biomarkers in non-diabetic adults: a pathway to impaired glucose metabolism? Environ Int 124 :370–392, PMID: , 10.1016/j.envint.2019.01.005.30660850
4. Gangwar RS, Bevan GH, Palanivel R, Das L, Rajagopalan S. 2020. Oxidative stress pathways of air pollution mediated toxicity: recent insights. Redox Biol 34 :101545, PMID: , 10.1016/j.redox.2020.101545.32505541
5. Lucht SA, Hennig F, Matthiessen C, Ohlwein S, Icks A, Moebus S, et al. 2018. Air Pollution and glucose metabolism: an analysis in non-diabetic participants of the Heinz Nixdorf Recall study. Environ Health Perspect 126 (4 ):047001, PMID: , 10.1289/EHP2561.29616776
6. Chen Z, Salam MT, Toledo-Corral C, Watanabe RM, Xiang AH, Buchanan TA, et al. 2016. Ambient air pollutants have adverse effects on insulin and glucose homeostasis in Mexican Americans. Diabetes Care 39 (4 ):547–554, PMID: , 10.2337/dc15-1795.26868440
7. Stone V, Miller MR, Clift MJD, Elder A, Mills NL, Møller P, et al. 2017. Nanomaterials versus ambient ultrafine particles: an opportunity to exchange toxicology knowledge. Environ Health Perspect 125 (10 ):106002, PMID: , 10.1289/EHP424.29017987
8. Schraufnagel DE. 2020. The health effects of ultrafine particles. Exp Mol Med 52 (3 ):311–317, PMID: , 10.1038/s12276-020-0403-3.32203102
9. Bai L, Chen H, Hatzopoulou M, Jerrett M, Kwong JC, Burnett RT, et al. 2018. Exposure to ambient ultrafine particles and nitrogen dioxide and incident hypertension and diabetes. Epidemiology 29 (3 ):323–332, PMID: , 10.1097/EDE.0000000000000798.29319630
10. Sørensen M, Poulsen AH, Hvidtfeldt UA, Frohn LM, Ketzel M, Christensen JH, et al. 2022. Exposure to source-specific air pollution and risk for type 2 diabetes: a nationwide study covering Denmark. Int J Epidemiol 51 (4 ):1219–1229, PMID: , 10.1093/ije/dyac040.35285908
11. Fairburn J, Schule SA, Dreger S, Karla Hilz L, Bolte G. 2019. Social inequalities in exposure to ambient air pollution: a systematic review in the WHO European Region. Int J Environ Res Public Health 16 (17 ):3127, PMID: , 10.3390/ijerph16173127.31466272
12. Raaschou-Nielsen O, Taj T, Poulsen AH, Hvidtfeldt UA, Ketzel M, Christensen JH, et al. 2022. Air pollution at the residence of Danish adults, by socio-demographic characteristics, morbidity, and address level characteristics. Environ Res 208 :112714, PMID: , 10.1016/j.envres.2022.112714.35031338
13. Sørensen M, Hvidtfeldt UA, Poulsen AH, Thygesen LC, Frohn LM, Ketzel M, et al. 2022. The effect of adjustment to register-based and questionnaire-based covariates on the association between air pollution and cardiometabolic disease. Environ Res 203 :111886, PMID: , 10.1016/j.envres.2021.111886.34411546
14. Chen H, Burnett RT, Kwong JC, Villeneuve PJ, Goldberg MS, Brook RD, et al. 2013. Risk of incident diabetes in relation to long-term exposure to fine particulate matter in Ontario, Canada. Environ Health Perspect 121 (7 ):804–810, PMID: , 10.1289/ehp.1205958.23632126
15. Andersen ZJ, Raaschou-Nielsen O, Ketzel M, Jensen SS, Hvidberg M, Loft S, et al. 2012. Diabetes incidence and long-term exposure to air pollution: a cohort study. Diabetes Care 35 (1 ):92–98, PMID: , 10.2337/dc11-1155.22074722
16. Weinmayr G, Hennig F, Fuks K, Nonnemacher M, Jakobs H, Mohlenkamp S, et al. Heinz Nixdorf Recall Investigator Group. 2015. Long-term exposure to fine particulate matter and incidence of type 2 diabetes mellitus in a cohort study: effects of total and traffic-specific air pollution. Environ Health 14 :53, PMID: , 10.1186/s12940-015-0031-x.26087770
17. Rajagopalan S, Brook RD. 2012. Air pollution and type 2 diabetes: mechanistic insights. Diabetes 61 (12 ):3037–3045, PMID: , 10.2337/db12-0190.23172950
18. Margina D, Ungurianu A, Purdel C, Tsoukalas D, Sarandi E, Thanasoula M, et al. 2020. Chronic Inflammation in the context of everyday life: dietary changes as mitigating factors. Int J Environ Res Public Health 17 (11 ):4135, PMID: , 10.3390/ijerph17114135.32531935
19. Byrd DA, Judd SE, Flanders WD, Hartman TJ, Fedirko V, Bostick RM. 2019. Development and validation of novel dietary and lifestyle inflammation scores. J Nutr 149 (12 ):2206–2218, PMID: , 10.1093/jn/nxz165.31373368
20. Clougherty JE. 2010. A growing role for gender analysis in air pollution epidemiology. Environ Health Perspect 118 (2 ):167–176, PMID: , 10.1289/ehp.0900994.20123621
21. Zhang JS, Gui ZH, Zou ZY, Yang BY, Ma J, Jing J, et al. 2021. Long-term exposure to ambient air pollution and metabolic syndrome in children and adolescents: a national cross-sectional study in China. Environ Int 148 :106383, PMID: , 10.1016/j.envint.2021.106383.33465664
22. Voss S, Schneider A, Huth C, Wolf K, Markevych I, Schwettmann L, et al. 2021. ENVINT-D-20-01309: long-term exposure to air pollution, road traffic noise, residential greenness, and prevalent and incident metabolic syndrome: results from the population-based KORA F4/FF4 cohort in Augsburg, Germany. Environ Int 147 :106364, PMID: , 10.1016/j.envint.2020.106364.33421766
23. Zare Sakhvidi MJ, Zare Sakhvidi F, Mehrparvar AH, Foraster M, Dadvand P. 2018. Association between noise exposure and diabetes: a systematic review and meta-analysis. Environ Res 166 :647–657, PMID: , 10.1016/j.envres.2018.05.011.30006240
24. Thacher JD, Poulsen AH, Hvidtfeldt UA, Raaschou-Nielsen O, Brandt J, Geels C, et al. 2021. Long-term exposure to transportation noise and risk for type 2 diabetes in a nationwide cohort study from Denmark. Environ Health Perspect 129 (12 ):127003, PMID: , 10.1289/EHP9146.34855467
25. Ohlwein S, Hennig F, Lucht S, Matthiessen C, Pundt N, Moebus S, et al. 2019. Indoor and outdoor road traffic noise and incident diabetes mellitus: results from a longitudinal german cohort study. Environ Epidemiol 3 (1 ):e037, PMID: , 10.1097/EE9.0000000000000037.33778334
26. Munzel T, Sorensen M, Daiber A. 2021. Transportation noise pollution and cardiovascular disease. Nat Rev Cardiol 18 (9 ):619–636, PMID: , 10.1038/s41569-021-00532-5.33790462
27. Klompmaker JO, Janssen NAH, Bloemsma LD, Gehring U, Wijga AH, van den Brink C, et al. 2019. Associations of combined exposures to surrounding green, air pollution, and road traffic noise with cardiometabolic diseases. Environ Health Perspect 127 (8 ):87003, PMID: , 10.1289/EHP3857.31393793
28. Thiering E, Markevych I, Brüske I, Fuertes E, Kratzsch J, Sugiri D, et al. 2016. Associations of residential long-term air pollution exposures and satellite-derived greenness with insulin resistance in German adolescents. Environ Health Perspect 124 (8 ):1291–1298, PMID: , 10.1289/ehp.1509967.26863688
29. Schmidt M, Pedersen L, Sørensen HT. 2014. The Danish Civil Registration System as a tool in epidemiology. Eur J Epidemiol 29 (8 ):541–549, PMID: , 10.1007/s10654-014-9930-3.24965263
30. Lynge E, Sandegaard JL, Rebolj M. 2011. The Danish National Patient Register. Scand J Publ Health 39 :30–33, PMID: , 10.1177/1403494811401482.21775347
31. Kildemoes HW, Sørensen HT, Hallas J. 2011. The Danish National Prescription Registry. Scand J Public Health 39 (suppl 7 ):38–41, PMID: , 10.1177/1403494810394717.21775349
32. Danish Health Data Agency. Register for Udvalgte Kroniske Sygdomme og Svære Psykiske Lidelser. (In Danish). https://www.esundhed.dk/Emner/Operationer-og-diagnoser/-/media/ADE56F96C9DB4777B7E52CAB5C15CE27.ashx [accessed 4 January 2023].
33. Knudsen JS, Knudsen SS, Hulman A, Witte DR, Gregg EW, Lauritzen T, et al. 2022. Changes in type 2 diabetes incidence and mortality associated with introduction of HbA1c as diagnostic option: a Danish 24-year population-based study. Lancet Reg Health Eur 14 :100291, PMID: , 10.1016/j.lanepe.2021.100291.35024680
34. Wittström F, Skajaa N, Bonnesen K, Pedersen L, Ekholm O, Strate L, et al. 2022. Type 2 diabetes and risk of diverticular disease: a Danish cohort study. BMJ open 12 (2 ):e059852, PMID: , 10.1136/bmjopen-2021-059852.35190447
35. Khan J, Kakosimos KE, Raaschou-Nielsen O, Brandt J, Jensen SS, Ellermann T, et al. 2019. Development and performance evaluation of new AirGIS – A GIS based air pollution and human exposure modelling system. Atmos Environ 198 :102–121, 10.1016/j.atmosenv.2018.10.036.
36. Brandt J, Silver JD, Frohn LM, Geels C, Gross A, Hansen AB, et al. 2012. An integrated model study for Europe and North America using the Danish Eulerian Hemispheric Model with focus on intercontinental transport of air pollution. Atmos Environ 53 :156–176, 10.1016/j.atmosenv.2012.01.011.
37. Brandt J, Christensen JH, Frohn LM, Berkowicz R. 2003. Air pollution forecasting from regional to urban street scale - implementation and validation for two cities in Denmark. Phys Chem Earth 28 :335–344, 10.1016/S1474-7065(03)00054-8.
38. Jensen SS, Plejdrup MS, Hillig K. 2019. GIS-based National Road and Traffic Database 1960–2020. Technical Report No. 151. Aarhus, Denmark: Aarhus University, Danish Centre for Environment and Energy.
39. Vignati E, Wilson J, Stier P. 2004. M7: an efficient size-resolved aerosol microphysics module for large-scale aerosol transport models. J Geophys Res 109 :D22202, 10.1029/2003JD004485.
40. Frohn LM, Ketzel M, Christensen JH, Brandt J, Im U, Massling A, et al. 2021. Modelling ultrafine particle number concentrations at address resolution in Denmark from 1979–2018 – part 1: regional and urban scale modelling and evaluation. Atmos Environ 264 :118631, 10.1016/j.atmosenv.2021.118633.
41. Ketzel M, Frohn LM, Christensen JH, Brandt J, Massling A, Andersen C, et al. 2021. Modelling ultrafine particle concentrations at address resolution in Denmark from 1979 to 2019 – part 2: street scale modelling and evaluation. Atmos Environ 264 :118633, 10.1016/j.atmosenv.2021.118633.
42. Hvidtfeldt UA, Ketzel M, Sorensen M, Hertel O, Khan J, Brandt J, et al. 2018. Evaluation of the Danish AirGIS air pollution modeling system against measured concentrations of PM2.5, PM10, and black carbon, Environ Epidemiol 2 :e014, 10.1097/EE9.0000000000000014.
43. Ketzel M, Berkowicz R, Hvidberg M, Jensen SS, Raaschou-Nielsen O. 2011. Evaluation of AirGIS: a GIS-based air pollution and human exposure modelling system. Int J Environ Pollution 47 :226–238, 10.1504/IJEP.2011.047337.
44. Textor J, van der Zander B, Gilthorpe MS, Liskiewicz M, Ellison GT. 2016. Robust causal inference using directed acyclic graphs: the R package ’dagitty’. Int J Epidemiol 45 (6 ):1887–1894, PMID: , 10.1093/ije/dyw341.28089956
45. Hajat A, MacLehose RF, Rosofsky A, Walker KD, Clougherty JE. 2021. Confounding by socioeconomic status in epidemiological studies of air pollution and health: challenges and opportunities. Environ Health Perspect 129 (6 ):65001, PMID: , 10.1289/EHP7980.34124937
46. Roffman CE, Buchanan J, Allison GT. 2016. Charlson Comorbidities Index. J Physiother 62 (3 ):171, PMID: , 10.1016/j.jphys.2016.05.008.27298055
47. Bendtsen H. 1999. The Nordic prediction method for road traffic noise. Sci Total Environ 235 :331–338, 10.1016/S0048-9697(99)00216-8.
48. Thacher JD, Poulsen AH, Raaschou-Nielsen O, Jensen A, Hillig K, Roswall N, et al. 2020. High-resolution assessment of road traffic noise exposure in Denmark. Environ Res 182 :109051, PMID: , 10.1016/j.envres.2019.109051.31896468
49. Levin G, Iosub CI, Jepsen MR. 2017. Basemap02. Technical Documentation of a Model for Elaboration of a Land-Use and Land-Cover Map for Denmark. Technical Report from DCE No. 95. Aarhus, Denmark: Aarhus University, Danish Centre for Environment and Energy.
50. Department of Enviromental Science AU, Denmark. Basemap02 for Download, https://envs.au.dk/en/research-areas/society-environment-and-resources/land-use-and-gis/basemap/basemap02-for-download [accessed 4 January 2023].
51. Bellemare F, Jeanneret A, Couture J. 2003. Sex differences in thoracic dimensions and configuration. Am J Respir Crit Care Med 168 (3 ):305–312, PMID: , 10.1164/rccm.200208-876OC.12773331
52. Iorga A, Cunningham CM, Moazeni S, Ruffenach G, Umar S, Eghbali M. 2017. The protective role of estrogen and estrogen receptors in cardiovascular disease and the controversial use of estrogen therapy. Biol Sex Differ 8 (1 ):33, PMID: , 10.1186/s13293-017-0152-8.29065927
53. Yan H, Yang W, Zhou F, Li X, Pan Q, Shen Z, et al. 2019. Estrogen improves insulin sensitivity and suppresses gluconeogenesis via the transcription factor Foxo1. Diabetes 68 (2 ):291–304, PMID: , 10.2337/db18-0638.30487265
54. Zhang YB, Chen C, Pan XF, Guo J, Li Y, Franco OH, et al. 2021. Associations of healthy lifestyle and socioeconomic status with mortality and incident cardiovascular disease: two prospective cohort studies. BMJ 373 :n604, PMID: , 10.1136/bmj.n604.33853828
55. Jørgensen ME, Ellervik C, Ekholm O, Johansen NB, Carstensen B. 2020. Estimates of prediabetes and undiagnosed type 2 diabetes in Denmark: the end of an epidemic or a diagnostic artefact? Scand J Public Health 48 (1 ):106–112, PMID: , 10.1177/1403494818799606.30222048
56. Moody A, Cowley G, Ng Fat L, Mindell JS. 2016. Social inequalities in prevalence of diagnosed and undiagnosed diabetes and impaired glucose regulation in participants in the Health Surveys for England series. BMJ Open 6 :e010155, PMID: , 10.1136/bmjopen-2015-010155.26857106
57. Ruiz PL, Hopstock LA, Eggen AE, Njølstad I, Grimnes G, Stene LC, et al. 2021. Undiagnosed diabetes based on HbA1c by socioeconomic status and healthcare consumption in the Tromsø Study 1994-2016. BMJ Open Diabetes Res Care 9 (2 ):e002423, PMID: , 10.1136/bmjdrc-2021-002423.34782335
58. Claypool KT, Chung MK, Deonarine A, Gregg EW, Patel CJ. 2020. Characteristics of undiagnosed diabetes in men and women under the age of 50 years in the Indian subcontinent: the national family health survey (NFHS-4)/Demographic Health Survey 2015-2016. BMJ Open Diabetes Res Care 8 (1 ):e000965, PMID: , 10.1136/bmjdrc-2019-000965.32098896
59. Reczek C, Gebhardt-Kram L, Kissling A, Umberson D. 2018. Healthcare work in marriage: how gay, lesbian, and heterosexual spouses encourage and coerce medical care. J Health Soc Behav 59 (4 ):554–568, PMID: , 10.1177/0022146518808718.30381973
|
PMC009xxxxxx/PMC9945560.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36821707
EHP11221
10.1289/EHP11221
Research
Impact of a Statewide Livestock Antibiotic Use Policy on Resistance in Human Urine Escherichia coli Isolates: A Synthetic Control Analysis
https://orcid.org/0000-0002-9809-4695
Casey Joan A. 1
Tartof Sara Y. 2 3
Davis Meghan F. 4 5 6
Nachman Keeve E. 4 7 8 9
Price Lance 10
Liu Cindy 10
Yu Kalvin 11
Gupta Vikas 11
Innes Gabriel K. 12
Tseng Hung Fu 2
Do Vivian 1
Pressman Alice R. 13 14
Rudolph Kara E. 15
1 Department of Environmental Health Sciences, Columbia University Mailman School of Public Health, New York, New York, USA
2 Department of Research & Evaluation, Kaiser Permanente Southern California, Pasadena, California, USA
3 Department of Health Systems Science, Kaiser Permanente Bernard J. Tyson School of Medicine, Pasadena, California, USA
4 Department of Environmental Health & Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
5 Department of Molecular and Comparative Pathobiology, Johns Hopkins School of Medicine, Baltimore, Maryland, USA
6 Division of Infectious Diseases, Johns Hopkins School of Medicine, Baltimore, Maryland, USA
7 Department of Health Policy and Management, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
8 Johns Hopkins Center for a Livable Future, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
9 Risk Sciences and Public Policy Institute, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
10 Milken Institute School of Public Health, The George Washington University, Washington, District of Columbia, USA
11 Medical and Scientific Affairs, Becton, Dickinson and Company, Franklin Lakes, New Jersey, USA
12 Yuma Center of Excellence for Desert Agriculture (YCEDA), University of Arizona, Yuma, Arizona, USA
13 Center for Health Systems Research, Sutter Health, Walnut Creek, California, USA
14 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA
15 Department of Epidemiology, Columbia University Mailman School of Public Health, New York, New York, USA
Address correspondence to Joan A. Casey, 722 W 168th St., Rm. 1206, New York, New York, 10032-3727 USA. Email: [email protected]
22 2 2023
2 2023
131 2 02700708 3 2022
12 1 2023
18 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
On 1 January 2018, California implemented Senate Bill 27 (SB27), banning, for the first time in the United States, routine preventive use of antibiotics in food-animal production and any antibiotic use without a veterinarian’s prescription.
Objectives:
Our objective was to assess whether SB27 was associated with decreased antimicrobial resistance among E. coli isolated from human urine.
Methods:
We used U.S. nationwide monthly state-level data from BD Insights Research Database (Becton, Dickinson, and Co.) spanning 1 January 2013 to 30 June 2021 on antibiotic-resistance patterns of 30-d nonduplicate E. coli isolated from urine. Tested antibiotic classes included aminoglycosides, extended-spectrum cephalosporins (ESC), fluoroquinolones, and tetracyclines. Counts of tested and not-susceptible (resistant and intermediate, hereafter resistant) urine isolates were available by sex, age group (<65, 65+ year), month, and state. We applied a synthetic control approach to estimate the causal effect of SB27 on resistance patterns. Our approach created a synthetic California based on a composite of other states without the policy change and contrasted its counterfactual postpolicy trends with the observed postpolicy trends in California.
Findings:
We included 7.1 million E. coli urine isolates, 90% among women, across 33 states. From 2013 to 2017, the median (interquartile range) resistance percentages in California were 11.9% (7.4, 17.6), 13.8% (5.8, 20.0), 24.6% (9.6, 36.4), 7.9% (2.1, 13.1), for aminoglycosides, ESC, fluoroquinolones, and tetracyclines, respectively. SB27 was associated with a 7.1% reduction in ESC resistance (p-value for joint null: <0.01), but no change in resistance to aminoglycosides, fluoroquinolones, or tetracyclines.
Discussion:
Further research is needed to determine the role of SB27 in the observed reduction in ESC resistance E. coli in human populations, particularly as additional states implement similar legislation. https://doi.org/10.1289/EHP11221
Supplemental Material is available online (https://doi.org/10.1289/EHP11221).
J.A.C., M.F.D., and S.Y.T. report grants from National Institutes of Allergy and Infectious Diseases, grants from National Institute of Environmental Health Sciences, grants from National Institutes of Health Office of the Director, grants from Johns Hopkins Berman Institute for Bioethics, and internal funding from Kaiser Permanente Southern California Department of Research & Evaluation. V.G. and K.Y. report being employees of BD and as such have stock in the BDX company. All other authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Antibiotic use in food animals is recognized as an important driver of antimicrobial-resistant colonization and infection in humans.1,2 In addition to antibiotic use, food animals are confined in crowded conditions where they often live atop their own waste. These environmental pressures from antibiotic use and production practices can select for resistant strains of bacteria at industrial food animal production sites.3 In the United States, antimicrobial-resistant (AMR) pathogens cause nearly 3million infections and 35,000 deaths annually.1 Despite modest declines,4 sales of antimicrobials for food-animal production represent approximately 65% of antibiotics sold in the United States.5 Antimicrobial-resistant pathogens may spread from industrial food animal production sites to people through multiple routes, including foodborne, occupational, and environmental exposures.6–9 Animals can harbor bacteria from the farm to the processing facility where they can contaminate the retail meat supply, where improperly handled or undercooked meat can lead to human infection.10–12 The European Union has enacted regulations to restrict antimicrobial use in food-animal production, resulting in a 35% (biomass-adjusted) reduction in use from 2011 to 2018.13 In 2017, the U.S. Food and Drug Administration (U.S. FDA) completed a voluntary process to end the marketing of antibiotics for promoting growth and improving feed conversion in food animals; these policies, however, generally allow for routine preventive administration of antimicrobials.13 Despite the 2017 voluntary U.S. FDA guidance, sales of medically important antibiotics for use in livestock grew during the period from 2017 to 2020.4
On 1 January 2018, California enacted Senate Bill 27 (SB27), the first U.S. law to address routine preventive uses of medically important antimicrobials in food animals.13 Medically important antibiotic classes include aminoglycosides, amphenicols, cephalosporins, diaminopyrimidines, fluoroquinolones, lincosamides, polymyxins, macrolides, penicillins, streptogramins, sulfoamides, and tetracyclines.14 SB27 effectively restricted availability of medically important antibiotics from over-the-counter to prescription-only, allowing use only under supervision of a veterinarian. No longitudinal data on antibiotic use in food animals are systematically collected in the United States, but the U.S. FDA collects nationwide data on sales of antimicrobial drugs. Of medically important antimicrobials, tetracycline dominates (66% of 2020 sales), followed by penicillins (13% of 2020 sales) and aminoglycosides (5% of 2020 sales).4 Nationally, cephalosporins constitute a small (<1%) portion of sales, but 80% of these are sold for use in cattle, one of the dominant food animals raised in California.15 The poultry industry is also important in the state.16 Within California, SB27 mandates that the California Department of Food and Agriculture (CDFA) gathers data on antimicrobial sales and on-farm use; however, the provision of these data to CDFA is voluntary and those collected have not been made available in a usable format for research.13
Some studies have indicated that restriction interventions reduce AMR bacteria on meat or in humans. A meta-analysis reported that externally imposed and voluntary restriction interventions have resulted in a −19% change [95% confidence interval (CI): −26, −11%] in the proportion of antibiotic-resistant isolates on meat and a −14% change (95% CI: −20%, −8%) in the proportion of AMR bacteria isolated from humans.6 However, not all longitudinal human studies have identified a reduction of AMR bacteria in people post intervention. For example, after a ban of the glycopeptide avoparcin in poultry production in Norway, an 8-y follow-up identified reductions in glycopeptide-resistant enterococci from poultry but not farmers’ feces.17 Only limited studies have assessed changes in resistance profiles of bacteria isolated from the general human population after external bans or voluntary reductions of antibiotic use in the food animal sector.18–22 Although several studies have suggested a link between foodborne exposures and AMR extraintestinal pathogenic Escherichia coli (E. coli) urinary tract infections (UTIs),23 no prior studies have evaluated changes in human E. coli resistance patterns from urine cultures after food-animal antimicrobial-use changes.6,24
We estimated the effect of SB27 on human AMR infection in California—a state that produced nearly 288 million broiler chickens (3.2% of U.S. total) and 5.2 million cattle and calves (5.6% of U.S. total) in 201716—using nationwide AMR data. Prior studies that evaluated European Union–based policies relied on comparisons of data from before and after legislation restricting the use of antibiotics in food animals in the same country.6 In this study, we used an augmented synthetic control approach to construct a synthetic California to predict the post-policy counterfactual resistance trends that would have occurred had SB27 not been implemented.25,26 We focused on UTI, one of the most common and costly outpatient bacterial infections ($3.5 billion per year in the United States).27
Methods
We used the augmented synthetic control method (ASCM26) to estimate the effect of SB27 on antibiotic-resistant E. coli UTI in California. The traditional synthetic control method (SCM28) constructs a weighted combination of control units to serve as the “synthetic control” to the treatment unit, where the weights are chosen such that the pre-policy control outcomes (and auxiliary covariates, if used) match the pre-policy treatment outcomes (and auxiliary covariates, if used) as closely as possible. When SCM was introduced, it was used to evaluate the effect of a tobacco control policy on cigarette sales in a single treated unit (California) using an optimally weighted combination of control units (other U.S. states) as a comparison.25 ASCM augments traditional SCM with a regularized outcome regression model (using ridge regression with the penalty parameter chosen via cross-validation) to improve pre-policy fit.26 In this analysis, control states are weighted—forming the synthetic control—such that their weighted average outcomes and covariates in the pre-SB27 period most closely match California’s outcomes and covariates in the pre-SB27 period. Including the outcome ridge regression further improved the fit of the synthetic control during the pre-SB27 period.
Using this approach, we estimated the expected prevalence of resistance (i.e., proportion of UTIs that were resistant) to four antibiotics among E. coli causing human UTI in the absence of SB27 implementation in California and compared these to the observed prevalence of resistance between 2013 and 2020. We considered the pre-SB27 period as dates prior to 1 January 2018 (SB27 implementation). We hypothesized that SB27 would result in reduced antibiotic resistance among E. coli isolates to antibiotic classes used in food animal production (e.g., tetracyclines), but not to those antibiotic classes used only or primarily in human medicine (e.g., fluoroquinolones). The institutional review boards at Kaiser Permanente Southern California (KPSC) and Columbia University approved this study (Protocol No: 11284 and Protocol No: AAAS9397, respectively).
Data Sources
We evaluated information on E. coli UTI in the ambulatory setting for the period spanning 1 January 2013 to 31 December 2020 from the BD Insights Research Database (Becton, Dickinson, and Co.). BD characterized the hospitals from which they received samples by bed size (<100, 100–300, >300), urban vs. rural (inside vs. outside a Core-Based Statistical Area), teaching status (major teaching hospital, graduate, limited, no affiliation) based on Healthcare Cost and Utilization Project definitions.29 Hospitals were defined as a major teaching hospital if they were a member of the Association of American Medical Colleges’ Council of Teaching Hospitals and Health Systems. BD then provided state-level counts of hospitals in each category combination. This database includes susceptibility testing on noncontaminant and nonduplicate (first isolate of a species within 30 d) urine cultures from 358 providers in 34 U.S. states and Washington D.C., with both small and large hospitals in urban and rural areas represented.30 Data include monthly, state-level counts of E. coli urine isolates that had at least one reported susceptibility result and counts of not-susceptible (resistant and intermediate, hereafter resistant) E. coli urine isolates by sex (female, male) and age group (<65, 65+ year). Reported antibiotic classes included: a) aminoglycosides (resistance to amikacin, gentamicin, or tobramycin); b) extended-spectrum cephalosporins (ESC; resistance to cefotaxime, ceftriaxone, ceftazidime, or cefepime); c) fluoroquinolones (resistance to ciprofloxacin or levofloxacin); and d) tetracyclines (resistance to tetracycline). We expected SB27 to reduce resistance to ESC and tetracycline but not aminoglycoside or fluoroquinolones, the latter of which has been banned for use in poultry in the United States since 2005,31 and therefore we considered it a negative outcome control.32 ESC are labeled for multiple indications in food animals, and although primarily for treatment,33 they may be employed for routine disease control, including intramammary ceftiofur use at dairy cow dry-off.34 In 2020, tetracyclines accounted for approximately two-thirds (∼4 million kg) of U.S. sales of medically important antibiotics for use in food animals.4 We excluded Maryland and Washington, D.C., from analyses because Maryland implemented a piece of legislation similar to SB27 in 2019, resulting in 33 states for final analyses.
In constructing the synthetic control, we included annual state-level variables that were hypothesized to be associated with antibiotic resistance and that were important predictors of resistance over time (receiving nonzero coefficients in a lasso model) as covariates.35 These included sex (percent female), racial/ethnic characteristics (percentage self-reported non-Hispanic White and percentage Hispanic), and age (percentage in the following categories: 0–14 y, 15–24 y, 25–44 y, 45–64 y, 65–84 y, 85+ y) from the 1-y American Community Survey (ACS) data beginning in 2013, prevalence of type 2 diabetes from the Centers for Disease Control and Prevention from 2013 to 2016, and the number of slaughterhouses in 2019 from the U.S. Department of Agriculture. From the BD Insights data, we also used the composition of hospitals from which they drew their data: urbanicity (urban vs. rural), size (greater vs. <300 beds), and teaching status [major medical center [yes/no] and graduate education (yes/no)]. We originally also included percentage non-Hispanic Black from the 1-y ACS data and poultry counts and concentrated animal feeding operation counts from the 2017 U.S. Department of Agriculture, but these variables received zero coefficients in the lasso model (Table S1). These covariates were averaged over the pre-SB27 period and were balanced alongside the pre-SB27 outcomes in constructing the optimal weights and were also incorporated into the ridge regression outcome model.
Outcomes
The primary outcomes in this study were the proportion of monthly state-level E. coli urine isolates resistant to each of the four tested antibiotic classes. Secondary outcomes were sex- and age-specific proportion of resistant E. coli urine isolates.
Statistical Analysis
We used ASCM,26 described at the beginning of the Methods section, to estimate the effect of SB27 on rates of antibiotic resistance in urine isolates (Figure S1). The weights were chosen to balance pre-SB27 outcomes and covariates, and the penalized outcome regression model included covariates and state-level fixed effects. The ASCM assumes that there are no post-SB27 changes influencing the outcome that differ between California and the comparison states forming the synthetic control. It also assumes there are no unobserved confounders of the decision to implement SB27 and the postlaw outcomes and assumes no serial correlation between postlaw and prelaw noise.26 For each of the antibiotic types and subgroups of interest, we estimated the difference in proportion of urine isolates resistant to the particular antibiotic class between California and its synthetic control for each year. We computed 95% point-wise CIs using the conformal inference approach of Cheronzhukov et al.,36 which has the advantage of retaining the correct coverage even in smaller, finite samples. In general, inference under this conformal approach tests the null hypothesis of whether the post-SB27 adjusted residual that contrasts the post-SB27 synthetic control with post-SB27 California is equal to the pre-SB27 residuals.
We conducted several sensitivity analyses. In the main analysis, we considered pre–1 January 2018 as pretreatment and post 1 January 2018 as posttreatment. In sensitivity analyses, we removed the 12 months prior and 6 months following SB27 enactment to allow for preemptive and delayed antibiotic use changes, which could have influenced our results. We also used data from the U.S. FDA National Antimicrobial Resistance Monitoring System (NARMS), which continuously and randomly collects retail chicken meat in NARMS states, to characterize where retail chicken meat processed (i.e., likely produced) in California is purchased (i.e., likely consumed) (Table S2). Among the 14 states where NARMS collects meat samples, California accounted for the highest proportion of purchase location for California processed chicken meat (65%), followed by New Mexico (12%), Colorado (9%), Washington (9%), and Oregon (5%). In a sensitivity analysis, we removed New Mexico, Oregon, and Washington from the synthetic control, and our data set did not include Colorado. Next, we conducted a falsification test, where we randomly assigned the date of SB27 to two additional dates (one pre-SB27 and one post-SB27 implementation) and repeated analysis. We would not expect to see a significant change in resistance related to these randomly selected dates. Finally, because changes in antibiotic-prescribing practices in human medicine in California could drive changes in resistance patterns among E. coli isolated from human urine samples, we partnered with KPSC to track antibiotic dispensing among patients with urinary tract infections (UTI). KPSC serves nearly 5 million patients in California. Between 2016 and 2020, we identified uncomplicated E. coli UTI events using diagnostic codes and laboratory testing37 from outpatient encounters [outpatient, synchronous virtual (e-visit, video, telephone advice visits), and emergency department] for members aged ≥18 y. Each year, we calculated the percentage of members diagnosed with an E. coli UTI who were dispensed aminoglycosides, ESC, and fluoroquinolones; KPSC does not routinely screen for tetracycline resistance, and thus it was not included.
We use the augsynth package to run ASCM analyses. Code to replicate our analyses are available in the Supplementary Material, “Supplemental Code” and on GitHub (https://github.com/kararudolph/code-for-papers/blob/master/SB27ASCM.R).
Results
The study used the augmented synthetic control method (ASCM) to compare California, which implemented legislation (SB27) addressing routine preventive uses of medically important antimicrobials in food animals, with other states that did not have such legislation during the period. Using ACSM, we constructed a synthetic California by reweighting other states that made up the pool of control states in such a way that the control weighted average pre-SB27 trend in antibiotic resistance of E. coli isolates closely matched that of California’s actual pre-SB27 trend. By comparing the estimated outcomes of the synthetic control in the post-SB27 period with California’s post-SB27 implementation outcomes, we estimated the effect of SB27 on E. coli resistance in California.
Using the BD Research Database, we identified 667,052 urinary E. coli isolates with antibiotic susceptibility testing collected in California and 6.5 million from 32 other states between 2013 and 2020. Most isolates were collected from women (88% in California and 90% elsewhere). On average, a slightly higher proportion of California isolates were resistant to the four tested antibiotic classes in comparison with isolates from the other 32 states (Table 1). Resistance to fluoroquinolones was most common (26% in California and 22% in all other states combined). In California, resistance to tetracycline increased from 14% in 2013 to 18% in 2020 but remained unchanged in other states (Figure S2). Resistance to fluoroquinolones decreased over the study period among isolates from older adults (65+ y) in non-California states. In California, it remained stable among isolates from older adults and increased among isolates from those <65 y of age.
Table 1 Distribution of outcome and predictor variables for California and control states from 2013 to 2020.
Variable California Other states
Mean SD Meana SD
Monthly E. coli isolates tested (n)
Female and <65y 219 6 35,737 10,319
Female and 65+ y 283 10 21,137 7,697
Male and <65 y 21 15 2,716 749
Male and 65+ y 55 15 3,775 1,360
Monthly antibiotic-resistant urinary E. coli isolatesb [n (%)]
Aminoglycosides 17 (14) 19 (10) 47 (9) 62 (4)
Extended-spectrum cephalosporins 17 (14) 20 (11) 33 (7) 45 (4)
Fluoroquinolones 34 (26) 42 (17) 106 (22) 138 (8)
Tetracyclines 17 (12) 23 (9) 43 (8) 79 (10)
a For other states, mean and SD calculated as a weighted mean where each state received the same weight.
b Defined as nonsusceptible (resistant and intermediate combined). An isolate was considered resistant to a class if resistance was reported to any of the following drugs within each class: aminoglycosides (resistance to amikacin, gentamicin, or tobramycin); extended-spectrum cephalosporins (resistance to cefotaxime, ceftriaxone, ceftazidime, or cefepime); fluoroquinolones (resistance to ciprofloxacin or levofloxacin); and tetracyclines (resistance to tetracycline).
California’s SB27 went into effect on 1 January 2018. ASCM results shown in Figure 1 indicated that after implementation, SB27 was associated with a 7.1% difference in ESC resistance (p-value for joint null: <0.01) between California and the synthetic California but no difference in resistance to aminoglycosides, fluoroquinolones, or tetracycline. When stratifying by sex and age, we saw similar trends in females and males and among those age 65+ y (Figure 2; Figure S3).
Figure 1. Difference between predicted and observed proportion of resistant urinary E. coli isolates by month from 2013–2020, before and after implementation of California’s Senate Bill 27 (SB27) for (A) Aminoglycosides; (B) Extended-spectrum cephalosporins; (C) Fluoroquinolones; and (D) Tetracyclines. The vertical dotted line indicates SB27 implementation on 1 January 2018.
Figures 1A to 1D are ribbon plus line graphs, plotting Difference in resistance prevalence (California-synthetic California), ranging from negative 0.02 to 0.04 in increments of 0.02; negative 0.15 to 0.00 in increments of negative 0.05; 0.00 to 0.05 in increments of 0.05; and negative 0.04 to 0.04 in increments of 0.04 (y-axis) across month of study, ranging from 30 to 90 in increments of 20 (x-axis) on January 1, 2018.
Figure 2. Difference between predicted and observed proportion of urinary E. coli isolates resistant to extended-spectrum cephalosporins (ESC) by month from 2013 to 2020, before and after implementation of California’s Senate Bill 27 (SB27) for (A) All women; (B) All men; (C) Women age 65+y; (D) Men age 65+y. The vertical dotted line indicates SB27 implementation on 1 January 2018.
Figures 2A to 2D are ribbon plus line graphs, plotting Difference in resistance prevalence (California-synthetic California), ranging from negative 0.15 to 0.00 in increments of negative 0.05; negative 0.20 to 0.00 in increments of negative 0.05; negative 0.10 to 0.05 in increments of 0.05; and negative 0.20 to 0.10 in increments of 0.10 (y-axis) across month of study, ranging from 30 to 90 in increments of 20 (x-axis) on January 1, 2018.
It is possible that changes in antibiotic use could have been made in anticipation of the law going into effect or that changes could have been phased in over time. With that in mind, we repeated the analyses removing the 12 months just prior to law enactment and 6 months postenactment (i.e., removing July 2017–December 2018). Results were similar to the primary analysis (Figures S4–S6) with the exception that SB27 enactment was no longer estimated to have a statistically significant effect on ESC resistance among older men. Not all retail chicken meat produced in California is consumed there, which could result in spillover effects of SB27. We identified the states where NARMS randomly purchased California-processed retail chicken meat and excluded them from the synthetic control (Table S2). These results did not differ from the main analysis (Figure S7). We also conducted a falsification test, randomly selecting two alternative California SB27 implementation dates and repeating the analysis; we observed no association between these dates and differences in resistance patterns (Figure S8A and S8B). Finally, we used KPSC electronic health record data to track annual trends in dispensing of aminoglycosides, ESC, and fluoroquinolones among members with a diagnosed E. coli UTI between 2016 and 2020 (Figure S9). These data suggested a modest increase in ESC prescribing; in 2016, 4.4% (n=1,590) of members with an E. coli UTI received an ESC and by 2020, 7.3% (n=2,600) did. We expect that an increase in prescribing of ESC in human medicine should be associated with an increase in ESC resistance, rather than the decrease that we observed post-SB27.
Discussion
California’s SB27 went into effect on 1 January 201813 and became the first state-level policy to restrict routine preventive use of antibiotics in food-animal production in the United States. We found evidence that implementation of SB27 co-occurred with a rapid reduction in resistance to ESC but not aminoglycosides, fluoroquinolones, or tetracyclines. SB27 could be related to changes in ESC resistance in multiple ways; exposures may occur through dietary and environmental pathways, including living near animal production sites.6–9 In 2019, third-generation cephalosporin-resistant E. coli accounted for the most worldwide deaths (n=59,900) of any pathogen–drug combination,38 and thus policies that may reduce resistance are critical.
Current uncomplicated UTI empiric therapy guidelines from the Infectious Diseases Society of America (IDSA) support the prescribing of certain antimicrobials, including trimethoprim-sulfamethoxazole, nitrofurantoin, beta-lactams, and quinolones.39 However, treatment with fluoroquinolones and ESC persists.40–42 Resistance to these antibiotics can lead to increased risk of clinical failure and recurrent UTI.43–45
We estimated a reduction in ESC resistance among E. coli concomitant with California’s SB27 implementation, which is consistent with timing of changes in AMR in European, Canadian, and Chinese examples. After bans of avoparcin in food-producing animals, the prevalence of vancomycin-resistant enterococci (VRE, a Gram-positive bacterium) in general-population human fecal samples dropped. Specifically, Germany banned avoparcin in 1996 and observed a drop in VRE from 12% to 3% from 1994 to 1997.20 In The Netherlands, VRE prevalence decreased from 12% in 1997 to 6% in 1999 after the 1997 European Commission ban of avoparcin.18 In Québec, the voluntary withdrawal (Q1-2005) and reintroduction (Q3-2007) of ceftiofur in chicken ovum provide another data point, this time for Gram-negative bacteria. From 2004 to 2006, Dutil et al. reported reduced ceftiofur resistance among Salmonella Heidelberg isolates from chickens (62% to 7%) and humans (36% to 8%); on partial reintroduction, ceftiofur resistance increased.21 Finally, in April 2017, the Chinese government banned colistin as a growth promoter. In the most comprehensive study to date, Wang et al. compared 2015–2016 to 2018–2019 and identified reductions in sales of colistin, colistin residues, and the mcr-1 gene (confers colistin resistance) on farms, and prevalence of colistin-resistant E. coli in livestock, causing human colonization (14.3% to 6.3%), and human infection (1.7% to 1.3%).22 These studies suggest that removal of the use of specific antibiotics from food-producing animals can result in a shift in resistance patterns for those same antimicrobial classes in bacteria isolated from humans.
Several factors could have biased our results toward the null result of no change in susceptibility. Given that SB27 was passed in 2015 but went into effect in 2018, producers may have preempted the change and shifted antimicrobial use practices sooner. Further, the U.S. FDA’s Veterinary Feed Directive (VFD) final rule, which requires veterinary oversight for use of antibiotics in feed nationally, was implemented in January 2017; thus California producers may have elected to change practices to be compliant with both the federal VFD rule and the SB27 during 2017.46 We see some evidence of this as prevalence of ESC resistance was estimated to decrease in the months leading up to 1 January 2018 and then stabilized at the new lower level. At the same time, some producers may have delayed or incompletely implemented new practices in response to SB27 because of perceived or actual benefits of continued antibiotic use47 or due to delayed guidance and enforcement from the CDFA. CDFA updated their guidance in 2019 and 2020.48 This lengthy period to complete guidance for implementation is one potential explanation for the lack of significant change in E. coli resistant to tetracycline estimated in our study, given that tetracycline is a major animal drug and accounted for 66% of domestic sales of medically important antimicrobials in food-producing animals in 2020.4 Because antibiotic use was not restricted outside of California and because retail meat processing and distribution is not restricted regionally,9 a continued influx of retail meat from animals raised and processed out of state (without the SB27 restrictions) could have impacted findings. Nationwide, tetracycline use in food-producing animals dropped by 30% from 2011 to 2020 (in comparison with a 1% drop for cephalosporins)4; this overall trend in tetracycline use could have masked any California-specific changes, especially when combined with nationwide food distribution. We did not, however, estimate reduced prevalence of tetracycline resistance among California urinary E. coli isolates post-SB27. This lack of reduction in tetracycline resistance could be explained by the near-ubiquity of tetracycline resistance genes in the human fecal microbiome49 or the exclusion of tetracycline from many clinical susceptibility panels. We know of no prior studies of policy interventions on agricultural antimicrobials that have documented changes in tetracycline resistance among bacteria isolated from humans. This absence of studies is possibly due to gaps in the collection of baseline (before antibiotic-use ban) zoonotic pathogen susceptibility data in some European Union member states.
Cephalosporins made up <1% of national food animal antimicrobial drugs sales by weight4 prior to SB27 (2011–2017), and factors other than the legislation could have driven the observed change in ESC resistance. For example, changes specific to California that occurred at the same time as SB27 implementation and were correlated with antimicrobial resistance could account for our results. Such changes could include California-specific antimicrobial prescribing practices in human medicine. In an attempt to measure such changes, we used electronic health record data from KPSC and observed an increase in prescribing of ESC among patients with E. coli UTI over the study period, which does not explain our findings. In this subanalysis, we tracked only ESC prescriptions related to E. coli UTI; overall clinical ESC prescribing could differ. Finally, changes in the distribution of E. coli strains that are associated with resistance in California vs. other states could potentially account for our results and remain unmeasured.
Ideally, we would have paired our results in human clinical isolates with information on changes in antimicrobial use practices in clinical medicine overall and on California farms. Currently, farm antimicrobial use data voluntarily submitted to CDFA are not available in a research-usable format.13 The lack of these on-farm use data constitutes a major limitation of the present research. At the federal level, the U.S. FDA only reports nationwide aggregated estimates of sales of medically important antimicrobials for sale or distribution for use in food animals.4 After implementation of SB27, 48% of surveyed California dairy farmers reported changes in antimicrobial-use practices, such as treating fewer animals or discontinuing the use of one or more antimicrobials.46 Another survey found that among conventional dairies, 43% reported reducing antimicrobial use of previously over-the-counter medications.50 Both surveys are limited by low response rates (∼10%) and we know of no similar reports from poultry farmers. Improved data availability in the future could allow for a more refined evaluation of the impact of reductions in use by drug class on human infections.
This study had limitations. Data from BD came from more than half of U.S. states with urban and rural and small and large health care settings but were not comprehensive. Urine cultures were from the outpatient setting, and we included only nonduplicate urine samples from outpatients without known recent hospitalization. If a patient had been hospitalized in a nonaffiliated health care center, however, we could have misclassified some samples. Further, although all U.S. laboratories follow Clinical and Laboratory Standards Institute and American Society for Microbiology guidelines, susceptibility testing results were based on local laboratory practices. Additionally, some laboratories suppress the results of susceptibility testing to prevent unnecessary use of certain antibiotics.51 Our analysis included all urine E. coli isolates with at least one reported susceptibility result and assumed that isolates were susceptible to unreported antibiotics as in prior work.52 This assumption likely led to underestimated overall antibiotic resistance prevalence (Figure S2), but we have no reason to believe hospital-specific suppression decisions would be differential with respect to SB27 implementation and therefore should not bias our overall findings of a reduction in ESC resistance post-SB27 in California. Because we only had access to phenotypic resistance by antibiotic class, we could not track changes in resistance to specific drugs, e.g., gentamicin and tobramycin within the aminoglycoside class or more refined changes in resistance as would be measured by minimum inhibitory concentrations. Due to lack of genetic typing, we also could not evaluate the mechanism of resistance among E. coli isolates or the circulating strain types over time. We conducted analyses at the state level, so ecological bias could exist. Although California-specific changes unrelated to SB27 that affected antimicrobial resistance could explain our results, the ASCM does control for temporal changes that affected all states in our analysis equally. This is an advantage of ASCM over alternative methods like difference-in-differences, which makes a parallel trend assumption.53 Our falsification tests, where we randomly assigned SB27 implementation dates, did not yield any significant findings, providing further support for a relationship between SB27 implementation and a reduction in ESC resistance. We also assume that similar programs did not take place in other states at the same time and that SB27 did not impact livestock antibiotic use in other states. Both assumptions seem plausible because a) SB27 was the first-of-its-kind legislation; b) we excluded Maryland and Washington, D.C., from the control pool since Maryland enacted a similar law in 2019; and c) in a sensitivity analysis we excluded Colorado, New Mexico, Oregon, and Washington from the control pool because NARMS reported purchasing California retail chicken meat in those states.
Unlike prior food-producing animal antibiotic policy evaluations,6 none of which included a control group, we used data from 32 states to construct a synthetic California. Our use of the Ridge ACSM was a strength, because it offered several advantages over traditional SCM in this case.26 First, the use of traditional SCM resulted in a less-than-ideal fit of the synthetic California in the prelaw period. Incorporating the outcome ridge regression markedly improved fit by allowing for negative weights while limiting the extent of extrapolation. Second, this approach generates point-wise confidence intervals of effect estimates in the postlaw period36—such inference is not part of the traditional SCM approach. We also used a negative control outcome (fluoroquinolone resistance), fake policy dates, and a sensitivity analysis excluding states that received a portion of California-processed retail chicken meat.
Conclusion
Policies such as SB27 could hold promise in helping to address the global challenge of antimicrobial resistance because antimicrobial use in food-animal production is predicted to increase 11.5% above 2017 levels by 2030 worldwide.54 In California, we estimated a decline in resistance to ESC—a drug class commonly used in food-animal production—concurrent with the implementation of SB27. Future studies, ideally using causal methods (like SCM), are needed to delineate the role of policies like SB27 in curbing antibiotic-resistant infections in people. Antimicrobial use patterns in both human medicine and animal production are essential. To date, California on-farm antimicrobial use data has not been available, hampering our ability to evaluate SB27. Closing these knowledge gaps and evaluating the impact of SB27 from farm to people will provide critical information for evidence-driven policymaking.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors also thank J. Hu at Kaiser Permanente for her assistance assembling the antibiotic prescribing data and B. Steiger at Columbia University Mailman School of Public Health for his assistance in preparing the supplemental figures. Finally, the authors thank the patients of Kaiser Permanente for their partnership with us to improve their health. Their information, collected through our electronic health record systems, leads to findings that help us improve care for our members and that can be shared with the larger community.
This work was supported by the National Institute of Allergy and Infectious Diseases at the National Institutes of Health (R01AI130066; S.Y.T.), National Institute of Environmental Health Sciences (P30ES009089; J.A.C.), the NIH Office of the Director (K01OD019918; M.F.D.), the Wellcome Trust (L.P. and C.L.); a grant from the Johns Hopkins Berman Institute for Bioethics (M.F.D.), and internal funding from KPSC Department of Research & Evaluation.
==== Refs
References
1. U.S. Centers for Disease Control and Prevention. 2019. Antibiotic Resistance Threats in the United States. https://www.cdc.gov/drugresistance/pdf/threats-report/2019-ar-threats-report-508.pdf [accessed 27 December 2021].
2. World Health Organization. 2019. Monitoring and evaluation of the global action plan on antimicrobial resistance. https://www.who.int/publications/i/item/monitoring-and-evaluation-of-the-global-action-plan-on-antimicrobial-resistance [accessed 27 December 2021].
3. Silbergeld EK, Graham J, Price LB. 2008. Industrial food animal production, antimicrobial resistance, and human health. Annu Rev Public Health 29 (1 ):151–169, PMID: , 10.1146/annurev.publhealth.29.020907.090904.18348709
4. U.S. FDA (U.S. Food and Drug Administration). 2021. Summary Report On Antimicrobials Sold or Distributed for Use in Food-Producing Animals, 2020. https://www.fda.gov/media/154820/download?utm_medium=email&utm_source=govdelivery [accessed 27 December 2021].
5. National Resources Defense Council. 2020. New Data: Animal vs. Human Antibiotic Use Remains Lopsided. https://www.nrdc.org/experts/david-wallinga-md/most-human-antibiotics-still-going-us-meat-production [accessed 27 December 2021].
6. Tang KL, Caffrey NP, Nóbrega DB, Cork SC, Ronksley PE, Barkema HW, et al. 2017. Restricting the use of antibiotics in food-producing animals and its associations with antibiotic resistance in food-producing animals and human beings: a systematic review and meta-analysis. Lancet Planet Health 1 (8 ):e316–e327, PMID: , 10.1016/S2542-5196(17)30141-9.29387833
7. Sun J, Liao X-P, D’Souza AW, Boolchandani M, Li S-H, Cheng K, et al. 2020. Environmental remodeling of human gut microbiota and antibiotic resistome in livestock farms. Nat Commun 11 (1 ):1427, PMID: , 10.1038/s41467-020-15222-y.32188862
8. White A, Hughes JM. 2019. Critical importance of a one health approach to antimicrobial resistance. Ecohealth 16 (3 ):404–409, PMID: , 10.1007/s10393-019-01415-5.31250160
9. Innes GK, Nachman KE, Abraham AG, Casey JA, Patton AN, Price LB, et al. 2021. Contamination of retail meat samples with multidrug-resistant organisms in relation to organic and conventional production and processing: a cross-sectional analysis of data from the United States National Antimicrobial Resistance Monitoring System, 2012–2017. Environ Health Perspect 129 (5 ):57004, PMID: , 10.1289/EHP7327.33978452
10. Davis GS, Waits K, Nordstrom L, Weaver B, Aziz M, Gauld L, et al. 2015. Intermingled Klebsiella pneumoniae populations between retail meats and human urinary tract infections. Clin Infect Dis 61 (6 ):892–899, PMID: , 10.1093/cid/civ428.26206847
11. Larsen J, Stegger M, Andersen PS, Petersen A, Larsen AR, Westh H, et al. 2016. Evidence for human adaptation and foodborne transmission of livestock-associated methicillin-resistant Staphylococcus aureus. Clin Infect Dis 63 (10 ):1349–1352, PMID: , 10.1093/cid/ciw532.27655995
12. Yamaji R, Friedman CR, Rubin J, Suh J, Thys E, McDermott P, et al. 2018. A population-based surveillance study of shared genotypes of Escherichia coli isolates from retail meat and suspected cases of urinary tract infections. mSphere 3 (4 ):e00179–e00218, PMID: , 10.1128/mSphere.00179-18.30111626
13. Wallinga D, Smit LAM, Davis MF, Casey JA, Nachman KE. 2022. A review of the effectiveness of current US policies on antimicrobial use in meat and poultry production. Curr Environ Health Rep 9 (2 ):339–354, PMID: , 10.1007/s40572-022-00351-x.35477845
14. California Department of Food and Agriculture. 2017. Antibiotic/antimicrobial livestock drugs in California. https://www.cdfa.ca.gov/is/ffldrs/pdfs/RLDvsMIAD.PDF [a ccessed 3 January 2022].
15. California Department of Food and Agriculture. 2022. California Agricultural Production Statistics. https://www.cdfa.ca.gov/Statistics/ [accessed 24 August 2022].
16. U. S. Department of Agriculture. 2019. 2017 Census of Agriculture. https://www.nass.usda.gov/Publications/AgCensus/2017/Full_Report/Volume_1,_Chapter_1_US/usv1.pdf [accessed 27 December 2021].
17. Sørum M, Johnsen PJ, Aasnes B, Rosvoll T, Kruse H, Sundsfjord A, et al. 2006. Prevalence, persistence, and molecular characterization of glycopeptide-resistant enterococci in Norwegian poultry and poultry farmers 3 to 8 years after the ban on avoparcin. Appl Environ Microbiol 72 (1 ):516–521, PMID: , 10.1128/AEM.72.1.516-521.2006.16391086
18. van den Bogaard AE, Bruinsma N, Stobberingh EE. 2000. The effect of banning avoparcin on VRE carriage in The Netherlands. J Antimicrob Chemother 46 (1 ):146–148, PMID: , 10.1093/jac/46.1.146.10882707
19. Skjøt-Rasmussen L, Ethelberg S, Emborg H-D, Agersø Y, Larsen LS, Nordentoft S, et al. 2009. Trends in occurrence of antimicrobial resistance in Campylobacter jejuni isolates from broiler chickens, broiler chicken meat, and human domestically acquired cases and travel associated cases in Denmark. Int J Food Microbiol 131 (2–3 ):277–279, PMID: , 10.1016/j.ijfoodmicro.2009.03.006.19345436
20. Klare I, Badstubner D, Konstabel C, Bohme G, Claus H, Witte W. 1999. Decreased incidence of VanA-type vancomycin-resistant enterococci isolated from poultry meat and from fecal samples of humans in the community after discontinuation of avoparcin usage in animal husbandry. Microb Drug Resist 5 (1 ):45–52, PMID: , 10.1089/mdr.1999.5.45.10332721
21. Dutil L, Irwin R, Finley R, Ng LK, Avery B, Boerlin P, et al. 2010. Ceftiofur resistance in Salmonella enterica serovar Heidelberg from chicken meat and humans, Canada. Emerg Infect Dis 16 (1 ):48–54, PMID: , 10.3201/eid1601.090729.20031042
22. Wang Y, Xu C, Zhang R, Chen Y, Shen Y, Hu F, et al. 2020. Changes in colistin resistance and mcr-1 abundance in Escherichia coli of animal and human origins following the ban of colistin-positive additives in China: an epidemiological comparative study. Lancet Infect Dis 20 (10 ):1161–1171, PMID: , 10.1016/S1473-3099(20)30149-3.32505232
23. Nordstrom L, Liu CM, Price LB. 2013. Foodborne urinary tract infections: a new paradigm for antimicrobial-resistant foodborne illness. Front Microbiol 4 :29, PMID: , 10.3389/fmicb.2013.00029.23508293
24. Nobrega DB, Tang KL, Caffrey NP, De Buck J, Cork SC, Ronksley PE, et al. 2021. Prevalence of antimicrobial resistance genes and its association with restricted antimicrobial use in food-producing animals: a systematic review and meta-analysis. J Antimicrob Chemother 76 (3 ):561–575, PMID: , 10.1093/jac/dkaa443.33146719
25. Abadie A, Diamond A, Hainmueller J. 2010. Synthetic control methods for comparative case studies: estimating the effect of California’s tobacco control program. J Am Stat Assoc 105 (490 ):493–505, 10.1198/jasa.2009.ap08746.
26. Ben-Michael E, Feller A, Rothstein J. 2021. The augmented synthetic control method. J Am Stat Assoc 116 (536 ):1789–1803, 10.1080/01621459.2021.1929245.
27. Flores-Mireles AL, Walker JN, Caparon M, Hultgren SJ. 2015. Urinary tract infections: epidemiology, mechanisms of infection and treatment options. Nat Rev Microbiol 13 (5 ):269–284, PMID: , 10.1038/nrmicro3432.25853778
28. Abadie A, Diamond A, Hainmueller J. 2015. Comparative politics and the synthetic control method. Am J Pol Sci 59 (2 ):495–510, 10.1111/ajps.12116.
29. Healthcare Cost and Utilization Project. 2008. HCUP NIS Description of Data Elements. https://www.hcup-us.ahrq.gov/db/vars/hosp_bedsize/nisnote.jsp [accessed 12 January 2023].
30. Kaye KS, Gupta V, Mulgirigama A, Joshi AV, Scangarella-Oman NE, Yu K, et al. 2021. Antimicrobial resistance trends in urine Escherichia coli isolates from adult and adolescent females in the United States from 2011 to 2019: rising ESBL strains and impact on patient management. Clin Infect Dis 73 (11 ):1992–1999, PMID: , 10.1093/cid/ciab560.34143881
31. U.S. FDA. 2001. Animal drugs, feeds, and related products; sarafloxacin for poultry; withdrawal of approval of NADAs (final rule). 21 CFR § 520, 522, 556. Fed Reg 66 (83 ):21282. https://www.govinfo.gov/content/pkg/FR-2001-04-30/html/01-10069.htm [accessed 24 October 2022].
32. Lipsitch M, Tchetgen Tchetgen E, Cohen T. 2010. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology 21 (3 ):383–388, PMID: , 10.1097/EDE.0b013e3181d61eeb.20335814
33. U.S. FDA. 2022. Extralabel Use and Antimicrobials. https://www.fda.gov/animal-veterinary/antimicrobial-resistance/extralabel-use-and-antimicrobials [accessed 26 August 2022].
34. Hallberg J, Wachowski M, Moseley W, Dame K, Meyer J, Wood S. 2006. Efficacy of intramammary infusion of ceftiofur hydrochloride at drying off for treatment and prevention of bovine mastitis during the nonlactating period. Vet Ther 7 (1 ):35–42, PMID: .16598682
35. Tibshirani R. 1996. Regression shrinkage and selection via the lasso. J R Stat Soc Series B Methodological 58 (1 ):267–288, 10.1111/j.2517-6161.1996.tb02080.x.
36. Chernozhukov V, Wüthrich K, Zhu Y. 2021. An exact and robust conformal inference method for counterfactual and synthetic controls. J Am Stat Assoc 116 (536 ):1849–1864, 10.1080/01621459.2021.1920957.
37. Bruxvoort KJ, Bider-Canfield Z, Casey JA, Qian L, Pressman A, Liang AS, et al. 2020. Outpatient urinary tract infections in an era of virtual healthcare: trends from 2008 to 2017. Clin Infect Dis 71 (1 ):100–108, PMID: , 10.1093/cid/ciz764.31437269
38. Murray CJ, Ikuta KS, Sharara F, Swetschinski L, Robles Aguilar G, Gray A, et al. 2022. Global burden of bacterial antimicrobial resistance in 2019: a systematic analysis. Lancet 399 (10325 ):629–655, 10.1016/S0140-6736(21)02724-0.35065702
39. Gupta K, Hooton TM, Naber KG, Wullt B, Colgan R, Miller LG, et al. 2011. International clinical practice guidelines for the treatment of acute uncomplicated cystitis and pyelonephritis in women: a 2010 update by the Infectious Diseases Society of America and the European Society for Microbiology and Infectious Diseases. Clin Infect Dis 52 (5 ):e103–e120, PMID: , 10.1093/cid/ciq257.21292654
40. Kabbani S, Palms D, Bartoces M, Stone N, Hicks LA. 2018. Outpatient antibiotic prescribing for older adults in the United States: 2011 to 2014. J Am Geriatr Soc 66 (10 ):1998–2002, PMID: , 10.1111/jgs.15518.30221746
41. Kobayashi M, Shapiro DJ, Hersh AL, Sanchez GV, Hicks LA. 2016. Outpatient antibiotic prescribing practices for uncomplicated urinary tract infection in women in the United States, 2002–2011. Open Forum Infect Dis 3 (3 ):ofw159, PMID: , 10.1093/ofid/ofw159.27704014
42. Durkin MJ, Keller M, Butler AM, Kwon JH, Dubberke ER, Miller AC, et al. 2018. An assessment of inappropriate antibiotic use and guideline adherence for uncomplicated urinary tract infections. Open Forum Infect Dis 5 (9 ):ofy198, PMID: , 10.1093/ofid/ofy198.30191156
43. Anesi JA, Lautenbach E, Nachamkin I, Garrigan C, Bilker WB, Omorogbe J, et al. 2018. Poor clinical outcomes associated with community-onset urinary tract infections due to extended-spectrum cephalosporin-resistant enterobacteriaceae. Infect Control Hosp Epidemiol 39 (12 ):1431–1435, PMID: , 10.1017/ice.2018.254.30375298
44. Anesi JA, Lautenbach E, Nachamkin I, Garrigan C, Bilker WB, Omorogbe J, et al. 2019. The role of extended-spectrum cephalosporin-resistance in recurrent community-onset enterobacteriaceae urinary tract infections: a retrospective cohort study. BMC Infect Dis 19 (1 ):163, PMID: , 10.1186/s12879-019-3804-y.30764770
45. Stapleton AE, Wagenlehner FME, Mulgirigama A, Twynholm M. 2020. Escherichia coli resistance to fluoroquinolones in community-acquired uncomplicated urinary tract infection in women: a systematic review. Antimicrob Agents Chemother 64 (10 ):e00862–e00920, PMID: , 10.1128/AAC.00862-20.32747356
46. Abdelfattah EM, Ekong PS, Okello E, Williams DR, Karle BM, Rowe JD, et al. 2021. 2019 Survey of antimicrobial drug use and stewardship practices in adult cows on California dairies: post Senate bill 27. Microorganisms 9 (7 ):1507, PMID: , 10.3390/microorganisms9071507.34361940
47. Price LB, Rogers L, Lo K. 2022. Policy reforms for antibiotic use claims in livestock. Science 376 (6589 ):130–132, PMID: , 10.1126/science.abj1823.35389795
48. California Department of Food and Agriculture. 2020. Guidelines for Veterinarians: Judicious Use of Antimicrobials in Livestock. https://www.cdfa.ca.gov/AHFSS/aus/docs/Guidelines_Veterinarians_Judicious_Use_of_Antimicrobials_Livestock.pdf [accessed 3 January 2022].
49. Carr VR, Witherden EA, Lee S, Shoaie S, Mullany P, Proctor GB, et al. 2020. Abundance and diversity of resistomes differ between healthy human oral cavities and gut. Nat Commun 11 (1 ):693, PMID: , 10.1038/s41467-020-14422-w.32019923
50. Ekong PS, Abdelfattah EM, Okello E, Williams DR, Lehenbauer TW, Karle BM, et al. 2021. 2018 Survey of factors associated with antimicrobial drug use and stewardship practices in adult cows on conventional California dairies: immediate post-Senate bill 27 impact. PeerJ 9 :e11596, PMID: , 10.7717/peerj.11596.34306825
51. Schreckenberger PC, Binnicker MJ. 2011. Optimizing antimicrobial susceptibility test reporting. J Clin Microbiol 49 (9 suppl ):S15–S19, 10.1128/JCM.00712-11.
52. Puzniak L, DePestel DD, Srinivasan A, Ye G, Murray J, Merchant S, et al. 2019. A combination antibiogram evaluation for Pseudomonas aeruginosa in respiratory and blood sources from intensive care unit (ICU) and non-ICU settings in U.S. hospitals. Antimicrob Agents Chemother 63 (4 ):e02564–e02618, PMID: , 10.1128/AAC.02564-18.30917987
53. Kreif N, Grieve R, Hangartner D, Turner AJ, Nikolova S, Sutton M. 2016. Examination of the synthetic control method for evaluating health policies with multiple treated units. Health Econ 25 (12 ):1514–1528, PMID: , 10.1002/hec.3258.26443693
54. Tiseo K, Huber L, Gilbert M, Robinson TP, Van Boeckel TP. 2020. Global trends in antimicrobial use in food animals from 2017 to 2030. Antibiotics 9 (12 ):918, PMID: , 10.3390/antibiotics9120918.33348801
|
PMC009xxxxxx/PMC9945578.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36821578
EHP11372
10.1289/EHP11372
Research
Metabolic Signatures of Youth Exposure to Mixtures of Per- and Polyfluoroalkyl Substances: A Multi-Cohort Study
https://orcid.org/0000-0001-6615-0472
Goodrich Jesse A. 1
Walker Douglas I. 2
He Jingxuan 1
Lin Xiangping 3
Baumert Brittney O. 1
Hu Xin 4
Alderete Tanya L. 5
Chen Zhanghua 1
Valvi Damaskini 3
Fuentes Zoe C. 3
Rock Sarah 1
Wang Hongxu 1
Berhane Kiros 6
Gilliland Frank D. 1
Goran Michael I. 7
Jones Dean P. 4
Conti David V. 1
Chatzi Leda 1
1 Department of Population and Public Health Sciences, Keck School of Medicine, University of Southern California, Los Angeles, California, USA
2 Gangarosa Department of Environmental Health, Rollins School of Public Health, Emory University, Atlanta, Georgia, USA
3 Department of Environmental Medicine and Public Health, Icahn School of Medicine at Mount Sinai, New York, New York, USA
4 Division of Pulmonary, Allergy, Critical Care and Sleep Medicine, Emory University, Atlanta, Georgia, USA
5 Department of Integrative Physiology, University of Colorado Boulder, Boulder, Colorado, USA
6 Department of Biostatistics, Columbia University, New York, New York, USA
7 Department of Pediatrics, Children’s Hospital Los Angeles, Saban Research Institute, Los Angeles, California, USA
Address correspondence to Jesse A. Goodrich, Southern California Environmental Health Sciences Center, Department of Population and Public Health Sciences, University of Southern California, 2001 N. Soto St., Los Angeles, CA 90032 USA. Email: [email protected]
22 2 2023
2 2023
131 2 02700506 4 2022
12 12 2022
09 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Exposure to per- and polyfluoroalkyl substances (PFAS) is ubiquitous and has been associated with an increased risk of several cardiometabolic diseases. However, the metabolic pathways linking PFAS exposure and human disease are unclear.
Objective:
We examined associations of PFAS mixtures with alterations in metabolic pathways in independent cohorts of adolescents and young adults.
Methods:
Three hundred twelve overweight/obese adolescents from the Study of Latino Adolescents at Risk (SOLAR) and 137 young adults from the Southern California Children’s Health Study (CHS) were included in the analysis. Plasma PFAS and the metabolome were determined using liquid-chromatography/high-resolution mass spectrometry. A metabolome-wide association study was performed on log-transformed metabolites using Bayesian regression with a g-prior specification and g-computation for modeling exposure mixtures to estimate the impact of exposure to a mixture of six ubiquitous PFAS (PFOS, PFHxS, PFHpS, PFOA, PFNA, and PFDA). Pathway enrichment analysis was performed using Mummichog and Gene Set Enrichment Analysis. Significance across cohorts was determined using weighted Z-tests.
Results:
In the SOLAR and CHS cohorts, PFAS exposure was associated with alterations in tyrosine metabolism (meta-analysis p=0.00002) and de novo fatty acid biosynthesis (p=0.03), among others. For example, when increasing all PFAS in the mixture from low (∼30th percentile) to high (∼70th percentile), thyroxine (T4), a thyroid hormone related to tyrosine metabolism, increased by 0.72 standard deviations (SDs; equivalent to a standardized mean difference) in the SOLAR cohort (95% Bayesian credible interval (BCI): 0.00, 1.20) and 1.60 SD in the CHS cohort (95% BCI: 0.39, 2.80). Similarly, when going from low to high PFAS exposure, arachidonic acid increased by 0.81 SD in the SOLAR cohort (95% BCI: 0.37, 1.30) and 0.67 SD in the CHS cohort (95% BCI: 0.00, 1.50). In general, no individual PFAS appeared to drive the observed associations.
Discussion:
Exposure to PFAS is associated with alterations in amino acid metabolism and lipid metabolism in adolescents and young adults. https://doi.org/10.1289/EHP11372
Supplemental Material is available online (https://doi.org/10.1289/EHP11372).
The authors have no conflicts of interest to report.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Per- and polyfluoroalkyl substances (PFAS) make up a prevalent class of persistent organic pollutants with endocrine and metabolism disrupting properties.1,2 PFAS are used in a broad range of industrial and consumer products, such as firefighting foams, nonstick pans, waterproof clothing, food packaging, and even cosmetic products, including lipstick.3,4 Because of their widespread use and resistance to chemical degradation, PFAS can be found in drinking water and food sources across the world. In the United States, an estimated 200 million people have drinking water with perfluorooctane sulfonic acid (PFOS) or perfluorooctanoic acid (PFOA) levels >1 ng/L,5 which is considerably higher than the U.S. Environmental Protection Agency 2022 safe drinking water health advisory levels of 4×10−6ng/L for PFOA and 2×10−6ng/L for PFOS.6,7 Because of their chemical properties, some PFAS also bioaccumulate in human tissues, with estimated half-lives of up to 25 y.8 Consequently, several legacy PFAS, including PFOS, PFOA, perfluorohexanesulfonic acid (PFHxS), perfluoroheptanesulfonic acid (PFHpS), perfluorononanoic acid (PFNA), and perfluorodecanoic acid (PFDA) are detectible in the blood of nearly all humans.9,10 Therefore, it is of utmost importance to determine how exposure to PFAS impacts human health to inform public health policy and reduce exposure levels for harmful PFAS.
Accumulating evidence suggests that PFAS exposure is associated with an increased risk of metabolic disorders.11 PFAS exposure during sensitive periods of development, such as childhood or adolescence, is of particular concern because this is an important developmental stage for cellular differentiation and development of metabolic tissues.12–15 Longitudinal studies have found that PFAS exposure during childhood is associated with the development of dysregulated glucose metabolism and insulin resistance,16,17 dyslipidemia,18 and adiposity.19,20 Similar associations have been observed in adults, suggesting that these associations persist into adulthood.21–23 However, the mechanisms linking PFAS exposure and metabolic disorders in humans remain unclear, especially in children.
Studies examining the associations of PFAS with alterations in targeted biomarkers have been used to examine the potential mechanisms linking PFAS with metabolic disorders. These targeted biomarker studies have focused primarily on metabolites known to be associated with a specific metabolic disease, with a particular focus on metabolites related to lipid metabolism. For example, cross-sectional and longitudinal human studies have consistently shown that PFAS exposure increases serum total and low-density lipoprotein cholesterol in both children and adults.11 Dyslipidemia is a hallmark of metabolic syndrome and can increase the risk for metabolic and cardiovascular disease,24 suggesting that PFAS-associated alterations in lipids may mediate the relationship between PFAS exposure and risk of metabolic disorders. However, using a targeted approach to study the impact of PFAS exposure on alterations in metabolites does not provide information about how PFAS alter metabolic networks, which is important for understanding the mechanisms linking PFAS exposure with metabolic disorders.
Untargeted metabolomics approaches can quantify thousands of metabolites, which can provide insight on the metabolic perturbations of PFAS exposure. Although previous studies have used metabolomics to examine the associations of PFAS exposure and alterations in metabolic pathways, important gaps in the literature remain.25 In vivo and in vitro models have shown that exposure to individual PFAS alters fatty acid metabolism, including in human cell lines,26 rodents,27–29 and zebrafish.30 In humans, studies using both targeted and untargeted metabolomics methods have reported similar associations between exposure to individual PFAS, including PFOS, PFHxS, and PFOA, and fatty acid metabolism.31–40 Other studies have reported associations between PFAS exposure and metabolic pathways such as amino acid31,34,37,40,41 or bile acid metabolism,39,42,43 both of which are important pathways that can contribute to the pathogenesis of diseases, such as type 2 diabetes and cancer.44–47 Together, these human studies present a myriad of sometimes conflicting results. Although differences in study populations and the use of targeted vs. untargeted metabolomic methods may account for some of the differences across studies, the majority of existing studies have examined the metabolic perturbations of PFAS congeners individually, using single exposure models. In reality, individuals are exposed to a complex mixture of potentially correlated PFAS compounds that may have synergistic effects on human metabolism.48 Previous studies examining the impact of PFAS mixtures on the metabolome have relied primarily on reducing the dimensionality of the correlated PFAS exposures prior to analysis using either principal components or by calculating the sum of PFAS congeners.25 However, it is difficult to draw conclusions about the effects of different components of the PFAS mixture from these studies owing to difficulties in interpreting principal components or summed exposure variables. Despite the analytical difficulties, examining the effects of PFAS mixtures on metabolic pathways is key to fully understand the consequences of exposure to these chemicals. Further, understanding how PFAS mixtures impact human metabolism can help to put previous studies on individual PFAS congeners in greater context.
In the present study, we aimed to examine the impact of PFAS mixtures on metabolic pathways in independent cohorts of children and young adults. By using two cohorts, we aimed to identify associations between PFAS mixtures and metabolic pathways that were consistent across cohorts despite varied background characteristics. We used high-resolution mass spectrometry (HRMS)–based untargeted metabolomics coupled with a Bayesian hierarchical regression with a g-prior specification and g-computation for modeling exposure mixtures to estimate the joint effects of PFAS mixtures on metabolic pathways.
Methods
Study Populations
Study of Latino Adolescents at Risk.
This study used data from the Study of Latino Adolescents at Risk (SOLAR). As described previously,16,49–51 the SOLAR cohort included 328 overweight/obese children recruited in two waves between 2001 and 2012. Participants completed yearly clinical visits at the General Clinical Research Center of the Clinical Trials Unit at the University of Southern California. At baseline, participants were between 8 and 13 years of age and were overweight or obese based on sex and age-specific body mass index (BMI) percentile >85%. Participants were included in the study if they were Hispanic or Latino based on all parents and grandparents self-reporting as Hispanic or Latino and if they had a direct family history of type 2 diabetes. Participants were excluded from the study if they had type 1 or type 2 diabetes or if they were on medications known to influence glucose or insulin metabolism. For the present analysis, participants were included if they completed a 2-h oral glucose tolerance test (OGTT) during their first or second visit, resulting in a total sample size of 312 participants. The institutional review board at the University of Southern California provided ethics approval for this study, and participants and their guardians provided written informed assent/consent prior to participation.
Southern California Children’s Health Study.
To examine the generalizability of findings from the SOLAR cohort to young adults, we analyzed data from the Metabolic and Asthma Incidence Research (Meta-AIR) study.52 Meta-AIR was a study including 172 young adults between the ages of 17 and 23 years of age who were part of the Southern California Children’s Health Study (CHS).53 Meta-AIR participants were recruited for a single clinical visit between 2014 and 2018. Clinical visits occurred at the Clinical Trials Unit or the Diabetes and Obesity Research institute at the University of Southern California, during which participants performed a 2-h OGTT and completed detailed questionnaires. CHS participants were selected for inclusion in the Meta-AIR study if they had overweight or obesity between 14 and 15 years of age. Overweight and obesity were based on the U.S. Centers for Disease Control and the U.S. Preventive Services Task Force guidelines, defined as having a sex- and age-specific BMI percentile >85%.54 Participants were excluded if they had a diagnosis of diabetes mellitus or if they took any medications that could influence glucose metabolism or insulin secretion. Meta-AIR participants were included in the present study if they provided consent for future use of biospecimens, resulting in a total sample size of 137. The University of Southern California institutional review board provided ethics approval for this study, and participants (and their guardians if participants were <18 years of age) provided written informed assent/consent prior to participation.
Covariates
Detailed information on the measurement of covariates in the SOLAR and CHS cohorts has been provided previously.16,49–53 Height (in meters) and weight (in kilograms) were measured at each visit in both cohorts and were used to calculate BMI as kilograms per meter squared. In both cohorts, participants completed questionnaires related to sociodemographics and individual and familial health history. In the SOLAR cohort, socioeconomic status (SES) was characterized using a modified version of the Hollingshead Four-Factor Index,55 as described previously.49 The Hollingshead Four-Factor Index was modified to provide a single score for children and takes into account the education, occupation, and marital status of parents or primary caregivers.49 In the CHS cohort, parental education was used to assess SES.52 For the primary analysis, SES scores were grouped into quantiles and analyzed numerically, with values ranging from 1 to 4, with 1 representing the lowest and 4 representing the highest quantile of SES. In the SOLAR cohort, SES was unavailable for 11% (n=35) of individuals. In these participants, we imputed the population median SES score; median imputation has been shown to perform similarly to more complex imputation methods when the proportion of missing values is relatively low.56 Further, as described previously, participants with missing SES data did not differ in metabolic or physical attributes, including sex (tested using a χ2 test; p=0.35), age (tested using an independent t-test; p=0.88), or BMI (tested using an independent t-test; p=0.94).16,51
In the SOLAR cohort, Tanner stages were determined by a physician during a physical exam to assess sexual maturity.57,58 Tanner stages range from 1 to 5 and were categorized as follows: stage 1: prepuberty, which is the developmental stage before secondary sex characteristics begin to develop; stages 2–4: puberty, which is the developmental stage when secondary sex characteristics begin developing and menarche occurs for most females; stage 5: postpuberty, the developmental stage where secondary sex characteristics, including pubic hair and secondary sex organs, reach maturity.
Covariates included in models were selected using directed acyclic graphs (DAGs; Figure S1) and included age (in years), sex (male/female, coded numerically as 0/1), BMI (in kilograms per meter squared), and SES (in quantiles, coded numerically as 1–4). In the SOLAR cohort, Tanner stage (Tanner stages 1–5, coded numerically as 1–5) and study wave (Wave 1/Wave 2, coded numerically as 0/1) were also included as covariates. No covariates for statistical models contained missing data, except SES in the SOLAR cohort (as described above).
Plasma PFAS
Six common PFAS (PFOS, PFHxS, PFHpS, PFOA, PFNA, and PFDA) were quantified in plasma samples collected at the 2-h OGTT time point. PFAS levels were quantified in batches of 70 study samples via liquid chromatography (LC) coupled to HRMS (LC-HRMS). The specific LC-HRMS protocol used to measure plasmas PFAS for this study has been described previously.59 Briefly, plasma samples were prepared by combining 40μL of plasma with 2.5μL of an internal standard. The internal standard contained thirty C13-labeled PFAS to obtain a final 5 ng/mL concentration. Proteins were then precipitated by adding 80μL of cold acetonitrile, then vortexing and centrifuging at 18,000×g for 15 min. LC-MS grade water was added to the resulting supernatant to achieve a dilution ratio of 2:1. PFAS analysis was completed by reverse phase (RP) chromatography with negative electrospray ionization, performed with a Thermo Scientific Vanquish Flex ultra-high-performance LC system with binary pump attached to a Thermo Scientific Q Exactive HF-X Orbitrap mass spectrometry system (Thermo Fisher Scientific).
After analysis of samples from both the SOLAR and CHS cohorts, peaks for PFAS and their corresponding C13-labeled PFAS were identified by matching mass m/z with a tolerance of 5 ppm, then extracted and integrated using TraceFinder 5.1 (Thermo Fisher Scientific). Quantification of PFAS was performed by dividing the peak for a given PFAS with its corresponding internal standard and comparing this value to a calibration curve with 6 points created using charcoal-stripped plasma. Replicate measures of National Institute of Standards and Technology (NIST) 1957 and NIST 1958 standard reference material were performed for every batch of 70 samples. Each batch also included instrumental and method blanks. The coefficient of variation for major PFAS was <15%, whereas the analyte recovery was >90%. The accuracy of the method was assessed by comparing with NIST standard reference materials. Method accuracy was also assessed by participating in the Center de toxicologie du Québec’s Arctic Monitoring and Assessment Program Ring Test for Persistent Organic Pollutants in Human Serum. The limits of detection (LODs) were as follows: PFOS: 0.43μg/L; PFHxS: 0.01μg/L; PFHpS: 0.05μg/L; PFOA: 0.01μg/L; PFNA: 0.01μg/L; and PFDA: 0.01μg/L. PFDA was the only PFAS included in the present study that had values below the LOD; these values were imputed as LOD divided by the square root of 2.
To contextualize exposure levels in the SOLAR and CHS cohorts, we compared the geometric mean and 95% confidence interval of PFAS concentrations to appropriate period- and age-matched PFAS concentrations from the National Health and Nutrition Examination Survey (NHANES).60 For the SOLAR cohort, PFAS concentrations were compared with those in young persons 12–19 years of age from the NHANES survey years 2007–2008, given that this was near the middle of the study period for this cohort. For the CHS cohort, PFAS concentrations were compared with those in young persons 12–19 years of age from the NHANES survey years 2017–2018, which overlapped in time with the CHS cohort and was the only survey year in which PFHpS concentrations were reported.
Untargeted Plasma Metabolomics
Untargeted plasma metabolomics were measured using samples collected at the 2-h OGTT time point. Previous studies have shown that metabolite concentrations following a glucose challenge are potentially more informative when examining alterations in metabolic pathways because the glucose challenge is a physiological stressor that requires metabolic flexibility.61,62 Measuring untargeted metabolomics was performed using established methods,63 as described previously.59 Briefly, LC-HRMS was performed with RP and hydrophilic interaction LC (HILIC) by employing a dual polarity/dual column approach. Analyses were performed with a Thermo Fisher Vanquish Duo LC system with dual pumps and columns with independent flow paths connected to a Thermo Fisher Q Exactive HF-X Orbitrap MS system. Samples were analyzed in RP and HILIC separately to allow for the analysis of all four LC-HRMS modes.
Before analysis, samples for RP were prepared using the methods described above in the section “Plasma PFAS”.59 Samples for HILIC were prepared using the methods described above, with minor alterations in the volumes as described previously. Specifically, for HILIC mode, 90μL of acetonitrile was added to 30μL of plasma before processing was continued as described above. For HILIC mode, 30μL of the processed supernatant was added to 90μL 1:1 (vol/vol) water/acetonitrile. As described previously, samples were analyzed with both positive and negative ionization by optimizing mobile phases.59 Separation for RP was performed using C18 (TARGA C18 5μm 50×2.1mm; Higgins Analytical), whereas separation for HILIC was performed using a SeQuant ZIC-HILIC column (3.5μm, 200A 4.6×50mm; MilliporeSigma) for positive mode, and an Amide ethylene bridged hybrid (BEH) HILIC column (3.5μm, 130A 3×50mm; Waters) in negative mode. Measurement of mass spectral data was performed for a mass-to-charge ratio (m/z) scan range of 85 to 1,275, and the resolution was 120,000 full width at half maximum (FWHM). Additional details on this method are provided by Goodrich et al.59
Following analysis of all samples from the SOLAR and CHS cohorts, mass spectral peaks for metabolites were extracted with apLCMS and xMSanalyzer.63,64 Extraction was performed across all study samples concurrently and was performed separately for each of the four LC-HRMS modes. After extracting LC-MS features, adjustment for inter- and intra-batch variation was performed using a random forest signal correction algorithm based on quality control samples run in tandem with study samples.65 As part of the random forest signal correction algorithm, features that were not detected in >25% of samples were removed from further analysis. Features were also removed from further analysis if the coefficient of variability in all quality control samples post-correction was >30%. After LC-MS data processing, the total number of features included in data analysis was 23,166, including 3,711 features from the C18-negative mode, 5,069 features from the C18-positive mode, 7,442 features from the HILIC-negative mode, and 6,944 features from the HILIC-positive mode.
Statistical Analysis
Differences in PFAS levels and participant characteristics between cohorts were examined using independent t-tests (for natural log-transformed PFAS concentrations and continuous participant characteristics) and chi-square tests (for categorical participant characteristics). Spearman correlation coefficients were calculated to examine associations between PFAS levels within each cohort. All analyses were performed with R (version 4.1.2; R Development Core Team).
Metabolome-Wide Association Study
To examine associations between PFAS mixtures and each metabolite, we performed a metabolome-wide association study (MWAS) using a Bayesian hierarchical regression modeling approach with g-computation (BHRM-g). We implemented a Bayesian g-computation approach66,67 to obtain both individual PFAS-specific estimates conditional on all other PFAS in the model and a single mixture effect estimate for the overall PFAS mixture. The approach is similar to quantile g-computation.68 Specifically, BHRM-g combines: a) a g-prior specification for the corresponding exposure effects to provide robust estimation of highly correlated exposures,69 b) a Bayesian stochastic selection procedure to estimate the posterior inclusion probability (PIP) of each PFAS in the PFAS mixtures,67 and c) Bayesian g-computation in a potential outcome framework for estimating the overall mixture effect based on two hypothetical exposure profiles (explained in further detail below).67 In contrast to Bayesian Kernel Machine Regression, which uses Gaussian process regression and models a nonlinear dose–response relationship, BHRM-g estimates a mixture effect from the additive terms from a regression model with each exposure and forces explicit specification of nonlinear effects.70 We modeled the dose–response with a linear function that allows for the calculation of a single monotonic effect estimate that is not dependent on the baseline exposure profile; this aids in interpretation and allows for additional downstream analysis. In this analysis, we independently fit the following model for each metabolite Y: Yi=αp+∑PγpβpXp+∑QδqUq+ϵi,
where Xp is a scaled variable for PFAS exposure p with corresponding estimate βp; γp is a binary variable indicating the inclusion of PFAS exposure p in the mixture; and Uq is a set of q covariates with corresponding effect estimates δq.
To obtain robust estimates in the presence of highly correlated exposures, we included a second-stage g-prior on the effect estimates of the form β∼NP(0,gσY2(X′X)−1).71–74 Here, σY2 is the variance of the outcome; g is a scalar with a specified hyper-g prior72,75,76; and X′X is the covariance matrix.77 In this model, the scalar g controls the shrinkage toward the prior mean of zero and the dispersion of the posterior covariance via a shrinkage factor of g/(1+g). Similar to regularized regression approaches, the resulting posterior estimate can be expressed as a function of this shrinkage β~=[g/(1+g)]β^, where β^ is the maximum likelihood estimate.
To facilitate model selection, we included a binary variable γp={0,1} in the first-stage linear predictor component of the model, indicating the inclusion of each PFAS. We included a beta-binomial prior for model selection such that γp ∼Bernoulli(π).78 Within this framework, the PIP on the individual γp is the posterior probability that the coefficient is nonzero. Inference for the Bayesian model was achieved with Markov chain Monte Carlo techniques using JAGs and the R package rjags.79,80 The model was initialized in an adaptive mode with 4,000 iterations to increase the efficiency, and we used Markov Chain for a burn-in period with 1,000 iterations. To generate the posterior distribution of the parameters, we updated the model with 5,000 iterations.
Finally, we used g-computation to yield a single effect estimate (i.e., a mixture effect) that captures the impact of 1-standard deviation(SD) increase in levels of all exposures simultaneously.66–68 This is performed by estimating the difference in the outcome between two hypothetical exposure profiles based on the distribution of PFAS in the study samples. In this study, the low exposure profile is defined as setting each log-transformed and standardized PFAS at a Z-score of −0.5 (i.e., the 30.8 percentile) and the high exposure profile is defined as setting each log-transformed and standardized PFAS at a Z-score of 0.5 (i.e., the 69.1 percentile). Specifically, we used posterior predictive distributions to estimate a single mixture risk difference (ψRD), such that ψRD=ψx*=0.5−ψx*=−0.5, where ψx*=∑Pγpβpxp* and x* is the counterfactual profile for the log-transformed and standardized exposures with all exposures set to “low” (30.8% levels with an x*=−0.5) and to “high” (69.1% levels with an x*=0.5). The PFAS concentrations for the low and high exposure profiles in the SOLAR and CHS cohorts are provided in Table S1.
Effect estimates from the BHRM-g models are reported as the posterior mean and 95% Bayesian credible intervals (BCIs). Associations between the PFAS mixture and individual metabolites were selected for further analysis based on a 95% BCI not containing zero. Inputs for the pathway enrichment analysis used the PFAS mixture effect (ψRD), dividing the posterior mean by the posterior variance to obtain a Wald test statistic and corresponding p-value. False discovery rate (FDR)–adjusted p-values were calculated with the Benjamini–Hochberg method to account for multiple comparisons.81 Adjusted p-values were calculated within each cohort separately and are represented in the text as q-values. Evidence of a significant association between the PFAS mixture and individual metabolites was defined as a 95% BCI for the PFAS mixture effect not containing zero. q<0.05 was considered as additional evidence of a significant association.
Prior to the MWAS, raw intensity values of LC-MS features were log2 transformed and normalized to a standard normal distribution to satisfy regression assumptions and obtain comparable effect estimates across all metabolites. Because metabolite intensity was log2 transformed and scaled prior to analysis, the effect estimate (ψRD) is also equivalent to a standardized mean difference calculated between a hypothetical group of individuals with all PFAS at the ∼70th percentile vs. a hypothetical group of individuals with all PFAS at the ∼30th percentile.
Finally, for BHRM-g to provide a mixture effect estimate interpretable as the change in the outcome when increasing all PFAS in the mixture by 1 SD, all exposures in the mixture must be scaled to a mean of zero and a standard deviation of one. In small sample sizes, extremely positively skewed exposures can also cause instability, and in our data, several PFAS, including PFHxS and PFNA, were highly positively skewed. Therefore, before analysis, PFAS were log2 transformed and scaled to a mean of zero and standard deviation of one. The R code for performing a single BHRM-g regression is provided in Supplemental Material, “Supplemental Code.” The R code for all analyses performed in this study is available on GitHub at https://github.com/chatzilab/PFAS_metabolomics_EHP_2022.
Metabolite Annotation and Pathway Enrichment
Following the BHRM-g MWAS, we used the PFAS mixture effect estimate to perform a pathway enrichment analysis using the MS peaks to pathways module from MetaboAnalyst (version 5.0).82 We used version 2 of the MS peaks to pathways module, which accounts for retention time for more accurate metabolite annotation.83 For this analysis, we included LC-MS features from both positive and negative ionization. We used a 5.0-ppm mass tolerance, a 0.05 p-value threshold, and the Human reference pathways of the MetaFishNet (MFN) database.84 We used the integrated metabolic pathway enrichment analysis,85 which determines pathway enrichment using an overrepresentation analysis86 and a Gene Set Enrichment Analysis.87 Significance of metabolic pathways across cohorts was determined by combining p-values using a weighted Z-test.88 For the pathway enrichment analysis, statistical significance was based on a p-value threshold of 0.05. Metabolic pathways identified as significantly enriched were included in follow-up analysis if they included at least four significant metabolites in either the SOLAR or CHS cohorts.
Sensitivity Analysis
Due to differences in the SOLAR and CHS cohorts, including differences in developmental stages and exposure profiles, the primary method for combining results across the two cohorts in this study was to perform a meta-analysis on the effect estimates obtained from study-specific regression models with study-specific covariates. This method avoids many of the potential pitfalls of pooled analysis.89 However, to assess the impact of this analytic choice, we conducted a sensitivity analysis by performing the entire analytic workflow using pooled data from the two cohorts. Because of differences in covariates between cohorts, a minimal set of covariates was used for this analysis, including sex, developmental stage (based on Tanner stage), parental education, and study wave/cohort. Given that Tanner stage was not measured in the CHS (young adult) cohort, Tanner stage was imputed as stage 5 for all individuals from this cohort. For the pooled analysis, study wave/cohort included three levels (Wave 1 SOLAR, Wave 2 SOLAR, and CHS) and was included in the model as two numeric dummy variables. For the pooled analysis, in addition to examining metabolites associated with significantly enriched metabolic pathways, we also further examined associations of the PFAS mixtures with individual metabolites associated with any of the significantly enriched metabolic pathways from the individual cohort analysis.
Results
Characteristics of the Study Population
Participant characteristics for the SOLAR and CHS cohorts are presented in Table 1, and plasma PFAS concentrations for the SOLAR and CHS cohorts are provided in Table 2. In both cohorts, PFOS, PFHxS, PFHpS, PFOA, and PFNA were detected in all participants. PFDA was detected in 99.4% of participants from the SOLAR cohort and 98.5% of participants from the CHS cohort. PFAS levels ranged from uncorrelated to strongly positively correlated, with Spearman correlation coefficients between 0.01 and 0.93 in the SOLAR cohort and between 0.10 and 0.93 in the CHS cohort (Figure S2). Concentrations of all PFAS were significantly higher in the SOLAR cohort compared with the CHS cohort (Table 2). In both cohorts, PFAS concentrations were similar to those reported in the appropriate time period and age-matched NHANES (Table S2).60
Table 1 Participant characteristics of adolescents from the Study of Latino Adolescents at Risk (SOLAR) cohort (recruited between 2001 and 2012) and young adults from the Southern California Children’s Health Study (CHS; recruited between 2014 and 2018).
General characteristics SOLAR CHS p-Value
Sample size (n) 312 137 —
Sex [female; n (%)] 133 (40) 61 (40) 0.79
Age [y (mean±SD)] 11.3±1.7 19.4±1.3 —
BMI [kg/m2 (mean±SD)] 28.2±5.8 29.6±4.7 0.0051
Puberty status [n (%)] —
Prepuberty (Tanner stage 1) 99 (32) —
Puberty (Tanner stages 2–4) 193 (62) —
Postpuberty (Tanner stage 5) 20 (6) —
Ethnicity [n (%)] <2×10−16
Hispanic 312 (100) 79 (58)
Non-Hispanic 0 (0) 58 (42)
Study wave [n (%)] —
Wave 1 (2001–2003) 234 (75) —
Wave 2 (2010–2012) 78 (25) —
Socioeconomic status
Modified Hollingshead Four-Factor Index (mean±SD) 17.9±10.1 — —
Household education level [n (%)] 7.6×10−31
Did not graduate high school 146 (47) 25 (18)
High school graduate 89 (29) 21 (15)
Partial college (≥1 y) or specialized training 36 (12) 36 (26)
Completed college/university 0 (0) 37 (27)
Graduate professional training 3 (1) 14 (10)
Missing 34 (11) 4 (3)
Note: p-Values not reported for variables that were only measured in one cohort. —, not applicable; BMI, body mass index; NA, not applicable; SD, standard deviation.
Table 2 PFAS Concentrations (μg/L) in overweight and obese adolescents from the SOLAR cohort and young adults from the CHS cohort.
PFAS subclass PFAS name SOLAR (n=312) CHS (n=137) p-Value
Geometric mean±GSD Arithmetic mean±SD Median (IQR) Geometric mean±GSD Arithmetic mean±SD Median (IQR)
Perfluorosulfonic acids PFOS 11.8±2.2 15.4±9.8 15.1 (14.1) 3.31±1.58 3.67±1.78 3.13 (1.90) 2.1×10−68
PFHxS 1.44±2.00 1.97±2.60 1.35 (1.20) 1.05±2.10 1.43±1.46 0.95 (1.00) 2.2×10−5
PFHpS 0.37±1.74 0.43±0.22 0.42 (0.31) 0.18±1.46 0.19±0.08 0.17 (0.10) 6.2×10−46
Perfluorocarboxylic acids PFOA 3.29±1.75 3.82±2.14 3.45 (2.46) 1.34±1.43 1.42±0.48 1.34 (0.67) 8.6×10−64
PFNA 0.59±1.40 0.63±0.31 0.57 (0.21) 0.48±1.32 0.49±0.14 0.46 (0.15) 1.2×10−11
PFDA 0.23±1.60 0.25±0.12 0.23 (0.11) 0.19±1.81 0.22±0.10 0.20 (0.12) 0.0012
Note: p-Value was calculated for the differences in PFAS concentrations between cohorts using independent t-tests on log-transformed PFAS concentrations. CHS, Children’s Health Study; GSD, geometric standard deviation; IQR, interquartile range; PFAS, per- and polyfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHpS, perfluoroheptanesulfonic acid; PFHxS, perfluorohexanesulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonic acid; SD, Arithmetic standard deviation; SOLAR, Study of Latino Adolescents at Risk.
PFAS Exposure Was Associated with Metabolic Pathways
In the SOLAR and CHS cohorts, we performed a MWAS with all 23,166 untargeted metabolite features to examine their association with a single mixture of all six PFAS compounds. In the SOLAR cohort, the MWAS identified 463 metabolite features associated with the PFAS mixture, defined as having a 95% BCI for the mixture effect not containing zero (Excel Table S1). Functional pathway analysis of the MWAS results identified significant enrichment of 11 metabolic pathways (Figure 1; Table S3). These pathways were primarily related to the metabolism of aromatic amino acids, nonaromatic amino acids, lipids, and cofactors and vitamins.
Figure 1. Metabolic pathways associated with exposure to a mixture of six PFAS in adolescents from the SOLAR cohort (n=312) and young adults from the CHS cohort (n=137). Metabolic pathways are grouped into super pathways as indicated on the right of the plot. Meta-analysis p values are provided for pathways identified as being associated with PFAS in both cohorts. Dot size for the SOLAR and CHS cohorts are proportional to the number of significant metabolites associated with each pathway. Only pathways that were significant in either the SOLAR cohort, the CHS cohort, or the meta-analysis are presented here; for complete results see Table S3. Note: CHS, Children’s Health Study; EPA, eicosapentaenoic acid; PFAS, per- and polyfluoroalkyl substances; Sig, Significant; SOLAR, Study of Latino Adolescents at Risk.
Figure is a dot graph with three columns titled Study of Latino Adolescents at Risk, Children’s Health Study, and Meta-analysis. There are five groups of pathways, named Other, which includes Nitrogen metabolism and Drug metabolism-cytochrome P 450; Metabolism of cofactors and vitamins, including Vitamin B6 (pyridoxine) metabolism and Porphyrin metabolism; Lipid metabolism, including Linoleate metabolism, Fatty acid metabolism, anti-inflammatory metabolism from Eicosapentaenoic acid, Prostaglandin formation from Arachidonate, and De novo fatty acid biosynthesis; Nonaromatic amino acid metabolism, including Lysine metabolism, Arginine and Proline metabolism, Urea cycle or amino group metabolism, and Glutathione metabolism; and Aromatic amino acid metabolism, including Tyrosine metabolism (y-axis) across negative log uppercase p, ranging from 0 to 6 in increments of 2 (x-axis) for analysis, individual cohort analysis, and meta-analysis, respectively.
In the CHS cohort, 200 metabolite features were associated with the PFAS mixture (Excel Table S2). Functional pathway analysis identified the tyrosine metabolism pathway, an aromatic amino acid, as significantly enriched (p=0.02; Figure 1; Table S3). In addition, four nonaromatic amino acid metabolism pathways, two lipid metabolism pathways, and one pathway related to the metabolism of cofactors were identified in the functional pathway enrichment analysis as having at least one metabolite associated with the PFAS mixture, although these pathways did not meet the threshold for statistical significance.
Meta-analysis of the p-values from the pathway enrichment analysis from both cohorts identified seven statistically significant metabolic pathways, including one aromatic amino acid metabolism pathway (tyrosine metabolism), four nonaromatic amino acid metabolism pathways (glutathione, urea cycle/amino group, arginine and proline, and lysine metabolism), one lipid metabolism pathway (de novo fatty acid biosynthesis), and one pathway related to metabolism of cofactors (porphyrin metabolism). In total, PFAS exposure was associated with alterations in 14 unique metabolic pathways across four super pathways in the SOLAR cohort, the CHS cohort, or in the meta-analysis (Figure 1; Table S3).
For the aromatic amino acid metabolism pathway, PFAS exposure was associated with several metabolites linked to important tyrosine metabolism subpathways across both cohorts. In the SOLAR cohort, 18 unique metabolites were linked to aromatic amino acid metabolism pathways, and seven of these associations remained significant after adjusting for multiple comparisons (Figure 2A; Table S4). In the CHS cohort, seven unique metabolites were linked to aromatic amino acid metabolism pathways, and one remained significant after adjusting for multiple comparisons (Figure 2B; Table S4). To gain additional insight about the functional implications of alterations in these metabolites, we grouped metabolites by subpathways within aromatic amino acid metabolism. These subpathways included catecholamine biosynthesis and degradation, tyrosine metabolism and degradation, thyroid hormone biosynthesis, phenylalanine metabolism, and melanin biosynthesis. Three key metabolites were positively associated with PFAS exposure in both cohorts. These included thyroxine (T4), the main thyroid hormone in circulation; l-glutamic acid, an amino acid associated with tyrosine metabolism; and hippuric acid, an acyl glycine associated with phenylalanine metabolism. Two metabolites were negatively associated with PFAS exposure in the SOLAR cohort but positively associated with PFAS exposure in the CHS cohort; these included vanylglycol, a methoxyphenol generated from the degradation of catecholamines, and acetoacetic acid, a product of the catabolism of tyrosine to fumaric acid.
Figure 2. Associations between PFAS mixtures and metabolites associated with aromatic amino acid metabolism in (A) adolescents from the SOLAR cohort (n=312) and (B) young adults from the CHS cohort (n=137). Metabolites are grouped by tyrosine metabolism subpathways as indicated on the right of the plot. Effect estimates for PFAS mixture (ψ) and the 95% Bayesian credible interval (BCI) estimate the change in metabolite levels (SD of the log-transformed feature intensity) when increasing all PFAS in the mixture from the 30th percentile to the 70th percentile. This estimate is also equivalent to a standardized mean difference calculated between a hypothetical group of individuals with all PFAS at the ∼70th percentile vs. a hypothetical group of individuals with all PFAS at the ∼30th percentile. Corresponding p-values and q-values are presented in Table S4. Note: CHS, Children’s Health Study; PFAS, per- and polyfluoroalkyl substances; SD, standard deviation; SOLAR, Study of Latino Adolescents at Risk.
Figure 2A is a coefficient plot titled Study of Latino Adolescents at Risk, plotting Thyroid hormone biosynthesis, including Thyroxine; Phenylalanine metabolism, including Phenylacetylglutamine, Phenylacetaldehyde, and Hippuric acid; Tyrosine metabolism and degradation, including Acetoacetic acid, 4-Hydroxyphenylacetaldehyde, Pyruvic acid, L-Glutamic acid, and Tyramine-O-sulfate; Catecholamine biosynthesis and degradation, including Homovanillin, Vanylglycol, 1,2-Dehydrosalsolinol, 3-O-Methyldopa, 3-Methoxytyramine, Norepinephrine, Norepinephrine sulfate, Metanephrine, and ascorbate (y-axis) across Per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 3 in unit increments (x-axis) for Same direction of association in Study of Latino Adolescents at Risk and Children’s Health Study, Only significant in one cohort, and Opposite direction of association in Study of Latino Adolescents at Risk versus Children’s Health Study. Figure 2B is a coefficient plot titled Children’s Health Study, plotting Melanin biosynthesis, including Dopaquinone; Thyroid hormone biosynthesis, including Thyroxine; Phenylalnine metabolism, including Hippuric acid; Tyrosine metabolism and degradation, including L-Glutamic acid and Acetoacetic acid; and Catecholamine biosynthesis and degradation, including Vanylglycol and Homovanillic acid (y-axis) across Per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 3 in unit increments (x-axis) for Same direction of association in Study of Latino Adolescents at Risk and Children’s Health Study, Only significant in one cohort, and Opposite direction of association in Study of Latino Adolescents at Risk versus Children’s Health Study.
In both the SOLAR and the CHS cohorts, PFAS exposure was also associated with key metabolites associated with lipid metabolism. In the SOLAR cohort, 14 unique metabolites were linked to lipid metabolism pathways, and 3 remained significant after adjusting for multiple comparisons (Figure 3A; Table S4). In the CHS cohort, 4 unique metabolites were linked to lipid metabolism pathways, and 2 remained significant after adjusting for multiple comparisons (Figure 3B; Table S4). Across the two cohorts there were similar positive associations between PFAS exposure and metabolites associated with de novo fatty acid biosynthesis and metabolites associated with prostaglandin formation from arachidonate. Of these metabolites, the most consistent association was observed with arachidonic acid, which was positively associated with PFAS exposure in both cohorts (Figure 3; Table S4).
Figure 3. Associations between PFAS mixtures and metabolites associated with lipid metabolism in (A) adolescents from the SOLAR cohort (n=312) and (B) young adults from the CHS cohort (n=137). Effect estimates for PFAS mixture (ψ) and the 95% Bayesian credible interval (BCI) estimate the change in metabolite levels (SD of the log-transformed feature intensity) when increasing all PFAS in the mixture from the 30th percentile to the 70th percentile. This estimate is also equivalent to a standardized mean difference calculated between a hypothetical group of individuals with all PFAS at the ∼70th percentile vs. a hypothetical group of individuals with all PFAS at the ∼30th percentile. Corresponding p-values and q-values are presented in Table S4. Note: CHS, Children’s Health Study; EPA, eicosapentaenoic acid; HPOT, hydroperoxyoctadecatrienoic acid; LysoPC, lysophosphatidylcholines; OxoODE, Octadecanienoic acid; PFAS, per- and polyfluoroalkyl substances; SD, standard deviation; SOLAR, Study of Latino Adolescents at Risk.
Figure 3A is a coefficient plot titled Study of Latino Adolescents at Risk, plotting Putative anti-inflammatory metabolites formation from Eicosapentaenoic acid, including Leukotriene C 5 and 15 Keto-prostaglandin E 2; Prostaglandin formation from Arachidonate, including Arachidonic acid and Prostaglandin E 2; Linoleate metabolism, including (E)-4-Hyfroxynon-2-enal, 12,13-Epoxy-9-alkoxy-10 E-octadecenoate, Lysophosphatidylcholines (18 to 1(9 Z)), Pelargonic acid, 13(S)-Hydroperoxyoctadecatrienoic acid, 13-Octadecanienoic acid, and Linoleic acid; Fatty acid metabolism, including Glycerol; and De novo fatty acid biosynthesis, including Elaidic acid and Dodecanoic acid (y-axis) across Per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 2 in unit increments (x-axis) for Same direction of association in Study of Latino Adolescents at Risk and Children’s Health Study and Only significant in one cohort. Figure 3B is a coefficient plot titled Children’s Health Study, plotting Prostaglandin formation from Arachidonate, including Arachidonic acid and 11-Hydroxyeicosatetraenoate glyceryl ester, and De novo fatty acid biosynthesis, including Docosahexaenoic acid and Behenic acid (y-axis) across per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 2 in unit increments (x-axis) for Same direction of association in Study of Latino Adolescents at Risk and Children’s Health Study and Only significant in one cohort.
PFAS exposure was also associated with alterations in metabolites associated with the metabolism of nonaromatic amino acids (Figure 4; Table S4). In the SOLAR cohort, eight unique metabolites were linked to nonaromatic amino acid metabolism pathways, and four remained significant after adjusting for multiple comparisons (Figure 4A; Table S4). In the CHS cohort, three unique metabolites were linked to nonaromatic amino acid metabolism pathways, although none remained significant after adjusting for multiple comparisons (Figure 4B; Table S4). Across cohorts, the most consistent associations were observed with aminoadipic acid, which was positively associated with PFAS exposure in both cohorts.
Figure 4. Associations between PFAS mixtures and metabolites associated with nonaromatic amino acid metabolism in (A) adolescents from the SOLAR cohort (n=312) and (B) young adults from the CHS cohort (n=137). Effect estimates for PFAS mixture (ψ) and the 95% Bayesian credible interval (BCI) estimate the change in metabolite levels (SD of the log-transformed feature intensity) when increasing all PFAS in the mixture from the 30th percentile to the 70th percentile. This estimate is also equivalent to a standardized mean difference calculated between a hypothetical group of individuals with all PFAS at the ∼70th percentile vs. a hypothetical group of individuals with all PFAS at the ∼30th percentile. Corresponding p-values and q-values are presented in Table S4. Note: CHS, Children’s Health Study; PFAS, per- and polyfluoroalkyl substances; SD, standard deviation; SOLAR, Study of Latino Adolescents at Risk.
Figure 4A is a coefficient plot titled Study of Latino Adolescents at Risk, plotting Urea cycle, including N-acetylornithine; Lysine metabolism, including L-Carnitine, Aminoadipic acid, 6-Amino-2-oxohexanoate; and Arginine and proline metabolism, including Aspartic acid, 5-Amino-2-oxopentanoic acid, Citrulline, and N-Acetylputrescine (y-axis) across Per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 2 in unit increments (x-axis) for Same direction of association in Study of Latino Adolescents at Risk and Children’s Health Study and Only significant in one cohort. Figure 4B is a coefficient plot titled Children’s Health Study, plotting Lysine metabolism, including 3-Dehydroxycarnitine and Aminoadipic acid; and Arginine and proline metabolism, including 5 prime-Methylthioadenosine (y-axis) across Per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 2 in unit increments (x-axis) for Same direction of association in Study of Latino Adolescents at Risk and Children’s Health Study and Only significant in one cohort.
In the SOLAR cohort, PFAS exposure was also positively associated with four metabolites linked to the metabolism of cofactors, including porphyrin metabolism and pyridoxine metabolism. Two of these associations remained significant after adjusting for multiple comparisons (Figure 5; Table S4).
Figure 5. Associations between PFAS mixtures and metabolites associated with metabolism of cofactors in adolescents from the SOLAR cohort (n=312). No significant associations were observed in the CHS cohort. Effect estimates for PFAS mixture (ψ) and the 95% Bayesian credible interval (BCI) estimate the change in metabolite levels (SD of the log-transformed feature intensity) when increasing all PFAS in the mixture from the 30th percentile to the 70th percentile. This estimate is also equivalent to a standardized mean difference calculated between a hypothetical group of individuals with all PFAS at the ∼70th percentile vs. a hypothetical group of individuals with all PFAS at the ∼30th percentile. Corresponding p-values and q-values are presented in Table S4. Note: CHS, Children’s Health Study; PFAS, per- and polyfluoroalkyl substances; SD, standard deviation; SOLAR, Study of Latino Adolescents at Risk.
Figure 5 is a coefficient plot, plotting Vitamin B 6 (pyridoxine) metabolism, including 4-Pyridoxic acid and Pyridoxamine and porphyrin metabolism, including Biliverdin and Bilirubin (y-axis) across Per- and polyfluoroalkyl substances mixture effect uppercase psi (95 percent Bayesian credible interval), ranging from negative 1 to 2 in unit increments (x-axis) for Only significant in one cohort.
Contribution of Individual PFAS on Individual Metabolites
To examine the potential contribution of individual PFAS to the overall mixture effect, we examined the PIP for all PFAS with each metabolite. The PIP is the posterior probability that the coefficient for each individual PFAS in the mixture is nonzero. PFAS that do not contribute to the overall mixture effect should have a PIP reflecting the prior probability of inclusion and =1/P, where P is the total number of PFAS in the mixture. In the present analysis, this corresponds to a PIP of 0.167. PIPs >0.167 indicate a greater likelihood that the individual PFAS has a nonzero effect on the overall mixture.
In the SOLAR cohort, the PIPs for individual exposures across all 44 significantly altered metabolites from enriched pathways ranged from 0.04 to 1.0 (Figures S2–S5). For 41 of the 44 metabolites (93%), two or more PFAS exhibited PIPs >0.167, indicating that for these metabolites, at least two PFAS had a nonzero effect on the overall mixture. Further examination of these PIPs revealed that no individual PFAS appeared to drive these associations across all metabolites, although some PFAS seemed more involved than others. Overall, PFDA and PFOS exhibited PIPs >0.167 for 73% and 71% of the significant metabolites, indicating that these PFAS played a role in the mixture effect for the majority of metabolites. In contrast, PFNA exhibited PIPs >0.167 for only 37% of metabolites, indicating that this PFAS may play less of a role in the association between PFAS exposure and alterations in metabolic pathways. Results were generally consistent within each super pathway (Figures S2–S5).
In the CHS cohort, the PIPs for individual exposures across all 14 significantly altered metabolites from enriched pathways ranged from 0.10 to 0.97. For 12 of the 14 metabolites (86%), two or more PFAS exhibited PIPs >0.167, indicating that for these metabolites, at least two PFAS had a nonzero effect on the overall mixture. Three PFAS exhibited PIPs >0.167 for >70% of metabolites, including PFHxS, PFHpS, and PFNA (Figure S3–S6).
Sensitivity Analysis
Results for the pooled analysis were similar to those of the meta-analysis. The MWAS on the pooled individual level data identified 464 metabolite features associated with the PFAS mixture, defined as having a 95% BCI for the mixture effect not containing zero (Excel Table S3). Functional pathway analysis of the MWAS results identified three metabolic pathways with four or more significant empirical compounds, two of which were significantly enriched (Table S5). These included tyrosine metabolism (p=0.02; significant in both cohorts individually) and urea cycle/amino group metabolism (p=0.04; significant in the meta-analysis). T4 and hippuric acid, two of the three key tyrosine metabolites associated with PFAS exposure in both cohorts, remained significantly associated with the PFAS mixture in the pooled analysis (Table S4).
Discussion
To our knowledge, this is the first study to comprehensively examine the effects of exposure to PFAS mixtures on human metabolic pathways. In two independent cohorts of children and young adults, we observed associations between PFAS exposure and alterations in aromatic amino acid metabolism, nonaromatic amino acid metabolism, and lipid metabolism pathways. These associations were present despite differences in levels of PFAS exposure between cohorts. Alterations in aromatic amino acid metabolism included changes in metabolites associated with catecholamine and thyroid hormone biosynthesis; alterations in nonaromatic amino acid metabolism included changes in metabolites related to arginine, proline, and lysine metabolism; and alterations in lipid metabolism included changes in metabolites related to de novo fatty acid biosynthesis and prostaglandin formation from arachidonate. Together, these results provide evidence that PFAS exposure is associated with alterations in several important metabolic pathways in children and young adults.
Previous studies examining metabolic perturbations of PFAS in humans have primarily relied on single exposure models. In reality, humans are exposed to a mixture of several PFAS compounds that may have synergistic effects.48 Although single exposure models provide insight into the health effects of individual PFAS compounds, these models do not account for co-exposure to multiple potentially correlated environmental exposures that may have synergistic effects. Here, we present the results of an innovative analytic strategy to examine the effects of exposure to chemical mixtures on human metabolic pathways. We used a Bayesian hierarchical regression modeling approach with a g-prior specification and Bayesian g-computation,66,67 which provided a flexible framework for obtaining robust mixture effect estimates in the presence of highly correlated exposures. In addition to providing a framework for future studies looking to examine the associations between exposure mixtures and omics scale data, our findings have important public health implications. Specifically, we did not find evidence that individual PFAS drove the associations between the PFAS mixture and metabolic pathways. Rather, alterations in metabolic pathways were primarily driven by mixtures of PFAS. Further, these effects were consistent across cohorts with different levels of exposure. As a result of changes in the regulation of PFAS in the United States starting in the early 2000s, the SOLAR cohort had higher levels of exposure to PFAS than the CHS cohort.90 These changes were especially pronounced for PFOS and PFOA, which saw decreases of >50% between cohorts. Despite this, we found similar, although slightly attenuated, results in the CHS cohort vs. the SOLAR cohort. This trend may suggest that the toxicological effects of PFAS exposure are more related to total PFAS levels, rather than individual PFAS compounds. Given the associations of PFAS exposure and metabolic pathways related to aromatic amino acid metabolism, nonaromatic amino acid metabolism, and lipid metabolism, our findings lend support to the argument that PFAS should be regulated as a chemical class rather than being regulated on a chemical-by-chemical basis.91
It is well established that PFAS exposure impacts thyroid function, but the downstream metabolic consequences of PFAS-associated thyroid toxicity is not well characterized.11,92 In both cohorts, we observed a positive association between PFAS and T4, the main thyroid hormone in circulation, which is consistent with previous human and animal studies.93 Thyroid hormones are key regulators of metabolism, including the regulation of anabolism and catabolism of lipids, carbohydrates, and proteins.94 Thyroid hormones upregulate de novo fatty acid biosynthesis through the transcription of lipogenic genes, including Acc1, Fasn, Me1, and Thrsp.95 In the present study, we observed positive associations between PFAS exposure and increased de novo fatty acid biosynthesis. Increased de novo fatty acid biosynthesis is a hallmark of metabolic disorders and can lead to obesity, insulin resistance, nonalcoholic fatty liver disease, and cancer.96 These diseases have also been associated with PFAS exposure11 and thyroid hormones.95,97–99 Together, our findings raise the possibility that increased risk of metabolic disorders associated with PFAS exposure are caused by alterations in thyroid hormones and mediated by changes in lipid metabolism.
Several animal studies have reported associations between PFAS exposure and catecholamine biosynthesis, but human data is lacking.29,100–106 Notably, exposure to an environmentally relevant PFAS mixture has been previously shown to alter dopamine levels, one of the three main catecholamines, in mice and in wild Bank voles.102,103 In mice, PFAS exposure has also been linked to alterations in the activity of tyrosine hydroxylase,103 which is the rate-limiting enzyme in the biosynthesis of catecholamines.107 Exposure to individual PFAS, including PFOS and PFOA, has also been shown to decrease tyrosine hydroxylase activity in pheochromocytoma (PC12) cells.108 Although previous animal studies have focused on the association of PFAS exposure on dopamine levels in brain regions, tyrosine hydroxylase is also expressed in the peripheral nervous system and the adrenal medulla, where a majority of peripheral catecholamine biosynthesis occurs.109 In addition to the potential metabolic consequences of alterations in peripheral catecholamine biosynthesis,110 alterations in central catecholamine biosynthesis could play a role in the potential neurotoxic effects of PFAS exposure.111
One of the proposed mechanisms linking PFAS exposure with a variety of diseases is an increase in inflammation and oxidative stress.11,112,113 In the present study, we observed positive associations of PFAS exposure with arachidonic acid, aminoadipic acid, and hippuric acid in both cohorts, each of which is associated with inflammation or oxidative stress. Arachidonic acid is a polyunsaturated fatty acid that contributes to inflammation and plays a role in carcinogenesis and cardiovascular disease.114,115 Aminoadipic acid is an amino acid involved in lysine metabolism that is a potential biomarker of oxidative stress and has been linked to a variety of diseases including type 2 diabetes.116,117 Hippuric acid is a gut-derived amino acid that disrupts redox homeostasis and contributes to oxidative stress, and which has been shown to be a uremic toxin.118,119 Together, these results lend support to the hypothesis that PFAS exposure impact inflammation and oxidative stress.
One factor that could play a role in our findings is diet. Although each of the metabolites identified in this study can be synthesized endogenously, some can also be obtained in the diet. For example, both arachidonic acid and metabolites associated with de novo fatty acid biosynthesis can be obtained via diet.120–122 However, previous literature has linked PFAS exposure with alterations in lipid metabolism and de novo lipogenesis in a variety of experimental studies, which suggests that PFAS are impacting the regulation of these fatty acids in circulation. PFAS exposure alters the expression of genes related to de novo lipogenesis in the liver of mice, resulting in an increase in circulating serum triglycerides.29,123 Similar results have been observed in rats,124 zebrafish,125 and chicken embryos.126 Mechanistically, in addition to PFAS-associated alterations in T4 levels, PFAS also interact with several peroxisome proliferator-activated receptors (PPARs) including PPAR-α and PPAR-γ,11 both of which are key nuclear receptors associated with the regulation of de novo lipogenesis in the liver.127 PPAR-γ also regulates the expression of Δ6-desaturase,128 the enzyme responsible for converting dietary linoleic acid to arachidonic acid. In conjunction with our findings, these studies suggest that PFAS exposure is associated with dysregulated lipid metabolism via alterations in PPAR activity. However, it is also possible that dietary intake of specific fatty acids may interact with PFAS exposure to cause further dysregulation of lipid metabolism,129 which should be examined in future studies.
Although many of the associations between PFAS mixtures and alterations in metabolic pathways were similar in the SOLAR and CHS cohorts, we also identified several differences, especially for lipid metabolism pathways. In the SOLAR cohort, three metabolic pathways (linoleate metabolism, fatty acid metabolism, and anti-inflammatory metabolite formation from eicosapentaenoic acid) were reported only in the SOLAR cohort. One possible explanation is that participants from the SOLAR cohort were either undergoing puberty or prepuberty, whereas participants in the CHS cohort were young adults postpuberty. Exposure to PFAS compounds during sensitive periods of development, such as adolescence, may be more likely to lead to deleterious health effects, including dyslipidemia, given that this is an important period of development for many metabolic tissues.12–15 These results are consistent with previous studies showing that PFAS exposure in childhood is associated with dysregulated lipid and fatty acid metabolism, which can greatly increase the risk of metabolic disorders and cardiovascular disease later in life.130
This study has some limitations. First, when using untargeted metabolomics, accurate identification of metabolites is often difficult because of the limited number of authentic standards available. Although we applied functional pathway enrichment that combines biological relationships among pathways to improve annotation confidence, in the present study the majority of metabolite annotations are limited to level 4, indicating accurate mass matching and molecular formula without informing on chemical structure.131 This is in contrast to targeted metabolomics, which have annotation levels 1 or 2, indicating a confirmed chemical structure. However, untargeted metabolomics has distinct advantages over targeted methods, including greater coverage of metabolites.132 This is ideal for hypothesis-generating studies, and allows for robust interpretation of functional activity at the pathway level.86 Second, it is difficult to assume temporality of associations because of the cross-sectional design. This also raises the possibility of exposure misclassification if plasma PFAS concentrations exhibit large changes in short periods of time. However, the half-life in plasma for PFAS included in this study is in years,133 which suggests that a single plasma measurement is likely to provide an accurate indication of a person’s long-term PFAS exposure. Third, the cohorts included in this study are relatively unique, which may limit generalizability to other populations. The SOLAR and CHS cohorts included primarily Hispanic overweight and obese children and young adults, and additional studies are required to determine whether these associations are generalizable to other populations. Fourth, there is a possibility that an unknown or unmeasured confounder could have impacted our results. For example, diet could be confounding the relationship between PFAS exposure and certain metabolites. However, dietary patterns only appeared to be a small portion of the total PFAS exposure in the general U.S. population between 2003 and 2014,134 which reduces the possibility that diet was a confounder for our study. Other potential confounders could be factors related to PFAS levels, such as plasma albumin levels,135 blood volume,136 menstruation,137 or kidney function.138 Finally, differences in covariates between cohorts may have caused some differences in the results between cohorts. For example, the measures of SES were different between cohorts. However, in the pooled analysis, which used a reduced set of covariates, we observed similar findings as those in the individual cohort analysis, especially those related to alterations in aromatic amino acid metabolism. Despite these limitations, we observed similar results across different populations with varied background characteristics and levels of PFAS exposure. Our findings are in line with previous experimental models in animals, lending additional validity to our results. The similarity of findings across different cohorts with distinct background characteristics suggests that metabolic perturbations associated with PFAS exposure are not restricted to a single cohort. In summary, this study provides evidence that PFAS exposure is associated with alterations in several key metabolic pathways, which could be an important mediating factor explaining the associations of PFAS exposure with risk of various diseases in humans.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
Contributions of the authors to this article are as follows: Concept and design—J.A.G., T.L.A., K.B., Z.C., F.D.G., M.I.G., D.P.J., D.V., D.I.W., D.V.C., and L.C. Acquisition, analysis, or interpretation of data—J.A.G., J.H., X.L., B.O.B., X.H., T.L.A., Z.C., D.V., Z.C.F., S.R., H.W., K.B., F.D.G., M.I.G., D.P.J., D.I.W., D.V.C., and L.C. Drafting of the manuscript—J.A.G., J.H., B.O.B., H.W, D.I.W., D.V.C., and L.C. Critical revision of the manuscript for important intellectual content—J.A.G., J.H., X.L., B.O.B., X.H., T.L.A, Z.C., D.V., Z.C.F., S.R., H.W, K.B., F.D.G., M.I.G., D.P.J., D.I.W., D.V.C., and L.C. Statistical analysis—J.A.G., J.H., B.O.B, H.W., D.V.C., and L.C. Administrative, technical, or material support—X.L., X.H., Z.C.F., S.R., D.P.J., and D.I.W.
The results reported herein correspond to specific aims of grant R01ES029944 from the National Institutes of Health/National Institute of Environmental Health Science (NIH/NIEHS) (to L.C.). Funding for the Study of Latino Adolescents at Risk (SOLAR) came from the NIH grant R01DK59211 (to M.I.G.), and funding for the MetaAir study came from the Southern California Children’s Environmental Health Center grants funded by the NIEHS (5P01ES022845-03, 5P30ES007048, 5P01ES011627), the U.S. Environmental Protection Agency (RD83544101), and the Hastings Foundation. Additional funding from NIH supported L.C. (R01ES030691, R01ES030364, R21ES029681, R21ES028903, and P30ES007048), J.A.G. (T32ES013678, R25GM143298), Z.C. (R00ES027870), D.V. (R01ES033688, R21ES029328, K12ES033594, P30ES023515), D.I.W. (U2CES030859, R01ES032831), D.V.C. (P01CA196569, P30ES007048, R01ES030691, R01ES030364, R21ES029681, R21ES028903), T.L.A (R00ES027853, P50MD017344), and D.P.J. (U2CES030163, P30ES019776, R24ES029490, R01ES032189, R21ES031824).
J.A.G. and L.C. had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
The data that support the findings of this study are available on request from the corresponding author, J.A.G. The data are not publicly available due to them containing information that could compromise research participant privacy/consent.
==== Refs
References
1. Mimoto MS, Nadal A, Sargis RM. 2017. Polluted pathways: mechanisms of metabolic disruption by endocrine disrupting chemicals. Curr Environ Health Rep 4 (2 ):208–222, PMID: , 10.1007/s40572-017-0137-0.28432637
2. Nadal A, Quesada I, Tudurí E, Nogueiras R, Alonso-Magdalena P. 2017. Endocrine-disrupting chemicals and the regulation of energy balance. Nat Rev Endocrinol 13 (9 ):536–546, PMID: , 10.1038/nrendo.2017.51.28524168
3. Sunderland EM, Hu XC, Dassuncao C, Tokranov AK, Wagner CC, Allen JG. 2019. A review of the pathways of human exposure to poly- and perfluoroalkyl substances (PFASs) and present understanding of health effects. J Expo Sci Environ Epidemiol 29 (2 ):131–147, PMID: , 10.1038/s41370-018-0094-1.30470793
4. Whitehead HD, Venier M, Wu Y, Eastman E, Urbanik S, Diamond ML, et al. 2021. Fluorinated compounds in North American cosmetics. Environ Sci Technol Lett 8 (7 ):538–544, 10.1021/acs.estlett.1c00240.
5. Andrews DQ, Naidenko OV. 2020. Population-wide exposure to per- and polyfluoroalkyl substances from drinking water in the United States. Environ Sci Technol Lett 7 (12 ):931–936, 10.1021/acs.estlett.0c00713.
6. U.S. EPA (U.S. Environmental Protection Agency). 2022. Interim Drinking Water Health Advisory: Perfluorooctanoic Acid (PFOA). CASRN 335-67-1. EPA/8222/R-22/003. https://www.epa.gov/system/files/documents/2022-06/interim-pfoa-2022.pdf [accessed 25 June 2022].
7. U.S. EPA. 2022. Interim Drinking Water Health Advisory: Perfluorooctane Sulfonic Acid (PFOS) CASRN 1763-23-1. EPA/822/R-22/004. https://www.epa.gov/system/files/documents/2022-06/interim-pfos-2022.pdf [accessed 25 June 2022].
8. Zhang Y, Beesoon S, Zhu L, Martin JW. 2013. Biomonitoring of perfluoroalkyl acids in human urine and estimates of biological half-life. Environ Sci Technol 47 (18 ):10619–10627, PMID: , 10.1021/es401905e.23980546
9. Kannan K, Corsolini S, Falandysz J, Fillmann G, Kumar KS, Loganathan BG, et al. 2004. Perfluorooctanesulfonate and related fluorochemicals in human blood from several countries. Environ Sci Technol 38 (17 ):4489–4495, PMID: , 10.1021/es0493446.15461154
10. Kato K, Wong LY, Jia LT, Kuklenyik Z, Calafat AM. 2011. Trends in exposure to polyfluoroalkyl chemicals in the U.S. population: 1999–2008. Environ Sci Technol 45 (19 ):8037–8045, PMID: , 10.1021/es1043613.21469664
11. Fenton SE, Ducatman A, Boobis A, DeWitt JC, Lau C, Ng C, et al. 2020. Per- and polyfluoroalkyl substance toxicity and human health review: current state of knowledge and strategies for informing future research. Environ Toxicol Chem 40 (3 ):606–630, PMID: , 10.1002/etc.4890.33017053
12. Casazza K, Goran MI, Gower BA. 2008. Associations among insulin, estrogen, and fat mass gain over the pubertal transition in African-American and European-American girls. J Clin Endocrinol Metab 93 (7 ):2610–2615, PMID: , 10.1210/jc.2007-2776.18349063
13. Heindel JJ. 2019. The developmental basis of disease: update on environmental exposures and animal models. Basic Clin Pharmacol Toxicol 125 (suppl 3 ):5–13, PMID: , 10.1111/bcpt.13118.30265444
14. Hines EP, White SS, Stanko JP, Gibbs-Flournoy EA, Lau C, Fenton SE. 2009. Phenotypic dichotomy following developmental exposure to perfluorooctanoic acid (PFOA) in female CD-1 mice: low doses induce elevated serum leptin and insulin, and overweight in mid-life. Mol Cell Endocrinol 304 (1–2 ):97–105, PMID: , 10.1016/j.mce.2009.02.021.19433254
15. Kelly LA, Lane CJ, Weigensberg MJ, Toledo-Corral CM, Goran MI. 2011. Pubertal changes of insulin sensitivity, acute insulin response and β-cell function in overweight Latino youth. J Pediatr 158 (3 ):442–446, PMID: , 10.1016/j.jpeds.2010.08.046.20888012
16. Goodrich JA, Alderete TL, Baumert BO, Berhane K, Chen Z, Gilliland FD, et al. 2021. Exposure to perfluoroalkyl substances and glucose homeostasis in youth. Environ Health Perspect 129 (9 ):097002, PMID: , 10.1289/EHP9200.34468161
17. Valvi D, Højlund K, Coull BA, Nielsen F, Weihe P, Grandjean P. 2021. Life-course exposure to perfluoroalkyl substances in relation to markers of glucose homeostasis in early adulthood. J Clin Endocrinol Metab 106 (8 ):2495–2504, PMID: , 10.1210/clinem/dgab267.33890111
18. Blomberg AJ, Shih YH, Messerlian C, Jørgensen LH, Weihe P, Grandjean P. 2021. Early-life associations between per- and polyfluoroalkyl substances and serum lipids in a longitudinal birth cohort. Environ Res 200 :111400, PMID: , 10.1016/j.envres.2021.111400.34081971
19. Domazet SL, Grøntved A, Timmermann AG, Nielsen F, Jensen TK. 2016. Longitudinal associations of exposure to perfluoroalkylated substances in childhood and adolescence and indicators of adiposity and glucose metabolism 6 and 12 years later: the European Youth Heart Study. Diabetes Care 39 (10 ):1745–1751, PMID: , 10.2337/dc16-0269.27489335
20. Høyer BB, Ramlau-Hansen CH, Vrijheid M, Valvi D, Pedersen HS, Zviezdai V, et al. 2015. Anthropometry in 5- to 9-year-old Greenlandic and Ukrainian children in relation to prenatal exposure to perfluorinated alkyl substances. Environ Health Perspect 123 (8 ):841–846, PMID: , 10.1289/ehp.1408881.25809098
21. Cardenas A, Hivert MF, Gold DR, Hauser R, Kleinman KP, Lin PID, et al. 2019. Associations of perfluoroalkyl and polyfluoroalkyl substances with incident diabetes and microvascular disease. Diabetes Care 42 (9 ):1824–1832, PMID: , 10.2337/dc18-2254.31296647
22. Qi W, Clark JM, Timme-Laragy AR, Park Y. 2020. Per- and polyfluoroalkyl substances and obesity, type 2 diabetes and non-alcoholic fatty liver disease: a review of epidemiologic findings. Toxicol Environ Chem 102 (1–4 ):1–36, PMID: , 10.1080/02772248.2020.1763997.33304027
23. Sun Q, Zong G, Valvi D, Nielsen F, Coull B, Grandjean P. 2018. Plasma concentrations of perfluoroalkyl substances and risk of type 2 diabetes: a prospective investigation among U.S. women. Environ Health Perspect 126 (3 ):037001, PMID: , 10.1289/EHP2619.29498927
24. Balkau B, Valensi P, Eschwège E, Slama G. 2007. A review of the metabolic syndrome. Diabetes Metab 33 (6 ):405–413, PMID: , 10.1016/j.diabet.2007.08.001.17981485
25. Guo P, Furnary T, Vasiliou V, Yan Q, Nyhan K, Jones DP, et al. 2022. Non-targeted metabolomics and associations with per- and polyfluoroalkyl substances (PFAS) exposure in humans: a scoping review. Environ Int 162 :107159, PMID: , 10.1016/j.envint.2022.107159.35231839
26. Peng S, Yan L, Zhang J, Wang Z, Tian M, Shen H. 2013. An integrated metabonomics and transcriptomics approach to understanding metabolic pathway disturbance induced by perfluorooctanoic acid. J Pharm Biomed Anal 86 :56–64, PMID: , 10.1016/j.jpba.2013.07.014.23978341
27. Kudo N, Suzuki-Nakajima E, Mitsumoto A, Kawashima Y. 2006. Responses of the liver to perfluorinated fatty acids with different carbon chain length in male and female mice: in relation to induction of hepatomegaly, peroxisomal β-oxidation and microsomal 1-acylglycerophosphocholine acyltransferase. Biol Pharm Bull 29 (9 ):1952–1957, PMID: , 10.1248/bpb.29.1952.16946516
28. Wang L, Wang Y, Liang Y, Li J, Liu Y, Zhang J, et al. 2014. PFOS induced lipid metabolism disturbances in BALB/c mice through inhibition of low density lipoproteins excretion. Sci Rep 4 (1 ):4582, PMID: , 10.1038/srep04582.24694979
29. Yu N, Wei S, Li M, Yang J, Li K, Jin L, et al. 2016. Effects of perfluorooctanoic acid on metabolic profiles in brain and liver of mouse revealed by a high-throughput targeted metabolomics approach. Sci Rep 6 :23963, PMID: , 10.1038/srep23963.27032815
30. Gebreab KY, Eeza MNH, Bai T, Zuberi Z, Matysik J, O’Shea KE, et al. 2020. Comparative toxicometabolomics of perfluorooctanoic acid (PFOA) and next-generation perfluoroalkyl substances. Environ Pollut 265 (pt A ):114928, PMID: , 10.1016/j.envpol.2020.114928.32540561
31. Alderete TL, Jin R, Walker DI, Valvi D, Chen Z, Jones DP, et al. 2019. Perfluoroalkyl substances, metabolomic profiling, and alterations in glucose homeostasis among overweight and obese Hispanic children: a proof-of-concept analysis. Environ Int 126 :445–453, PMID: , 10.1016/j.envint.2019.02.047.30844580
32. Chen Z, Yang T, Walker DI, Thomas DC, Qiu C, Chatzi L, et al. 2020. Dysregulated lipid and fatty acid metabolism link perfluoroalkyl substances exposure and impaired glucose metabolism in young adults. Environ Int 145 :106091, PMID: , 10.1016/j.envint.2020.106091.32892005
33. Hu X, Li S, Cirillo PM, Krigbaum NY, Tran V, Jones DP, et al. 2019. Metabolome wide association study of serum poly and perfluoroalkyl substances (PFASs) in pregnancy and early postpartum. Reprod Toxicol 87 :70–78, PMID: , 10.1016/j.reprotox.2019.05.058.31121237
34. Kingsley SL, Walker DI, Calafat AM, Chen A, Papandonatos GD, Xu Y, et al. 2019. Metabolomics of childhood exposure to perfluoroalkyl substances: a cross-sectional study. Metabolomics 15 (7 ):95, PMID: , 10.1007/s11306-019-1560-z.31227916
35. Li Y, Lu X, Yu N, Li A, Zhuang T, Du L, et al. 2021. Exposure to legacy and novel perfluoroalkyl substance disturbs the metabolic homeostasis in pregnant women and fetuses: a metabolome-wide association study. Environ Int 156 :106627, PMID: , 10.1016/j.envint.2021.106627.33991873
36. Lu Y, Gao K, Li X, Tang Z, Xiang L, Zhao H, et al. 2019. Mass spectrometry-based metabolomics reveals occupational exposure to per- and polyfluoroalkyl substances relates to oxidative stress, fatty acid β-oxidation disorder, and kidney injury in a manufactory in China. Environ Sci Technol 53 (16 ):9800–9809, PMID: , 10.1021/acs.est.9b01608.31246438
37. Mitro SD, Liu J, Jaacks LM, Fleisch AF, Williams PL, Knowler WC, et al. 2021. Per- and polyfluoroalkyl substance plasma concentrations and metabolomic markers of type 2 diabetes in the Diabetes Prevention Program trial. Int J Hyg Environ Health 232 :113680, PMID: , 10.1016/j.ijheh.2020.113680.33348273
38. Salihovic S, Fall T, Ganna A, Broeckling CD, Prenni JE, Hyötyläinen T, et al. 2019. Identification of metabolic profiles associated with human exposure to perfluoroalkyl substances. J Expo Sci Environ Epidemiol 29 (2 ):196–205, PMID: , 10.1038/s41370-018-0060-y.30185940
39. Sen P, Qadri S, Luukkonen PK, Ragnarsdottir O, McGlinchey A, Jäntti S, et al. 2022. Exposure to environmental contaminants is associated with altered hepatic lipid metabolism in non-alcoholic fatty liver disease. J Hepatol 76 (2 ):283–293, PMID: , 10.1016/j.jhep.2021.09.039.34627976
40. Stratakis N, Conti DV, Jin R, Margetaki K, Valvi D, Siskos AP, et al. 2020. Prenatal exposure to perfluoroalkyl substances associated with increased susceptibility to liver injury in children. Hepatology 72 (5 ):1758–1770, PMID: , 10.1002/hep.31483.32738061
41. Jin R, McConnell R, Catherine C, Xu S, Walker DI, Stratakis N, et al. 2020. Perfluoroalkyl substances and severity of nonalcoholic fatty liver in children: an untargeted metabolomics approach. Environ Int 134 :105220, PMID: , 10.1016/j.envint.2019.105220.31744629
42. Li S, Cirillo P, Hu X, Tran V, Krigbaum N, Yu S, et al. 2020. Understanding mixed environmental exposures using metabolomics via a hierarchical community network model in a cohort of California women in 1960’s. Reprod Toxicol 92 :57–65, PMID: , 10.1016/j.reprotox.2019.06.013.31299210
43. Matta K, Lefebvre T, Vigneau E, Cariou V, Marchand P, Guitton Y, et al. 2022. Associations between persistent organic pollutants and endometriosis: a multiblock approach integrating metabolic and cytokine profiling. Environ Int 158 :106926, PMID: , 10.1016/j.envint.2021.106926.34649050
44. Holeček M. 2018. Branched-chain amino acids in health and disease: metabolism, alterations in blood plasma, and as supplements. Nutr Metab (Lond) 15 (1 ):33, PMID: , 10.1186/s12986-018-0271-1.29755574
45. Prawitt J, Caron S, Staels B. 2011. Bile acid metabolism and the pathogenesis of type 2 diabetes. Curr Diab Rep 11 (3 ):160–166, PMID: , 10.1007/s11892-011-0187-x.21431855
46. Di Ciaula A, Wang DQH, Molina-Molina E, Lunardi Baccetto R, Calamita G, Palmieri VO, et al. 2017. Bile acids and cancer: direct and environmental-dependent effects. Ann Hepatol 16 (suppl 1 ):S87–S105, PMID: , 10.5604/01.3001.0010.5501.29080344
47. Wei Z, Liu X, Cheng C, Yu W, Yi P. 2021. Metabolism of amino acids in cancer. Front Cell Dev Biol 8 :603837, PMID: , 10.3389/fcell.2020.603837.33511116
48. Wang Z, DeWitt JC, Higgins CP, Cousins IT. 2017. A never-ending story of per- and polyfluoroalkyl substances (PFASs)? Environ Sci Technol 51 (5 ):2508–2518, PMID: , 10.1021/acs.est.6b04806.28224793
49. Alderete TL, Habre R, Toledo-Corral CM, Berhane K, Chen Z, Lurmann FW, et al. 2017. Longitudinal associations between ambient air pollution with insulin sensitivity, β-cell function, and adiposity in Los Angeles Latino children. Diabetes 66 (7 ):1789–1796, PMID: , 10.2337/db16-1416.28137791
50. Goran MI, Bergman RN, Avila Q, Watkins M, Ball GDC, Shaibi GQ, et al. 2004. Impaired glucose tolerance and reduced β-cell function in overweight Latino children with a positive family history for type 2 diabetes. J Clin Endocrinol Metab 89 (1 ):207–212, PMID: , 10.1210/jc.2003-031402.14715851
51. Baumert BO, Goodrich JA, Hu X, Walker DI, Alderete TL, Chen Z, et al. 2022. Plasma concentrations of lipophilic persistent organic pollutants and glucose homeostasis in youth populations. Environ Res 212 (pt B ):113296, PMID: , 10.1016/j.envres.2022.113296.35447156
52. Kim JS, Chen Z, Alderete TL, Toledo-Corral C, Lurmann F, Berhane K, et al. 2019. Associations of air pollution, obesity and cardiometabolic health in young adults: the Meta-AIR study. Environ Int 133 (pt A ):105180, PMID: , 10.1016/j.envint.2019.105180.31622905
53. McConnell R, Shen E, Gilliland FD, Jerrett M, Wolch J, Chang CC, et al. 2015. A longitudinal cohort study of body mass index and childhood exposure to secondhand tobacco smoke and air pollution: the Southern California Children’s Health Study. Environ Health Perspect 123 (4 ):360–366, PMID: , 10.1289/ehp.1307031.25389275
54. US Preventive Services Task Force, Grossman DC, Bibbins-Domingo K, Curry SJ, Barry MJ, Davidson KW, et al. 2017. Screening for obesity in children and adolescents: US Preventive Services Task Force recommendation statement. JAMA 317 (23 ):2417–2426, PMID: , 10.1001/jama.2017.6803.28632874
55. Hollingshead AB. 2011. Four factor index of social status. Yale J Sociol 8 :21–52.
56. Berkelmans GFN, Read SH, Gudbjörnsdottir S, Wild SH, Franzen S, van der Graaf Y, et al. 2022. Population median imputation was noninferior to complex approaches for imputing missing values in cardiovascular prediction models in clinical practice. J Clin Epidemiol 145 :70–80, PMID: , 10.1016/j.jclinepi.2022.01.011.35066115
57. Marshall WA, Tanner JM. 1969. Variations in pattern of pubertal changes in girls. Arch Dis Child 44 (235 ):291–303, PMID: , 10.1136/adc.44.235.291.5785179
58. Marshall WA, Tanner JM. 1970. Variations in the pattern of pubertal changes in boys. Arch Dis Child 45 (239 ):13–23, PMID: , 10.1136/adc.45.239.13.5440182
59. Goodrich JA, Walker D, Lin X, Wang H, Lim T, McConnell R, et al. 2022. Exposure to perfluoroalkyl substances and risk of hepatocellular carcinoma in a multiethnic cohort. JHEP Rep 4 (10 ):100550, PMID: , 10.1016/j.jhepr.2022.100550.36111068
60. CDC (Centers for Disease Control and Prevention). 2019. Fourth Report on Human Exposure to Environmental Chemicals, Updated Tables. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention.
61. Goodpaster BH, Sparks LM. 2017. Metabolic flexibility in health and disease. Cell Metab 25 (5 ):1027–1036, PMID: , 10.1016/j.cmet.2017.04.015.28467922
62. Huo S, Sun L, Zong G, Shen X, Zheng H, Jin Q, et al. 2021. Changes in plasma metabolome profiles following oral glucose challenge among adult Chinese. Nutrients 13 (5 ):1474, PMID: , 10.3390/nu13051474.33925473
63. Liu KH, Nellis M, Uppal K, Ma C, Tran V, Liang Y, et al. 2020. Reference standardization for quantification and harmonization of large-scale metabolomics. Anal Chem 92 (13 ):8836–8844, PMID: , 10.1021/acs.analchem.0c00338.32490663
64. Liu Q, Walker D, Uppal K, Liu Z, Ma C, Tran V, et al. 2020. Addressing the batch effect issue for LC/MS metabolomics data in data preprocessing. Sci Rep 10 (1 ):13856, PMID: , 10.1038/s41598-020-70850-0.32807888
65. Luan H, Ji F, Chen Y, Cai Z. 2018. statTarget: a streamlined tool for signal drift correction and interpretations of quantitative mass spectrometry-based omics data. Anal Chim Acta 1036 :66–72, PMID: , 10.1016/j.aca.2018.08.002.30253838
66. Keil AP, Daza EJ, Engel SM, Buckley JP, Edwards JK. 2018. A Bayesian approach to the g-formula. Stat Methods Med Res 27 (10 ):3183–3204, PMID: , 10.1177/0962280217694665.29298607
67. Keil AP, Buckley JP, Kalkbrenner AE. 2021. Bayesian g-computation for estimating impacts of interventions on exposure mixtures: demonstration with metals from coal-fired power plants and birth weight. Am J Epidemiol 190 (12 ):2647–2657, PMID: , 10.1093/aje/kwab053.33751055
68. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 128 (4 ):47004, PMID: , 10.1289/EHP5838.32255670
69. Maruyama Y, George EI. 2011. Fully Bayes factors with a generalized g-prior. Ann Statist 39 (5 ):2740–2765, 10.1214/11-AOS917.
70. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16 (3 ):493–508, PMID: , 10.1093/biostatistics/kxu058.25532525
71. Bottolo L, Richardson S. 2010. Evolutionary stochastic search for Bayesian model exploration. Bayesian Anal 5 (3 ):583–618, 10.1214/10-BA523.
72. Hanson TE, Branscum AJ, Johnson WO. 2014. Informative g-priors for logistic regression. Bayesian Anal 9 (3 ):597–612, 10.1214/14-BA868.
73. Newcombe PJ, Conti DV, Richardson S. 2016. JAM: a scalable Bayesian framework for joint analysis of marginal SNP effects. Genet Epidemiol 40 (3 ):188–201, PMID: , 10.1002/gepi.21953.27027514
74. Zellner A. 1986. On assessing prior distributions and Bayesian regression analysis with g-prior distributions. In: Bayesian Inference and Decision Techniques: Essays in Honor of Bruno de Finetti. Goel P, Zellner A, eds. New York, NY: Elsevier Science Publishers, 233–243.
75. Liang F, Paulo R, Molina G, Clyde MA, Berger JO. 2008. Mixtures of g priors for Bayesian variable selection. J Am Stat Assoc 103 (481 ):410–423, 10.1198/016214507000001337.
76. Li Y, Clyde MA. 2018. Mixtures of g-priors in generalized linear models. J Am Stat Assoc 113 (524 ):1828–1845, 10.1080/01621459.2018.1469992.
77. Greenland S. 1992. A semi-Bayes approach to the analysis of correlated multiple associations, with an application to an occupational cancer-mortality study. Stat Med 11 (2 ):219–230, PMID: , 10.1002/sim.4780110208.1579760
78. George E, McCulloch R. 1995. Stochastic search variable selection. In: Markov Chain Monte Carlo in Practice. Gilks WR, Richardson S, Spiegelhalter D, eds. Boca Raton, FL: Chapman and Hall/CRC, 203–214.
79. Plummer M, Stukalov A, Denwood M. 2021. rjags: Bayesian graphical models using MCMC. https://CRAN.R-project.org/package=rjags [accessed 18 January 2022].
80. Perrakis K, Ntzoufras I. 2018. Bayesian variable selection using the hyper-g prior in WinBUGS. Wiley Interdiscip Rev Comput Stat 10 (6 ):e1442, 10.1002/wics.1442.
81. Benjamini Y, Hochberg Y. 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol 57 (1 ):289–300, 10.1111/j.2517-6161.1995.tb02031.x.
82. Pang Z, Chong J, Zhou G, de Lima Morais DA, Chang L, Barrette M, et al. 2021. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res 49 (W1 ):W388–W396, PMID: , 10.1093/nar/gkab382.34019663
83. Pang Z, Chong J, Li S, Xia J. 2020. MetaboAnalystR 3.0: toward an optimized workflow for global metabolomics. Metabolites 10 (5 ):186, PMID: , 10.3390/metabo10050186.32392884
84. Li S, Pozhitkov A, Ryan RA, Manning CS, Brown-Peterson N, Brouwer M. 2010. Constructing a fish metabolic network model. Genome Biol 11 (11 ):R115, PMID: , 10.1186/gb-2010-11-11-r115.21114829
85. Chong J, Wishart DS, Xia J. 2019. Using MetaboAnalyst 4.0 for comprehensive and integrative metabolomics data analysis. Curr Protoc Bioinformatics 68 (1 ):e86, PMID: , 10.1002/cpbi.86.31756036
86. Li S, Park Y, Duraisingham S, Strobel FH, Khan N, Soltow QA, et al. 2013. Predicting network activity from high throughput metabolomics. PLOS Comput Biol 9 (7 ):e1003123, PMID: , 10.1371/journal.pcbi.1003123.23861661
87. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. 2005. Gene Set Enrichment Analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102 (43 ):15545–15550, PMID: , 10.1073/pnas.0506580102.16199517
88. Zaykin DV. 2011. Optimally weighted Z-test is a powerful method for combining probabilities in meta-analysis. J Evol Biol 24 (8 ):1836–1841, PMID: , 10.1111/j.1420-9101.2011.02297.x.21605215
89. Bravata DM, Olkin I. 2001. Simple pooling versus combining in meta-analysis. Eval Health Prof 24 (2 ):218–230, PMID: , 10.1177/01632780122034885.11523387
90. Brennan NM, Evans AT, Fritz MK, Peak SA, von Holst HE. 2021. Trends in the regulation of per- and polyfluoroalkyl substances (PFAS): a scoping review. Int J Environ es Public Health 18 (20 ):10900, PMID: , 10.3390/ijerph182010900.34682663
91. Kwiatkowski CF, Andrews DQ, Birnbaum LS, Bruton TA, DeWitt JC, Knappe DRU, et al. 2020. Scientific basis for managing PFAS as a chemical class. Environ Sci Technol Lett 7 (8 ):532–543, PMID: , 10.1021/acs.estlett.0c00255.34307722
92. Preston EV, Webster TF, Claus Henn B, McClean MD, Gennings C, Oken E, et al. 2020. Prenatal exposure to per- and polyfluoroalkyl substances and maternal and neonatal thyroid function in the Project Viva Cohort: a mixtures approach. Environ Int 139 :105728, PMID: , 10.1016/j.envint.2020.105728.32311629
93. Coperchini F, Croce L, Ricci G, Magri F, Rotondi M, Imbriani M, et al. 2021. Thyroid disrupting effects of old and new generation PFAS. Front Endocrinol (Lausanne) 11 :612320, PMID: , 10.3389/fendo.2020.612320.33542707
94. Boelaert K, Franklyn JA. 2005. Thyroid hormone in health and disease. J Endocrinol 187 (1 ):1–15, PMID: , 10.1677/joe.1.06131.16214936
95. Sinha RA, Singh BK, Yen PM. 2018. Direct effects of thyroid hormones on hepatic lipid metabolism. Nat Rev Endocrinol 14 (5 ):259–269, PMID: , 10.1038/nrendo.2018.10.29472712
96. Ameer F, Scandiuzzi L, Hasnain S, Kalbacher H, Zaidi N. 2014. De novo lipogenesis in health and disease. Metabolism 63 (7 ):895–902, PMID: , 10.1016/j.metabol.2014.04.003.24814684
97. Wang B, Wang B, Yang Y, Xu J, Hong M, Xia M, et al. 2021. Thyroid function and non-alcoholic fatty liver disease in hyperthyroidism patients. BMC Endocr Disord 21 (1 ):27, PMID: , 10.1186/s12902-021-00694-w.33602203
98. Khan SR, Chaker L, Ruiter R, Aerts JGJV, Hofman A, Dehghan A, et al. 2016. Thyroid function and cancer risk: the Rotterdam Study. J Clin Endocrinol Metab 101 (12 ):5030–5036, PMID: , 10.1210/jc.2016-2104.27648963
99. Kalra S, Aggarwal S, Khandelwal D. 2019. Thyroid dysfunction and type 2 diabetes mellitus: screening strategies and implications for management. Diabetes Ther 10 (6 ):2035–2044, PMID: , 10.1007/s13300-019-00700-4.31583645
100. Eggers Pedersen K, Basu N, Letcher R, Greaves AK, Sonne C, Dietz R, et al. 2015. Brain region-specific perfluoroalkylated sulfonate (PFSA) and carboxylic acid (PFCA) accumulation and neurochemical biomarker responses in east Greenland polar bears (Ursus maritimus). Environ Res 138 :22–31, PMID: , 10.1016/j.envres.2015.01.015.25682255
101. Foguth RM, Flynn RW, de Perre C, Iacchetta M, Lee LS, Sepúlveda MS, et al. 2019. Developmental exposure to perfluorooctane sulfonate (PFOS) and perfluorooctanoic acid (PFOA) selectively decreases brain dopamine levels in Northern leopard frogs. Toxicol Appl Pharmacol 377 :114623, PMID: , 10.1016/j.taap.2019.114623.31195004
102. Grønnestad R, Schlenk D, Krøkje Å, Jaspers VLB, Jenssen BM, Coffin S, et al. 2021. Alteration of neuro-dopamine and steroid hormone homeostasis in wild Bank voles in relation to tissue concentrations of PFAS at a Nordic skiing area. Sci Total Environ 756 :143745, PMID: , 10.1016/j.scitotenv.2020.143745.33250251
103. Grønnestad R, Johanson SM, Müller MHB, Schlenk D, Tanabe P, Krøkje Å, et al. 2021. Effects of an environmentally relevant PFAS mixture on dopamine and steroid hormone levels in exposed mice. Toxicol Appl Pharmacol 428 :115670, PMID: , 10.1016/j.taap.2021.115670.34371090
104. Long Y, Wang Y, Ji G, Yan L, Hu F, Gu A. 2013. Neurotoxicity of perfluorooctane sulfonate to hippocampal cells in adult mice. PLoS One 8 (1 ):e54176, PMID: , 10.1371/journal.pone.0054176.23382877
105. Salgado R, López-Doval S, Pereiro N, Lafuente A. 2016. Perfluorooctane sulfonate (PFOS) exposure could modify the dopaminergic system in several limbic brain regions. Toxicol Lett 240 (1 ):226–235, PMID: , 10.1016/j.toxlet.2015.10.023.26529483
106. Salgado RM, Sheard AC, Vaughan RA, Parker DL, Schneider SM, Kenefick RW, et al. 2017. Mitochondrial efficiency and exercise economy following heat stress: a potential role of uncoupling protein 3. Physiol Rep 5 (3 ):e13054, PMID: , 10.14814/phy2.13054.28174343
107. Daubner SC, Le T, Wang S. 2011. Tyrosine hydroxylase and regulation of dopamine synthesis. Arch Biochem Biophys 508 (1 ):1–12, PMID: , 10.1016/j.abb.2010.12.017.21176768
108. Slotkin TA, MacKillop EA, Melnick RL, Thayer KA, Seidler FJ. 2008. Developmental neurotoxicity of perfluorinated chemicals modeled in vitro. Environ Health Perspect 116 (6 ):716–722, PMID: , 10.1289/ehp.11253.18560525
109. Berends AMA, Eisenhofer G, Fishbein L, Horst-Schrivers ANAVD, Kema IP, Links TP, et al. 2019. Intricacies of the molecular machinery of catecholamine biosynthesis and secretion by chromaffin cells of the normal adrenal medulla and in pheochromocytoma and paraganglioma. Cancers (Basel) 11 (8 ):1121, PMID: , 10.3390/cancers11081121.31390824
110. Goldstein DS, Eisenhofer G, Kopin IJ. 2003. Sources and significance of plasma levels of catechols and their metabolites in humans. J Pharmacol Exp Ther 305 (3 ):800–811, PMID: , 10.1124/jpet.103.049270.12649306
111. Patel R, Bradner JM, Stout KA, Caudle WM. 2016. Alteration to dopaminergic synapses following exposure to perfluorooctane sulfonate (PFOS), in vitro and in vivo. Med Sci (Basel) 4 (3 ):13, PMID: , 10.3390/medsci4030013.29083377
112. Bonato M, Corrà F, Bellio M, Guidolin L, Tallandini L, Irato P, et al. 2020. PFAS environmental pollution and antioxidant responses: an overview of the impact on human field. Int J Environ Res Public Health 17 (21 ):8020, PMID: , 10.3390/ijerph17218020.33143342
113. Omoike OE, Pack RP, Mamudu HM, Liu Y, Strasser S, Zheng S, et al. 2021. Association between per and polyfluoroalkyl substances and markers of inflammation and oxidative stress. Environ Res 196 :110361, PMID: , 10.1016/j.envres.2020.110361.33131681
114. Tallima H, El Ridi R. 2018. Arachidonic acid: physiological roles and potential health benefits—a review. J Adv Res 11 :33–41, PMID: , 10.1016/j.jare.2017.11.004.30034874
115. Wang B, Wu L, Chen J, Dong L, Chen C, Wen Z, et al. 2021. Metabolism pathways of arachidonic acids: mechanisms and potential therapeutic targets. Signal Transduct Target Ther 6 (1 ):94, PMID: , 10.1038/s41392-020-00443-w.33637672
116. Wijekoon EP, Skinner C, Brosnan ME, Brosnan JT. 2004. Amino acid metabolism in the Zucker diabetic fatty rat: effects of insulin resistance and of type 2 diabetes. Can J Physiol Pharmacol 82 (7 ):506–514, PMID: , 10.1139/y04-067.15389298
117. Zeitoun-Ghandour S, Leszczyszyn OI, Blindauer CA, Geier FM, Bundy JG, Stürzenbaum SR. 2011. C. elegans metallothioneins: response to and defence against ROS toxicity. Mol Biosyst 7 (8 ):2397–2406, PMID: , 10.1039/c1mb05114h.21647514
118. Duranton F, Cohen G, De Smet R, Rodriguez M, Jankowski J, Vanholder R, et al. 2012. Normal and pathologic concentrations of uremic toxins. J Am Soc Nephrol 23 (7 ):1258–1270, PMID: , 10.1681/ASN.2011121175.22626821
119. Lim YJ, Sidor NA, Tonial NC, Che A, Urquhart BL. 2021. Uremic toxins in the progression of chronic kidney disease and cardiovascular disease: mechanisms and therapeutic targets. Toxins 13 (2 ):142, PMID: , 10.3390/toxins13020142.33668632
120. Calder PC. 2007. Dietary arachidonic acid: harmful, harmless or helpful? Br J Nutr 98 (3 ):451–453, PMID: , 10.1017/S0007114507761779.17705889
121. Lewis RJ. 2007. Hawley’s Condensed Chemical Dictionary. 15th ed. Hoboken, NJ: Wiley.
122. Roe M, Pinchen H, Church S, Elahi S, Walker M, Farron-Wilson M, et al. 2013. Trans fatty acids in a range of UK processed foods. Food Chem 140 (3 ):427–431, PMID: , 10.1016/j.foodchem.2012.08.067.23601386
123. Schlezinger JJ, Hyötyläinen T, Sinioja T, Boston C, Puckett H, Oliver J, et al. 2021. Perfluorooctanoic acid induces liver and serum dyslipidemia in humanized PPARα mice fed an American diet. Toxicol Appl Pharmacol 426 :115644, PMID: , 10.1016/j.taap.2021.115644.34252412
124. Lai KP, Li JW, Cheung A, Li R, Billah MB, Chan TF, et al. 2017. Transcriptome sequencing reveals prenatal PFOS exposure on liver disorders. Environ Pollut 223 :416–425, PMID: , 10.1016/j.envpol.2017.01.041.28131474
125. Cheng J, Lv S, Nie S, Liu J, Tong S, Kang N, et al. 2016. Chronic perfluorooctane sulfonate (PFOS) exposure induces hepatic steatosis in zebrafish. Aquat Toxicol 176 :45–52, PMID: , 10.1016/j.aquatox.2016.04.013.27108203
126. Jacobsen AV, Nordén M, Engwall M, Scherbak N. 2018. Effects of perfluorooctane sulfonate on genes controlling hepatic fatty acid metabolism in livers of chicken embryos. Environ Sci Pollut Res Int 25 (23 ):23074–23081, PMID: , 10.1007/s11356-018-2358-7.29860686
127. Souza-Mello V. 2015. Peroxisome proliferator-activated receptors as targets to treat non-alcoholic fatty liver disease. World J Hepatol 7 (8 ):1012–1019, PMID: , 10.4254/wjh.v7.i8.1012.26052390
128. Saliani N, Darabi M, Yousefi B, Baradaran B, Khaniani MS, Darabi M, et al. 2013. PPARγ agonist-induced alterations in Δ6-desaturase and stearoyl-CoA desaturase 1: role of MEK/ERK1/2 pathway. World J Hepatol 5 (4 ):220–225, PMID: , 10.4254/wjh.v5.i4.220.23671727
129. Roth K, Imran Z, Liu W, Petriello MC. 2020. Diet as an exposure source and mediator of per- and polyfluoroalkyl substance (PFAS) toxicity. Front Toxicol 2 :601149, PMID: , 10.3389/ftox.2020.601149.35296120
130. Rappazzo KM, Coffman E, Hines EP. 2017. Exposure to perfluorinated alkyl substances and health outcomes in children: a systematic review of the epidemiologic literature. Int J Environ Res Public Health 14 (7 ):691, PMID: , 10.3390/ijerph14070691.28654008
131. Schymanski EL, Jeon J, Gulde R, Fenner K, Ruff M, Singer HP, et al. 2014. Identifying small molecules via high resolution mass spectrometry: communicating confidence. Environ Sci Technol 48 (4 ):2097–2098, PMID: , 10.1021/es5002105.24476540
132. Schrimpe-Rutledge AC, Codreanu SG, Sherrod SD, McLean JA. 2016. Untargeted metabolomics strategies—challenges and emerging directions. J Am Soc Mass Spectrom 27 (12 ):1897–1905, PMID: , 10.1007/s13361-016-1469-y.27624161
133. Olsen GW, Burris JM, Ehresman DJ, Froehlich JW, Seacat AM, Butenhoff JL, et al. 2007. Half-life of serum elimination of perfluorooctanesulfonate, perfluorohexanesulfonate, and perfluorooctanoate in retired fluorochemical production workers. Environ Health Perspect 115 (9 ):1298–1305, PMID: , 10.1289/ehp.10009.17805419
134. Susmann HP, Schaider LA, Rodgers KM, Rudel RA. 2019. Dietary habits related to food packaging and population exposure to PFASs. Environ Health Perspect 127 (10 ):107003, PMID: , 10.1289/EHP4092.31596611
135. Forsthuber M, Kaiser AM, Granitzer S, Hassl I, Hengstschläger M, Stangl H, et al. 2020. Albumin is the major carrier protein for PFOS, PFOA, PFHxS, PFNA and PFDA in human plasma. Environ Int 137 :105324, PMID: , 10.1016/j.envint.2019.105324.32109724
136. Sagiv SK, Rifas-Shiman SL, Fleisch AF, Webster TF, Calafat AM, Ye X, et al. 2018. Early-pregnancy plasma concentrations of perfluoroalkyl substances and birth outcomes in Project Viva: confounded by pregnancy hemodynamics? Am J Epidemiol 187 (4 ):793–802, PMID: , 10.1093/aje/kwx332.29155920
137. Nyström J, Benskin JP, Plassmann M, Sandblom O, Glynn A, Lampa E, et al. 2022. Demographic, life-style and physiological determinants of serum per- and polyfluoroalkyl substance (PFAS) concentrations in a national cross-sectional survey of Swedish adolescents. Environ Res 208 :112674, PMID: , 10.1016/j.envres.2022.112674.34998808
138. Jain RB, Ducatman A. 2019. Perfluoroalkyl substances follow inverted U-shaped distributions across various stages of glomerular function: implications for future research. Environ Res 169 :476–482, PMID: , 10.1016/j.envres.2018.11.033.30530087
|
PMC009xxxxxx/PMC9945580.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36821708
EHP11072
10.1289/EHP11072
Research
Oral Exposure to Polystyrene Microplastics of Mice on a Normal or High-Fat Diet and Intestinal and Metabolic Outcomes
https://orcid.org/0000-0001-7269-1697
Okamura Takuro 1
https://orcid.org/0000-0002-8651-4445
Hamaguchi Masahide 1
Hasegawa Yuka 1
https://orcid.org/0000-0002-8794-0550
Hashimoto Yoshitaka 1
Majima Saori 1
Senmaru Takafumi 1
https://orcid.org/0000-0003-1031-4380
Ushigome Emi 1
Nakanishi Naoko 1
Asano Mai 1
Yamazaki Masahiro 1
Sasano Ryoichi 2
Nakanishi Yuki 3
Seno Hiroshi 3
Takano Hirohisa 4
Fukui Michiaki 1
1 Department of Endocrinology and Metabolism, Kyoto Prefectural University of Medicine, Graduate School of Medical Science, Kyoto, Japan
2 AiSTI Science Co., Ltd., Wakayama, Japan
3 Department of Gastroenterology and Hepatology, Kyoto University Graduate School of Medicine, Kyoto, Japan
4 Environmental Health Sciences, Graduate School of Global Environmental Studies, Kyoto University, Kyoto, Japan
Address correspondence to Michiaki Fukui, Department of Endocrinology and Metabolism, Kyoto Prefectural University of Medicine, Graduate School of Medical Science, 465 Kajii-cho, Kamigyo-ku, Kyoto-city, Kyoto 621-8585, Japan. Telephone: +81-75-251-5505; Fax: +81-75-252-3721. Email: [email protected]
22 2 2023
2 2023
131 2 02700608 2 2022
05 1 2023
13 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Microplastics (MPs) are small particles of plastic (≤5mm in diameter). In recent years, oral exposure to MPs in living organisms has been a cause of concern. Leaky gut syndrome (LGS), associated with a high-fat diet (HFD) in mice, can increase the entry of foreign substances into the body through the intestinal mucosa.
Objectives:
We aimed to evaluate the pathophysiology of intestinal outcomes associated with consuming a high-fat diet and simultaneous intake of MPs, focusing on endocrine and metabolic systems.
Methods:
C57BL6/J mice were fed a normal diet (ND) or HFD with or without polystyrene MP for 4 wk to investigate differences in glucose tolerance, intestinal permeability, gut microbiota, as well as metabolites in serum, feces, and liver.
Results:
In comparison with HFD mice, mice fed the HFD with MPs had higher blood glucose, serum lipid concentrations, and nonalcoholic fatty liver disease (NAFLD) activity scores. Permeability and goblet cell count of the small intestine (SI) in HFD-fed mice were higher and lower, respectively, than in ND-fed mice. There was no obvious difference in the number of inflammatory cells in the SI lamina propria between mice fed the ND and mice fed the ND with MP, but there were more inflammatory cells and fewer anti-inflammatory cells in mice fed the HFD with MPs in comparison with mice fed the HFD without MPs. The expression of genes related to inflammation, long-chain fatty acid transporter, and Na+/glucose cotransporter was significantly higher in mice fed the HFD with MPs than in mice fed the HFD without MPs. Furthermore, the genus Desulfovibrio was significantly more abundant in the intestines of mice fed the HFD with MPs in comparison with mice fed the HFD without MPs. Muc2 gene expression was decreased when palmitic acid and microplastics were added to the murine intestinal epithelial cell line MODE-K cells, and Muc2 gene expression was increased when IL-22 was added.
Discussion:
Our findings suggest that in this study, MP induced metabolic disturbances, such as diabetes and NAFLD, only in mice fed a high-fat diet. These findings suggest that LGS might have been triggered by HFD, causing MPs to be deposited in the intestinal mucosa, resulting in inflammation of the intestinal mucosal intrinsic layer and thereby altering nutrient absorption. These results highlight the need for reducing oral exposure to MPs through remedial environmental measures to improve metabolic disturbance under high-fat diet conditions. https://doi.org/10.1289/EHP11072
Supplemental Material is available online (https://doi.org/10.1289/EHP11072).
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Over the past few decades, the use of plastics has increased dramatically.1 The production, use, and consumption of plastics, which has been ongoing since the 1950s, has caused major environmental problems globally; in 1960, approximately 500,000 tons of production of synthetic fibers were emitted worldwide.2 The tonnage has since increased exponentially and was 348 million tons in 2017.3 Approximately 50% of the plastics produced annually is of the disposable type.4 Thus, the proliferation of plastics in the environment continues at an alarming rate. Plastic particles have been found to be persistent and ubiquitous pollutants in a variety of environments, including seawater, freshwater, soil, and air.5–9 Plastic accounts for 60%–80% of marine litter,10 which indicates that some of the plastic spilled due to inadequate management on land is washed up to the shore.
Microplastics (MPs) are small particles of plastic (≤5mm in diameter). They can be classified into two categories: primary and secondary. Primary MPs are small plastic beads designed for commercial use, such as in toothpaste, facial cleansers, cosmetics, and industrial abrasives.11 They are also used as raw materials in the manufacture of various commonly used plastic products. Secondary MPs are the plastics that have been discarded into the environment and are gradually degraded and disintegrated by external factors (especially ultraviolet rays and other environmental factors) into small fragments (≤5mm).12–14 Plastics have a stable polymeric structure; however, upon discharge into the ocean, they are degraded and miniaturized mainly by photolysis and thermal oxidative degradation.15 Human exposure to plastics has been considered to occur only through direct use for eating and drinking purposes.16,17 However, indirect exposure to additives with large hydrophobic properties, which is attributable to the miniaturization of plastics in the environment, and their uptake into the digestive system of organisms, dissolution into digestive juices containing oil, bioaccumulation, and the food chain are routes for the entry of MPs. Such indirect exposures and their routes of entry are considered to have the most significant impact on humans.18–20 From a toxicological perspective, the route of exposure (oral, respiratory, or dermal) is important, and oral exposure is the main route of uptake for MPs into the body. Exposure to polystyrene MPs (PS-MPs) reduced the reproductive capacity of oysters21 and induced the expression of antioxidant enzymes in rotifers.22 In addition to the global threat to environment from MP pollution, the issue of their potential toxicity to humans has raised serious concerns.23,24 In mammals, a pioneering study on PS-MPs was conducted on mice by Deng et al.25 who found that daily exposure to 5 or 20μm of fluorescent PS-MPs resulted in the accumulation of these particles in the liver, kidney, and intestine. Changes in metabolic profiles revealed that 5μm of PS-MPs affect energy metabolism, lipid metabolism, and oxidative stress in the liver of mice.25 PS is one of the major polymer types in plastic products, along with the accompanying wastes; PS-specific MPs were commonly found in MP fields.15,26,27 PS-MPs of 0.5–7μm in diameter have also been widely applied in bioassays to investigate biological interactions and toxicity in living organisms.28,29,18 Moreover, exposure of C57BL6/J mice to high concentrations of MPs was reported to increase the number and diversity of gut microbiota in the intestine. Inflammation and increased expression of TLR4, AP-1, and IRF5 was reported in the intestines of mice fed high concentrations of MPs.30
Dysbiosis has been known to cause thinning of the mucin layer and loosening of the tight junction in the mucosal epithelium of the small intestine, which allowed toxic substances in the intestinal tract to enter the body.31,32 In a study using mice, oral intake of MPs worsened dysbiosis.30 Moreover, mice fed a Western diet that includes a lot of fat, that is, a high-fat diet, developed dysbiosis and leaky gut syndrome (LGS).33,34 However, most of the papers on MP-induced metabolic disturbances have described experiments conducted under normal diet feeding conditions, and the reports conducted under high-fat diet feeding conditions have not been compared with normal diet feeding conditions.35 Thus, we hypothesized that the toxicity of MPs might be more marked in LGS induced by a high-fat diet than under a normal diet.
On the other hand, disruption of the mucus barrier has been previously reported to change the innate immunity of the intestinal tract.36 Among the cells involved in innate immunity, innate lymphoid cells (ILCs) are a type of lymphocyte that compose the T cell innate immune system and include ILC1, 2, and 3. They secrete cytokines that respond rapidly to pathogenic tissue damage and are postured to form subsequent adaptive immunity.37 Disruption of the mucus barrier alters the number of ILC3s, an important regulator of inflammation and infection in the mucosal barrier.38 ILC3-derived IL-22 maintained intestinal epithelial barrier function.39–42 On the other hand, in our previous study, we reported that intestinal ILC1 increased in inflammation of the intestinal tract caused by a high-fat diet.33,43,44 Taken together, to evaluate the inflammation caused by a high-fat diet and MPs in the intestine, it was necessary to evaluate ILC1 and 3 and associated M1/M2 macrophages in the intestinal mucosa.
We have previously shown that changes in diet alter metabolites, gene expression of nutrient transporters, and inflammatory cells in the upper small intestine, i.e., jejunum, in mice.33,43,44 Furthermore, because it is well known that nutrient absorption occurs primarily in the small intestine,45 we thought it more appropriate to evaluate the small intestine rather than the large intestine in this study to clarify the relationship between MPs and metabolic disturbances caused by HFD. Moreover, liver is directly connected to the intestinal tract through the portal blood flow, and portal influx of lipopolysaccharide and endotoxin due to impaired intestinal barrier function in the small intestine has been shown to contribute to the development of nonalcoholic fatty liver disease (NAFLD) and nonalcoholic steatohepatitis (NASH).46,47 Furthermore, because the metabolic syndrome was identified as a strong predictor of fatty liver disease,48 it is important to evaluate the liver in the study of endocrine and metabolic pathophysiology through the synergistic effects of MPs and HFD. In this study, we aimed to elucidate the pathophysiology of intestinal deterioration associated with consuming a Western diet and simultaneous intake of MPs, focusing on the endocrine and metabolic systems.
Materials and Methods
Mice
All experimental procedures were approved by the Committee for Animal Research, Kyoto Prefectural University of Medicine, Japan (approval number: M2021-61). We purchased 7-wk-old C57BL/6J (wild type) male mice from Shimizu Laboratory Supplies and kept them in a pathogen-free controlled environment. Littermate mice that were born in Shimizu Laboratory Supplies were used in the experiments. The mice were fed a normal diet (ND; 345 kcal/100g, fat kcal 4.6%; CLEA) or a high-fat diet [HFD; 459 kcal/100g, 20% protein, 20% carbohydrate, and 60% fat (lard); D12492, Research Diets Inc.] for 4 wk, starting at 8 wk of age, and equal amounts of feed and water were supplied for pair feeding. One mouse was kept per cage. In our previous studies, the group fed the normal diet with MPs added tended to have the lowest food and water intake among the four groups. Therefore, food and water intakes were measured every 3 d, and the food and water intakes of all mice were normalized based on the lowest intake (group fed a normal diet with MPs added). Moreover, in the MP exposure group, carboxyl group-modified fluorescent PS particles (F-K1 050; 0.45–0.53μm polystyrene COOH; Green Fluorescent Protein (GFP) fluorescence; Merck) were dissolved in water at 1,000μg/L, and water was provided ad libitum.49,50 Water with dissolved MPs was sonicated at 20 kHz for 15 min, and the jugs were wrapped in aluminum foil to shield them from light. The water was changed every 3 d. To verify whether MPs were uniformly dispersed in the water, we kept the animals in the cage for 3 d; collected 200μL of water from the jug on days 0, 1, and 2; counted the number of microplastics using a fluorescence microscope (BZ-X710; Keyence); measured the fluorescence intensity of the microplastic using an Orion L microplate luminometer (490 nm excitation, 520 nm emission) (Berthold Detection Systems); and repeated these procedures thrice. The mice were divided into four groups (n=10 per group) according to diet and whether or not they were given MPs. The body weights of the mice were measured weekly. After an overnight fast, the mice were euthanized at 12 wk of age by exposure to anesthesia (4.0mg/kg midazolam, 0.3mg/kg medetomidine, and 5.0mg/kg butorphanol)51 (Figure 1A).
Figure 1. Body weight, food, and water intake; and iPGTT and ITT in mice exposed to ND or HFD±MPs from 8 wk to 12 wk of age; intestinal permeability in mice exposed to ND or HFD±MPs at 12 wk of age. (A) Exposure to ND or HFD±MPs started at 8-weeks of age. (B) Body weight (n=10) and (C and D) intake of food and water over the course of the experiment (n=10). (E and F) Results of iPGTT (2g/kg body weight) for 12-wk-old mice and the AUC analysis (n=5). Blood glucose was monitored 0, 15, 30, 60, and 120 min after injection. (G and H) Results of ITT (0.75U/kg body weight) for 12-wk-old mice and the AUC analysis (n=5). Blood glucose was monitored 0, 15, 30, 60, and 120 min after injection. (I) GFP signals from GFP-MP before and after oral gavage of FITC-Dextran for 12-wk-old mice (n=10). Normalized to the levels of ND-fed mice. Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Summary data can be found in Table S2. Note: ANOVA, analysis of variance; a.u., arbitrary unit; AUC, area under the curve; FITC, fluorescein isothiocyanate; GFP, green fluorescent protein; HFD, high-fat diet; iPGTT, intraperitoneal glucose tolerance test; ITT, insulin tolerance test; min, minutes; MPs, microplastics; ND, normal diet; SD, standard deviation.
Figure 1A is an illustration that depicts mice exposed to a normal diet, a normal diet plus microplastics, a high-fat diet, and a high-fat diet plus microplastics from 8 weeks of age until 12 weeks of age. Figures 1B to 1D are line graphs, plotting Body weight (gram), ranging from 15 to 35 in increments of 5; Oral intake (gram per day), ranging from 2.4 to 3.0 in increments of 0.1; and Water intake (gram per day), ranging from 6.4 to 7.4 in increments of 0.2 (y-axis) across Time (week), ranging from 8 to 12 in unit increments; 9 to 12 in unit increments; and 9 to 12 in unit increments (x-axis) for normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics. Figure 1E and 1G are line graphs, plotting blood glucose (milligram per deciliter), ranging from 0 to 600 in increments of 200 and 0 to 400 in increments of 100 (y-axis) across Time (minutes), ranging from 0 to 30 in increments of 15; 30 to 60 in increments of 30; and 60 to 120 in increments of 60 (x-axis) for normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics. Figures 1F and 1H are bar graphs, plotting area under the curve (milligram per deciliter asterisk minutes), ranging as 0, 1 times 10 begin superscript 4 end superscript, 2 times 10 begin superscript 4 end superscript, 3 times 10 begin superscript 4 end superscript, 4 times 10 begin superscript 4 end superscript (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis). Figure 1I is a graph, plotting Fluorescence (astronomical unit), ranging from 0 to 3 in unit increments (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis) for before and after.
Analytical Procedures and Glucose and Insulin Tolerance Tests
Twelve-week-old mice were subjected to intraperitoneal glucose tolerance testing (iPGTT) (2g/kg of body weight) after a 16-h fast (2 d before sacrifice) and insulin tolerance testing (ITT) (0.5U/kg body weight) after a 5-h fast (3 d before sacrifice). Blood was sampled via the tail vein. Blood glucose was measured using a glucometer (Gultest mintII; Sanwa Kagaku Kenkyusho). iPGTT and ITT were performed in different mice. Blood glucose was monitored 0, 15, 30, 60, and 120 min after injection. The area under the curve (AUC) of the iPGTT and ITT results was analyzed (n=5).
Measurement of Intestinal Permeability
Fluorescein isothiocyanate (FITC)-labeled dextran solution (25mg/mL, 20mL/kg per mouse) (Chondrex, Inc.) was administered via oral gavage at 12 wk of age after a 4-h fast (5 d before sacrifice), and blood was collected by retro-orbital puncture before and 3 h after gavage. The collected blood was immediately subjected to density gradient centrifugation at 1,500×g for 15 min. Plasma was collected and diluted with phosphate-buffered saline (PBS) at a ratio of 1:2. Plasma dextran levels were estimated from luminescence measurements using an Orion L microplate luminometer (490 nm excitation, 520 nm emission) (Berthold Detection Systems) (n=10). It was difficult to separate the FITC signal from FITC-dextran and GFP signal from GFP-MP using absorbance spectrometry. Thus, we measured the GFP signal in serum before administration and FITC-Dextran signal 3 h after administration. We used the plasma signal after FITC-Dextran administration minus the plasma signal before FITC-Dextran administration as a surrogate index of intestinal permeability.
Blood Biochemistry
Blood samples were taken from fasted mice by cardiac puncture during euthanasia, and the serum samples were collected after centrifugation at 14,000 rpm for 10 min at 4°C. The collected serum was stored at −30 °C until they were mailed to the subcontractor. The levels of alanine aminotransferase (ALT) were measured via the standardization support method described by the Japanese Society for Clinical Chemistry.52 ALT catalyzes the transfer reaction between the α-keto group of α-ketoglutarate and the amino group of L-alanine to produce pyruvate and glutamate. Conjugated to this reaction, lactate dehydrogenase converts β-nicotinamide adenine dinucleotide reduced form (β-NADH) to β-nicotinamide adenine dinucleotide oxidized form (β-NAD) in the presence of the generated pyruvic acid. The rate of decrease of β-NADH at that time was measured at a wavelength of 330–350 nm to obtain the ALT activity value. Triglyceride (TG),52 nonesterified fatty acid (NEFA),53 and low-density lipoprotein (LDL)-cholesterol levels54 were measured via enzymatic methods [TG; glycerol kinase (GK)-GPO·glycerol blanking method; NEFA; Acyl-CoA synthetase; acyl-CoA oxidase-3; methyl-N-ethyl-N-(beta-Hydroxyethyl)aniline; LDL-cholesterol, cholesterol oxidase]. Free glycerol in the sample is eliminated by the reaction of GK and adenosine-5′-diphosphate (ADP), which is generated simultaneously, and is converted to adenosine-5′-triphosphate (ADP) by pyruvate kinase and phosphoenolpyruvate (PEP) and to adenosine triphosphate (ATP) by pyruvate kinase (PK) and PEP. This elimination reaction removes the effect of free glycerol. TG in the sample is then hydrolyzed to glycerol and fatty acids by lipoprotein lipase, and in the presence of ATP, glycerol-3-phosphate is generated from glycerol by the action of GK, simultaneously producing ADP. In the presence of ADP and glucose, glucose-6-phosphate and adenosine-5′-monophosphate are generated by the action of ADP-dependent hexokinase. G-6-P is further converted to 6-phosphogluconate by the action of glucose-6-phosphate dehydrogenase in the presence of β-NAD. The TG concentration is determined by measuring the increase in the amount of β-NADH produced at the same time at a wavelength of 330–350 nm. In the presence of coenzyme A (CoA) and ATP, NEFA in the sample generates acyl-CoA, AMP, and pyrophosphate by the action of acyl-CoA synthase (ACS). The generated acyl-CoA is oxidized by the action of acyl-CoA oxidase, simultaneously generating 2,3-trans-enoyl-CoA and hydrogen peroxide. The generated hydrogen peroxide quantitatively oxidizes and condenses MEHA and 4-aminoantipyrine by the action of peroxidase to produce a blue-violet dye. The NEFA concentration in the sample is determined by measuring the absorbance of this blue-violet color. LDL and calixarene sulfate form a soluble complex and stabilize LDL. The soluble complex is then degraded by intestinal lamina propria (LPL) in the presence of cholic acid. The free ester and free cholesterol are then oxidized in the presence of β-NAD, cholesterol dehydrogenase, and LPL. The amount of β-NADH produced during this process is measured by measuring the absorbance at a wavelength of 330 nm to 350 nm to determine the concentration of LDL-cholesterol in the sample. The biochemical examinations were performed at the FUJIFILM Wako Pure 18 Chemical Corporation (n=10).
Measurement of Free Fatty Acids in Serum, Feces, and Liver Tissues
Collected samples were stored at −30 °C until being used in experiments. Serum (25μL) obtained via cardiac puncture during euthanasia, feces from the small intestine (15μg), and liver tissue (15μg) samples were used for measuring free fatty acids. A fatty acid methylation kit (Nacalai Tesque) was used to analyze the methylation of samples. Gas chromatography–mass spectrometry (GC-MS) was performed using an Agilent 7890B/7000D system (Agilent Technologies) to measure palmitic acid levels in murine sera, feces, and liver tissues (n=10). The final product was loaded onto a Varian capillary column (DB-FATWAX UI; Agilent Technologies). The capillary column used for fatty acid separation was CP-Sil 88 for FAME [100m×0.25mm (inner diameter)×0.20μm (membrane thickness); Agilent Technologies]. The column was maintained at 100°C for 4 min, and the temperature was then increased gradually by 3°C/min to 240°C and held for 7 min. The sample was injected in split mode with a split ratio of 5:1. Each fatty acid methyl ester was detected in the selected ion-monitoring mode. All the results were normalized to the peak height for the C17:0 internal standard.55
Histology and Immunochemical Analysis of the Small Intestine
Small intestines removed from mice were immediately fixed in 10% buffered formaldehyde for 24 h at 22°C, embedded in paraffin, cut into 4μm-thick sections, and stained with hematoxylin and eosin (H&E) stain and periodic acid Schiff (PAS) stain with Carnoy’s solution. Two pieces of small intestine (one for H&E and one for PAS) were used from each mouse. Images were captured using a fluorescence microscope (BZ-X710; Keyence). The villus height/width and crypt depth were visualized on the H&E–stained sections and were measured at 5 locations per slide for each group of 10 animals using ImageJ software (version 1.53 k; National Institutes of Health). Mucin grains and goblet cells (PAS+) were enumerated and reported as the average number of goblet cells (PAS+) per 10 crypts using the ImageJ software, as reported previously (n=10).56 Moreover, the amount of MP deposition in the intestinal mucosa was evaluated based on the area represented by the green fluorescent protein (GFP)+ region in the slice, and the ratio of the GFP+ region to the total intestinal epithelial region was calculated (469 nm excitation, 525 nm emission) (n=10). Because the fluorescent dye was stable in the body for 4 d57 and has been shown to retain up to 50% fluorescence intensity for up to 3 d under ethanol dialysis conditions,58 small intestine pathology was evaluated 2 d after sacrifice. For immunochemistry, sections of the small intestine were prepared and blocked with Blocking One Histo (Nacalai Tesque, Inc.) for 30 min. Subsequently, the samples were stained with monoclonal primary anti-Muc2 (1/500 dilution; ab272692, Abcam)59 at 4°C overnight. After washing for 5 min in PBS, the samples were subsequently stained with a Texas-red-conjugated antimouse secondary antibody (1/1,000 dilution; Jackson ImmunoResearch) in the dark at room temperature for 60 min. After washing for 5 min in PBS, nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI; Sigma-Aldrich). Images were captured using the BZ-X710 fluorescence microscope (540 nm excitation, 605 nm emission) (Keyence), and the fluorescence intensity of myotube cells was analyzed using ImageJ software (n=10).
Isolation of Mononuclear Cells from the Small Intestine of Mice
To prevent blood contamination in the small intestine, systemic perfusion with heparinized saline was performed before harvesting or washing the tissue with PBS. Samples were stored in cold 2% Fetal Bovine Serum (FBS) with Roswell Park Memorial Institute (RPMI) until being used for experiments. The following experiments were performed on the day of euthanasia. LPL mononuclear cells were isolated using the Lamina Propria Dissociation Kit (130-097-410; Miltenyi Biotec) following the manufacturer’s instructions. Cell pellets were resuspended in 5mL of 40% Percoll® and slowly poured to the upper portion of centrifuge tubes, which contained a bottom layer of 5mL of 80% Percoll®. Density gradient centrifugation (420×g, 20 min) was performed, and mononuclear cells in the middle layer were gently extracted with a 1-mL PIPETMAN. The extracted mononuclear cells were washed twice with 2% FBS/PBS.
Tissue Preparation and Flow Cytometry
The cell suspension obtained in the previous section was preincubated with Mouse BD Block purified antimouse CD16/CD32 mAb (394,656; clone: 2.4G2; 1/100; BD Biosciences) for 10 min at 22°C. The following antibodies were used for the gating of the innate lymphoid cells. Cell suspensions were incubated with a mixture of Biotin-CD3e (100,304; clone: 145-2C11; 1/200; eBioscience, Inc.), Biotin-CD45R/B220 (103,204; clone: RA3–6B2; 1/200; eBioscience, Inc.), Biotin-Gr-1 (108,404; clone: RB6-8C5; 1/200; eBioscience), Biotin-CD11c (117,304; clone: N418; 1/200; eBioscience, Inc.), Biotin-CD11b (101,204; clone: M1/70; 1/200; eBioscience, Inc.), Biotin-Ter119 (116,204; clone: TER-119; 1/200; eBioscience, Inc.), Biotin-FceRIa (134,304; clone: MAR-1; 1/200; eBioscience, Inc.), Brilliant Violet 510-Streptavidin (405,233; 1/500; eBioscience, Inc.), PE-Cy7-CD127 (135,014; clone: A7R34; 1/100; eBioscience, Inc.), Pacific Blue-CD45 (103,116; clone: 30-F11; 1/100; eBioscience, Inc.), and Fixable Viability Dye eFluor 780 (1/400; eBioscience, Inc.) for 20 min at 4°C. The cell suspension was washed twice with 2% FBS/PBS and fixed with fixation buffer (420,801; BioLegend, Inc.) for 30 min. After washing with 2% FBS/PBS, the cell suspension was incubated with the mixture of PE-GATA-3 (clone TWAJ, 1/50; eBioscience, Inc.), APC-RORγ (clone AFKJS-9, 1/50, eBioscience, Inc.), and FITC-T-bet (clone 4B10, 1/50; BioLegend, Inc.)60,61 (Figure S1). We used the following antibodies for gating of M1 and M2 macrophages: FITC-CD206 (MA516870; clone: MR5D3, 1/50, eBioscience, Inc.), PE-F4/80 (12,480,182; clone: BM8, 1/50, eBioscience, Inc.), APC- CD45.2 (17,045,482; clone: 104, 1/50; eBioscience, Inc.), PE-Cy7-CD11c (25,011,482; clone: N418, 1/50, eBioscience, Inc.), and APC-Cy7-CD11b (47,011,282; clone: M1/70, 1/50; eBioscience, Inc.)62 (Figure S2). We analyzed the stained cells using flow cytometry Canto II, and the data were analyzed using FlowJo (version 10; TreeStar, Inc.) (n=10).
Measurement of Short-Chain Fatty Acids (SCFAs) Levels in the Feces Samples
Samples obtained during euthanasia from the small intestines of mice were stored at −30 °C until being used in experiments. The levels of SCFAs in the feces were analyzed using GC/MS on an Agilent 7890B/7000D system (Agilent Technologies). We homogenized the rectal fecal samples (20mg) in 500μL of acetonitrile and 500μL of distilled water by grinding them (4,000 rpm for 2 min) in a ball mill. We then shook the samples at 1,000 rpm for 30 min at 37°C and centrifuged them at 14,000 rpm for 3 min at room temperature. We separated the supernatant (500μL), added it to 500μL of acetonitrile, and shook the mixture at 1,000 rpm for 3 min at 37°C. After centrifugation at 14,000 rpm for 3 min at room temperature, the pH was adjusted to 8 with 0.1 mol/L NaOH, and SCFAs were extracted.
SCFA concentration measurements were automatically measured via GC-MS and an online solid-phase extraction (SPE) method (SGI-M100 SPE-GC system; AiSTI Science Co., Ltd.). After the vials were filled with samples and placed in the autosampler tray, injection into the SPE and GC-MS systems occurred automatically. Solid stratification was performed using Flash-SPE ACXs (AiSTI Science Co., Ltd.). Fifty microliter aliquots of each sample extract were loaded onto the solid phase and rinsed with a 1:1 mixture of water and acetonitrile. Subsequently, the products were dehydrated with acetone, saturated with 4μL N-tert-butyldimethylsilyl-N-methyltrifluoroacetamide (MTBSTFA)-toluene solution (1:3), and eluted with hexane after derivatization in the solid phase. Injection of the final product was performed with a programmed temperature vaporizer injector LVI-S250 (AiSTI Science Co., Ltd.). The temperature was maintained at 150°C for 0.5 min, increased gradually to 290°C at a rate of 25°C/min, and then maintained at this temperature for 16 min. A Vf-5ms capillary column [30m×0.25mm (inner diameter)×0.25μm (membrane thickness); Agilent Technologies] was used. The column was maintained at 60°C for 3 min, then gradually increased to 100°C at 10°C/min, then to 310°C at 20°C/min, and finally maintained at 310°C for 7 min. The sample was injected in split mode at a ratio of 20:1. Each SCFA was detected in the scan mode (m/z: 70–470). All SCFA results were standardized using the height of the peak of tetradeuteroacetic acid (0.02 nmol/μL)44 (n=10).
Liver Histology
Liver tissue was obtained, immediately fixed with 10% buffered formaldehyde, and embedded in paraffin. After 24 h of fixation at room temperature, liver sections (4μm thick) were prepared, and H&E stained. We prepared an Oil Red O stock solution in isopropanol (0.25g/100mL) and heated it to 100°C for 10 min. We fixed liver sections with 4% paraformaldehyde for 30 min and then rinsed with PBS. We then prepared a 60% Oil Red O stock solution diluted with distilled water and immersed the sections in the solution for 30 min. We rinsed the stained sections in PBS until the background became clear. We captured the pictures of the liver sections using a BZ-X710 fluorescence microscope (Keyence). Additionally, to assess the severity of NAFLD, we determined the NAFLD activity score (NAS),63 which is a well-known standard used for measuring the severity of nonalcoholic steatohepatitis (NASH) and changes in NAFLD. The scoring system comprised 14 histological features, of which 4 [steatosis (0–3), lobular inflammation (0–2), hepatocellular ballooning (0–2), and fibrosis (0–4)] were evaluated semiquantitatively (n=10).
Gene Expression Analysis in Murine Jejunum and Liver
The jejunum and liver of mice fasted for 16 h were excised and instantly frozen in liquid nitrogen. The samples were homogenized in ice-cold QIAzol Lysis reagent (Qiagen) at 4,000 rpm for 2 min in a ball mill, and total RNA was extracted according to the manufacturer’s instructions and quantified and qualified using the Qubit RNA Assay (Invitrogen). A High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) was used for reverse transcription of the total RNA (0.5μg) to first-strand cDNA, according to the manufacturer’s instructions. The mRNA expression levels of Tnfa, Il6, Il1b, Il22, Ffar2, Ffar3, Cd36, Sglt2, and Muc2 in the jejunum and of Tnfa, Il6, Il1b, Scd1, Elovl6, and Fasn in the liver were quantified via real-time reverse transcription-polymerase chain reaction (RT-PCR), and TaqMan Fast Advanced Master Mix (Applied Biosystems) was used according to the manufacturer’s instructions. The PCR conditions were as follows: 1 cycle of 2 min at 50°C and 20 s at 95°C, followed by 40 cycles of 1 s at 95°C and 20 s at 60°C. The relative expression levels of each target gene were normalized to the Gapdh threshold cycle (Ct) values and quantified via the comparative threshold cycle 2-ΔΔCt method. Signals from ND-fed mice were assigned a relative value of 1.0. Expression levels from six mice from each group were determined, and RT-PCR was performed in triplicate for each sample (n=10). Primer sequences (TaqMan probe primers; Applied Biosystems) for each of the genes are presented in Table S1.
16s rRNA Sequencing
A QIAamp DNA Feces Mini Kit (Qiagen) was used for extracting microbial DNA from frozen appendicular fecal samples to ensure sufficient feces volume according to the manufacturer’s instructions. The V3-V4 region of the 16S rRNA gene was amplified from DNA using a bacterial universal primer set (341F and 806R). PCR was performed using EF-Taq (Solgent) with 20 ng of genomic DNA as a template in a 30μL reaction mixture with the following cycles: 95°C for 2 min for activating Taq polymerase, followed by 35 cycles at 95°C, 55°C, and 72°C for 1 min each, and a final 10-min step at 72°C. The amplification products were purified using a multiscreen filter plate (Millipore Corp.). According to the manufacturer’s instructions (Macrogen), a MiSeq sequencer (Illumina) was used for 16S rRNA sequencing. For sequence quality filtering, we used QIIME (version 1.9.1).64 Barcodes or primers with scores of <75% were excluded from the files. Using the UCLUST algorithm at 97% similarity, the number of operational taxonomic units (OTUs) was determined.65 In addition, BLAST (UNITE 2017) was used for taxonomic assignment of 16S rRNAs with the UNITE sequence set of the Greengenes core set aligned with UCLUST and ITS.
The relative abundances of phyla in the groups were evaluated by one-way analysis of variance (ANOVA) with Holm–Šídák’s multiple comparisons test. In addition, alpha diversity (defined as the diversity within an individual sample) was analyzed using the Chao1,66 Shannon,67 and Gini–Simpson indices.68
The relative abundances of bacterial genera between the groups were evaluated by linear discriminant analysis (LDA) coupled with effect size measurements (LEfSe) (http://huttenhower.sph.harvard.edu/lefse/).69 With a normalized relative abundance matrix, LEfSe showed taxa with significantly different abundances, and the effect size of the feature was estimated via LDA. A p-value threshold of 0.05 (Wilcoxon rank sum test) and an effect size threshold of 2 were used for all biomarkers discussed in this study.
MODE-K Cell Experiments
The murine intestinal epithelial MODE-K cell line (sex: female) was kindly provided by Prof. Richard Blumberg (Harvard Medical School, Boston, USA). MODE-K cells were cultured in Dulbecco’s modified Eagle’s medium (Nacalai Tesque) supplemented with 10% FBS, 2 mM of glutamine, 100U/ml penicillin, and 100μg/mL streptomycin, in an atmosphere of 95% air and 5% CO2.70 Cells were spread in 96-well plates at 1×104 cells/well, and 1,000μg/L of MP, 200μM of palmitic acid (PA),33 and 50 ng/mL of recombinant mouse IL-22 protein (NBP2-35122, Funakoshi) were added for 24 h on day 5.71 After 24 h, the cells were then washed twice with cold PBS, detached with 2.5g/l-Trypsin/1 mmol/l-EDTA solution (Nacalai Tesque), and centrifuged at 300×g for 5 min at 4°C, and the supernatant was discarded. RT-PCR was performed as described in the “Gene Expression Analysis in Murine Jejunum and Liver” section, and the mRNA expression level of Muc2 was quantified (n=6). To evaluate the intracellular accumulation of MPs, the pellet was counted after centrifugation. Briefly, 1×104 MPs were fractionated and then diluted into 200μL of PBS. The cell lysate was transferred to a black 96-well plate, and luminescence was measured using an Orion L microplate luminometer (Berthold Detection Systems) (n=6). Signals from cells without MPs, PAs, and IL-22 were assigned a relative value of 1.0.
Statistical Analysis
The data were analyzed using JMP software (version 13.0; SAS Institute, Inc.). One-way ANOVA with Holm–Šídák’s multiple comparisons test was used to compare the results of different groups. Statistical significance was set at p<0.05. Figures were generated using GraphPad Prism (version 9.0; GraphPad Software Inc.).
Results
Verification of MP Uniformity in Water
First, to verify whether MPs were uniformly dispersed in the water, we kept the animals in the cage for 3 d with a jug of water, collected 200μL of water from the jug on days 0, 1, and 2, counted the number of MPs using a fluorescence microscope, and repeated the procedure three times. There were no statistically significant differences in the number and fluorescence intensity of MPs at the three time points, and we considered that MPs were sufficiently available for consumption without fluorescence intensity diminishing or precipitation even under ad libitum conditions (Figure S3).
Body Weight, Food and Water Intake, and iPGTT and ITT in Mice Exposed to ND or HFD with and without MPs from 8 Wk to 12 Wk of Age
Body weight and intake of food and water were measured. We observed no difference in the body weight of mice fed a ND and a ND with MPs added and of mice fed a HFD and a HFD with MPs added (Figure 1B). Intake of food and water was not different among the four groups (Figure 1C and 1D). Glucose tolerance was assessed by iPGTT and ITT. In iPGTT, blood glucose levels were higher in mice fed a ND with MPs added and mice fed HFD with MPs added than in both ND and HFD mice (Figure 1E and 1F). On the contrary, in ITT, blood glucose levels were not different between ND and mice fed a ND with MPs added, whereas blood glucose levels in mice fed mice fed a HFD with MPs added were higher than those in mice fed a HFD (Figure 1G and 1H).
Intestinal Permeability in Mice Exposed to ND or High-Fat Diet with and without MPs at 12 Wk of Age
Intestinal permeability was assessed using fluorescent-stained dextran. FITC-labeled dextran was administered orally, and blood was collected before the administration of dextran and 3 h later to measure the plasma GFP-MP and dextran levels as a proxy for intestinal permeability. Because of the interference in signal detection of the GFP-MP accumulation in these mice, we assumed the difference in signal intensity before and 3 h after administration of FITC-dextran represented intestinal permeability and absorbance. There was no difference in GFP signal between those recorded before and after the administration in the group fed a ND and group fed a ND with MPs, whereas GFP signal after the administration was significantly higher in the group fed an HFD and the group fed an HFD containing MPs in comparison with the signal before the administration (Figure 1I). In addition, the GFP signal was higher in the HFD-fed mice than in the ND-fed mice. There was no significant difference in plasma GFP expression between the ND-fed group and the group fed the ND containing MPs. In contrast, plasma GFP expression was significantly higher in the mice fed an HFD that included MPs than in HFD-fed mice fed an HFD that did not contain MPs (Figure S3D and S3E).
The Serum Levels of Liver Enzymes and Lipids in Mice Exposed to a ND or an HFD with and without MPs at 12 Wk of Age
Next, the serum levels of hepatic enzymes and lipids were investigated. Serum ALT, TG, NEFA, LDL-cholesterol, and palmitic acid levels had no obvious differences between mice fed the ND and the ND containing MPs, whereas the serum levels in mice fed the HFD containing MPs were significantly higher in HFD-fed mice (Figure S4A–E).
Histological Evaluation of Jejunum and Immune Cells Involved in Innate Immunity in LPL of Small Intestine
Histological evaluation of the jejunum was performed. Representative histological images of the jejunum are shown in Figure 2A. We found no obvious differences in the height and width of villi between mice fed ND and ND containing MPs; on the other hand, these parameters in mice fed HFD containing MPs were significantly lower than those in the HFD mice (Figure 2B,C). In addition, exposure to MPs resulted in no apparent difference in crypt depth in ND-fed mice, but crypt depth was significantly higher in mice fed HFD that included MPs than in HFD mice (Figure 2D). The total number of goblet cells was counted in the PAS-stained images. The number of goblet cells in mice fed the ND containing MPs and the those fed HFD containing MPs was significantly lower than in ND and HFD mice, respectively (Figure 2E), and mucin layer thickness analyzed by immunochemical staining for Muc2 was smaller in mice fed the ND containing MPs and mice fed the HFD containing MPs in comparison with ND- and HFD-fed mice, respectively (Figure 2F). Moreover, the amount of MP deposition in the intestinal mucosa was calculated based on the area represented by the GFP-positive region in the image. The ratio of GFP-positive region to epithelial region in ND-fed mice did not differ, with or without MP exposure, whereas the ratio of GFP-positive regions was significantly higher in Mice fed HFD containing MPs in comparison with HFD-fed mice (Figure 2G).
Figure 2. Histological evaluation of jejunum and immune cells involved in innate immunity in LPL of small intestine. (A) Representative images of HE- and PAS-stained, GFP-positive, and Muc2-immunostained jejunum sections. Jejunum tissue was collected at 12 wk of age. In the GFP fluorescence image, MPs are enlarged and indicated by arrows. The scale bars show 100μm (50μm for Muc2 image). (B) Villus height (n=10). (C) Villus width (n=10). (D) Crypt depth (n=10). (E) Total goblet cells/area (mm2 of jejunum) (n=10). (F) Mucus layer thickness (n=10). (G) GFP-positive area (n=10). Ratio of (H) ILC1s to CD45-positive cells, (I) T-bet positive ILC3s to CD45-positive cells, (J) M1 macrophages to M2 macrophages in the small intestine, and (K) ILC3s to CD45-positive cells (n=10 in each case). Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. Summary data can be found in Table S2. *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Note: ANOVA, analysis of variance; GFP, green fluorescent protein; H&E, hematoxylin and eosin; HFD, high-fat diet; ILCs, innate lymphoid cells; MPs, microplastics; ND, normal diet; PAS, periodic acid Schiff; SD, standard deviation.
Figure 2A is a stained tissue displays hematoxylin and eosin, periodic acid-Schiff stained, green fluorescent protein, and Muc2-immunostained jejunum sections (columns) and normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (rows). Figures 2B to 2K are bar graphs, plotting villus height (microgram), ranging from 0 to 800 in increments of 200; villus width (micrometer), ranging 0 to 250 in increments of 50; Crypt depth (micrometer), ranging from 0 to 250 in increments of 50; total goblet cells or area (millimeter squared of jejunum), ranging from 0 to 500 in increments of 100; mucus thickness (micrometer), ranging from 0 to 30 in increments of 10; green fluorescent protein positive area (millimeter cubed), ranging from 0 to 3 in unit increments; 1 innate lymphoid cells per C D 45 plus cells in small intestine, ranging from 0.0 to 2.5 in increments of 0.5; T-bet positive 3 innate lymphoid cells per C D 45 plus cells in small intestine, ranging from 0.0 to 2.0 in increments of 0.5; M1 or M2 macrophages uppercase phi ratio in small intestine, ranging from 0.0 to 2.5 in increments of 0.5; 3 innate lymphoid cells per C D 45 plus cells in small intestine, ranging 0.0 to 2.0 in increments of 0.5 (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis).
Next, the number of cells involved in innate immunity in the LPL of the small intestine was measured by flow cytometry.33,38,44,72–76 The ratio of ILC1s and T-bet positive ILC3s in CD45+ cells and the ratio of M1 macrophages in M2 macrophages of ND mice were not significantly different from those of mice fed the ND containing MPs; on the other hand, those of mice fed the HFD with MPs were significantly higher than those of HFD-fed mice (Figure 2H,I, and J). Furthermore, there was no significant difference in the ratio of ILC3 cells between mice fed the ND and mice fed the ND containing MPs, but the ratio was lower in the mice fed the HFD containing MPs in comparison with mice fed the HFD that did not contain MPs (Figure 2K).
Metabolites in Feces of Mice Exposed to ND or HFD with and without MPs at 12 Wk of Age
Next, the concentration of the selected metabolites in the feces in the small intestine was determined. Palmitic acid concentrations in feces did not differ between mice fed the ND and mice fed the ND containing MPs, whereas they were lower in the feces of mice fed the HFD containing MPs in comparison with mice fed the HFD that did not contain MPs (Figure 3A). It is notable that the concentration of palmitic acid was considerably higher in mice fed the HFD than in the ND-fed mice, whereas mice fed the HFD containing MPs exhibited levels similar to the ND and mice fed the ND containing MPs. Furthermore, the concentrations of SCFAs such as acetic acid, propanoic acid, and butanoic acid did not differ between ND mice and mice fed the ND containing MPs; on the other hand, they were significantly lower in the feces of mice fed the HFD containing MPs in comparison with HFD-fed mice (Figure 3B–D).
Figure 3. Metabolites in feces of mice exposed to ND or HFD±MPs at 12 wk of age. The concentrations of (A) palmitic acid, (B) acetic acid, (C) propanoic acid, and (D) butanoic acid in the feces (n=10). Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. Summary data can be found in Table S2. **p<0.01 and ****p<0.0001. Note: ANOVA, analysis of variance; HFD, high-fat diet; MPs, microplastics; ND, normal diet; SD, standard deviation.
Figures 3A to 3D are bar graphs, plotting Palmitic acid concentration in feces (nanomole per microgram), ranging from 0 to 8 in increments of 2; Acetic acid concentration in feces (nanomole per microgram), ranging from 0 to 5 in unit increments; Propanoic acid concentration in feces (nanomole per microgram), ranging from 0.0 to 2.0 in increments of 0.5; and Butanoic acid concentration in feces (nanomole per microgram), ranging from 0.0 to 0.6 in increments of 0.2 (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis).
Histological Evaluation of Liver and Palmitic Acid Concentration in Liver of Mice Exposed to the ND or the HFD containing MPs at 12 Wk of Age
The effects of MP on the liver were evaluated. Wet weight of liver was not different between the mice fed the ND and mice fed the ND containing MPs, whereas the weight of mice fed the HFD containing MPs was lower than that of HFD mice (Figure 4A). Although showing a similar trend, the ratio of liver weight to body weight did not differ significantly between ND- and HFD-fed mice, with or without exposure to MPs (Figure 4B). The representative histological images of the liver were shown in Figure 4C. The NAS was zero in ND-fed mice, with and without exposure to MPs, whereas NAS in mice fed the HFD containing MPs was higher than that in mice fed the HFD without MPs (Figure 4D). The area of Oil Red O–stained region was not different between ND-fed mice, with or without exposure to MPs, whereas the area in mice fed the HFD containing MPs was higher than that of HFD-fed mice (Figure 4E). Moreover, palmitic acid concentration in the liver was not different between ND-fed mice, with and without exposure to MPs, but the concentration in mice fed the HFD containing MPs was higher than that in HFD-fed mice (Figure 4F).
Figure 4. Histological evaluation of liver and palmitic acid concentration in liver of mice exposed to ND or HFD±MPs at 12 wk of age. (A and B) Absolute and relative liver weight (n=10). (C) Representative images of HE- and Oil Red O–stained liver sections. Liver tissue was collected at 12 wk of age. The scale bars show 100μm. (D) Nonalcoholic fatty liver disease (NAFLD) activity scores (n=10). (E) Area of Oil Red O–stained region (n=10). (F) The concentration of palmitic acid in the liver (n=10). Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. Summary data can be found in Table S2. *p<0.05, **p<0.01, and ****p<0.0001. Note: HE, hematoxylin & eosin; HFD, high-fat diet; MPs, microplastics; ND, normal diet; SD, standard deviation.
Figures 4A, 4B, 4D, 4E, 4F are bar graphs, plotting Liver weight (milligram), ranging from 0.0 to 2.0 in increments of 0.5; Liver weight or Body weight ratio, ranging from 0.00 to 0.08 in increments of 0.02; Nonalcoholic fatty liver disease activity score, ranging from 0 to 6 in increments of 2; Area of oil red O staining (micrometer cubed), ranging as 0.0, 5.0 times 10 begin superscript 4 end superscript, 1.0 times 10 begin superscript 5 end superscript, 1.5 times 10 begin superscript 5 end superscript, 2.0 times 10 begin superscript 5 end superscript, 2.5 times 10 begin superscript 5 end superscript; Palmitic acid concentration in liver (nanomole per microgram), ranging from 0 to 15 in increments of 5 (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis). Figure 4C is a stained tissue displays hematoxylin and eosin and Oil red O-stained liver sections (columns) and normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (rows).
Gene Expression in the Small Intestine of Mice Exposed to ND or to the HFD Containing MPs at 12 Wk of Age
Gene expression in small intestine was investigated (Table 1). The relative expression of genes related to inflammation, such as Tnfa, Il6, and Il1b, in small intestine of mice fed the HFD with MPs was higher than that of mice fed the HFD without MPs. The relative expression of Il22, a cytokine produced by ILC3, which acts on intestinal epithelial cells and embryonic cells to promote the secretion of antimicrobial peptides and mucus and is involved in the intestinal barrier function,77 in small intestine of mice fed the HFD with MPs was significantly lower than that of HFD mice. Moreover, the relative expression of Ffar2 and Ffar3, which are free fatty acid receptors that use SCFAs, such as acetic acid, propanoic acid, and butanoic acid as ligands,78,79 in small intestine of mice fed the HFD with MPs was higher than that of mice fed the HFD without MPs. The relative expression of Cd36, which is a long-chain fatty acid transporter, in small intestine of mice fed the HFD with MPs was higher than that of mice fed the HFD without MPs. It is interesting to note that the relative expression of Sglt1, a Na+/glucose cotransporter, in small intestine of mice fed the HFD with MPs was higher than that of mice fed the HFD without MPs. On the other hand, their expressions were not different between ND-fed mice and mice fed the ND with MPs (Table 1).
Table 1 Gene expression of small intestine and liver.
Organ Gene ND ND+MP HFD HFD+MP
Small intestine Tnfa 1 (0.09) 1.07 (0.04) 1.66 (0.15)*,† 2.53 (0.04)*,†,‡
Il6 1 (0.18) 0.86 (0.05) 2.58 (0.23)*,† 3.93 (0.06)*,†,‡
Il1b 1 (0.21) 0.98 (0.13) 2.65 (0.24)*,† 4.04 (0.06)*,†,‡
Il22 1 (0.18) 0.88 (0.09) 0.47 (0.04)*,† 0.27 (0.08)*,†,‡
Ffar2 1 (0.02) 0.95 (0.04) 2.43 (0.09)*,† 6.26 (0.13)*,†,‡
Ffar3 1 (0.05) 0.93 (0.15) 2.45 (0.19)*,† 3.28 (0.11)*,†,‡
Cd36 1 (0.11) 1.13 (0.06) 1.47 (0.09)*,† 7.63 (2.48)*,†,‡
Sglt1 1 (0.14) 1.22 (0.28) 2.31 (1.08)*,† 6.11 (0.71)*,†,‡
Muc2 1 (0.06) 0.97 (0.03) 0.48 (0.14)*,† 0.18 (0.12)*,†,‡
Liver Tnfa 1 (0.21) 1.05 (0.28) 2.44 (0.83)*,† 5.93 (1.18)*,†,‡
Il6 1 (0.19) 1.01 (0.24) 2.22 (0.36)*,† 5.51 (0.50)*,†,‡
Il1b 1 (0.19) 1.01 (0.24) 1.66 (0.23)*,† 6.17 (1.72)*,†,‡
Scd1 1 (0.12) 1.16 (0.09) 1.22 (0.20)*,† 3.03 (0.27)*,†,‡
Elovl6 1 (0.12) 1.16 (0.09) 1.44 (0.24)*,† 4.28 (0.95)*,†,‡
Fasn 1 (0.12) 1.16 (0.09) 2.05 (0.09)*,† 9.82 (0.90)*,†,‡
Note: Relative mRNA expression of Tnfa, Il6, Il1b, Il22, Ffar2, Ffar3, Cd36, and Sglt1 in the jejunum normalized to the expression of Gapdh (n=10). Relative mRNA expression of Tnfa, Il6, Il1b, Scd1, Elovl6, and Fasn in the liver normalized to the expression of Gapdh (n=10). Data are presented as mean±SD values. The statistical analyses among four groups were performed by one-way ANOVA with Holm-Šídák’s multiple comparisons test: *, p<0.05 vs. ND; †, p<0.05 vs. ND+MP; ‡, p<0.05 vs. HFD group. HFD, high-fat diet; MPs, microplastics; ND, normal diet; SD, standard deviation.
Gene Expression in the Liver of Mice Exposed to ND or HFD with MPs at 12 Wk of Age
Gene expression in liver was also evaluated. The relative expression of Tnfa, Il6, and Il1b in liver of mice fed the HFD with MPs was higher than that of mice fed the HFD without MPs. Similarly, the relative expression of Scd1, Elovl6, and Fasn, which are fatty acid synthases, in liver of mice fed the HFD with MPs was higher than that in mice fed the HFD without MPs. On the other hand, their expressions were not different between ND mice and mice fed the ND with MPs (Table 1).
16s rRNA Sequence of Gut Microbiota
In 16s rRNA sequence of gut microbiota, the relative abundance of phyla was investigated (Figure 5A). Although there was no clear difference in the relative abundance of phyla between mice fed the ND and mice fed the ND containing MPs, the mice fed the HFD with MPs had a lower abundance of phylum Bacteroidetes and a higher abundance of phylum Proteobactria, in comparison with mice fed the HFD without MPS (Table 2). There was no difference in OTUs between the mice fed the ND and mice fed the ND with MPs and HFD-fed mice and mice fed the HFD with MPs, respectively (Figure 5B). On the other hand, in alpha diversity index, Chao1 index, Shannon index, and Gini-Simpson index in mice fed the HFD with MPs were lower than those in mice fed the HFD without MPs, whereas there was no difference between ND and mice fed the ND with MPs (Figure 5C–E). We also used the LEfSe algorithm to identify the specific taxa that were variably distributed between mice fed the ND without MPs and mice fed the ND with MPs, and between mice fed the HFD without MPS and mice fed the HFD with MPs, respectively. Six taxa were overrepresented (including the phylum Proteobacteria, the genus Parasutterella, the family Sutterellaceae, the class Betaproteobacteria) and five were underrepresented (including the order Bacteroidales, the class Bacteroidia, the phylum Bacteroidetes, and the genus Tannerella) in mice fed the ND with MPs in comparison with that in mice fed the ND without MPs (Figure 5F and H). Moreover, seven taxa were underrepresented (including the family Prevotellaceae, the order Bacteroidales, the class Bacteoidia, the phylum Bacteroidetes, the genus Alloprevotella, and the family Ruminococcaceae) and the genus Desulfovibrio were overrepresented in mice fed the HFD with MPs in comparison with that in mice fed the HFD without MPs (Figure 5G and I).
Figure 5. 16s rRNA sequence of gut microbiota of mice exposed to ND or HFD±MPs at 12 wk of age. (A) The relative abundance of phyla (%). Summary data for this graph shown in Table 2. (B) The number of OTUs (n=6). (C) Chao1 index (n=6). (D) Shannon index (n=6). (E) Gini-Simpson index (n=6). (F and H) LDA score (Log10) and LEfSe cladogram of ND and ND+MP mice; (Red) taxa enriched in ND mice; (Green) taxa enriched in mice fed the ND with MPs (n=6). (G) LDA score and LEfSe cladogram of HFD and mice fed the HFD with MPs; (Red) taxa enriched in HFD mice; (Green) taxa enriched in mice fed the HFD with MPs (n=6). Only taxa with a significant LDA threshold value >2 are shown. Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. *p<0.05 and **p<0.01. Summary data can be found in Table S2. Note: ANOVA, analysis of variance; HFD, high-fat diet; LDA, linear discriminant analysis; LEfSe, LDA coupled with effect size measurements; MPs, microplastics; ND, normal diet; OTUs, operational taxonomic units; SD, standard deviation.
Figure 5A is a stacked bar graph, plotting Abundance (percentage), ranging from 0 to 100 in increments of 50 (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis) for Actinobacteria, Bacteroidetes, Deferribacteres, Proteobacteria, Verrucomicrobia, Candidatus Saccharibacteria, Cyanobacteria or Chloroplast, and Firmicutes. Figures 5B to 5E are bar graphs, plotting operational taxonomic units, ranging from 0 to 400 in increments of 100; Chao1 index 1, ranging from 0 to 400 in increments of 100; Shannon index, ranging from 0 to 8 in increments of 2; and Gini-Simpson index, ranging from 0.0 to 1.5 in increments of 0.5 (y-axis) across normal diet, normal diet plus microplastics, high-fat diet, and high-fat diet plus microplastics (x-axis). Figure 5F is a set of two horizontal bar graphs, plotting Tannerella, uncultured bacterium, Bacteroidetes, bacteroidia, bacteroidales, burkholderrales, betaproteobacteria, sutterellaceae, unculturedbacterium, parasutterella, and proteobacteria; and unculturedbacterium, alloprevotella, prevotellaceae, bacteroidia, Bacteroidetes, bacteroidales, ruminococcaceae, and desulfovibrio (y-axis) across linear discriminant analysis score (log 10), ranging from negative 10.0 to 8.0 in increments of 1.8 (x-axis) for high-fat diet and high-fat diet plus microplastics. Figure 5H is a Linear discriminant analysis Effect Size cladogram depicts the taxa enriched in normal diet mice, including a: Tannerella, b: Bacteroidales, and taxa enriched in normal diet plus microplastics mice, including c: Parasuttterella, d: Sutterellaceae, e: Burkholderiales. Figure 5I is a Linear discriminant analysis Effect Size cladogram depicts the taxa enriched in normal diet mice, including a: alloprevotella, b: prevotellaceae, and taxa enriched in normal diet plus microplastics mice, including c: bacteroidales, d: ruminococcaceae, and e: desulfovibrio.
Table 2 The relative abundance of phyla in the feces of mice exposed to ND or HFD±MPs at 12 wk of age.
ND ND+MP HFD HFD+MP
Phylum Mean (%) SD Mean (%) SD p-Value Mean (%) SD Mean (%) SD p-Value
Actinobacteria 0.08 0.01 0.05 0.02 0.959 0.05 0.06 0.01 0.02 0.392
Bacteroidetes 71.85 6.18 68.56 10.53 0.859 49.77 2.93 29.12 9.74 0.026
Deferribacteres 0.28 0.05 0.95 0.72 0.847 4.90 2.11 8.65 2.12 0.035
Proteobacteria 0.28 0.07 0.71 0.19 0.995 14.66 6.69 44.98 9.76 0.001
Verrucomicrobia 0.12 0.21 0.22 0.12 0.579 0.01 0.02 0.00 0.00 0.989
Candidatus Saccharibacteria 0.02 0.02 0.01 0.01 0.750 0.00 0.00 0.00 0.00 >0.9999
Cyanobacteria/Chloroplast 0.04 0.03 0.04 0.04 0.990 0.03 0.03 0.01 0.00 0.622
Firmicutes 26.66 5.54 27.17 8.06 0.996 30.30 9.63 19.90 4.65 0.220
Note: Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. ANOVA, analysis of variance; HFD, high-fat diet; MPs, microplastics; ND, normal diet; SD, standard deviation.
MP and Murine Intestinal Epithelial Cell Line
Using the murine intestinal epithelial cell line MODE-K cells, we investigated the mechanism by which a high-fat diet promotes MP deposition in the small intestinal epithelium. To reproduce the intestinal environment caused by a HFD, PA, a type of saturated fatty acid, was administered to the cells. The mucus that covers and protects the intestinal epithelium is built around its major structural component, the gel-forming MUC2 mucin,80 and Muc2-deficient mice were shown to have increased intestinal permeability.81 Therefore, we investigated the expression of Muc2 by RT-PCR as an indicator of mucin secretion and intestinal permeability. The expression of Muc2 was not different between control group and MP alone group, whereas the expression in both the PA alone and PA with MPs groups was lower than in the control and MP groups (Figure 6A). In addition, treatment with recombinant protein IL-22, which acts on small intestinal epithelial cells to stimulate secretion of the mucin layer,77 increased Muc2 gene expression in both the PA alone and PA with MPs groups, in comparison with that in the nonIL-22-treated group (Figure 6B).
Figure 6. Differences in Muc2 gene expression and MP accumulation in MODE-K cells exposed to MP and saturated fatty acids. (A) Relative mRNA expression of Muc2 in MODE-K cells without/with MP and PA (n=10). (B) Relative mRNA expression of Muc2 in MODE-K cells with MP without/with PA and IL-22 (n=10). (C) The fold differences in accumulation of MPs in MODE-K cells without/with MP and PA (n=10). (D) The fold differences in accumulation of MPs in MODE-K cells with MP without/with PA and IL-22 (n=10). Data are presented as mean±SD values. Data were analyzed using one-way ANOVA with Holm-Šídák’s multiple comparisons test. Summary data can be found in Table S2. *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001. Note: MP, microplastics; PA, palmitic acid; SD, standard deviation.
Figures 6A to 6D are bar graphs, plotting Relative expression of Muc2 normalized to Gapdh in M O D E-K cells, ranging from 0.0 to 1.5 in increments of 0.5; Relative expression of Muc2 normalized to Gapdh in M O D E-K cells, ranging from 0 to 3 in unit increments; Fold change (G F P expression), ranging from 0 to 25 in increments of 5; and Fold change (G F P expression), ranging from 0 to 10 in increments of 2 (y-axis) across microplastic (negative, positive, negative, positive), palmitic acid (negative, negative, positive, positive); microplastic (positive, positive, positive, positive), palmitic acid (negative, negative, positive, positive), interleukin-22 (negative, positive, negative, positive); microplastic (negative, positive, negative, positive), palmitic acid (negative, negative, positive, positive); and microplastic (positive, positive, positive, positive), palmitic acid (negative, negative, positive, positive), interleukin-22 (negative, positive, negative, positive) (x-axis).
Next, the amount of fluorescently labeled MPs accumulated in the cells was measured by absorbance spectrometer. The PA with MPs group accumulated more MP than the MP alone group. However, the accumulation of MP was higher in the MP alone group than in the control group. (Figure 6C). On the other hand, cells treated with IL-22 had significantly less microplastic accumulation in the PA with MPs group in comparison with the MP alone group (Figure 6D).
Discussion
In this study, findings suggestive of metabolic disturbances induced by MP, including deterioration of glucose tolerance and increased fat accumulation in the liver, which were similar to symptoms seen in diabetes and NAFLD, were observed only in mice fed the HFD. Mice fed the HFD with MPs exhibited dysbiosis, thinning of the intestinal mucin layer, indications of inflammation of the intestinal tract, and different gene expression of nutrient transporters in the intestine. To the best of our knowledge, this study is the first to demonstrate that a modern lipid-rich diet and MPs can be associated with findings suggestive of metabolic disturbances, such as diabetes and NAFLD.
MPs are a cause of concern for global pollution and have been found to be toxic to Mytilus edulis,23 fish,22 and mammals.24 In vivo evidence on the immunotoxicity of MPs suggested that immune cells, including those of the intestinal immune system, might be the targets of plastic-induced damage. Indeed, Li et al.30 reported that dysbiosis and intestinal inflammation in mice exposed to MPs. Moreover, in a human study, a positive correlation was observed between MPs in feces and the inflammatory bowel disease (IBD) status, indicating that MP exposure might be related to the pathogenesis of IBD or that IBD might exacerbate the retention of MPs.82 In this study, intestinal permeability was approximated by measuring plasma dextran 4 h after oral administration. We found that plasma dextran was significantly higher in the HFD group in comparison with the ND group and in the mice fed the HFD with MPs in comparison with the mice fed the HFD without MPs, suggesting that intestinal permeability was higher in these groups. In addition, the number of goblet cells was significantly lower in the HFD group in comparison with that in the ND group, and this was even lower in mice exposed to MPs. Pathological images showed very little deposition of MPs in the intestinal mucosa of mice in the ND group unlike in the HFD group. IL-1β and TNF-α were markedly elevated in inflammatory conditions of the intestine, such as IBD.83,84 Physiological concentrations of IL-1β and TNF-α were associated with markedly increased permeability of tight junctions in intestinal epithelial cells in vitro.85,86 In this study, the expression of IL-1β and TNF-α in small intestine was higher following exposure to MPs. These results suggest that the reduced mucin layer in the HFD-fed mice allowed MPs to enter the intestinal mucosa, causing inflammation in the LPL of the small intestine.
We also analyzed the dynamics of immune cells involved in innate immunity of small intestine. Disruption of the mucosal barrier in mice with colitis altered the number of ILC3s, which are key regulators of inflammation and infection at the mucosal barrier.87 ILC3-derived IL-22 has been revealed to promote STAT3-dependent expression of antimicrobial peptides and play an important role in maintaining the barrier function of the intestinal epithelium in mice lacking IL-22–producing NKp46+cells.38–40 Conversely, loss of ILC3s results in decreased expression of IL-22 and reduced levels of antimicrobial peptides expressed by intestinal epithelial cells.88 ILC3s exhibited plasticity and their function was altered by the expression of the transcription factors RORγt and T-bet in mice with colitis.41 When stimulated with cytokines, such as IL-12 and IL-18, ex-RORγt-positive ILC3s with T-bet-positive features, that is, ex-ILC3s, increased and RORγt-positive ILC3s decreased, indicating that ILC3s could respond to environmental cues. Previous studies have shown that T-bet-positive ILC3s produced IFN-γ and inhibited the production of IL-17 and IL-22 in mice with inflammatory bowel disease.75 Thus, T-bet-positive ILC3s exerted a function similar to that of ILC1. ILCs were shown to express receptors for SCFAs, such as G protein–coupled receptor (GPR) 41 [also known as a free fatty acid receptor (FFAR) 3] and GPR43 (FFAR2), which are important for their proliferation, and phosphatidylinositol-3 kinase (PI3K).89,90 They stimulated the activation of signal transducer and activator of transcription 3 (Stat3), Stat5, and mechanistic target of rapamycin (mTOR) in mice with acute Clostridium difficile infection.91 In fact, several studies of mice,92,93 rats,94–96 and humans97–99 have reported that administration of SCFA improves intestinal inflammation and protected mice from diet-induced obesity and insulin resistance.93,96,100 Therefore, we hypothesized that the adverse effects of MPs on various metabolic abnormalities may be related to the inflammatory effects of innate immunity by decreasing the production of SCFAs in the gut. Furthermore, in the cell experiments of this study, MODE-K cells, a cell line of small intestinal epithelial cells, when exposed to with IL-22, which is secreted from ILC3 upon stimulation of SCFA, had higher gene expression of Muc2; this expression was lower when cells were exposed to MP and PA treatment, and these cells exhibited significantly lower MP accumulation in the cells. In summary, it is suggested that the production of SCFA enhanced the intestinal barrier function and prevented MP-induced intestinal inflammation, resulting in the induction of various MP-induced metabolic disorders only in the HFD-treated group.
We previously reported that the expression of CD36, a long-chain fatty acid transporter, was reduced in association with the improvement in intestinal inflammation by an increase in SCFAs in mice.44 In the present study, the expression of CD36 in the small intestine was also significantly higher in mice fed the HFD with MPs in comparison with that in HFD-fed mice. The concentration of palmitic acid, a saturated fatty acid, in fecal excretion was significantly lower in mice fed the HFD with MPs in comparison with that in the other three groups, and the serum and intrahepatic concentrations of palmitic acid were significantly higher, which could be expected to aggravate existing NAFLD. Furthermore, we hypothesized that the MP-induced differences in Cd36 expression in the small intestine occurred only in the HFD group because the thinning of the mucin layer caused by HFD treatment led to MP deposition in the small intestinal epithelium, which exacerbated the inflammation in the intestinal tract.
In the analyses of gut microbiota, differences in gut microbiota were observed after microplastic administration. The phylum Proteobacteria, the genus Parasutterella, the family Sutterellaceae, and the class Betaproteobacteria were more frequently present in mice fed the ND with MPs mice than in ND-fed mice. In human study, obese individuals had higher amounts of Proteobacteria and a positive correlation between Proteobacteria and fat intake.101,102 The family Sutterellaceae have been reported to play a part in the pathogenesis of inflammatory bowel disease.103–105 Moreover, in an animal study, the class Betaproteobacteria was reported to be correlated with weight gain.106 Taken together, the abundance of gut microbiota involved in dysbiosis was significantly higher in mice fed the ND with MPs than in mice fed the ND without MPs. The phylum Proteobacteria, the genus Parasutterella, the family Sutterellaceae, and the class Betaproteobacteria were more frequently present in mice fed the HFD with MPs than in ND-fed mice. The family Prevotellaceae, the order Bacteroidales, the class Bacteroidia, the phylum Bacteroidetes, the genus Alloprevotella, and the family Ruminococcaceae were less frequently present in mice fed the HFD with MPs than in mice fed the HFD without MPs. In a human study, the order Bacteroidales, the class Bacteoidia, the phylum Bacteroidetes were associated with weight loss,107 and the family Ruminococcaceae have been described as SCFA producers with beneficial effect on the intestinal barrier in a human study.108
Moreover, the genus Alloprevotella have been reported to produce acetic acid in a human study109 and decreased by the administration of HFD in an animal study.110 On the other hand, it has been observed in several animal models and human studies that the genus Parasutterella were significantly reduced by HFD administration.111–114 The family Ruminococcaceae have been reported to be associated with obesity115 and diabetes116 in animal studies. Moreover, the genus Desulfovibrio, which was frequently present in the feces of mice fed an HFD with MPs, have been reported to increase in individuals with type II diabetes and obesity117 and upregulate CD36 expression.118 In the present study, the expression of Cd36 in the small intestine of mice fed the HFD with MPs was higher than that of mice fed the HFD without MPs. It was suggested that this alteration of the gut microbiota might have caused various metabolic disorders due to increased absorption of saturated fatty acids from the intestinal tract. Based on these results, the frequency of gut microbiota involved in dysbiosis was higher and that in production of SCFAs was lower in both mice fed the ND with MPs and mice fed the HFD with MPs in comparison with ND- and HFD-fed mice. On the other hand, the diversity of gut microbiota was not different between ND mice and mice fed the ND with MPs, whereas that in mice fed the HFD with MPs was lower than that in HFD-fed mice. Decreased gut microbiota richness have been reported to be associated with various physiological markers of obesity and metabolic syndrome.119 This decrease might be one possible reason why we did not find differences in metabolic disturbances between ND-fed mice and mice fed the ND with MPs that we observed between HFD-fed mice and mice fed the HFD with MPs, although there were differences in gut microbiota abundance between ND-fed mice and mice fed the ND with MPs.
We also previously showed that HFD-fed wild-type mice experienced dysbiosis glucose intolerance, associated with up-regulation of the expression of Sglt1, a Na+/glucose cotransporter, in the small intestine.44 In the present study, the expression of Sglt1 in the small intestine was significantly higher in mice fed the HFD with MPs in comparison with that in HFD-fed mice. An increase in glucose absorption from the small intestine was thought to be one of the reasons for the deterioration in glucose tolerance induced upon exposure to MPs. In contrast, mice fed the ND with MPs did not show significant deterioration in many metabolism-related parameters in comparison with ND-fed mice, but postprandial blood glucose was elevated in the iPGTT and serum lipid levels were higher in the mice fed the ND with MPs than in the group fed only the ND, although the difference was not statistically significant in the present sample size. Considering several reports in zebrafish that oral exposure to MPs increases the deposition of microplastics in the pancreas,120–122 it is possible that there was a deposition of MPs in the pancreases of mice fed the ND with MPs in this study, which affected the early secretion of insulin. In this study, Red O staining area of the liver was higher in the group fed the ND with MPs than in the ND-only group, although not significantly, suggesting that the ND group was also affected by MP-induced metabolic disturbances to a small extent.
There was a limitation of this study. In iPGTT, we did not measure the plasma insulin levels. If we had had the data, more accurate assessment of insulin resistance would have been possible. In addition, the evaluation of dose response and verification with other size and concentration combinations of microplastics has not been conducted, and this is a future research topic.
In conclusion, to the best of our knowledge, the present study is the first to suggest that MPs cause metabolic disturbances under HFD intake conditions, characteristic of a modern diet. Changes in nutrient absorption might be involved in promoting an inflammatory shift of innate immunity in the intestine, accompanied by the deposition of MPs in the intestinal mucosa and a decrease in SCFA production. This study highlights the need for reducing oral exposure to MPs through remedial environmental measures to improve metabolic disturbances under HFD conditions. Further clinical studies are needed to assess the translational potential of the reported findings for metabolic disturbances.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank all the staff members of the Kyoto Prefectural University of Medicine. The authors also thank Editage (www.editage.com) for English-language editing.
T. O. originated and designed the study, researched the data, and wrote the manuscript. M.H. and Y.H. originated and designed the study, researched the data, and reviewed the manuscript. Y.H., S.M., T.S., E.U., N.N., M.A., and M.Y. researched the data and contributed to the discussion. R.S., Y.N., H.S., and H.T. provided technical cooperation. H.T. and M.F. originated and designed the study, researched the data, and reviewed and edited the manuscript. M.F. is the guarantor of this work and, as such, had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors were involved in the writing of the manuscript and approved the final version of this article.
This study is supported by JST-CREST (JPMJCR19H3).
The authors declare the following funding received outside the submitted work: M.H. has received grants from Asahi Kasei Pharma, Nippon Boehringer Ingelheim Co., Ltd.; Mitsubishi Tanabe Pharma Corporation; Daiichi Sankyo Co., Ltd.; Sanofi K.K.; Takeda Pharmaceutical Co., Ltd.; Astellas Pharma Inc.; Kyowa Kirin Co., Ltd.; Sumitomo Dainippon Pharma Co., Ltd.; Novo Nordisk Pharma Ltd.; and Eli Lilly Japan K.K.
The authors declare the following funding received outside the submitted work: Y.H. has received grants from Kowa company Ltd.; Takeda Pharmaceutical Co., Ltd.; Ono Pharmaceutical Co., Ltd.; Sumitomo Dainippon Pharma Co., Ltd.; personal fees from Daiichi Sankyo Co., Ltd.; personal fees from Mitsubishi Tanabe Pharma Corp.; personal fees from Sanofi K.K.; and personal fees from Novo Nordisk Pharma Ltd.
The authors declare the following funding received outside the submitted work: T.S. has received personal fees from Ono Pharma Co., Ltd.; Mitsubishi Tanabe Pharma Co.; Astellas Pharma Inc.; Kyowa Hakko Kirin Co., Ltd.; Sanofi K.K.; MSD K.K.; Kowa Pharmaceuticals Co., Ltd.; Taisho Toyama Pharma Co., Ltd.; Takeda Pharma Co., Ltd.; Kissei Pharma Co., Ltd.; Novo Nordisk Pharma Ltd.; and Eli Lilly Japan K.K.
The authors declare the following funding received outside the submitted work: E.U. has received grants from the Japanese Study Group for Physiology and Management of Blood Pressure and the Astellas Foundation for Research on Metabolic Disorders and personal fees from AstraZeneca PLC; Astellas Pharma Inc.; Daiichi Sankyo Co., Ltd.; Kowa Pharmaceuticals Co., Ltd.; MSD K.K.; Mitsubishi Tanabe Pharma Corp.; Novo Nordisk Pharma Ltd.; Taisho Toyama Pharmaceutical Co., Ltd.; Nippon Boehringer Ingelheim Co., Ltd.; and Sumitomo Dainippon Pharma Co., Ltd., outside the submitted work. M.A. received personal fees from Novo Nordisk Pharma Ltd.; Abbott Japan Co., Ltd.; AstraZeneca PLC; Kowa Pharmaceutical Co., Ltd.; Ono Pharmaceutical Co., Ltd.; and Takeda Pharmaceutical Co., Ltd.
The authors declare the following funding received outside the submitted work: M.Y. reports personal fees from MSD K.K.; Sumitomo Dainippon Pharma Co., Ltd.; Kowa Phramaceutical Co., Ltd.; AstraZeneca PLC; Takeda Pharmaceutical Co., Ltd.; Kyowa Hakko Kirin Co., Ltd.; Daiichi Sankyo Co., Ltd.; Kowa Pharmaceuticals Co., Ltd.; and Ono Pharma Co., Ltd.
The authors declare the following funding received outside the submitted work: M.F. has received grants from Nippon Boehringer Ingelheim Co., Ltd.; Kissei Pharma Co., Ltd.; Mitsubishi Tanabe Pharma Co.; Daiichi Sankyo Co., Ltd.; Sanofi K.K.; Takeda Pharma Co., Ltd.; Astellas Pharma Inc.; MSD K.K.; Kyowa Hakko Kirin Co., Ltd., Sumitomo Dainippon Pharma Co., Ltd.; Kowa Pharmaceuticals Co., Ltd.; Novo Nordisk Pharma Ltd., Ono Pharma Co., Ltd.; Sanwa Kagaku Kenkyusho Co., Ltd.; Eli Lilly Japan K.K.; Taisho Pharma Co., Ltd.; Terumo Co.; Teijin Pharma Ltd.; Nippon Chemiphar Co., Ltd.; and Johnson & Johnson K.K. Medical Co.; Abbott Japan Co., Ltd.; and received personal fees from Nippon Boehringer Ingelheim Co., Ltd.; Kissei Pharma Co., Ltd.; Mitsubishi Tanabe Pharma Corp.; Daiichi Sankyo Co.; Ltd., Sanofi K.K.; Takeda Pharma Co., Ltd.; Astellas Pharma Inc.; MSD K.K.; Kyowa Kirin Co., Ltd.; Sumitomo Dainippon Pharma Co., Ltd.; Kowa Pharmaceuticals Co., Ltd.; Novo Nordisk Pharma Ltd.; Ono Pharma Co., Ltd.; Sanwa Kagaku Kenkyusho Co., Ltd.; Eli Lilly Japan K.K.; Taisho Pharma Co., Ltd.; Bayer Yakuhin, Ltd.; AstraZeneca K.K.; Mochida Pharma Co., Ltd.; Abbott Japan Co., Ltd.; Medtronic Japan Co., Ltd.; Arkley Inc.; Teijin Pharma Ltd.; and Nipro Corp.
==== Refs
References
1. Geyer R, Jambeck JR, Law KL. 2017. Production, use, and fate of all plastics ever made. Sci Adv 3 (7 ):e1700782, PMID: , 10.1126/SCIADV.1700782.28776036
2. Wu P, Huang J, Zheng Y, Yang Y, Zhang Y, He F, et al. 2019. Environmental occurrences, fate, and impacts of microplastics. Ecotoxicol Environ Saf 184 :109612, PMID: , 10.1016/J.ECOENV.2019.109612.31476450
3. Plastics-the Facts 2018 An analysis of European plastics production, demand and waste data.
4. UNEP (United Nations Environmental Programme). Single-Use Plastics: A Roadmap for Sustainability | International Environmental Technology Centre. https://www.unep.org/ietc/ja/node/53?%2Fresources%2Fpublication%2Fsingle-use-plastics-roadmap-sustainability= [accessed 3 May 2022].
5. Alimba CG, Faggio C. 2019. Microplastics in the marine environment: current trends in environmental pollution and mechanisms of toxicological profile. Environ Toxicol Pharmacol 68 :61–74, PMID: , 10.1016/j.etap.2019.03.001.30877952
6. Ferreira I, Venâncio C, Lopes I, Oliveira M. 2019. Nanoplastics and marine organisms: what has been studied? Environ Toxicol Pharmacol 67 :1–7, PMID: , 10.1016/j.etap.2019.01.006.30685594
7. Koelmans AA, Mohamed Nor NH, Hermsen E, Kooi M, Mintenig SM, de France J. 2019. Microplastics in freshwaters and drinking water: critical review and assessment of data quality. Water Res 155 :410–422, PMID: , 10.1016/j.watres.2019.02.054.30861380
8. Rillig MC. 2012. Microplastic in terrestrial ecosystems and the soil? Environ Sci Technol 46 (12 ):6453–6454, PMID: , 10.1021/es302011r.22676039
9. Prata JC. 2018. Airborne microplastics: consequences to human health? Environ Pollut 234 :115–126, PMID: , 10.1016/j.envpol.2017.11.043.29172041
10. Derraik JGB. 2002. The pollution of the marine environment by plastic debris: a review. Mar Pollut Bull 44 (9 ):842–852, PMID: , 10.1016/S0025-326X(02)00220-5.12405208
11. Boucher J, Friot D. 2017. Primary Microplastics in the Oceans: A Global Evaluation of Sources. Gland, Switzerland: IUCN. 10.2305/IUCN.CH.2017.01.en.
12. Cooper DA, Corcoran PL. 2010. Effects of mechanical and chemical processes on the degradation of plastic beach debris on the island of Kauai, Hawaii. Mar Pollut Bull 60 (5 ):650–654, PMID: , 10.1016/j.marpolbul.2009.12.026.20106491
13. Yakimets I, Lai D, Guigon M. 2004. Effect of photo-oxidation cracks on behaviour of thick polypropylene samples. Polym Degrad Stab 1 (86 ):59–67, 10.1016/J.POLYMDEGRADSTAB.2004.01.013.
14. Turton TJ, White JR. 2001. Effect of stabilizer and pigment on photo-degradation depth profiles in polypropylene. Polym Degrad Stab 74 (3 ):559–568, 10.1016/S0141-3910(01)00193-8.
15. Andrady AL. 2011. Microplastics in the marine environment. Mar Pollut Bull 62 (8 ):1596–1605, PMID: , 10.1016/j.marpolbul.2011.05.030.21742351
16. Mason SA, Welch VG, Neratko J. 2018. Synthetic polymer contamination in bottled water. Front Chem 6 :407, PMID: , 10.3389/FCHEM.2018.00407.30255015
17. Schymanski D, Goldbeck C, Humpf HU, Fürst P. 2018. Analysis of microplastics in water by micro-Raman spectroscopy: release of plastic particles from different packaging into mineral water. Water Res 129 :154–162, PMID: , 10.1016/j.watres.2017.11.011.29145085
18. Farrell P, Nelson K. 2013. Trophic level transfer of microplastic: Mytilus edulis (L.) to Carcinus maenas (L.). Environ Pollut 177 :1–3, PMID: , 10.1016/j.envpol.2013.01.046.23434827
19. Nelms SE, Galloway TS, Godley BJ, Jarvis DS, Lindeque PK. 2018. Investigating microplastic trophic transfer in marine top predators. Environ Pollut 238 :999–1007, PMID: , 10.1016/j.envpol.2018.02.016.29477242
20. Zhao S, Ward JE, Danley M, Mincer TJ. 2018. Field-based evidence for microplastic in marine aggregates and mussels: implications for trophic transfer. Environ Sci Technol 52 (19 ):11038–11048, PMID: , 10.1021/acs.est.8b03467.30156835
21. Sussarellu R, Suquet M, Thomas Y, Lambert C, Fabioux C, Pernet MEJ, et al. 2016. Oyster reproduction is affected by exposure to polystyrene microplastics. Proc Natl Acad Sci U S A 113 (9 ):2430–2435, PMID: , 10.1073/pnas.1519019113.26831072
22. Jeong C-B, Won E-J, Kang H-M, Lee M-C, Hwang D-S, Hwang U-K, et al. 2016. Microplastic size-dependent toxicity, oxidative stress induction, and p-JNK and p-p38 activation in the monogonont rotifer (Brachionus koreanus). Environ Sci Technol 50 (16 ):8849–8857, PMID: , 10.1021/acs.est.6b01441.27438693
23. Rochman CM, Browne MA, Halpern BS, Hentschel BT, Hoh E, Karapanagioti HK, et al. 2013. Policy: classify plastic waste as hazardous. Nature 494 (7436 ):169–171, PMID: , 10.1038/494169a.23407523
24. Browne MA, Dissanayake A, Galloway TS, Lowe DM, Thompson RC. 2008. Ingested microscopic plastic translocates to the circulatory system of the mussel, Mytilus edulis (L). Environ Sci Technol 42 (13 ):5026–5031, PMID: , 10.1021/es800249a.18678044
25. Deng Y, Zhang Y, Lemos B, Ren H. 2017. Tissue accumulation of microplastics in mice and biomarker responses suggest widespread health risks of exposure. Sci Rep 7 :46687, PMID: , 10.1038/SREP46687.28436478
26. Faure F, Demars C, Wieser O, Kunz M, de Alencastro LF. 2015. Plastic pollution in Swiss surface waters: nature and concentrations, interaction with pollutants. Environ Chem 12 (5 ):582–591, 10.1071/EN14218.
27. Di M, Wang J. 2018. Microplastics in surface waters and sediments of the Three Gorges Reservoir, China. Sci Total Environ 616–617 :1620–1627, PMID: , 10.1016/j.scitotenv.2017.10.150.29050832
28. Cole M, Lindeque P, Fileman E, Halsband C, Goodhead R, Moger J, et al. 2013. Microplastic ingestion by zooplankton. Environ Sci Technol 47 (12 ):6646–6655, PMID: , 10.1021/es400663f.23692270
29. Lei L, Liu M, Song Y, et al. 2018. Polystyrene (nano)microplastics cause size-dependent neurotoxicity, oxidative damage and other adverse effects in Caenorhabditis elegans. Environ Sci Nano 5 (8 ):2009–2020, 10.1039/C8EN00412A.
30. Li B, Ding Y, Cheng X, Sheng D, Xu Z, Rong Q, et al. 2020. Polyethylene microplastics affect the distribution of gut microbiota and inflammation development in mice. Chemosphere 244 :125492, PMID: , 10.1016/J.CHEMOSPHERE.2019.125492.31809927
31. Ghosh SS, Wang J, Yannie PJ, Ghosh S. 2020. Intestinal barrier dysfunction, LPS translocation, and disease development. J Endocr Soc 4 (2 ):bvz039, PMID: , 10.1210/JENDSO/BVZ039.32099951
32. Vancamelbeke M, Vermeire S. 2017. The intestinal barrier: a fundamental role in health and disease. Expert Rev Gastroenterol Hepatol 11 (9 ):821–834, PMID: , 10.1080/17474124.2017.1343143.28650209
33. Okamura T, Hashimoto Y, Majima S, Senmaru T, Ushigome E, Nakanishi N, et al. 2021. Trans fatty acid intake induces intestinal inflammation and impaired glucose tolerance. Front Immunol 12 :669672, PMID: , 10.3389/fimmu.2021.669672.33995404
34. Cheng L, Jin H, Qiang Y, Wu S, Yan C, Han M, et al. 2016. High fat diet exacerbates dextran sulfate sodium induced colitis through disturbing mucosal dendritic cell homeostasis. Int Immunopharmacol 40 :1–10, PMID: , 10.1016/j.intimp.2016.08.018.27567245
35. Li L, Xu M, He C, Wang H, Hu Q. 2022. Polystyrene nanoplastics potentiate the development of hepatic fibrosis in high fat diet fed mice. Environ Toxicol 37 (2 ):362–372, PMID: , 10.1002/tox.23404.34755918
36. Constantinides MG, McDonald BD, Verhoef PA, Bendelac A. 2014. A committed precursor to innate lymphoid cells. Nature 508 (7496 ):397–401, PMID: , 10.1038/nature13047.24509713
37. Bando JK, Colonna M. 2016. Innate lymphoid cell function in the context of adaptive immunity. Nat Immunol 17 (7 ):783–789, PMID: , 10.1038/ni.3484.27328008
38. Satoh-Takayama N, Vosshenrich CAJ, Lesjean-Pottier S, Sawa S, Lochner M, Rattis F, et al. 2008. Microbial flora drives interleukin 22 production in intestinal NKp46+ cells that provide innate mucosal immune defense. Immunity 29 (6 ):958–970, PMID: , 10.1016/j.immuni.2008.11.001.19084435
39. Sonnenberg GF, Fouser LA, Artis D. 2011. Border patrol: regulation of immunity, inflammation and tissue homeostasis at barrier surfaces by IL-22. Nat Immunol 12 (5 ):383–390, PMID: , 10.1038/ni.2025.21502992
40. Zheng Y, Valdez PA, Danilenko DM, Hu Y, Sa SM, Gong Q, et al. 2008. Interleukin-22 mediates early host defense against attaching and effacing bacterial pathogens. Nat Med 14 (3 ):282–289, PMID: , 10.1038/nm1720.18264109
41. Figliuolo VR, Dos Santos LM, Abalo A, Nanini H, Santos A, Brittes NM, et al. 2017. Sulfate-reducing bacteria stimulate gut immune responses and contribute to inflammation in experimental colitis. Life Sci 189 :29–38, PMID: , 10.1016/j.lfs.2017.09.014.28912045
42. Aparicio-Domingo P, Romera-Hernandez M, Karrich JJ, Cornelissen F, Papazian N, Lindenbergh-Kortleve DJ, et al. 2015. Type 3 innate lymphoid cells maintain intestinal epithelial stem cells after tissue damage. J Exp Med 212 (11 ):1783–1791, PMID: , 10.1084/jem.20150318.26392223
43. Okamura T, Hamaguchi M, Mori J, Yamaguchi M, Mizushima K, Abe A, et al. 2022. Partially hydrolyzed guar gum suppresses the development of sarcopenic obesity. Nutrients 14 (6 ):1157, PMID: , 10.3390/NU14061157.35334814
44. Kawano R, Okamura T, Hashimoto Y, Majima S, Senmaru T, Ushigome E, et al. 2021. Erythritol ameliorates small intestinal inflammation induced by high-fat diets and improves glucose tolerance. Int J Mol Sci 22 (11 ):5558, 10.3390/ijms22115558.34074061
45. Kiela PR, Ghishan FK. 2016. Physiology of intestinal absorption and secretion. Best Pract Res Clin Gastroenterol 30 (2 ):145–159, PMID: , 10.1016/j.bpg.2016.02.007.27086882
46. Kolodziejczyk AA, Zheng D, Shibolet O, Elinav E. 2019. The role of the microbiome in NAFLD and NASH. EMBO Mol Med 11 (2 ):e9302, PMID: , 10.15252/EMMM.201809302.30591521
47. Dai X, Wang B. 2015. Role of gut barrier function in the pathogenesis of nonalcoholic fatty liver disease. Gastroenterol Res Pract 2015 :287348, PMID: , 10.1155/2015/287348.25945084
48. Hamaguchi M, Kojima T, Takeda N, Nakagawa T, Taniguchi H, Fujii K, et al. 2005. The metabolic syndrome as a predictor of nonalcoholic fatty liver disease. Ann Intern Med 143 (10 ):722–728, PMID: , 10.7326/0003-4819-143-10-200511150-00009.16287793
49. Yang YF, Chen C-Y, Lu T-H, Liao C-M. 2019. Toxicity-based toxicokinetic/toxicodynamic assessment for bioaccumulation of polystyrene microplastics in mice. J Hazard Mater 366 :703–713, PMID: , 10.1016/j.jhazmat.2018.12.048.30583240
50. Jin Y, Xia J, Pan Z, Yang J, Wang W, Fu Z. 2018. Polystyrene microplastics induce microbiota dysbiosis and inflammation in the gut of adult zebrafish. Environ Pollut 235 :322–329, PMID: , 10.1016/j.envpol.2017.12.088.29304465
51. Kawai S, Takagi Y, Kaneko S, Kurosawa T. 2011. Effect of three types of mixed anesthetic agents alternate to ketamine in mice. Exp Anim 60 (5 ):481–487, PMID: , 10.1538/expanim.60.481.22041285
52. McGowan MW, Artiss JD, Strandbergh DZ, Zak B. 1983. A peroxidase-coupled method for the colorimetric determination of serum triglycerides. Clin Chem 29 (3 ):538–542, , 10.1093/clinchem/29.3.538.6825269
53. Christmass MA, Mitoulas LR, Hartmann PE, Arthur PG. 1998. A semiautomated enzymatic method for determination of nonesterified fatty acid concentration in milk and plasma. Lipids 33 (10 ):1043–1049, PMID: , 10.1007/s11745-998-0304-9.9832086
54. Allain CC, Poon LS, Chan CS, Richmond W, Fu PC. 1974. Enzymatic determination of total serum cholesterol. Clin Chem 20 (4 ):470–475, PMID: , 10.1093/clinchem/20.4.470.4818200
55. Okamura T, Nakajima H, Hashimoto Y, Majima S, Senmaru T, Ushigome E, et al. 2021. Low circulating dihomo-gamma-linolenic acid is associated with diabetic retinopathy: a cross sectional study of KAMOGAWA-DM cohort study. Endocr J 68 (4 ):421–428, PMID: , 10.1507/endocrj.EJ20-0564.33361692
56. Motta J-P, Flannigan KL, Agbor TA, Beatty JK, Blackler RW, Workentine ML, et al. 2015. Hydrogen sulfide protects from colitis and restores intestinal microbiota biofilm and mucus production. Inflamm Bowel Dis 21 (5 ):1006–1017, PMID: , 10.1097/MIB.0000000000000345.25738373
57. Keinänen O, Dayts EJ, Rodriguez C, Sarrett SM, Brennan JM, Sarparanta M, et al. 2021. Harnessing PET to track micro- and nanoplastics in vivo. Sci Rep 11 (1 ):11463, PMID: , 10.1038/S41598-021-90929-6.34075133
58. Tenuta T, Monopoli MP, Kim JA, Salvati A, Dawson KA, Sandin P, et al. 2011. Elution of labile fluorescent dye from nanoparticles during biological use. PLoS One 6 (10 ):e25556, PMID: , 10.1371/journal.pone.0025556.21998668
59. Widner DB, Liu C, Zhao Q, Sharp S, Eber MR, Park SH, et al. 2021. Activated mast cells in skeletal muscle can be a potential mediator for cancer-associated cachexia. J Cachexia Sarcopenia Muscle. 12 (4 ):1079–1097, PMID: , 10.1002/JCSM.12714.34008339
60. Molofsky AB, Nussbaum JC, Liang H-E, Van Dyken SJ, Cheng LE, Mohapatra A, et al. 2013. Innate lymphoid type 2 cells sustain visceral adipose tissue eosinophils and alternatively activated macrophages. J Exp Med 210 (3 ):535–549, PMID: , 10.1084/jem.20121964.23420878
61. Wang S, Li J, Wu S, Cheng L, Shen Y, Ma W, et al. 2018. Type 3 innate lymphoid cell: a new player in liver fibrosis progression. Clin Sci (Lond) 132 (24 ):2565–2582, PMID: , 10.1042/CS20180482.30459204
62. Ono Y, Nagai M, Yoshino O, Koga K, Nawaz A, Hatta H, et al. 2018. CD11c+ M1-like macrophages (MΦs) but not CD206+ M2-like MΦ are involved in folliculogenesis in mice ovary. Sci Rep 8 (1 ):8171, PMID: , 10.1038/s41598-018-25837-3.29802255
63. Kleiner DE, Brunt EM, Van Natta M, Behling C, Contos MJ, Cummings OW, et al. 2005. Design and validation of a histological scoring system for nonalcoholic fatty liver disease. Hepatology 41 (6 ):1313–1321, PMID: , 10.1002/hep.20701.15915461
64. Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, Costello EK, et al. 2010. QIIME allows analysis of high-throughput community sequencing data. Nat Methods 7 (5 ):335–336, PMID: , 10.1038/nmeth.f.303.20383131
65. Edgar RC. 2010. Search and clustering orders of magnitude faster than BLAST. Bioinformatics 26 (19 ):2460–2461, PMID: , 10.1093/bioinformatics/btq461.20709691
66. Chao A, Chazdon RL, Colwell RK, Shen TJ. 2006. Abundance-based similarity indices and their estimation when there are unseen species in samples. Biometrics 62 (2 ):361–371, PMID: , 10.1111/j.1541-0420.2005.00489.x.16918900
67. Shannon CE, Weaver W. 1949. The Mathematical Theory of Communication. Champagne, IL: University of Illinois Press, 1–117.
68. Simpson EH. 1949. Measurement of diversity. Nature 163 (4148 ):688–688, 10.1038/163688a0.
69. Segata N, Izard J, Waldron L, Gevers D, Miropolsky L, Garrett WS, et al. 2011. Metagenomic biomarker discovery and explanation. Genome Biol 12 (6 ):R60, PMID: , 10.1186/gb-2011-12-6-r60.21702898
70. Nakanishi Y, Duran A, L’Hermitte A, Shelton PM, Nakanishi N, Reina-Campos M, et al. 2018. Simultaneous loss of both atypical protein kinase C genes in the intestinal epithelium drives serrated intestinal cancer by impairing immunosurveillance. Immunity 49 (6 ):1132–1147.e7, PMID: , 10.1016/j.immuni.2018.09.013.30552022
71. Chiang HY, Lu HH, Sudhakar JN, et al. 2022. IL-22 initiates an IL-18-dependent epithelial response circuit to enforce intestinal host defence. Nat Commun 13 (1 ):874, PMID: , 10.1038/S41467-022-28478-3.35169117
72. Bernink JH, Peters CP, Munneke M, Te Velde AA, Meijer SL, Weijer K, et al. 2013. Human type 1 innate lymphoid cells accumulate in inflamed mucosal tissues. Nat Immunol 14 (3 ):221–229, PMID: , 10.1038/ni.2534.23334791
73. Forkel M, van Tol S, Höög C, Michaëlsson J, Almer S, Mjösberg J. 2019. Distinct Alterations in the composition of mucosal innate lymphoid cells in newly diagnosed and established Crohn’s disease and ulcerative colitis. J Crohns Colitis 13 (1 ):67–78, PMID: , 10.1093/ecco-jcc/jjy119.30496425
74. Rankin LC, Girard-Madoux MJH, Seillet C, Mielke LA, Kerdiles Y, Fenis A, et al. 2016. Complementarity and redundancy of IL-22-producing innate lymphoid cells. Nat Immunol 17 (2 ):179–186, PMID: , 10.1038/ni.3332.26595889
75. Ijssennagger N, van der Meer R, van Mil SWC. 2016. Sulfide as a mucus Barrier-Breaker in inflammatory bowel disease? Trends Mol Med 22 (3 ):190–199, PMID: , 10.1016/j.molmed.2016.01.002.26852376
76. Mills CD, Kincaid K, Alt JM, Heilman MJ, Hill AM. 2000. M-1/M-2 macrophages and the Th1/Th2 paradigm. J Immunol 164 (12 ):6166–6173, PMID: , 10.4049/jimmunol.164.12.6166.10843666
77. Yang W, Yu T, Huang X, Bilotta AJ, Xu L, Lu Y, et al. 2020. Intestinal microbiota-derived short-chain fatty acids regulation of immune cell IL-22 production and gut immunity. Nat Commun 11 (1 ):4457, PMID: , 10.1038/S41467-020-18262-6.32901017
78. Tolhurst G, Heffron H, Lam YS, Parker HE, Habib AM, Diakogiannaki E, et al. 2012. Short-chain fatty acids stimulate glucagon-like peptide-1 secretion via the G-protein-coupled receptor FFAR2. Diabetes 61 (2 ):364–371, PMID: , 10.2337/db11-1019.22190648
79. Brown AJ, Goldsworthy SM, Barnes AA, Eilert MM, Tcheang L, Daniels D, et al. 2003. The Orphan G protein-coupled receptors GPR41 and GPR43 are activated by propionate and other short chain carboxylic acids. J Biol Chem 278 (13 ):11312–11319, PMID: , 10.1074/jbc.M211609200.12496283
80. Schütte A, Ermund A, Becker-Pauly C, Johansson MEV, Rodriguez-Pineiro AM, Bäckhed F, et al. 2014. Microbial-induced meprin β cleavage in MUC2 mucin and a functional CFTR channel are required to release anchored small intestinal mucus. Proc Natl Acad Sci U S A 111 (34 ):12396–12401, PMID: , 10.1073/pnas.1407597111.25114233
81. Borisova MA, Achasova KM, Morozova KN, Andreyeva EN, Litvinova EA, Ogienko AA, et al. 2020. Mucin-2 knockout is a model of intercellular junction defects, mitochondrial damage and ATP depletion in the intestinal epithelium. Sci Rep 10 (1 ):21135, PMID: , 10.1038/S41598-020-78141-4.33273633
82. Yan Z, Liu Y, Zhang T, Zhang F, Ren H, Zhang Y. 2021. Analysis of microplastics in human feces reveals a correlation between fecal microplastics and inflammatory bowel disease status. Environ Sci Technol 56 (1 ):414–421, PMID: , 10.1021/ACS.EST.1C03924.34935363
83. Al-Sadi R, Boivin M, Ma T. 2009. Mechanism of cytokine modulation of epithelial tight junction barrier. Front Biosci (Landmark Ed) 14 (7 ):2765–2778, PMID: , 10.2741/3413.19273235
84. Ludwiczek O, Vannier E, Borggraefe I, Kaser A, Siegmund B, Dinarello CA, et al. 2004. Imbalance between interleukin-1 agonists and antagonists: relationship to severity of inflammatory bowel disease. Clin Exp Immunol 138 (2 ):323–329, PMID: , 10.1111/j.1365-2249.2004.02599.x.15498044
85. Al-Sadi RM, Ma TY. 2007. IL-1beta causes an increase in intestinal epithelial tight junction permeability. J Immunol 178 (7 ):4641–4649, PMID: , 10.4049/jimmunol.178.7.4641.17372023
86. Ma TY, Iwamoto GK, Hoa NT, Akotia V, Pedram A, Boivin MA, et al. 2004. TNF-alpha-induced increase in intestinal epithelial tight junction permeability requires NF-kappa B activation. Am J Physiol Gastrointest Liver Physiol 286 (3 ):G367–G376, PMID: , 10.1152/AJPGI.00173.2003.14766535
87. Artis D, Spits H. 2015. The biology of innate lymphoid cells. Nature 517 (7534 ):293–301, PMID: , 10.1038/nature14189.25592534
88. Turner JE, Stockinger B, Helmby H. 2013. IL-22 mediates goblet cell hyperplasia and worm expulsion in intestinal helminth infection. PLoS Pathog 9 (10 ):e1003698, PMID: , 10.1371/JOURNAL.PPAT.1003698.24130494
89. Chun E, Lavoie S, Fonseca-Pereira D, Bae S, Michaud M, Hoveyda HR, et al. 2019. Metabolite-sensing receptor Ffar2 regulates colonic group 3 innate lymphoid cells and gut immunity. Immunity 51 (5 ):871–884.e6, PMID: , 10.1016/j.immuni.2019.09.014.31628054
90. Sepahi A, Liu QY, Friesen L, Kim CH. 2021. Dietary fiber metabolites regulate innate lymphoid cell responses. Mucosal Immunol 14 (2 ):317–330, PMID: , 10.1038/s41385-020-0312-8.32541842
91. Fachi JL, Sécca C, Rodrigues PB, Pinheiro de Mato FC, Di Luccia B, de Souza Felipe J, et al. 2020. Acetate coordinates neutrophil and ILC3 responses against C. difficile through FFAR2. J Exp Med 217 (3 ):jem.20190489, PMID: , 10.1084/jem.20190489.31876919
92. Vieira ELM, Leonel AJ, Sad AP, Beltrão NRM, Costa TF, Ferreira TMR, et al. 2012. Oral administration of sodium butyrate attenuates inflammation and mucosal lesion in experimental acute ulcerative colitis. J Nutr Biochem 23 (5 ):430–436, PMID: , 10.1016/j.jnutbio.2011.01.007.21658926
93. Gao Z, Yin J, Zhang J, Ward RE, Martin RJ, Lefevre M, et al. 2009. Butyrate improves insulin sensitivity and increases energy expenditure in mice. Diabetes 58 (7 ):1509–1517, PMID: , 10.2337/db08-1637.19366864
94. Rodríguez-Cabezas ME, Gálvez J, Lorente MD, Concha A, Camuesco D, Azzouz S, et al. 2002. Dietary fiber down-regulates colonic tumor necrosis factor alpha and nitric oxide production in trinitrobenzenesulfonic acid-induced colitic rats. J Nutr 132 (11 ):3263–3271, PMID: , 10.1093/jn/132.11.3263.12421838
95. Tarrerias AL, Millecamps M, Alloui A, Beaughard C, Kemeny JL, Bourdu S, et al. 2002. Short-chain fatty acid enemas fail to decrease colonic hypersensitivity and inflammation in TNBS-induced colonic inflammation in rats. Pain 100 (1–2 ):91–97, PMID: , 10.1016/S0304-3959(02)00234-8.12435462
96. Yamashita H, Fujisawa K, Ito E, Idei S, Kawaguchi N, Kimoto M, et al. 2007. Improvement of obesity and glucose tolerance by acetate in type 2 diabetic Otsuka Long-Evans Tokushima fatty (OLETF) rats. Biosci Biotechnol Biochem 71 (5 ):1236–1243, PMID: , 10.1271/bbb.60668.17485860
97. Scheppach W, Sommer H, Kirchner T, Paganelli GM, Bartram P, Christl S, et al. 1992. Effect of butyrate enemas on the colonic mucosa in distal ulcerative colitis. Gastroenterology 103 (1 ):51–56, PMID: , 10.1016/0016-5085(92)91094-K.1612357
98. Hamer HM, Jonkers DMAE, Vanhoutvin SALW, Troost FJ, Rijkers G, de Bruïne A, et al. 2010. Effect of butyrate enemas on inflammation and antioxidant status in the colonic mucosa of patients with ulcerative colitis in remission. Clin Nutr 29 (6 ):738–744, PMID: , 10.1016/j.clnu.2010.04.002.20471725
99. Vernia P, Annese V, Bresci G, d’Albasio G, D’Incà R, Giaccari S, et al. 2003. Topical butyrate improves efficacy of 5-ASA in refractory distal ulcerative colitis: results of a multicentre trial. Eur J Clin Invest 33 (3 ):244–248, PMID: , 10.1046/j.1365-2362.2003.01130.x.12641543
100. Lin H. V, Frassetto A, Kowalik EJ Jr, Nawrocki AR, Lu MM, Kosinski JR, et al. 2012. Butyrate and propionate protect against diet-induced obesity and regulate gut hormones via free fatty acid receptor 3-independent mechanisms. PLoS One 7 (4 ):e35240, PMID: , 10.1371/journal.pone.0035240.22506074
101. Shin NR, Whon TW, Bae JW. 2015. Proteobacteria: microbial signature of dysbiosis in gut microbiota. Trends Biotechnol 33 (9 ):496–503, PMID: , 10.1016/j.tibtech.2015.06.011.26210164
102. Méndez-Salazar EO, Ortiz-López MG, Granados-Silvestre MDLÁ, Palacios-González B, Menjivar M. 2018. Altered Gut microbiota and compositional changes in Firmicutes and Proteobacteria in Mexican undernourished and obese children. Front Microbiol 9 :2494, PMID: , 10.3389/FMICB.2018.02494.30386323
103. Mangin I, Bonnet R, Seksik P, Rigottier-Gois L, Sutren M, Bouhnik Y, et al. 2004. Molecular inventory of faecal microflora in patients with Crohn’s disease. FEMS Microbiol Ecol 50 (1 ):25–36, PMID: , 10.1016/j.femsec.2004.05.005.19712374
104. Gophna U, Sommerfeld K, Gophna S, Doolittle WF, Veldhuyzen Van Zanten SJO. 2006. Differences between tissue-associated intestinal microfloras of patients with Crohn’s disease and ulcerative colitis. J Clin Microbiol 44 (11 ):4136–4141, PMID: , 10.1128/JCM.01004-06.16988016
105. Lavelle A, Lennon G, O’Sullivan O, Docherty N, Balfe A, Maguire A, et al. 2015. Spatial variation of the colonic microbiota in patients with ulcerative colitis and control volunteers. Gut 64 (10 ):1553–1561, PMID: , 10.1136/gutjnl-2014-307873.25596182
106. Volynets V, Louis S, Pretz D, Lang L, Ostaff MJ, Wehkamp J, et al. 2017. Intestinal barrier function and the gut microbiome are differentially affected in mice fed a Western-style diet or drinking water supplemented with fructose. J Nutr 147 (5 ):770–780, PMID: , 10.3945/jn.116.242859.28356436
107. Cho KY. 2021. Lifestyle modifications result in alterations in the gut microbiota in obese children. BMC Microbiol 21 (1 ):10, PMID: , 10.1186/S12866-020-02002-3.33407104
108. Schnorr SL, Candela M, Rampelli S, Consolandi C, Basaglia G, Turroni S, et al. 2014. Gut microbiome of the Hadza hunter-gatherers. Nat Commun 5 :3654, PMID: , 10.1038/NCOMMS4654.24736369
109. Downes J, Dewhirst FE, Tanner ACR, Wade WG. 2013. Description of Alloprevotella rava gen. nov., sp. nov., isolated from the human oral cavity, and reclassification of Prevotella tannerae Moore et al. 1994 as Alloprevotella tannerae gen. nov., comb. nov. Int J Syst Evol Microbiol 63 (Pt 4 ):1214–1218, PMID: , 10.1099/ijs.0.041376-0.22753527
110. Kong C, Gao R, Yan X, Huang L, Qin H. 2019. Probiotics improve gut microbiota dysbiosis in obese mice fed a high-fat or high-sucrose diet. Nutrition 60 :175–184, PMID: , 10.1016/j.nut.2018.10.002.30611080
111. Kreutzer C, Peters S, Schulte DM, Fangmann D, Türk K, Wolff S, et al. 2017. Hypothalamic Inflammation in human obesity is mediated by environmental and genetic factors. Diabetes 66 (9 ):2407–2415, PMID: , 10.2337/db17-0067.28576837
112. Zhang C, Zhang M, Pang X, Zhao Y, Wang L, Zhao L. 2012. Structural resilience of the gut microbiota in adult mice under high-fat dietary perturbations. ISME J 6 (10 ):1848–1857, PMID: , 10.1038/ismej.2012.27.22495068
113. Xie G, Wang X, Huang F, Zhao A, Chen W, Yan J, et al. 2016. Dysregulated hepatic bile acids collaboratively promote liver carcinogenesis. Int J Cancer 139 (8 ):1764–1775, PMID: , 10.1002/ijc.30219.27273788
114. Danneskiold-Samsøe NB, Andersen D, Radulescu ID, Normann-Hansen A, Brejnrod A, Kragh M, et al. 2017. A safflower oil based high-fat/high-sucrose diet modulates the gut microbiota and liver phospholipid profiles associated with early glucose intolerance in the absence of tissue inflammation. Mol Nutr Food Res 61 (5 ), PMID: , 10.1002/MNFR.201600528.28012235
115. Ibrahim KS, Bourwis N, Dolan S, Lang S, Spencer J, Craft JA. 2021. Characterisation of gut microbiota of obesity and type 2 diabetes in a rodent model. Biosci Microbiota Food Health 40 (1 ):65–74, PMID: , 10.12938/bmfh.2019-031.33520571
116. Krych Ł, Nielsen DS, Hansen AK, Hansen CHF. 2015. Gut microbial markers are associated with diabetes onset, regulatory imbalance, and IFN-γ level in NOD mice. Gut Microbes 6 (2 ):101–109, PMID: , 10.1080/19490976.2015.1011876.25648687
117. Qin J, Li Y, Cai Z, Li S, Zhu J, Zhang F, et al. 2012. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature 490 (7418 ):55–60, PMID: , 10.1038/nature11450.23023125
118. Petersen C, Bell R, Klag KA, Lee S-H, Soto R, Ghazaryan A, et al. 2019. T cell-mediated regulation of the microbiota protects against obesity. Science 365 (6451 ):eaat9351, PMID: , 10.1126/science.aat9351.31346040
119. Le Chatelier E, Nielsen T, Qin J, Prifti E, Hildebrand F, Falony G, et al. 2013. Richness of human gut microbiome correlates with metabolic markers. Nature 500 (7464 ):541–546, PMID: , 10.1038/nature12506.23985870
120. Pitt JA, Kozal JS, Jayasundara N, Massarsky A, Trevisan R, Geitner N, et al. 2018. Uptake, tissue distribution, and toxicity of polystyrene nanoparticles in developing zebrafish (Danio rerio). Aquat Toxicol 194 :185–194, PMID: , 10.1016/j.aquatox.2017.11.017.29197232
121. Pitt JA, Trevisan R, Massarsky A, Kozal JS, Levin ED, di Giulio RT. 2018. Maternal transfer of nanoplastics to offspring in zebrafish (Danio rerio): a case study with nanopolystyrene. Sci Total Environ 643 :324–334, PMID: , 10.1016/j.scitotenv.2018.06.186.29940444
122. Brun NR, van Hage P, Hunting ER, Haramis A-PG, Vink SC, Vijver MG, et al. 2019. Polystyrene nanoplastics disrupt glucose metabolism and cortisol levels with a possible link to behavioural changes in larval zebrafish. Commun Biol 2 (1 ): PMID: , 10.1038/s42003-019-0629-6.31646185
|
PMC009xxxxxx/PMC9969990.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36847817
EHP11076
10.1289/EHP11076
Research
Mechanisms and Consequences of Variable TRPA1 Expression by Airway Epithelial Cells: Effects of TRPV1 Genotype and Environmental Agonists on Cellular Responses to Pollutants in Vitro and Asthma
Rapp Emmanuel 1
Lu Zhenyu 1
Sun Lili 1
Serna Samantha N. 1
Almestica-Roberts Marysol 1
Burrell Katherine L. 1
Nguyen Nam D. 1
Deering-Rice Cassandra E. 1
https://orcid.org/0000-0002-5006-1982
Reilly Christopher A. 1
1 Department of Pharmacology and Toxicology, Center for Human Toxicology, University of Utah, Salt Lake City, Utah, USA
Address correspondence to Christopher A. Reilly, Department of Pharmacology and Toxicology, Center for Human Toxicology, University of Utah, 30 S. 2000 E., Room 201 Skaggs Hall, Salt Lake City, UT 84112 USA. Telephone: (801) 581-5236. Email: [email protected]
27 2 2023
2 2023
131 2 02700908 2 2022
20 1 2023
20 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Transient receptor potential ankyrin-1 [transient receptor potential cation channel subfamily A member 1 (TRPA1)] and vanilloid-1 [transient receptor potential cation channel subfamily V member 1 (TRPV1)] detect inhaled irritants, including air pollutants and have roles in the development and exacerbation of asthma.
Objectives:
This study tested the hypothesis that increased expression of TRPA1, stemming from expression of the loss-of-function TRPV1 (I585V; rs8065080) polymorphic variant by airway epithelial cells may explain prior observations of worse asthma symptom control among children with the TRPV1 I585I/V genotype, by virtue of sensitizing epithelial cells to particulate materials and other TRPA1 agonists.
Methods:
TRP agonists, antagonists, small interfering RNA (siRNA), a nuclear factor kappa light chain enhancer of activated B cells (NF-κB) pathway inhibitor, and kinase activators and inhibitors were used to modulate TRPA1 and TRPV1 expression and function. Treatment of genotyped airway epithelial cells with particulate materials and analysis of asthma control data were used to assess consequences of TRPV1 genotype and variable TRPA1 expression on cellular responses in vitro and asthma symptom control among children as a function of voluntarily reported tobacco smoke exposure.
Results:
A relationship between higher TRPA1 expression and function and lower TRPV1 expression and function was revealed. Findings of this study pointed to a mechanism whereby NF-κB promoted TRPA1 expression, whereas NF-κB–regulated nucleotide-binding oligomerization domain, leucine rich repeat and pyrin domain containing 2 (NLRP2) limited expression. Roles for protein kinase C and p38 mitogen activated protein kinase were also demonstrated. Finally, the TRPV1 I585I/V genotype was associated with increased TRPA1 expression by primary airway epithelial cells and amplified responses to selected air pollution particles in vitro. However, the TRPV1 I585I/V genotype was not associated with worse asthma symptom control among children exposed to tobacco smoke, whereas other TRPA1 and TRPV1 variants were.
Discussion:
This study provides insights on how airway epithelial cells regulate TRPA1 expression, how TRPV1 genetics can affect TRPA1 expression, and that TRPA1 and TRPV1 polymorphisms differentially affect asthma symptom control. https://doi.org/10.1289/EHP11076
Supplemental Material is available online (https://doi.org/10.1289/EHP11076).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Transient receptor potential ankyrin-1 [transient receptor potential cation channel subfamily A member 1 (TRPA1)] and vanilloid-1 [transient receptor potential cation channel subfamily V member 1 (TRPV1)] are cation-permeable (primarily calcium) channels involved in sensory physiology and pain.1,2 These receptors are also proximal sensors of selected inhaled chemicals and particulate materials (PMs), and coordinate responses of lung cells exposed to environmental irritants and pneumotoxins.3–6
Asthma is a chronic airway inflammatory disease that is affected by exposure to environmental irritants and pollutants, including volatile/semivolatile chemicals and PM released during the combustion of various materials (e.g., petroleum products/fuels, coal, biomass/biomass-based materials, cigarettes/E-cigarettes).7–13 TRPA1 and TRPV1 are important in the pathogenesis and exacerbation of asthma,1,2,14–17 and the expression and activity of TRPA118 and TRPV119,20 can be elevated in the airways of people with asthma and chronic cough. Studies in mice have demonstrated a need for TRPA1 in the development of Th2-high allergic asthma phenotypes,21 with similar findings reported for TRPV1.22,23 Several representative PMs, including diesel exhaust particles (DEP),4,5 wood smoke particles (WSPM),6,24,25 and coal fly ash (CFA),3,15,26 activate TRPA1 and TRPV1. Moreover, in airway epithelial cells (AECs), TRP activation by PM can trigger pro-inflammatory responses and pathological endoplasmic reticulum stress, leading to cell death. In addition, TRPA1 activation by WSPM can promote mucin 5AC (MUC5AC) expression and secretion by AECs, which could contribute to asthma exacerbation and symptomatology.24
Previous studies by our group demonstrated that the predominantly co-inherited TRPA1 gain-of-function polymorphisms R3C and R58T (rs13268757 and rs16937976) correlated with reduced likelihood (relative risk=0.27; p=0.14) of achieving optimal asthma symptom control in children.26 Studies have also shown the TRPV1 I585V (rs8065080) variant to be less responsive to canonical TRPV1 agonists and stimuli27 and to be associated with reduced cough in people with asthma.27,28 However, our research has shown that the TRPV1 I585I/V genotype occurs at a higher frequency in children with moderate-to-severe steroid-resistant asthma, specifically correlating with more frequent asthma symptoms (e.g., cough/wheeze, nighttime awakenings, inhaler use, care needs; odds ratio=2.04; p=0.03) and decreased overall asthma control when compared with those with either the I585I/I or I585V/V genotypes or combinations of two additional common TRPV1 polymorphisms (i.e., I315M; rs222747 and T469I; rs224534).15 A basis for this clinical association was hypothesized to result from higher TRPA1 expression by AECs of individuals with the I585I/V genotype, potentially increasing the sensitivity of such individuals to common asthma triggers that activate TRPA1.
Currently, neither how TRPA1 expression is regulated by AECs, nor the basis for how the TRPV1 I585I/V genotype promotes TRPA1 expression are understood. Moreover, the potential impact of elevated TRPA1 expression on asthma is not fully understood. Goals of this work were to a) understand how TRPA1 expression is regulated and specifically promoted by the TRPV1 I585I/V genotype in AECs, b) determine the importance of elevated TRPA1 expression in regulating responses of AECs to pneumotoxic PM treatment, and c) evaluate the contribution of the I585I/V genotype and TRPA1 expression on asthma symptom control as a function of tobacco smoke exposure among children with asthma.
Materials and Methods
Chemicals and Other Materials
N-(4-tert-butylbenzyl)-N-(1-[3-fluoro-4-(methylsulfonylamino)phenyl]ethyl)thiourea (LJO-328) was provided by J. Lee of Seoul National University. The structure has been published.29 2-Mercaptoethanol, n-vanillylnonanamide (nonivamide; a capsaicin analog), allyl isothiocyanate (AITC), ionomycin calcium salt, and phorbol 12-myristate 13-acetate (PMA) were from Sigma-Aldrich. Dimethyl sulfoxide (DMSO) was from Fisher Scientific. BMS-345541 and Go6983 were from Tocris, and recombinant human tumor necrosis factor-alpha (TNFα), interleukin-1 alpha (IL1α), IL1β, IL6, and IL13 were from Peprotech. 4-(4-Fluorophenyl)-2-(4-nitrophenyl)-5-(4-pyridyl)-1H-imidazole (PD169316) was from Cayman Chemical.
WSPM, DEP, and CFA
Preparation of pine WSPM,6 and the sources and properties of the DEP4,5 and CFA3 have been described. Briefly, WSPM was prepared by burning ∼10g of Austrian pine (from a tree growing in the Salt Lake Valley; 1.5cm long×0.2–0.5cm wide) using a pipe furnace at 750°C with constant air flow. WSPM was collected using an Anderson cascade impactor operated at 1L/min, and fractions 6 and 7 (0.65–1.1μm and 0.43–0.65μm) were used. For experiments, WSPM concentrate was suspended in DMSO at 115mg/mL and diluted to 0.076mg/mL in media containing ≤0.2% DMSO to achieve a 20-μg/cm2 area dose in a single well of a 6-well plate. Features of pine WSPM, and the effects that it has on AECs (i.e., calcium flux, pro-inflammatory, cytostatic, and cytotoxic) have been described.24,25,30,31 Specifically, the material was shown to contain TRPA1 agonists, including resin acids, perinaphthenone, coniferaldehyde, ethyl phenols, and substituted xylenols; ethyl phenols and xylenols also activate TRPV3.6,30 The DEP was collected from idling diesel-powered vehicles during emissions testing in the Salt Lake City area.4 The PM was suspended in media containing ≤0.2% DMSO and applied to cells at a final area dose of 10 μg/cm2. The DEP used in this study was shown to contain the TRPA1 agonists 2,4-di-tert-butylphenol (also a TRPV3 agonist) and several quinones (benzo and naptho), as well as perinanpthenone.4 Finally, CFA was collected from the Hunter power plant in Castle Dale, Utah, and fractionated to <10μm. The ash was from low-sulfur bituminous coal, which was previously reported to consist of primarily insoluble oxides and salts of silicon, calcium, aluminum, and iron, with ∼3% elemental carbon and ∼1% unspecified organic carbon32 (presumably polycyclic aromatic hydrocarbons as described for other CFA samples33). Effects of CFA on AECs and TRP channels have also been described.3,26,34
Cell Culture
BEAS-2B and HEK-293 cells were from ATCC. These and other cells were maintained in a humidified cell culture incubator at 37°C with a 95% air:5% carbon dioxide (CO2) atmosphere. Human TRPV1-overexpressing HEK-293 and BEAS-2B cells were generated as previously described.15,35 Briefly, cells were transfected with a pcDNA3.1 plasmid (ThermoFisher) harboring human TRPV1, selected using Geneticin (300μg/mL), and expanded from a single colony. Overexpression was verified using calcium flux assays and western blots. TRPV1-overexpressing HEK-293 cells were cultured in Dulbecco’s Modified Eagle Medium/Ham’s F12 (DMEM:F12) media containing 5% fetal bovine serum, 1× penicillin/streptomycin, and 300μg/mL Geneticin (ThermoFisher), and BEAS-2B cells were cultured in LHC-9 medium fortified with 300μg/mL Geneticin. Normal (NHBE; CC-2540 and CC-2541) and diseased (DHBE; CC-2540 and 00194911) human bronchial epithelial cells (HBEs) were from Lonza and were cultured in bronchial epithelial cell growth medium (Lonza). NHBEs immortalized with cyclin-dependent kinase 4 (CDK4) and telomerase reverse transcriptase (hTERT), or HBEC3-KT cells, were from ATCC (CRL-4051) and were grown in airway epithelial cell basal medium supplemented with bronchial epithelial cell growth kit, 30μg/mL Geneticin, and 250 ng/mL puromycin (ATCC). Human small airway epithelial cells (SAECs; CC-2547) were from Lonza and were grown in BronchiaLife epithelial airway medium complete kit.
Cell Genotyping
NHBE and DHBE cells were genotyped using TaqMan Genotyping Master Mix and assays for TRPV1 I585V (ThermoFisher #C_11679656_10) I315M (ThermoFisher #C_1093688_10), T469I (ThermoFisher #C_1093674), TRPA1 R3C (ThermoFisher #C_2175739_10), and R58T (ThermoFisher #C_25646603_10). Genomic DNA was isolated from cells using the GeneElute Mammalian Genomic DNA Miniprep Kit (Sigma-Aldrich). DNA was then quantified by ultraviolet (UV) absorbance on a Nanodrop Onec (ThermoFisher) and assayed using a Life Technologies QuantStudio 6 Flex instrument (ThermoFisher) and the polymerase chain reaction (PCR) program specified by the supplier for the TaqMan Genotyping Master Mix. Genotypes were differentiated by relative probe fluorophore intensity using pcDNA3.1 plasmids harboring human TRPV1, human TRPA1, and the target single nucleotide polymorphism (SNP) variant mutations.15,26 Genotyping results and donor lot identifications for the cells used in this work are provided in Table S1. HBEC3-KT cells have the TRPV1 I585I/V genotype, BEAS-2B cells the I585I/I genotype, SAECs the I585I/V genotype, and NHBEs have variable genotypes.
Quantitative Real-Time PCR
Cells were plated at 10,000 cells/cm2 in 6-well plates and treated at ∼90% confluence (3 d post plating with feeding on day 2). After treatment, total RNA was isolated using the PureLink RNA Mini Kit (Invitrogen). Total RNA (2μg) was quantified by UV absorbance on a Nanodrop Onec (ThermoFisher) and complementary DNA (cDNA) was synthesized using the ABI High-Capacity cDNA Synthesis Kit with RNase inhibitor (Applied Biosystems). The cDNA was then subjected to analysis by quantitative real-time PCR (q-PCR) using TaqMan Gene Expression Master Mix (ThermoFisher) and a Life Technologies QuantStudio 6 Flex instrument. The following TaqMan probe-based assays were used: human glyceraldehyde 3-phosphate dehydrogenase (GAPDH; Hs99999905_m1), human TRPA1 (Hs00175798_m1), human TRPV1 (Hs00218912_m1), human TRPV3 (Hs00376854_m1), human NLRP2 (Hs01546932_m1), human IGFBP2 (Hs01040719_m1), human DNA damage-inducible transcript-3 (DDIT3; Hs_00358796_g1), and human interleukin-8 (IL8; Hs00174103_m1). The PCR programs used were according to the supplier for the TaqMan Gene Expression Master Mix and probe assays. mRNA expression was normalized to the housekeeping gene, human β2-microglobulin (β2M; Hs00984230_m1), and the average value of control samples (i.e., the comparative ΔΔCt method)36 with relative quantification (Rq) reported.
Small Interfering RNA Studies
Small interfering RNA (siRNA) transfections were performed according to the ThermoFisher Stealth/siRNA Transfection Lipofectamine 2000 Protocol. HBEC3-KT cells were plated in 6-well plates at ∼5,000 cells/cm2 and transfected with siRNA at 30%–50% confluence (2 d post plating), as per the manufacturer protocol. Briefly, siRNAs were reconstituted to a concentration of 50 pmol/μL in nuclease-free water. An aliquot containing 100 or 500 pmol/mL negative control (Silencer Negative Control siRNA No.1; ThermoFisher #AM4611), positive control GAPDH (ThermoFisher #4404024), NLRP2 siRNA-1 or -2 (ThermoFisher siRNA ID 25505; AM16708), or TRPV1 siRNA (ThermoFisher siRNA ID 105495; AM16708) was diluted in antibiotic and serum-free HBEC3-KT media and incubated in the dark at room temperature. After 5 min, the siRNA was mixed 1:1 with the diluted Lipofectamine 2000 and incubated for 25 min in the dark at room temperature. After incubation, the siRNA:Lipofectamine complex was diluted with 700μL of antibiotic-free and serum-free HBEC3-KT media to a final volume of 1mL. This solution was then added to an individual well in a 6-well plate and incubated for 6 h, upon which the transfection solution was replaced with fresh HBEC3-KT media. Isolation and quantification of mRNA after siRNA transfection was performed as described above, 24 h posttransfection. For western blotting, cells were plated at ∼5,000 cells/cm2 in 75-cm2 flasks and the siRNA volumes and concentrations were adjusted according to the supplier protocol.
Western Blotting
Western blots were performed as previously described by Nguyen et al.25 For isolating nuclear protein, the NE-PER Nuclear and Cytoplasmic Extraction Reagents kit and protocol (ThermoFisher) was used. Cells were grown to confluence in 75-cm2 flasks. Total protein was harvested on ice using radioimmunoprecipitation buffer, supplemented with 6 M urea, 1% sodium dodecyl sulfate, and Halt protease inhibitor. Lysates were sonicated on ice using 10×1-s pulses at 100 W, repeated 10 times and clarified by centrifugation at 13,000×g for 15 min at 4°C. Protein concentrations were determined using the bicinchoninic acid protein assay kit (ThermoFisher), and 30μg was loaded into each well of a 4%–12% Bolt Bis-Tris 12-well gel and resolved by electrophoresis for 1.5 h at 120 V. The Precision Plus Protein Dual Color Standards (5μL) ladder was also used (BioRad). Following electrophoresis, the proteins were transferred to a polyvinylidene fluoride membrane using the iBlot 2 gel transfer device. After transfer, the membrane was incubated in SuperBlock (ThermoFisher) for 1 h at room temperature. Primary rabbit monoclonal antibodies against nuclear factor kappa light chain enhancer of activated B cells (NF-κB)/p65 (#3034; Cell Signaling Technology) and NF-κB-phospho-S536 (#3033; Cell Signaling Technology) were used at 1:5,000 dilution. Polyclonal antibodies against NLRP2 (ThermoFisher #15182-1-AP) and rabbit nuclear matrix protein (p84; ThermoFisher, PA5-69083) were used at 1:1,000 dilution. A mouse monoclonal antibody against β-actin (8H10D10; Cell Signaling Technology #3700) was used at 1:10,000 dilution. Rabbit GAPDH (D16H11) monoclonal antibody (Cell Signaling Technology #5174) was used at 1:1000 dilution. All primary antibodies were prepared in 5% bovine serum albumin with 0.1% sodium azide and incubated with the membranes at 4°C for 16 h. Horseradish peroxidase-conjugated sheep-antimouse (NA931) and anti-rabbit (NA934) secondary antibodies (GE Health Sciences) were used at 1:10,000 in SuperBlock and were also applied for 16 h at 4°C. SuperSignal West Dura Extended Duration Substrate (ThermoFisher) was added to the membrane and visualized using a FluorChem M imager with the chemiluminescence plus markers setting. Bands were quantified using densitometry analysis of 8-bit images in ImageJ37 and normalized to the respective gene, as outlined in the figure legends.
RNA Sequencing
RNA sequencing was performed as previously described15 at the High Throughput Genomics Core Facility at the Huntsman Cancer Institute, University of Utah. NHBE cells from the four donors (14359 and 14664 for TRPV1 I585I/I, and 9853 and unknown for TRPV1 I585I/V) were used. Total RNA was extracted using the RNeasy Mini kit (Qiagen) with on-column DNase digestion. RNA quality was assessed by RNA nanochip technology, and library construction was performed using the Illumina TruSeq Stranded mRNA Sample Preparation kit using established protocols. The sequencing libraries (18 pM) were then chemically denatured and applied to an Illumina TruSeq version 3 single-read flow cell using an Illumina cBot. Hybridized molecules were clonally amplified and annealed to sequencing primers with reagents from an Illumina TruSeq SR Cluster kit, version 3-cBot-HS. Following transfer of the flow cell to an Illumina HiSeq instrument, a 50-cycle single-read sequence run was performed using TruSeq SBS version 3 sequencing reagents. Data were processed at the University of Utah Bioinformatics core. The data presented in this publication have been deposited in the National Center for Biotechnology Information’s (NCBI’s) Gene Expression Omnibus (GEO) and are accessible through GEO Series accession no. GSE85447 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE85447).
Calcium Flux/TRP Activity Assays
Calcium flux in HBEC3-KT cells was measured using the Fluo-4 Direct assay kit and imaging on an EVOS FL Auto microscope at 10× magnification using a green fluorescent protein (GFP) filter.3–5,25,26,30 Briefly, cells were plated in a flat-bottom 96-well plate at ∼10,000 cells/cm2, fed on day 2 post plating, and assayed at 80%–90% confluence 3 d post plating. Pretreatment of cells with nonivamide (10μM), LJO-328 (25μM), pine WSPM (20 μg/cm2), TNFα (50 ng/mL), and BMS-345541 (10μM) occurred 12 or 24 h before the assay. Prior to the assay, the cells were loaded with 1× Fluo-4 diluted 1:1 in LHC-9 at 37°C for 1 h. After 1 h, the Fluo-4 was removed and the cells were washed with LHC-9 containing 1 mM probenecid and 0.75 mM trypan red (ATT Bioquest). The assays were performed in an on-stage environmental chamber maintained at 37°C in a 95% air:5% CO2 atmosphere. Agonist treatments (AITC 25 or 150μM) were added to cells at 3× the desired final concentration in LHC-9, which contained 111.1μM calcium. Images were captured every 6 s for 72 s. Changes in fluorescence were quantified using a custom MATLAB program, as previously described.3,5 Reported values are from the 60-s time point and were corrected by subtracting the fluorescence response to a blank media control (i.e., no agonist) and then normalized to the response value at 72 s following ionomycin (10μM) treatment applied after the 60-s image was taken.
Calcium flux in HEK-293 cells stably overexpressing TRPV1 (HEKV1OE cells) was measured essentially as above using a BMG Labtech NOVOStar fluorescence plate reader. The following differences applied: HEKV1OE cells were plated in flat-bottom 96-well plates coated with 1% gelatin at ∼30,000 cells/well and assayed at 100% confluence 1–2 d post plating. Pretreatments (12 h) included Go6983 (10μM) or media containing ≤0.2% DMSO (control). Changes in fluorescence were determined using the NOVOStar analysis software (MARS version 2.41). The values were normalized to the response elicited by LHC-9 media and are reported as the maximum change in fluorescence intensity from plots of change in fluorescence vs. time (60 s total assay time).
Asthma Cohort Studies
Details on the cohort used in this study have been published,15,26,38 and updated demographic and other information relevant to asthma are summarized in Table S2. Briefly, participants were recruited from the emergency department and inpatient wards at Primary Children’s Hospital, University of Utah, Salt Lake City, Utah, as part of an institutional review board–approved study of factors influencing asthma symptom control. All patients/parents/guardians provided informed consent prior to enrollment, DNA sampling, and data collection and analysis. Saliva samples were obtained prospectively from children 2–17 years of age with a physician-confirmed diagnosis of asthma. Information on chronic medical conditions, concomitant medication use, and the chief complaint at the time of enrollment was collected during enrollment and subsequent medical chart abstraction. The level of asthma control in each subject was assessed using a questionnaire based on guidelines modified from the National Heart, Lung, and Blood Institute’s Expert Panel Report 339 (Table S3), as described by Stockman et al.38 The questionnaire consisted of five questions scored on a four-point scale (i.e., 0, 1, 2 or 3). The asthma control score equals the sum of the five item scores, where 0 represents well controlled and 15 represents poorly/not controlled.
Saliva Collection, Genomic DNA Extraction, and SNP Genotyping
DNA was collected using either an Oragene DNA Kit (DNA Genotek) or Zymo DNA/RNA shield collection tube with swab (Zymo Research) and extracted using the GeneElute Mammalian Genomic DNA Miniprep Kit (Sigma-Aldrich). DNA was then quantified by Nanodrop and assayed for TRP SNPs using custom 64- or 128-feature TaqMan Open Array cards (ThermoFisher) containing assays for the TRPV1 I315M, T469I, I585V, and TRPA1 R3C and R58T SNPs, among others, at the University of Utah Genomics core facility. Prior to genotyping, 4 ng of genomic DNA was amplified using a custom TaqMan PreAmp Mastermix specific to the array card features. The TaqMan reactions were cycled as recommended by the manufacturer on a Life Technologies QuantStudio 12K instrument. Data clustering and SNP identification analysis were performed using TaqMan Genotyper software (v1.3.1; Life Technologies).
Statistical Analysis and Graphics
Graphing and statistical analyses were performed using GraphPad Prism (version 7.03). Values are represented as the mean±standard deviation (SD). One-tailed unpaired Student’s t-tests; one-way analysis of variance (ANOVA) and Tukey’s multiple comparisons test; two-way ANOVA and either a Bonferroni, Tukey’s, or Dunnett’s multiple comparisons test; and repeated measures two-way ANOVA and Bonferroni’s multiple comparisons test were used as specified in the figure legends. A p-value of ≤0.05 was considered significant for all experiments. Schematics were prepared using Adobe Illustrator (v24.4.8).
Results
TRPA1 mRNA Expression in AECs with the TRPV1 I585I/V Genotype
Transcriptome analysis comparing NHBEs with either the TRPV1 I585I/I or I585I/V genotype revealed an enrichment of TRPA1 mRNA in cells with the I585I/V genotype (Figure 1A), as previously reported.15 Using 13 unique donor lots of NHBEs and qPCR, cells with the TRPV1 I585I/I genotype expressed on average 3.9-fold (p=0.0183) lower TRPA1 mRNA than cells with the I585I/V genotype, whereas TRPV1 was equivalent (Figure 1B,C). Stratification of these data based on the TRPV1 I315M and T469I SNPs demonstrated essentially no difference in TRPA1 mRNA expression, but TRPV1 mRNA was on average 1.8-fold (p=0.0083) more abundant in cells with the TRPV1 T469I/T genotype (Figure S1). Agreement between changes in TRPA1 and TRPV1 mRNA and activity/protein in lung epithelial cells has previously been shown.15,25,29,40 In cells with the TRPV1 I585I/I genotype, mRNA for the NF-κB-induced and NF-κB regulatory genes NLRP241,42 and IGFBP243–45 were enriched; NLRP2 and IGFBP2 mRNA expression was 2.2- (p=0.0016) and 5.7-fold (p=0.0070) lower in cells with the TRPV1 I585I/V genotype compared with cells with the I585 I/I genotype (Figure 1D,E).
Figure 1. (A) Volcano plot of RNA sequencing data from NHBE cells with either the TRPV1 I585I/I or I585I/V genotypes. Summary data can be found in Excel Table S1. (B) TRPA1, (C) TRPV1, (D) NLRP2, and (E) IGFBP2 mRNA expression in NHBE cells having either the TRPV1 I585I/I or I585I/V genotype (n=6 or 7 donors/genotype). Data (Rq) are the mean±SD, relative to β2M mRNA and the average for target gene expression in NHBEs with the I585I/I genotype. *p≤0.05 and **p<0.01 using a one-tailed unpaired Student’s t-test. Summary data can be found in Excel Table S2 (B,C) Excel Table S3 (D,E). Note: IGFBP2, insulin like growth factor binding protein 2; NHBE, normal human bronchial epithelial (cells); NLRP2, nucleotide-binding oligomerization domain, leucine rich repeat and pyrin domain containing 2; Rq, relative quantification; SD, standard deviation; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; β2M, β2-microglobulin.
Figure 1A is a volcano plot, plotting negative 10 [log (lowercase p value)], ranging from 0 to 450 in increments of 50 (y-axis) across log to the base 2 (fold change), ranging from negative 2 to 2 in unit increments (x-axis) for enriched in I 585 I or I cells and enriched in I 585 I or V cells. Figures 1B to 1E are bar graphs, plotting T R P A 1 messenger ribonucleic acid (relative quantity) (fold T R P V 1 I 5805 I or I average), ranging from 0 to 10 in increments of 2; T R P V 1 messenger ribonucleic acid (relative quantity) (fold T R P V 1 I 5805 I or I average), ranging from 0.0 to 2.0 in increments of 0.5; N L R P 2 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 3.0 in increments of 0.5; and I G F B P 2 messenger ribonucleic acid (relative quantity), ranging from 0 to 8 in increments of 2 (y-axis) across T R P V 1 genotype, ranging as I 585 I or I and I 585 I or V (x-axis).
Effect of TRPV1 Activity on TRPA1 mRNA Expression
The role of variable TRPV1 activity as a basis for differences in TRPA1 mRNA expression was tested by treating HBEC3-KT cells for 24 h with the TRPV1 agonist nonivamide (more active TRPV1, akin to TRPV1 I585I/I) or the antagonist LJO-328 (less active TRPV1, akin to TRPV1 I585I/V). Protracted stimulation of TRPV1 with nonivamide lowered TRPA1 mRNA expression 1.3-fold (p=0.1997) compared with media+DMSO-treated control cells, whereas 1.8-fold higher TRPA1 mRNA expression (p<0.0001) was observed when TRPV1 was inhibited by LJO-328 (Figure 2A, white bars). The difference between nonivamide and LJO-328 treatment was 2.3-fold (p<0.0001) and TRPV1 mRNA expression was not different between cells treated with nonivamide and those treated with LJO-328 (Figure 2B, white bars). Similar differences in TRPA1 mRNA expression were observed using TRPV1-overexpressing BEAS-2B cells (Figure S2); cells treated with nonivamide had 5.1-fold lower TRPA1 mRNA expression (p=0.6425) and cells treated with LJO-328 had 12.5-fold higher expression (p<0.0001).
Figure 2. (A) TRPA1 and (B) TRPV1 mRNA expression in HBEC3-KT cells following 24-h treatment with nonivamide (10μM) or LJO-328 (25μM; white bars), and co-treatment with pine WSPM in media (20 μg/cm2; gray hashed bars). Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and cells treated with media containing 0.2% DMSO only (n=3–6). Control and WSPM co-treated groups were analyzed independently comparing all treatment groups using one-way ANOVA and Tukey’s multiple comparisons test. *p≤0.05, ***p<0.001, and ****p<0.0001. ####Indicates significant difference (p<0.0001) from all WSPM co-treated groups using two-way ANOVA and a Bonferroni multiple comparisons test comparing the corresponding ±WSPM groups. Summary data can be found in Excel Table S4. Note: ANOVA, analysis of variance; DMSO, dimethyl sulfoxide; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); LJO-328, N-(4-tert-butylbenzyl)-N-(1-[3-fluoro-4-(methylsulfonylamino)phenyl]ethyl)thiourea; nonivamide, 2-mercaptoethanol, n-vanillylnonanamide; Rq, relative quantity; SD, standard deviation; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; WSPM, wood smoke particulate matter; β2M, β2-microglobulin.
Figures 2A and 2B are bar graphs, plotting T R P A 1 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 2.5 in increments of 0.5 and 5 to 25 in increments of 5, and T R P V 1 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 2.0 in increments of 0.5 (y-axis) across treatment, ranging as control, nonivamide, L J O-328, control, nonivamide, and L J O-328 (x-axis) for media and W S P M.
TRPA1 mRNA Expression and Effects of TRPV1 Agonists and Antagonists on AECs Treated with Pro-Inflammatory and Cytotoxic Agents
The effects of concurrent treatment of cells with pine WSPM, a pro-inflammatory and pneumotoxic environmental pollutant that activates TRPA1,6,24,25 on the ability of TRPV1 agonists and antagonists to alter TRPA1 mRNA expression was also evaluated. HBEC3-KT cells exposed to WSPM (24 h) alone had 10.9-fold higher TRPA1 mRNA expression (p<0.0001; Figure 2A, gray hashed bars). As above, cells co-treated with nonivamide and WSPM had ∼1.9-fold lower TRPA1 mRNA expression (p<0.0102), whereas cells co-treated with LJO-328 and WSPM had 1.9-fold higher expression (p<0.0003; Figure 2A, gray hashed bars). Also as above, TRPV1 mRNA expression was similar to control cells for cells co-treated with nonivamide and WSPM, but it was 1.6-fold lower (p=0.0396) in cells co-treated with LJO-328 and WSPM (Figure 2B, gray hashed bars). Cells were also treated with alternative inflammatory stimuli, including TNFα, IL1α, IL1β, IL6, and IL13. Cells treated with all but IL13 had ∼1.5- to2-fold higher TRPA1 mRNA expression 4 h posttreatment (Figure S3).
TRP mRNA Expression with WSPM/TRPA1 Agonist Challenge
The temporal profiles of TRPA1 and TRPV1 mRNA expression following WSPM treatment were distinct and opposite (Figure 3A,B), revealing an inverse relationship. As previously observed,25 TRPA1 mRNA increased in control cells provided fresh media at time=0, between 6 and 20 h. This did not occur with WSPM treatment. Accordingly, TRPA1 mRNA was markedly lower relative to the media-treated control cells when normalized at each time point. However, after ∼24h, as the WSPM-treated cells visibly recovered from stress/damage and restored a monolayer, TRPA1 mRNA was higher (1.6-fold; p=0.0131) compared with control cells, as in Figure 2A. The suppression of TRPA1 mRNA expression was also observed in cells treated with the TRPA1 agonists AITC, coniferaldehyde, and 2,4-di-tert-butylphenol (Figure S4), albeit the magnitude and duration of the effect varied by treatment, with WSPM producing the most robust effect and monolayer damage. Concurrently, TRPV1 mRNA was higher in cells treated with WSPM between 0 and 24 h (Figures 3B and S4), and similar time-dependent perturbations to mRNA expression for TRPV3 (higher expression across at all time points where TRPA1 was lower, similar to TRPV1), TRPV4 (higher expression at 0 to 8 h and lower at other times), and TRPM8 (lower at all time points up to 24 h, paralleling TRPA1) following WSPM treatment and treatment with other TRPA1 agonists. Like TRPA1 and TRPV1, variations in TRPV3 mRNA also manifest as changes in protein expression and activity.25,31
Figure 3. Temporal changes in (A) TRPA1 and (B) TRPV1 mRNA expression in HBEC3-KT cells following treatment with media containing 0.2% DMSO (open circles, solid lines) or pine WSPM in media (20 μg/cm2; closed circles, dashed lines). Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the control (n=3). *p≤0.05 and ****p<0.0001 using repeated measures two-way ANOVA comparing the control and WSPM-treated groups at each time and correction using Bonferroni’s multiple comparisons test. Summary data can be found in Excel Table S6. Note: ANOVA, analysis of variance; DMSO, dimethyl sulfoxide; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); Rq, relative quantification; SD, standard deviation; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; WSPM, wood smoke particulate matter; β2M, β2-microglobulin.
Figures 3A and 3B are line graphs, T R P A 1 messenger ribonucleic acid (relative quantity), ranging from 0 to 15 in increments of 5 and T R P V 1 messenger ribonucleic acid (relative quantity), ranging from 0 to 6 in increments of 2 (y-axis) across Time (hour), ranging from 0 to 24 in increments of 4 (x-axis) for untreated control and W S P M.
TRPA1 mRNA Expression in Cells with Suppression of TRPV1 Expression/Function Using siRNA
To further link TRPV1 expression/function with TRPA1 mRNA expression, TRPV1 siRNA was used. Cells treated with TRPV1 siRNA had 2.7- (p<0.004) and 3.4-fold (p=0.002) higher TRPA1 mRNA compared with the scramble and GAPDH siRNA controls, respectively (Figure 4, white bars). In this experiment, TRPV1 mRNA was ∼1.6-fold (p=0.0151) lower relative to both controls (p<0.0001 for both, gray hashed bars). For comparison, cells treated with GAPDH siRNA had 6.3-fold (p<0.0001) and ∼1.5-fold (p=0.0023) lower GAPDH mRNA and protein expression (Figure S5).
Figure 4. TRPA1 (white bars) and TRPV1 (gray hashed bars) mRNA expression in HBEC3-KTs transfected with scramble, GAPDH, and TRPV1 siRNA (500 pmol/mL) 24 h posttransfection. Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the average for cells transfected with scramble siRNA (n=3). Data for TRPA1 and TRPV1 were analyzed independently comparing all three siRNAs using one-way ANOVA and Tukey’s multiple comparisons test. ***p<0.001 and ****p<0.0001. Summary data can be found in Excel Table S7. Note: ANOVA, analysis of variance; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); Rq, relative quantification; SD, standard deviation; siRNA, small interfering RNA; TRP, transient receptor potential; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; β2M, β2-microglobulin.
Figure 4 is a bar graph, plotting T R P messenger ribonucleic acid (relative quantity), ranging from 0 to 4 in unit increments (y-axis) across scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, T R P V 1 small interfering ribonucleic acid, scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, and T R P V 1 small interfering ribonucleic acid (x-axis) for T R P A 1 and T R P V 1.
NF-κB and TRPA1, TRPV1, and IL8 mRNA Expression
A role for NF-κB in regulating TRPA1 mRNA expression was tested by treating HBEC3-KT cells (24 h) with the Inhibitor of Nuclear Factor Kappa-B Kinase Subunits Alpha and Beta (IKKα/IKKβ; NF-κB pathway) inhibitor BMS-345541. Treated cells had 7.7-fold (p<0.0001) lower TRPA1 mRNA expression compared with control cells (Figure 5A). Alternatively, cells treated with TNFα, an NF-κB pathway activator, had 2.6- (p=0.0353) and ∼6.9-fold (p=0.0001) higher TRPA1 mRNA expression at 4 and 24 h, respectively (Figure 5B). Consistent with the inverse relationship between TRPA1 and TRPV1 mRNA expression, cells treated with TNFα expressed lower TRPV1 mRNA with time (1.8-fold lower with 24-h treatment; p=0.1425; Figure 5C), consistent with reports that NF-κB inhibits TRPV1 expression.46 Simultaneously mRNA for another NF-κB target gene, IL8,47,48 was 24.3-fold (p<0.0001) and 140.4-fold (p=0.1241) higher at 4 and 24 h, respectively (Figure 5D). TRPA1 activity was also higher in HBEC3-KT cells treated for 24 h with TNFα, as indicated by 2.7-fold higher AITC-induced changes in intracellular calcium (p=0.0001; Figure 5E). In addition, pretreatment with BMS-345541 alone and in combination with TNFα, resulted in ∼1.6-fold (p=0.3787) and ∼1.4-fold (p=0.5550) lower AITC-induced calcium flux relative to the control, as well as 4.4- (p<0.0001) and 4-fold (p=0.0001) lower response compared with TNFα-treated cells. Figure S6 shows raw images and antagonist inhibition data confirming that the AITC-induced calcium flux was due to TRPA1.
Figure 5. (A) TRPA1 mRNA expression in HBEC3-KT cells following 24-h treatment with media containing 0.2% DMSO or the NF-κB inhibitor BMS-345541 (BMS; 10μM). ****p<0.0001 using a one-tailed unpaired Student’s t-test. (B) TRPA1, (C) TRPV1, and (D) IL8 mRNA expression in HBEC3-KT cells following 4- and 24-h treatment with TNFα (50 ng/mL) or 0.01% BSA in media (n=3). Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the control. *p≤0.05, **p<0.01, and ****p<0.0001 using two-way ANOVA and Tukey’s multiple comparisons test comparing all treatments groups. (E) Calcium flux in HBEC3-KT cells treated for 24 h with either media containing 0.01% BSA and 0.2% DMSO, TNFα (50 ng/mL), BMS-345541 (10μM), or TNFα and BMS subsequently stimulated by the addition of AITC (150μM). Data were normalized to ionomycin (n=5). Raw images are shown in Figure S6. ****p<0.0001 using one-way ANOVA and Tukey’s multiple comparison test comparing all groups. Summary data can be found in Excel Table S8. Note: AITC, allyl isothiocyanate; ANOVA, analysis of variance; BSA, bovine serum albumin; DMSO, dimethyl sulfoxide; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); IL8, interleukin-8; NF-κB, nuclear factor kappa light chain enhancer of activated B cells; Rq, relative quantification; SD, standard deviation; TNFα, tumor necrosis factor-alpha; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; β2M, β2-microglobulin.
Figures 5A to 5E are bar graphs, plotting T R P A 1 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 1.5 in increments of 0.5; T R P A 1 messenger ribonucleic acid (relative quantity), ranging from 0 to 10 in increments of 2; T R P V 1 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 2.0 in increments of 0.5; interleukin-8 messenger ribonucleic acid (relative quantity), ranging from 0 to 5 in unit increments, 30 to 70 in increments of 10; and calcium flux (percentage ionomycin), ranging from 0.0 to 2.5 in increments of 0.5 (y-axis) across treatment, including control, B M S; control, T N F lowercase alpha, control T N F lowercase alpha; control, T N F lowercase alpha, control T N F lowercase alpha; control, T N F lowercase alpha, control T N F lowercase alpha; and control, T N F lowercase alpha, B MS, T N F lowercase alpha plus B M S (x-axis) for 4 hours and 24 hours.
Nuclear Localization of NF-κB/p65 and TRPA1 mRNA Expression
Nuclear localization of NF-κB is central to transcriptional regulation by NF-κB.49 Western blot analysis of NF-κB proteins from HBEC3-KTs treated with TNFα, nonivamide, and LJO-328 showed 2-fold (p=0.0275) higher nuclear phospho-NF-κB (pNF-κB) with TNFα treatment and 1.8 (p=0.4248) lower and essentially equivalent (i.e., 1.2-fold; p=0.9333) nuclear pNF-κB with nonivamide and LJO-328 treatment, respectively (Figure 6A,B), consistent TRPV1-dependent modulation of NF-κB to alter TRPA1 expression. Unprocessed western blot images are shown in Figure S7.
Figure 6. (A) Quantification of nuclear pNF-κB/p65 in HBEC3-KTs treated for 24 h with media, TNFα (50 ng/mL), nonivamide (10μM), or LJO-328 (25μM). pNF-κB/p65 intensity was normalized to p84, and NF-κB (total) was normalized to β-actin prior to calculating the pNF-κB/NF-κB ratio. Data represents the mean±SD (n=3). **p<0.01 using one-way ANOVA and Tukey’s multiple comparisons test comparing all groups. (B) Representative western blot image where, from left to right, are the molecular weight standard (MW), control (1), TNFα (2), nonivamide (3), and LJO-328 treatments (4). Raw western blot data are shown in Figure S7. Summary data can be found in Excel Table S9. Note: ANOVA, analysis of variance; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); LJO-328, N-(4-tert-butylbenzyl)-N-(1-[3-fluoro-4-(methylsulfonylamino)phenyl]ethyl)thiourea; nonivamide, 2-mercaptoethanol, n-vanillylnonanamide; NF-κB, nuclear factor kappa light chain enhancer of activated B cells; p84, rabbit nuclear matrix protein; pNF-κB, phospho-nuclear factor kappa light chain enhancer of activated B cells; SD, standard deviation; TNFα, tumor necrosis factor-alpha.
Figure 6A is a bar graph, plotting (phospho-nuclear factor-kappa uppercase b per lowercase p 84) over (nuclear factor-kappa uppercase b per lowercase beta-actin), ranging from 0 to 5 in unit increments (y-axis) across treatment, ranging as control, T N F lowercase alpha, nonivamide, and L J O-328 (x-axis). Figure 6B is a western blot, depicting molecular weight, including 1, 2, 3, 4 (columns) and phospho-nuclear factor-kappa uppercase b, lowercase p 84, nuclear factor-kappa uppercase b, and lowercase beta-actin (rows).
Relationship between Protein Kinase C, p38 Mitogen-Activated Protein Kinase, and TRPA1 mRNA Expression
Protein kinase C (PKC)50 and p38 mitogen-activated protein kinase (MAPK)51,52 promote NF-κB pathway activity and enhance target gene (e.g., IL8) expression.47,48 PKC also acutely sensitizes TRPV1.53,54 HBEC3-KTs cells treated with the PKC activator and TRPV1 sensitizer PMA exhibited dose- and time-dependent lower TRPA1 mRNA expression at 4 and 12 h (Figure 7A) while simultaneously exhibiting higher TRPV1 and IL8 mRNA expression at 4 h, but less so at 12 h (Figure 7B,C). Consistent with a role for PKC and p38 MAPK activity in modulating NF-κB signaling, AECs treated for 12 h with the PKC inhibitor Go6983 and the p38 MAPK inhibitor PD169316 also had lower TRPA1 mRNA expression: 3.3- (p<0.0001) and 2.9-fold (p<0.0001), respectively (Figure 7D), while simultaneously having 1.4-fold and 2.2-fold higher TRPV1 (p=0.0331) and IL8 (p<0.0001) mRNA expression (Figure 7E). Finally, it is possible that the increase in TRPV1 mRNA expression associated with protracted PKC inhibition in nonstimulated cells by Go6983 (Figure 7E) was driven in part by a reduction in basal TRPV1 activity given that 12-h Go6983 treatment attenuated nonivamide-induced TRPV1 activation in TRPV1-overexpressing HEK-293 cells (Figure 7F). This effect was characterized by a shift in the 50% effective concentration (EC50) from 0.94±0.05 to 2.13±0.04μM and a decrease in the Hill slope from 2.6±0.3 to 1.1±0.1 (raw data are shown in Figure S8).
Figure 7. (A) TRPA1, (B) TRPV1, and (C) IL8 mRNA expression in HBEC3-KT cells following 4- or 12-h treatment with media containing 0.2% DMSO or the PKC activator and the TRPV1 sensitizer PMA. Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the control (n=3). ***p<0.001 and ****p<0.0001 using two-way ANOVA and Dunnett’s multiple comparison test comparing to the 4 and 12 h gene-specific control. (D) Effects of 12-h treatment with the PKC inhibitor Go6983 (10μM) and p38 MAPK inhibitor PD169316 (10μM) on TRPA1 mRNA expression in HBEC3-KT and SAECs compared with cells treated with media containing 0.2% DMSO. Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the control. ****p<0.0001 using multiple t-tests comparing treatment vs. the respective cell type–specific control. (E) TRPV1 and IL8 mRNA expression in HBEC3-KT cells following 12-h treatment with media containing 0.2% DMSO or Go6983 (10μM). *p≤0.05 and ****p<0.0001 using two-way ANOVA and Bonferroni’s multiple comparisons test to compare the gene-specific control and treatment group. (F) TRPV1-mediated calcium flux in HEK-293 cells stably overexpressing human TRPV1 with and without 12-h treatment with media containing 0.2% DMSO or the PKC inhibitor Go6983 (10μM). Data are the mean±SD for change in fluorescence relative to media-treated cells normalized to the maximum response (100%) and fit using the log[agonist] vs. normalized response-variable slope equation (n=3). **p<0.01 and ****p<0.0001 using two-way ANOVA and a Bonferroni test. Raw data are graphed in Figure S8. Summary data can be found in Excel Table S10 (A–C) and Excel Table S11 (D–F). Note: ANOVA, analysis of variance; DMSO, dimethyl sulfoxide; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); IL8, interleukin-8; p38 MAPK, p38 mitogen-activated protein kinase; PD169316, a p38 MAPK inhibitor; PKC, protein kinase C; PMA, phorbol 12-myristate 13-acetate; Rq, relative quantification; SAECs, small airway epithelial cells; SD, standard deviation; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; β2M, β2-microglobulin.
Figures 7A to 7E are clustered bar graphs, plotting T R P A 1 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 1.5 in increments of 0.5; T R P V 1 messenger ribonucleic acid (relative quantity), ranging from 0 to 5 in unit increments; interleukin-8 messenger ribonucleic acid (relative quantity), ranging from 0 to 1,500 in increments of 500; T R P A 1 messenger ribonucleic acid abundance (fold control), ranging from 0.0 to 1.5 in increments of 0.5; and target messenger ribonucleic acid (relative quantity), ranging from 0 to 3 in unit increments (y-axis) across phorbol 12-myristate 13-acetate (nanomole), including 0, 125, 250, and 500; phorbol 12-myristate 13-acetate (nanomole), including 0, 125, 250, and 500; phorbol 12-myristate 13-acetate (nanomole), including 0, 125, 250, and 500; treatment, including G o 6983 and P D 169316; treatment, including control, G o 6983; control, G o 6983 (x-axis) for 4 hours and 12 hours; control and treated; and T R P V 1 and interleukin-8. Figure 7F is a line graph, plotting calcium flux (normalized response), ranging from 0 to 150 in increments of 25 (y-axis) across log [Nonivamide] (micrometer), ranging from negative 2 to 2 in unit increments (x-axis) for vehicle and G o 6983.
Relationship between NF-κB, NLRP2, TRPA1, TRPV1, and IL8 mRNA Expression
Treatment of HBEC3-KT cells with the NF-κB pathway inhibitor BMS-345541 resulted in ∼5-fold (p<0.0001) lower NLRP2 mRNA expression (Figure 8A). The role for NLRP2 in TRPA1 mRNA expression was further evaluated. HBEC3-KT cells transfected with NLRP2 siRNA-1 and -2 exhibited 1.7 (p=0.0078) and 2.2-fold (p=0.005) lower NLRP2 protein compared with the scramble siRNA control (Figure 8B,C) and 2.7-fold (p=0.0001; Figure 8D) lower NLRP2 mRNA (NLRP2 siRNA-2 shown). Unprocessed western blot images for NLRP2 are shown in Figure S9. Finally, cells transfected with NLRP2 siRNA-2 had 7.4-fold (p=0.0077) higher TRPA1 (Figure 8E), 2-fold higher TRPV1 (p<0.001), and 4.3-fold higher (p=0.0006) IL8 mRNA (Figure 8G). Cells treated with TRPV1 siRNA also had ∼29-fold (p<0.0001) higher IL8 mRNA (Figure 8H), supporting a role for NLRP2 in regulating the NF-κB-dependent expression of TRPA1 and IL8 and the suppression of TRPV1.
Figure 8. (A) NLRP2 mRNA expression in HBEC3-KT cells following 24-h treatment with media containing 0.2% DMSO or the NF-κB inhibitor BMS-345541 (10μM). Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the control (n=3). ****p<0.0001 using a one-tailed, unpaired Student’s t-test. (B) Representative western blot image for NLRP2 in HBEC3-KT cells transfected with 100 pmol/mL NLRP2 siRNA, where, from left to right, are the molecular weight standards (MW), scramble siRNA (1), GAPDH siRNA (2), and NLRP2 siRNA 1 and 2 transfected cell lysates (3 and 4). (C) Quantification of NLRP2 protein in siRNA-transfected HBEC3-KTs. Raw data are provided in Figure S9. Data are the mean±SD of the ratio of NLRP2 to β-actin band density (n=3). *p≤0.05 and **p<0.01 using one-way ANOVA and Tukey’s multiple comparison test comparing all groups. (D–G) NLRP2, TRPA1, TRPV1, and IL8 mRNA expression in HBEC3-KTs 24 h after NLRP2 siRNA-2 (100 pmol) transfection, and (H) IL8 mRNA expression following TRPV1 siRNA (500 pmol) transfection compared with the respective control siRNA and GAPDH siRNA groups. Data (Rq) are the mean±SD for target gene mRNA expression relative to β2M mRNA and the scramble control (n=3). **p<0.01, ***p<0.001, and ****p<0.0001 using one-way ANOVA and Tukey’s multiple comparison test comparing all groups. Summary data can be found in Excel Table S12. Note: ANOVA, analysis of variance; BMS, BMS-345541; DMSO, dimethyl sulfoxide; GAPDH, glyceraldehyde 3-phosphate dehydrogenase; HBEC3-KT, telomerase reverse transcriptase and CDK4–immortalized normal human bronchial epithelial (cells); IL8, interleukin-8; NF-κB, nuclear factor kappa light chain enhancer of activated B cells; NLRP2, nucleotide-binding oligomerization domain, leucine rich repeat and pyrin domain containing 2; Rq, relative quantification; SD, standard deviation; siRNA, small interfering RNA; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; β2M, β2-microglobulin.
Figures 8A, 8C, 8D, 8E, 8F, 8G, and 8H are bar graphs, plotting N L R P 2 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 1.5 in increments of 0.5; N L R P 2 per lowercase beta-actin, ranging from 0.00 to 1.25 in increments of 0.25; N L R P 2 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 1.5 in increments of 0.5; T R P A 1 messenger ribonucleic acid (relative quantity), ranging from 0 to 12 in increments of 3; T R P V 1 messenger ribonucleic acid (relative quantity), ranging from 0.0 to 2.5 in increments of 0.5; interleukin-8 messenger ribonucleic acid (relative quantity), ranging from 0 to 8 in increments of 2; and interleukin-8 messenger ribonucleic acid (relative quantity), ranging from 0 to 40 in increments of 10 (y-axis) across treatment, including control and B M S; scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, N L R P 2 small interfering ribonucleic acid 2, and N L R P 2 small interfering ribonucleic acid 1; treatment, including scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, and N L R P 2 small interfering ribonucleic acid 2; treatment, including scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, and N L R P 2 small interfering ribonucleic acid 2; treatment, including scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, and N L R P 2 small interfering ribonucleic acid 2; treatment, including scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, and N L R P 2 small interfering ribonucleic acid 2; and treatment, including scramble small interfering ribonucleic acid, G A P D H small interfering ribonucleic acid, and T R P V 1 small interfering ribonucleic acid (x-axis). Figure 8B is a western blot, depicting molecular weight, including 1, 2, 3, 4 (columns) and N L R P 2 and lowercase beta-actin (rows).
TRPV1 I585I/V Genotype and Responses of NHBE Cells to PM in Vitro
NHBE cells from multiple donors (Figure 1B–D) were treated with DEP and CFA. DDIT3 mRNA expression was used as a biomarker for TRPA1 activation and pathological endoplasmic reticulum stress,25 and IL8 mRNA as a marker of inflammation and likely NF-κB activity.29,40 Both DEP and CFA treatment resulted in higher IL8 and DDIT3 mRNA expression relative to the respective untreated donor line control (Figure 9A,B,D,E). In addition, the average fold difference in DDIT3 and IL8 mRNA was greater for cells with the I585I/V genotype: 2.3-fold (p=0.0047) for DEP and 3.6-fold (p=0.2123) for CFA (Figure 9C). For IL8, responses were also greater in I585I/V expressing cells, but not significant (p=0.285 and 0.2029, respectively for DEP and CFA; Figure 9F). Correlation analysis confirmed an association between TRPV1 I585I/V genotype and TRPA1 mRNA expression (0.54), as well as for DEP-induced DDIT3 mRNA expression (0.93; Figure S10). A negative correlation between TRPV1 mRNA expression and the I585I/V genotype (−0.34) was also observed. Finally, stratification of the DDIT3 and IL8 mRNA expression data as a function of either the TRPV1 I315M (Figure S11) or T469I (Figure S12) genotypes did not show evidence of an effect on in vitro cellular responses to DEP or CFA.
Figure 9. (A–C) DDIT3 and (D–F) IL8 mRNA expression in NHBE cells having either the TRPV1 I585I/I or I585I/V genotype 24 h after treatment with media containing either 0.2% DMSO, DEP (CFA 10 μg/cm2), or CFA (180 μg/cm2) treatment (n≥4 donors/genotype). (A,B,D,E) Data (Rq) are the mean±SD for target gene mRNA expression normalized to β2M mRNA, analyzed using a paired one-tailed t-test. Fold change in (C) DDIT3 and (F) IL8 mRNA expression in DEP- and CFA-treated cells normalized to media-treated controls for each donor. *p≤0.05 using multiple t-tests to compare genotype effects for each particle. Summary data can be found in Excel Table S13. Note: CFA, coal fly ash; DDIT3, DNA damage-inducible transcript-3; DEP, diesel exhaust particles; DMSO, dimethyl sulfoxide; IL8, interleukin-8; NHBE, normal human bronchial epithelial (cells); Rq, relative quantification; SD, standard deviation; TRPV1, transient receptor potential cation channel subfamily V member 1; β2M, β2-microglobulin.
Figures 9A, 9B, 9D, and 9E are line graphs, plotting D D I T 3 messenger ribonucleic acid (relative quantity), ranging from 0 to 15 in increments of 5; D D I T 3 messenger ribonucleic acid (relative quantity), ranging from 0 to 10 in increments of 2; interleukin-8 messenger ribonucleic acid (relative quantity), ranging from 0 to 20 in increments of 5; and interleukin-8 messenger ribonucleic acid (relative quantity), ranging from 0 to 50 in increments of 10 (y-axis) across T R P V 1 genotype and treatment, including I 585 I or I Control, I 585 I or I D E P, I 585 I or V Control, and I 585 I or V D E P; treatment, including I 585 I or I Control, I 585 I or I C F A, I 585 I or V Control, and I 585 I or V C F A; treatment, including I 585 I or I Control, I 585 I or I D E P, I 585 I or V Control, and I 585 I or V D E P; and treatment, including I 585 I or I Control, I 585 I or I C F A, I 585 I or V Control, and I 585 I or V C F A (x-axis). Figures 9C and 9F are bar graphs, plotting fold D D I T 3 messenger ribonucleic acid difference, ranging from 0 to 20 in increments of 5 and fold interleukin-8 messenger ribonucleic acid difference, ranging from 0 to 20 in increments of 5 and 50 to 300 in increments of 50 (y-axis) across treatment, including D E P and C F A (x-axis) for I 585 I or I and I 585 I or V.
Gain and Loss-of-Function TRPA1 and TRPV1 SNPs and Asthma Symptom Control as a Function of Cigarette Smoke Exposure
Stratifying asthma control data for TRPV1 I585V allele expression and tobacco smoke exposure among children with asthma revealed worse asthma control for the I585I/I+V/V genotype combination with smoke exposure (p=0.0346; Figure 10A). In fact, an unexpected trend toward improved control for the I585I/V genotype was observed for smoke exposure. In addition, the essentially co-inherited TRPA1 R3C and R58T SNPs,26 or ≥1 allelic copy of the TRPV1 I315M or T469I SNPs were associated with poorer symptom control (p=0.0684, 0.0143, and 0.0031, respectively; Figure 10B–D) compared with individuals of the same genotype without smoke exposure. For comparison, the effects of several other SNPs previously reported to be associated with asthma risk or asthma symptoms associated with tobacco smoke including rs1871042 and rs6591255 (glutathione S-transferase pi 1),55,56 rs4073 (IL8),57 rs1800925 (IL13),58,59 rs1042713 (adrenergic receptor beta-1),60 and rs2243250 (IL4)61 were also evaluated (Figure S13). In all cases, tobacco smoke exposure was associated with worse asthma symptom control, having similar effect sizes as the TRPA1 and TRPV1 SNPs, but genotype-specific associations were not observed as they were for the TRP SNPs.
Figure 10. Average asthma control score as a function of (A) TRPV1 I585V, (B) TRPA1 R3C/R58T, (C) TRPV1 I315M, and (D) TRPV1 T469I genotype and as a function of voluntarily reported tobacco smoke exposure. Data are the means±95% confidence intervals (CIs) using one-way ANOVA and Bonferroni’s multiple comparison test. Subject numbers and p-values are shown. Summary data can be found in Excel Table S16. Note: ANOVA, analysis of variance; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1.
Figures 10A to 10D are box and whisker plots, plotting Asthma score (mean plus 95 percent confidence intervals), ranging from negative 3 to 18 in increments of 3 (y-axis) across I 585 I or I plus V or V, I 585 I or I plus V or V smoke, I 585 I or V, and I 585 I or V smoke; R 3 R or plus R 585 R or R, R 3 R or plus R 585 R or R smoke, R 3 C plus R 5 8 T, and R 3 C plus R 5 8 T smoke; I 3 15 I or I, I 3 15 I or I smoke, greater than or equal to 1 M allele, and greater than or equal to 1 M allele smoke; and T 469 T or T, T 469 T or T smoke, greater than or equal to 1 I allele, and greater than or equal to 1 I allele smoke (x-axis) for T R P V 1 I 585 V r s 8065080; T R P A 1 R 3 C plus R 58 T r s 13268757 and r s 16937976; T R P V 1 I 315 M r s 222747; and T R P V 1 469 I r s 224534.
Discussion
From 2020 data, the Centers for Disease Control and Prevention estimates that >25 million people in the United States have asthma.62 Literature shows that exposure to environmental pollution increases risks for developing asthma among children and that asthmatics are more susceptible to exacerbation by environmental pollutants.7–13 Previous in vitro studies by our group have demonstrated that both TRPA1 and TRPV1 were variably activated by captured PM, including WSPM,24,25 DEP,4,5 CFA3,15 and cigarette smoke PM (CSPM)6. Depending upon source and composition, WSPM, DEP, and CSPM can be potent TRPA1 agonists, whereas CFA was shown to be a weak TRPA1 agonist,26 as well as an agonist of both TRPV13,15 and TRPM8.3,34 Because both TRPA1 and TRPV1 are implicated in the pathogenesis of asthma, this study tested the hypothesis that differences in TRPA1 expression and activity as a function of TRPV1 genetics may contribute to variations in AEC responses to PM challenge in vitro and asthma control, particularly as a function of one’s environmental exposure profile (i.e., their exposome or envirome), via altered sensitivity of AECs to PM and TRPA1 agonists.
The first objective of this study was to determine the mechanism driving higher TRPA1 expression by AECs having the TRPV1 I585I/V genotype. In vitro results suggest an integrated network consisting of PKC, p38 MAPK, NF-κB, and NLRP2 in regulating basal and dynamic TRPA1 expression in AECs. The activity of this network was TRPV1-genotype and -activity dependent, and TRPA1 activation itself served as a catalyst for variable expression. Multiple endogenous pro-inflammatory stimuli that directly or indirectly activate NF-κB (e.g., TNFα, IL1α, IL1β) and orchestrate pulmonary inflammation, also affected TRPA1 expression. A hypothetical mechanism for how the TRPV1 genotype and associated changes in basal activity affect TRPA1 expression is presented in Figure 11. Further, the role of TRPA1 activation as a trigger for dynamic expression of TRPA1 and other TRPs (e.g., TRPV3) is shown in Figure 12.
Figure 11. Schematic summarizing the authors hypothesis for how TRPV1, NF-κB, PKC, and p38 MAPK may regulate TRPA1, TRPV1, and NLRP2 expression in AECs. The summary is based on the cumulative results of this and other referenced studies. (A) TRPV1 I585I/I and (B) TRPV1 I585I/V. Note: AECs, airway epithelial cells; Ca2+, calcium ions; Go6983, a PKC inhibitor; NF-κB, nuclear factor kappa light chain enhancer of activated B cells; NLRP2, nucleotide-binding oligomerization domain, leucine rich repeat and pyrin domain containing 2; MAPK, p38 mitogen-activated protein kinase; PD169316, a p38 MAPK inhibitor; PKC, protein kinase C; PMA, phorbol 12-myristate 13-acetate; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1.
Figure 11A is a schematic diagram depicting the hypothesis on T R P V 1 I 585 I or I. T R P V 1 agonists or activators with calcium ions are shown on the cell surface in the diagram. In the cytosol, P M A leads to P K C and T R P V 1. P K C is linked to T R P V 1 agonists or activators as well as T R P V 1. P K C and G o 6983 lead to M A P K, and M A P K and N L R P 2 lead to nuclear factor-kappa uppercase b. TRPV-1 is in the endoplasmic reticulum. T R P A 1 and N L R P 2 are formed in the nucleus by nuclear factor-kappa uppercase b and B M S-345541. Figure 11B is a schematic diagram depicting the hypothesis on T R P V 1 I 585 I or V. T R P V 1 agonists or activators with calcium ions are shown on the cell surface in the diagram. In the cytosol, P M A is connected to P K C. P K C is linked to T R P V 1 agonists or activators as well as T R P V 1 in the endoplasmic reticulum. P K C with G o 6983 and P D 169316 leads to M A P K. M A P K and N L R P 2 lead to nuclear factor-kappa uppercase b. Nuclear factor-kappa uppercase b with B M S 345541 forms T R P A 1 and N L R P 2 in the nucleus.
Figure 12. Schematic summarizing the authors hypothesis for how TRPA1 activation may affect the expression/function of TRPs involved in calcium handling (TRPV1 and TRPV3 shown) and cytotoxic ERS, thus modulating responses to TRPA1 stimuli. The summary is based on the cumulative results of this and other referenced studies. (A) No TRPA1 activation; (B) TRPA1 activation. A timeline representing the relative activities of NF-κB and NLRP2 following TRPA1 stimulation is also shown. Western blot and mRNA expression data supporting this timeline are shown in Figure S14. Note: ATF3, activating transcription factor 3; Ca2+, calcium ions; CSPM, cigarette smoke particulate matter; DDIT3, DNA damage-inducible transcript-3; DEP, diesel exhaust particles; ERS, endoplasmic reticulum stress; IL8, interleukin-8; NF-κB, nuclear factor kappa light chain enhancer of activated B cells; NLRP2, nucleotide-binding oligomerization domain, leucine rich repeat and pyrin domain containing 2; PKC, protein kinase C; PM, particulate material; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; TRP, transient receptor potential; TRPA1, transient receptor potential cation channel subfamily A member 1; TRPV1, transient receptor potential cation channel subfamily V member 1; TRPV3, transient receptor potential cation channel subfamily V member 3; WSPM, wood smoke particulate matter.
Figure 12A is a schematic diagram depicting the hypothesis on T R P A 1 activation. Agonists (D E P, W S P M, C S P M, ambient PM 2.5) with T R P A 1 and calcium ions act on T R P A 1, leading to calcium ions and P K C. T R P V 1 and T R P V 3 in the endoplasmic reticulum with calcium ions lead to the nucleus, leading to E R S (D D I T 3), nuclear factor-kappa uppercase b (interleukin-8, T R P V 1, N L R P 2), and T R P V 3. Figure 12B is a schematic diagram depicting the hypothesis for T R P A 1 activation. Agonists or P M act on T R P A 1 and increase calcium ions in the cytosol. T R P V 1 also increases calcium ions in the cytosol. Calcium ions lead to P K C. In the endoplasmic reticulum, T R P V 1 and T R P V 3 move calcium ions to the cytosol. In the nucleus, there are E R S (D D I T 3), T R P A 1, A T F 3, N L R P 2, chaperones, etc. At the bottom, there is a line graphs, plotting relative activity, ranging from 0 to 100 percent in increments of 100 (y-axis) across time (hour), ranging from 0 hour to 24 or 48 hours (x-axis) for nuclear factor-kappa uppercase b and N L R P 2.
The role of NF-κB in regulating TRPA1 expression has previously been shown in keratinocytes,63 A549 cells,64 and synoviocytes.63–66 The present study identified a comparable mechanism regulating basal and dynamic TRPA1 expression in AECs and further identified a relationship wherein loss or attenuation expression and function paradoxically promoted TRPA1 expression and function. Conversely, when TRPV1 was activated or more abundant, TRPA1 expression/function was lower. This relationship was also reciprocal in that TRPA1 activation using multiple agonists, including WSPM and AITC, led to transiently higher levels of TRPV1 mRNA and lower TRPA1 mRNA relative to control cells. Regarding this paradigm, the TRPV1 I585I/V genotype of HBEC3-KT cells was likely paramount, although dynamic expression of TRPA1 (and other TRPs) in BEAS-2B and NHBE AECs with the TRPV1 I585I/I genotype was/has also been observed.15,25,31 Regardless, this connection between TRP channels and TRP channel agonists is likely to have important consequences with respect to AEC and individual sensitivity to specific environmental stimuli and asthma triggers, as well as the ability to therapeutically manipulate TRPA1 or TRPV1 for pain, inflammation, and other purposes. Accordingly, a general recommendation is that this relationship be considered when studying TRPA1 and TRPV1 in AECs and possibly other cell types, particularly when evaluating the effects of pollutants or chemicals that may target one or both receptors.
A key aspect of TRPA1 regulation seemingly involved basal and temporal differences in NF-κB activity and associated NLRP2 expression as a function of cell status. In TRPV1 I585I/I cells, the markedly higher levels basal NLRP2 and IGFBP2 expression could suggest a higher level of basal NF-κB activity. Although not proven, it is tempting to hypothesize that higher NLRP2 and IGFBP2 expression may represent a mechanism to control basal inflammation mediated by NF-κB and TRPV1. Specifically, higher/more active TRPV1 could promote NF-κB signaling and increased IL8, NLRP2, and IGFBP2 expression, leading to the suppression of TRPA1 and cellular effects associated with changes in TRPA1 activity. Conversely, lower expression of NLRP2 in TRPV1 I585I/V cells could reflect a basally suppressed inflammatory state, due to less active TRPV1, and less need to negatively suppress NF-κB–regulated genes by NLRP2 feedback inhibition,41,42 including IL8 and TRPA1. Of significance, multiple TRP and pro-inflammatory stimuli that promote NF-κB signaling impacted this network, including TNFα, IL1α/β, and modifiers of PKC and p38 MAPK. PKC activation is regulated by intracellular calcium, which is a likely consequence of variable TRP activity. Here, activating PKC by PMA,53,54 which would also promote TRPV1 activity53,54 and NF-κB signaling led to a rapid but diminishing increase in IL8 and TRPV1 expression, as well as a delayed reduction in TRPA1 expression. This likely occurred by short term stimulation of NF-κB-driven transcription, followed by a period of attenuated activity due to NLRP2-dependent negative feedback on NF-κB.41,42 Interestingly, inhibiting both PKC and p38 MAPK, which would attenuate basal NF-κB activity, also resulted in lower TRPA1 mRNA expression and higher TRPV1 expression, collectively demonstrating that the balance of NF-κB activation and the expression of NF-κB–regulated genes, including NLRP2, determined the TRPA1/TRPV1/IL8 dynamic. More work is necessary to validate and fully unravel the scope of interactions involved in the control of TRPA1 and TRPV1 expression/function by PKC, p38 MAPK, and presumably other kinases/phosphatases, and specifically how NRLP2 and NF-κB activities are altered. Such work should also address the limitation of this work by defining the kinetics of these interactions at the protein/activity level. Regardless, the finding that these entities communicated to affect TRP activity and expression basally and during inflammation/cell damage provides key insights into the complex, but orchestrated, regulation of TRP signaling during basal and pathological states, including following exposure of AECs to potential asthma triggers. Given the vital role of NF-κB in regulating pulmonary homeostasis, it is reasonable to conclude that the balance of this mechanistic hub may also play a key role in shaping cellular responses to various pro-inflammatory agents relevant to asthma.
A second objective of this study was to determine whether the I585I/V genotype and elevated TRPA1 expression would be consequential with respect to the effects of environmental pollutants on AECs in vitro and, ultimately, on asthma control. A theme that was consistent throughout this study was the inverse relationship between TRPA1 and TRPV1 mRNA expression and activity. The discovery of this balance, and the degree to which it is influenced by multiple TRPV1 genotypes and TRPA1 and TRPV1 agonists is unique, intriguing, and seemingly indicative of a broad mechanism for regulating inflammation and general cellular homeostasis in which TRPA1 and TRPV1 may play different, but defined roles in a context-dependent manner. Specifically, cellular calcium homeostasis is critical, with acute and chronic perturbations serving as a catalyst for many effects of stimuli, including those driven by PKC, p38 MAPK, and NF-κB. Results here suggest that regulation of an intracellular calciome is in part dependent upon TRPA1, TRPV1, and other dynamically regulated TRPs such as TRPV3, which seemingly act in concert to shape how cells respond to TRPA1 agonists, environmental stimuli, and even endogenous pro-inflammatory mediators relevant to asthma and airway physiology/pathophysiology in general. This conceptual framework is supported by both our prior work demonstrating the modulation of TRPA1, endoplasmic reticulum stress, and growth/repair in HBEC3-KT and other AECs by TRPV3,25,31 as well as by findings here that TRP genetics, and changes in TRPA1 and TRPV1 expression and function affect acute cellular responses to PM in vitro and asthma control as a function of tobacco smoke exposure.
Regarding the consequences of variable TRPA1 or TRPV1 expression/function on asthma control, results from a cohort analysis did not support the hypothesized association between the TRPV1 I585I/V genotype and worse asthma control as a function of smoke exposure. Rather, an unexpected trend indicative of better symptom control for individuals with the I585I/V genotype was observed, opposite that of individuals with the I585I/I+I585V/V and other TRPA1 and TRPV1 genotypes evaluated. Despite limitations related to the use of self- and parental/guardian-reported smoke exposure (i.e., exposure is likely underreported), and the levels of expression/activity of the TRPs and other elements of the proposed regulatory network were not evaluated, several findings suggest the observed genotype–phenotype trends could be clinically relevant. First, higher TRPA1 expression and activity associated with the TRPV1 I585I/V genotype and tobacco smoke exposure would be expected to promote calcium-dependent PKC activation,67 acute sensitization of TRPV1,53,54,67 and TRPA1 suppression. With time, protracted PKC activation could also desensitize TRPV1. Second, despite the possibility that TRPV1 may become induced by TRPA1 agonists, at least with acute exposures, loss of function associated with TRPV1 I585V expression may render this effect irrelevant, whereas higher levels of expression of more active TRPV1 I315M and T469I15 may sensitize asthmatics to a broader array of stimuli, including non-TRPA1 agonists. This could explain why the TRPV1 I585V/V genotype has previously been associated with improved asthma symptom control15,27,28 and why the I315M and T469M genotypes have been associated with poorer symptom control with smoking. Interestingly, the effects of several other SNPs previously reported to be associated with asthma risk or asthma symptoms also showed associations with tobacco smoke exposure, but genotype-specific associations with asthma control were not observed, suggesting unique effects of TRP SNPs in regulating asthma control.
Finally, a person’s individual exposures must be considered with respect to the mechanisms described herein, and in understanding discrepancies between the in vitro observations related to the TRPA1/TRPV1 dynamic and asthma control. Specifically, the in vitro studies provide insight into how potential asthma-exacerbating stimuli could acutely perturb the NF-κB/TRP/IL8 balance, which may be more applicable to people who are not regularly exposed to TRPA1 agonists such as tobacco smoke but then are subsequently exposed. Alternatively, the cohort data provides insights on how chronic exposure to tobacco smoke (or perhaps other pollutants/TRPA1 agonists) may impact this dynamic; specifically, that TRPA1 signaling may become suppressed, whereas TRPV1 may become amplified.
To summarize, asthma symptom control is the product of an individual’s genetics, responses to therapeutics, medication adherence, lifestyle (diet and exercise), infection status, and exposure to indoor and outdoor triggers. Air pollutants can promote and exacerbate existing asthma by activating TRPA1 and TRPV1. A mechanism was demonstrated by which TRPV1 genotype and TRPV1 activity, as well as activation of TRPA1 itself, influenced TRPA1, TRPV1, other TRP expression by AECs through NF-κB signaling. Further, it was shown that the balance of TRPA1 and TRPV1 expression and activity was consequential in that higher TRPA1 expression was associated with higher acute cellular responses to selected model pollutants in vitro in AECs with the I585I/V genotype. However, analysis of asthma cohort data indicated a more complex relationship between TRPA1 and TRPV1, likely driven by variations in exposure to distinct types of asthma triggers. Overall, this study provides early but intriguing insight into how the TRPV1 I585I/V genotype and other SNPs, variable TRP expression, and exposure to certain types of pollutants may coordinately affect in vitro cellular responses to pollutant challenges and a person’s asthma in a variable environment.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by the National Institutes of Health/National Institute of Environmental Health Sciences [grants ES017431 and ES027015 (to C.A.R.)] and National Institute of General Medical Sciences [grant GM121648 (to C.A.R.)].
==== Refs
References
1. Basbaum AI, Bautista DM, Scherrer G, Julius D. 2009. Cellular and molecular mechanisms of pain. Cell 139 (2 ):267–284, PMID: , 10.1016/j.cell.2009.09.028.19837031
2. Julius D. 2013. TRP channels and pain. Annu Rev Cell Dev Biol 29 :355–384, PMID: , 10.1146/annurev-cellbio-101011-155833.24099085
3. Deering-Rice CE, Johansen ME, Roberts JK, Thomas KC, Romero EG, Lee J, et al. 2012. Transient receptor potential vanilloid-1 (TRPV1) is a mediator of lung toxicity for coal fly ash particulate material. Mol Pharmacol 81 (3 ):411–419, PMID: , 10.1124/mol.111.076067.22155782
4. Deering-Rice CE, Memon T, Lu Z, Romero EG, Cox J, Taylor-Clark T, et al. 2019. Differential activation of TRPA1 by diesel exhaust particles: relationships between chemical composition, potency, and lung toxicity. Chem Res Toxicol 32 (6 ):1040–1050, PMID: , 10.1021/acs.chemrestox.8b00375.30945539
5. Deering-Rice CE, Romero EG, Shapiro D, Hughen RW, Light AR, Yost GS, et al. 2011. Electrophilic components of diesel exhaust particles (DEP) activate transient receptor potential ankyrin-1 (TRPA1): a probable mechanism of acute pulmonary toxicity for DEP. Chem Res Toxicol 24 (6 ):950–959, PMID: , 10.1021/tx200123z.21591660
6. Shapiro D, Deering-Rice CE, Romero EG, Hughen RW, Light AR, Veranth JM, et al. 2013. Activation of transient receptor potential ankyrin-1 (TRPA1) in lung cells by wood smoke particulate material. Chem Res Toxicol 26 (5 ):750–758, PMID: , 10.1021/tx400024h.23541125
7. Tuazon JA, Kilburg-Basnyat B, Oldfield LM, Wiscovitch-Russo R, Dunigan-Russell K, Fedulov AV, et al. 2022. Emerging insights into the impact of air pollution on immune-mediated asthma pathogenesis. Curr Allergy Asthma Rep 22 (7 ):77–92, PMID: , 10.1007/s11882-022-01034-1.35394608
8. Grant TL, Wood RA. 2022. The influence of urban exposures and residence on childhood asthma. Pediatr Allergy Immunol 33 (5 ):e13784, PMID: , 10.1111/pai.13784.35616896
9. Chen Z, Salam MT, Eckel SP, Breton CV, Gilliland FD. 2015. Chronic effects of air pollution on respiratory health in Southern California children: findings from the Southern California Children’s Health Study. J Thorac Dis 7 (1 ):46–58, PMID: , 10.3978/j.issn.2072-1439.2014.12.20.25694817
10. Madaniyazi L, Xerxes S. 2021. Outdoor air pollution and the onset and exacerbation of asthma. Chronic Dis Transl Med 7 (2 ):100–106, PMID: , 10.1016/j.cdtm.2021.04.003.34136769
11. Orellano P, Quaranta N, Reynoso J, Balbi B, Vasquez J. 2017. Effect of outdoor air pollution on asthma exacerbations in children and adults: systematic review and multilevel meta-analysis. PLoS One 12 (3 ):e0174050, PMID: , 10.1371/journal.pone.0174050.28319180
12. Orellano P, Quaranta N, Reynoso J, Balbi B, Vasquez J. 2018. Association of outdoor air pollution with the prevalence of asthma in children of Latin America and the Caribbean: a systematic review and meta-analysis. J Asthma 55 (11 ):1174–1186, PMID: , 10.1080/02770903.2017.1402342.29211546
13. Thurston GD, Balmes JR, Garcia E, Gilliland FD, Rice MB, Schikowski T, et al. 2020. Outdoor air pollution and new-onset airway disease. An official American Thoracic Society workshop report. Ann Am Thorac Soc 17 (4 ):387–398, PMID: , 10.1513/AnnalsATS.202001-046ST.32233861
14. Andrade EL, Meotti FC, Calixto JB. 2012. TRPA1 antagonists as potential analgesic drugs. Pharmacol Ther 133 (2 ):189–204, PMID: , 10.1016/j.pharmthera.2011.10.008.22119554
15. Deering-Rice CE, Stockmann C, Romero EG, Lu Z, Shapiro D, Stone BL, et al. 2016. Characterization of transient receptor potential vanilloid-1 (TRPV1) variant activation by coal fly ash particles and associations with altered transient receptor potential ankyrin-1 (TRPA1) expression and asthma. J Biol Chem 291 (48 ):24866–24879, PMID: , 10.1074/jbc.M116.746156.27758864
16. Jordt SE. 2021. TRPA1: an asthma target with a zing. J Exp Med 218 (4 ):e20202507, PMID: , 10.1084/jem.20202507.33625497
17. Bessac BF, Jordt SE. 2008. Breathtaking TRP channels: TRPA1 and TRPV1 in airway chemosensation and reflex control. Physiology (Bethesda) 23 :360–370, PMID: , 10.1152/physiol.00026.2008.19074743
18. Long L, Yao H, Tian J, Luo W, Yu X, Yi F, et al. 2019. Heterogeneity of cough hypersensitivity mediated by TRPV1 and TRPA1 in patients with chronic refractory cough. Respir Res 20 (1 ):112, PMID: , 10.1186/s12931-019-1077-z.31170994
19. Groneberg DA, Niimi A, Dinh QT, Cosio B, Hew M, Fischer A, et al. 2004. Increased expression of transient receptor potential vanilloid-1 in airway nerves of chronic cough. Am J Respir Crit Care Med 170 (12 ):1276–1280, PMID: , 10.1164/rccm.200402-174OC.15447941
20. McGarvey LP, Butler CA, Stokesberry S, Polley L, McQuaid S, Abdullah H, et al. 2014. Increased expression of bronchial epithelial transient receptor potential vanilloid 1 channels in patients with severe asthma. J Allergy Clin Immunol 133 (3 ):704–712.e4, PMID: , 10.1016/j.jaci.2013.09.016.24210884
21. Caceres AI, Brackmann M, Elia MD, Bessac BF, del Camino D, D’Amours M, et al. 2009. A sensory neuronal ion channel essential for airway inflammation and hyperreactivity in asthma. Proc Natl Acad Sci USA 106 (22 ):9099–9104, PMID: , 10.1073/pnas.0900591106.19458046
22. Baker K, Raemdonck K, Dekkak B, Snelgrove RJ, Ford J, Shala F, et al. 2016. Role of the ion channel, transient receptor potential cation channel subfamily V member 1 (TRPV1), in allergic asthma. Respir Res 17 (1 ):67, PMID: , 10.1186/s12931-016-0384-x.27255083
23. Song J, Kang J, Lin B, Li J, Zhu Y, Du J, et al. 2017. Mediating role of TRPV1 ion channels in the co-exposure to PM2.5 and formaldehyde of Balb/c mice asthma model. Sci Rep 7 (1 ):11926, PMID: , 10.1038/s41598-017-11833-6.28931832
24. Memon TA, Nguyen ND, Burrell KL, Scott AF, Almestica-Roberts M, Rapp E, et al. 2020. Wood smoke particles stimulate MUC5AC overproduction by human bronchial epithelial cells through TRPA1 and EGFR signaling. Toxicol Sci 174 (2 ):278–290, PMID: , 10.1093/toxsci/kfaa006.31944254
25. Nguyen ND, Memon TA, Burrell KL, Almestica-Roberts M, Rapp E, Sun L, et al. 2020. Transient receptor potential ankyrin-1 and vanilloid-3 differentially regulate endoplasmic reticulum stress and cytotoxicity in human lung epithelial cells after pneumotoxic wood smoke particle exposure. Mol Pharmacol 98 (5 ):586–597, PMID: , 10.1124/molpharm.120.000047.32938721
26. Deering-Rice CE, Shapiro D, Romero EG, Stockmann C, Bevans TS, Phan QM, et al. 2015. Activation of transient receptor potential ankyrin-1 by insoluble particulate material and association with asthma. Am J Respir Cell Mol Biol 53 (6 ):893–901, PMID: , 10.1165/rcmb.2015-0086OC.26039217
27. Cantero-Recasens G, Gonzalez JR, Fandos C, Duran-Tauleria E, Smit LAM, Kauffmann F, et al. 2010. Loss of function of transient receptor potential vanilloid 1 (TRPV1) genetic variant is associated with lower risk of active childhood asthma. J Biol Chem 285 (36 ):27532–27535, PMID: , 10.1074/jbc.C110.159491.20639579
28. Sadofsky LR, Cantero-Recasens G, Wright C, Valverde MA, Morice AH. 2017. TRPV1 polymorphisms influence capsaicin cough sensitivity in men. J Thorac Dis 9 (3 ):839–840, PMID: , 10.21037/jtd.2017.03.50.28449493
29. Reilly CA, Johansen ME, Lanza DL, Lee J, Lim JO, Yost GS. 2005. Calcium-dependent and independent mechanisms of capsaicin receptor (TRPV1)-mediated cytokine production and cell death in human bronchial epithelial cells. J Biochem Mol Toxicol 19 (4 ):266–275, PMID: , 10.1002/jbt.20084.16173059
30. Deering-Rice CE, Nguyen N, Lu Z, Cox JE, Shapiro D, Romero EG, et al. 2018. Activation of TRPV3 by wood smoke particles and roles in pneumotoxicity. Chem Res Toxicol 31 (5 ):291–301, PMID: , 10.1021/acs.chemrestox.7b00336.29658714
31. Burrell KL, Nguyen ND, Deering-Rice CE, Memon TA, Almestica-Roberts M, Rapp E, et al. 2021. Dynamic expression of transient receptor potential vanilloid-3 and integrated signaling with growth factor pathways during lung epithelial wound repair following wood smoke particle and other forms of lung cell injury. Mol Pharmacol 100 (3 ):295–307, PMID: , 10.1124/molpharm.121.000280.34290137
32. Smith KR, Veranth JM, Kodavanti UP, Aust AE, Pinkerton KE. 2006. Acute pulmonary and systemic effects of inhaled coal fly ash in rats: comparison to ambient environmental particles. Toxicol Sci 93 (2 ):390–399, PMID: , 10.1093/toxsci/kfl062.16840564
33. Cao X, Xu X, Cui W, Xi Z. 2001. Development and certification of a coal fly ash certified reference material for selected polycyclic aromatic hydrocarbons. Fresenius J Anal Chem 370 (8 ):1035–1040, PMID: , 10.1007/s002160100939.11583083
34. Lamb JG, Romero EG, Lu Z, Marcus SK, Peterson HC, Veranth JM, et al. 2017. Activation of human transient receptor potential melastatin-8 (TRPM8) by calcium-rich particulate materials and effects on human lung cells. Mol Pharmacol 92 (6 ):653–664, PMID: , 10.1124/mol.117.109959.29038158
35. Reilly CA, Taylor JL, Lanza DL, Carr BA, Crouch DJ, Yost GS. 2003. Capsaicinoids cause inflammation and epithelial cell death through activation of vanilloid receptors. Toxicol Sci 73 (1 ):170–181, PMID: , 10.1093/toxsci/kfg044.12721390
36. Livak KJ, Schmittgen TD. 2001. Analysis of relative gene expression data using real-time quantitative PCR and the 2–ΔΔC(T) method. Methods 25 (4 ):402–408, PMID: , 10.1006/meth.2001.1262.11846609
37. Schneider CA, Rasband WS, Eliceiri KW. 2012. NIH image to ImageJ: 25 years of image analysis. Nat Methods 9 (7 ):671–675, PMID: , 10.1038/nmeth.2089.22930834
38. Stockmann C, Fassl B, Gaedigk R, Nkoy F, Uchida DA, Monson S, et al. 2013. Fluticasone propionate pharmacogenetics: CYP3A4*22 polymorphism and pediatric asthma control. J Pediatr 162 (6 ):1222–1227.e2, PMID: , 10.1016/j.jpeds.2012.11.031.23290512
39. National Asthma Education and Prevention Program. 2007. Expert Panel Report 3 (EPR-3): Guidelines for the Diagnosis and Management of Asthma—Summary Report 2007. J Allergy Clin Immunol 120 (suppl 5 ):S94–S138, PMID: , 10.1016/j.jaci.2007.09.043.17983880
40. Johansen ME, Reilly CA, Yost GS. 2006. TRPV1 antagonists elevate cell surface populations of receptor protein and exacerbate TRPV1-mediated toxicities in human lung epithelial cells. Toxicol Sci 89 (1 ):278–286, PMID: , 10.1093/toxsci/kfi292.16120755
41. Fontalba A, Gutierrez O, Fernandez-Luna JL. 2007. NLRP2, an inhibitor of the NF-κB pathway, is transcriptionally activated by NF-κB and exhibits a nonfunctional allelic variant. J Immunol 179 (12 ):8519–8524, PMID: , 10.4049/jimmunol.179.12.8519.18056399
42. Tilburgs T, Meissner TB, Ferreira LMR, Mulder A, Musunuru K, Ye J, et al. 2017. NLRP2 is a suppressor of NF-κB signaling and HLA-C expression in human trophoblasts. Biol Reprod 96 (4 ):831–842, PMID: , 10.1093/biolre/iox009.28340094
43. Gao S, Sun Y, Zhang X, Hu L, Liu Y, Chua CY, et al. 2016. IGFBP2 activates the NF-κB pathway to drive epithelial–mesenchymal transition and invasive character in pancreatic ductal adenocarcinoma. Cancer Res 76 (22 ):6543–6554, PMID: , 10.1158/0008-5472.CAN-16-0438.27659045
44. Ben-Shmuel A, Shvab A, Gavert N, Brabletz T, Ben-Ze’ev A. 2013. Global analysis of L1-transcriptomes identified IGFBP-2 as a target of ezrin and NF-κB signaling that promotes colon cancer progression. Oncogene 32 (27 ):3220–3230, PMID: , 10.1038/onc.2012.340.22869145
45. Guo Q, Yu DY, Yang ZF, Liu DY, Cao HQ, Liao XW. 2020. IGFBP2 upregulates ZEB1 expression and promotes hepatocellular carcinoma progression through NF-κB signaling pathway. Dig Liver Dis 52 (5 ):573–581, PMID: , 10.1016/j.dld.2019.10.008.31818638
46. Zhang BY, Zhang YL, Sun Q, Zhang PA, Wang XX, Xu GY, et al. 2020. Alpha-lipoic acid downregulates TRPV1 receptor via NF-κB and attenuates neuropathic pain in rats with diabetes. CNS Neurosci Ther 26 (7 ):762–772, PMID: , 10.1111/cns.13303.32175676
47. Schins RP, Donaldson K. 2000. Nuclear factor kappa-B activation by particles and fibers. Inhal Toxicol 12 (suppl 3 ):317–326, PMID: , 10.1080/08958378.2000.11463241.26368631
48. Simeonova PP, Luster MI. 1996. Asbestos induction of nuclear transcription factors and interleukin 8 gene regulation. Am J Respir Cell Mol Biol 15 (6 ):787–795, PMID: , 10.1165/ajrcmb.15.6.8969274.8969274
49. Giridharan S, Srinivasan M. 2018. Mechanisms of NF-κB p65 and strategies for therapeutic manipulation. J Inflamm Res 11 :407–419, PMID: , 10.2147/JIR.S140188.30464573
50. Kim SW, Schifano M, Oleksyn D, Jordan CT, Ryan D, Insel R, et al. 2014. Protein kinase C-associated kinase regulates NF-κB activation through inducing IKK activation. Int J Oncol 45 (4 ):1707–1714, PMID: , 10.3892/ijo.2014.2578.25096806
51. Saha RN, Jana M, Pahan K. 2007. MAPK p38 regulates transcriptional activity of NF-κB in primary human astrocytes via acetylation of p65. J Immunol 179 (10 ):7101–7109, PMID: , 10.4049/jimmunol.179.10.7101.17982102
52. Refsnes M, Skuland T, Låg M, Schwarze PE, Øvrevik J. 2014. Differential NF-κB and MAPK activation underlies fluoride- and TPA-mediated CXCL8 (IL-8) induction in lung epithelial cells. J Inflamm Res 7 :169–185, PMID: , 10.2147/JIR.S69646.25540590
53. Bhave G, Hu HJ, Glauner KS, Zhu W, Wang H, Brasier DJ, et al. 2003. Protein kinase C phosphorylation sensitizes but does not activate the capsaicin receptor transient receptor potential vanilloid 1 (TRPV1). Proc Natl Acad Sci USA 100 (21 ):12480–12485, PMID: , 10.1073/pnas.2032100100.14523239
54. Mandadi S, Numazaki M, Tominaga M, Bhat MB, Armati PJ, Roufogalis BD. 2004. Activation of protein kinase C reverses capsaicin-induced calcium-dependent desensitization of TRPV1 ion channels. Cell Calcium 35 (5 ):471–478, PMID: , 10.1016/j.ceca.2003.11.003.15003856
55. Islam T, Berhane K, McConnell R, Gauderman WJ, Avol E, Peters JM, et al. 2009. Glutathione-S-transferase (GST) P1, GSTM1, exercise, ozone and asthma incidence in school children. Thorax 64 (3 ):197–202, PMID: , 10.1136/thx.2008.099366.18988661
56. Joubert BR, Reif DM, Edwards SW, Leiner KA, Hudgens EE, Egeghy P, et al. 2011. Evaluation of genetic susceptibility to childhood allergy and asthma in an African American urban population. BMC Med Genet 12 :25, PMID: , 10.1186/1471-2350-12-25.21320344
57. Charrad R, Kaabachi W, Rafrafi A, Berraies A, Hamzaoui K, Hamzaoui A. 2017. IL-8 gene variants and expression in childhood asthma. Lung 195 (6 ):749–757, PMID: , 10.1007/s00408-017-0058-6.28993876
58. Sadeghnejad A, Karmaus W, Arshad SH, Kurukulaaratchy R, Huebner M, Ewart S. 2008. IL13 gene polymorphisms modify the effect of exposure to tobacco smoke on persistent wheeze and asthma in childhood, a longitudinal study. Respir Res 9 (1 ):2, PMID: , 10.1186/1465-9921-9-2.18186920
59. Sadeghnejad A, Meyers DA, Bottai M, Sterling DA, Bleecker ER, Ohar JA. 2007. IL13 promoter polymorphism –1112C/T modulates the adverse effect of tobacco smoking on lung function. Am J Respir Crit Care Med 176 (8 ):748–752, PMID: , 10.1164/rccm.200704-543OC.17615386
60. Wang Z, Chen C, Niu T, Wu D, Yang J, Wang B, et al. 2001. Association of asthma with β2-adrenergic receptor gene polymorphism and cigarette smoking. Am J Respir Crit Care Med 163 (6 ):1404–1409, PMID: , 10.1164/ajrccm.163.6.2001101.11371409
61. Kousha A, Mahdavi Gorabi A, Forouzesh M, Hosseini M, Alexander M, Imani D, et al. 2020. Interleukin 4 gene polymorphism (–589C/T) and the risk of asthma: a meta-analysis and met-regression based on 55 studies. BMC Immunol 21 (1 ):55, PMID: , 10.1186/s12865-020-00384-7.33087044
62. CDC (Centers for Disease Control and Prevention). 2022. Most recent national asthma data. Last reviewed 13 December 2022. https://www.cdc.gov/asthma/most_recent_national_asthma_data.htm [accessed 1 July 2022].
63. Luostarinen S, Hämäläinen M, Moilanen E. 2021. Transient receptor potential ankyrin 1 (TRPA1)-an inflammation-induced factor in human HaCaT keratinocytes. Int J Mol Sci 22 (7 ):3322, PMID: , 10.3390/ijms22073322.33805042
64. Luostarinen S, Hämäläinen M, Hatano N, Muraki K, Moilanen E. 2021. The inflammatory regulation of TRPA1 expression in human A549 lung epithelial cells. Pulm Pharmacol Ther 70 :102059, PMID: , 10.1016/j.pupt.2021.102059.34302984
65. Hatano N, Itoh Y, Suzuki H, Muraki Y, Hayashi H, Onozaki K, et al. 2012. Hypoxia-inducible factor-1α (HIF1α) switches on transient receptor potential ankyrin repeat 1 (TRPA1) gene expression via a hypoxia response element-like motif to modulate cytokine release. J Biol Chem 287 (38 ):31962–31972, PMID: , 10.1074/jbc.M112.361139.22843691
66. Hatano N, Matsubara M, Suzuki H, Muraki Y, Muraki K. 2021. HIF-1α dependent upregulation of ZIP8, ZIP14, and TRPA1 modify intracellular Zn2+ accumulation in inflammatory synoviocytes. Int J Mol Sci 22 (12 ):6349, PMID: , 10.3390/ijms22126349.34198528
67. Huang KP. 1989. The mechanism of protein kinase C activation. Trends Neurosci 12 (11 ):425–432, PMID: , 10.1016/0166-2236(89)90091-x.2479143
|
PMC009xxxxxx/PMC9970290.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12836
10.1289/EHP12836
Erratum
Erratum: Exposure to Air Pollution during Pre-Hypertension and Subsequent Hypertension, Cardiovascular Disease, and Death: A Trajectory Analysis of the UK Biobank Cohort
Zhang Shiyu
Qian Zhengmin Min
Chen Lan
Zhao Xing
Cai Miao
Wang Chongjian
Zou Hongtao
Wu Yinglin
Zhang Zilong
Li Haitao
https://orcid.org/0000-0002-3643-9408
Lin Hualiang
27 2 2023
2 2023
27 2 2023
131 2 02900102 2 2023
03 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Environ Health Perspect 131(1):017008 (2023), https://doi.org/10.1289/EHP10967
==== Body
pmcThe email of the corresponding author, Hualiang Lin, was incorrectly stated and should be [email protected]. In addition, the funding recipient was incorrectly listed as H.T.L. Funding support was awarded to H.L.L. Finally, the hazard ratio (HR) listed for the association of PM2.5 with transitioning from pre-hypertensive status (HTN) to HTN in the fifth sentence of the Results, “Associations between Air Pollution and Dynamic CVD Progression Using Cox Regressions” was incorrectly reported as 1.00. This sentence should read “PM2.5 exposure was positively associated with transitioning from pre-HTN to HTN (HR=1.100; 95% CI: 1.079, 1.120)…”.
|
PMC009xxxxxx/PMC9973346.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36853096
EHP12232
10.1289/EHP12232
Focus
Breathing Room: Cleaner Fuels for Home Cooking in LMICs
Seltenrich Nate
28 2 2023
2 2023
131 2 02200104 10 2022
30 11 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
A woman stokes fire in small wood stove with cooking pans on top.
==== Body
pmcCooking with solid fuels is risky, not just for the estimated 2.4–2.8 billion people1–3 who rely on these fuels at home, but also for the future of the planet. Household air pollution—which includes the noxious fumes from incomplete combustion of wood, charcoal, coal, crop residues, dung, and others—is considered one of the leading environmental causes of disease.4 The World Health Organization (WHO) estimates household air pollution is responsible for around 3.2 million premature deaths each year.1 In addition, collecting solid fuels contributes to land degradation and deforestation5 and imposes an inordinate burden on women and children.4,6,7 Finally, burning these fuels produces at least 2% of global carbon-equivalent emissions.8
Traditional solid fuels such as wood, charcoal, crop residues, and dried dung are used for home cooking by billions of people, contributing to adverse health outcomes, climate change, and environmental degradation. On Russia’s Yamal Peninsula (shown here), as in many places around the world, women maintain cooking fires and prepare meals, exposing themselves and others to indoor air pollutants. Image: © Elena Chernyshova/Panos Pictures.
A woman stokes fire in small wood stove with cooking pans on top.
In many low- and middle-income countries (LMICs), cleaner-burning natural gas—often in the form of liquid petroleum gas (LPG)—has long been pursued as a feasible solution to reducing household air pollution and other impacts from solid cooking fuels. The recently completed Household Air Pollution Intervention Network (HAPIN) trial9 offers robust evidence that in a best-case scenario, with LPG stoves and cylinders provided free of charge, residents’ exposures to fine particulate matter (PM2.5) can be brought below interim guidelines established by the WHO.10
LPG cookstoves are far from a perfect solution, however. Fossil fuels—including natural gas—must be rapidly phased out to avoid the worst impacts of climate change, according to the Intergovernmental Panel on Climate Change.11 Gas stoves emit pollutants that are harmful to human health.12 Finally, provision of gas relies on distribution networks that are subject to interruption13 and a global market with fluctuating prices,14 so low-income, rural communities cannot rely on a steady and affordable supply.
As a result, some in the clean-cooking field have begun to advocate leapfrogging past LPG and prioritizing a shift from the dirtiest fuels to electric cooking devices. Access to electricity is expanding in LMICs,15 and such a switch would not only drastically reduce air pollution inside homes, they argue, but also help pave the way for a decarbonized global energy sector.
Could it be that easy?
Electrification May Not Mean Clean Cooking
Perhaps the biggest bet yet on electric cooking has come from the British government’s UK Aid Direct program, which in 2019 committed approximately $50 million toward an initiative called Modern Energy Cooking Services, or MECS. Led by England’s Loughborough University and the World Bank’s Energy Sector Management Assistance Program (ESMAP), the 7-year program aims to rapidly accelerate the transition from biomass-based cooking to, as its name states, modern energy cooking services—a term that encompasses not just clean energy but also efficiency, convenience, safety, reliability, and affordability.16
Three years in, MECS continues to advocate for electric cooking as a multi-benefit solution to the climate, health, and social problems caused by widespread use of solid fuels. In August 2022, at a webinar hosted by the Colorado State University–based consortium Advancing Sustainable Household Energy Solutions (ASHES), MECS research director Ed Brown stated that after two to three decades of effort, the provision of cleaner cooking fuels is barely keeping up with population growth. “Over these years,” he said, “we’ve come to the conclusion that the really big potential game changer—indeed, we think it is far more than ‘potential’—is electricity.”17
A key element undergirding this logic is that global access to electricity has been expanding dramatically. Over the last decade, tens of billions of dollars have been invested annually in building and maintaining new connections, and an estimated 1.3 billion people globally gained electricity access between 2010 and 2020.15 In Asia overall, connectivity approaches 90%, up from 66% in 2000;15,18 in sub-Saharan Africa, the regional average is now above 50%, although large populations lack access, especially in Nigeria, the Democratic Republic of Congo, and Ethiopia.15,18
However, progress in electricity access has not always been accompanied by adoption of clean electric cooking, Brown says. For example, in Kampala, Uganda, a city of 1.7 million people, nearly all households are connected to a reliable electrical grid,19 yet only 8% used electric stoves as of 2014—over three-fourths still cooked with charcoal, and most of the rest used firewood, LPG, or kerosene. Similar situations exist in Southeast Asia. Globally, at least 1.8 billion people who have access to electricity continue to cook with highly polluting traditional fuels.20
Households with access to electricity may continue to cook with charcoal (shown here) and other solid fuels for many reasons, including inadequate or inconsistent power supply. Charcoal stoves, for example, are cheap and widely available, and charcoal is less bulky to transport and cook with than wood.21 Image: MECS, The Kenya eCookbook (CC BY 4.0).
Two large pots on charcoal cookstoves outdoors.
One reason for this, Brown suggests, is that energy access and clean cooking have been siloed into separate campaigns, with electricity often being advanced primarily as a means to achieve better lighting and “productive use” (including powering and charging electronic devices), to the exclusion of cooking.
For example, Colorado State University professor of mechanical engineering John Volckens, who co-leads the ASHES consortium, is running the Sustainable Household Energy Adoption in Rwanda (SHEAR) study alongside Maggie Clark of the University of Colorado. This study will bring solar-powered electricity into remote rural households for the first time—but not for cooking. The electricity is mainly for lighting, because the trial’s low-voltage power supply is not sufficient to support cooking appliances.
“We could have created a more powerful grid if we wanted to, but that still wouldn’t have worked for cooking,” Volckens says. He explains that batteries to store electrical energy from solar power are still prohibitively expensive in LMICs. “Solar power is only cost effective when the sun is shining, so participants cannot cook at night or early in the morning,” he says, pointing out that instead, households in the SHEAR study will be provided with LPG stoves. Researchers will investigate the combined health benefits of simultaneously switching both lighting and cooking to cleaner fuels.
Still, Volckens agrees that electricity is the future of clean cooking. “It’s absolutely critical, both for health and for climate, to move away from fossil fuels,” he says. “We know that although burning natural gas is cleaner than wood, it’s not clean. Natural gas creates nitrogen oxides, it creates ultrafine particles, and these are things that we know can do harm. Moving to electric cooking is the direction we want to go.”
A related reason for the slower adoption of electric cooking is infrastructure, says Jennifer Peel, a professor of epidemiology at Colorado State University and one of three principal investigators for HAPIN (along with William Checkley of The Johns Hopkins University and Thomas Clasen of Emory University). Although great progress is being made in electrification, grids in some rural or poorer communities are not up to the task of handling an electric stove or pressure cooker in every home, she explains. Such grids may not be stable or reliable enough to support everyone cooking at the same time, or, as with SHEAR’s grid, they may not run on the necessary current. In such cases, LPG likely remains the best available clean-cooking option—for now.
“It’s not as easy a choice as it may seem when you have to deal with the realities in low-resource settings,” she says. “I think we have to continue to be both dynamic and aspirational. Everybody has the right to clean cooking and clean air.”
Natural Gas—Cleaner Indoor Air in the Interim
HAPIN investigator Kalpana Balakrishnan, of India’s Sri Ramachandra Institute of Higher Education and Research, acknowledges the proven benefits of electric cooking for both health and climate. But, like Peel, she is wary of singling out electric cooking as the solution of choice when so many people are still using solid fuels, and electricity is either out of reach, unreliable, or too expensive for many of them.
“Nobody is arguing the benefits of [renewables], but the perfect can’t be the enemy of the good,” Balakrishnan says. “We shouldn’t throw out everything we know about LPG stoves just because we’re looking forward to electricity.”
Balakrishnan and others see LPG stoves, such as this one in Uttar Pradesh, India, as a bridge toward cleaner cooking, helping reduce use of solid fuels as electrification efforts are under way. Image: © Bloomberg/Getty Images.
Woman with bangles on wrist lighting a clean gas cooking burner.
Rather, she says, LPG remains positioned to serve as a valuable bridge fuel in India and elsewhere. “LPG is the most scalable near-term proposition to achieve health and climate benefits,” she emphasizes. “If you don’t enable the biomass-using households to scale the energy ladder and transition to LPG and [instead,] keep waiting for the perfect solution to emerge, then the energy inequities are just going to widen beyond repair.” Indeed, a national campaign in India to promote LPG for home cooking has already demonstrated that the fuel can be brought to scale rapidly given the right setting and investment, Balakrishnan states. As recently as 10 years ago, roughly 75% of the population cooked primarily with solid fuel, she says; today, that figure is down to 45%–50%. Much of this progress is attributable to general economic growth in combination with the government’s campaign, which has included infrastructure development, public messaging and campaigning, cost subsidies, and free connections to households below the poverty line.
Meanwhile, the HAPIN study has demonstrated more conclusively than ever that meaningful reductions in household air pollution can be achieved through consistent use of LPG stoves.22 “We were really pleased to see the proportion of the intervention measurements, about 70%, that were below the WHO Interim 1 standard for fine particulate matter of 35μg/m3,” says Peel. “That’s pretty remarkable, especially compared to previous similar trials that were trying to reduce exposure.”
In homes that switched from solid fuels to LPG, HAPIN researchers found associations between lower exposures to both PM2.5 and black carbon and higher birthweight and weight-for-gestational-age among babies born to mothers living in those homes, suggesting that using the new stoves may have also meaningfully improved health outcomes.22 As of February 2023, these results had not yet been peer-reviewed but were available as a preprint.
Exposure to PM2.5 is associated with birth outcomes that can predispose a child to future health problems. Reducing household air pollution through effective cooking interventions promises some relief from these challenges to children’s growth and healthy development. Image: © Aubrey Wade/Panos Pictures.
African nurse weighing baby on a scale in a clinic.
Still, PM2.5 concentrations inside homes that received the intervention were not as low as the WHO’s ambitious new guideline of 5μg/m3.23 “If a community can bypass LPG and get to electric cooking—induction stoves—that’s probably the way to go,” Peel suggests. “The question is, how many communities can get there, and when?”
Scaling Up Electricity for the Long Term
MECS sees a solution in grouping cooking together with other household needs supported by electrification. To achieve its goals, the program publishes original research and other resources, such as case studies, country-specific market assessments, factsheets on the viability of electric cooking, and cuisine-specific cookbooks tailored for the use of electric appliances (https://mecs.org.uk/publications/). In one oft-cited albeit non–peer-reviewed finding from MECS research, published in The Kenya eCookbook in 2018,21 cooking beans with an efficient electric pressure cooker cost seven times less than cooking with charcoal and took half as long. It was also far cheaper than three alternatives: kerosene, LPG, and a traditional electric-coil hotplate.21
Another tool employed by the program is a challenge fund to support research projects geared toward safe and efficient modern energy cooking systems. Competitions, each with their own focus and objectives,24 are open to companies and organizations of all sizes, as well as participants from academia.
MECS has launched a global call to action dubbed “40,60 by 2030.”25 It proposes that by 2030, 40% of all households connected to grid or off-grid electricity should use it for cooking, and 60% of households cooking with electricity should use power generated from low-carbon sources, which include wind, solar, nuclear, or hydropower.
Sheila Oparaocha, director of Energia, a Netherlands-based nonprofit that supports gender equality in energy access throughout Africa and Asia, has partnered with MECS to promote 40,60 by 2030. In announcing the program during a 2022 webinar, she said electricity will provide the long-term solution to decarbonizing cooking as the international community embraces the greening of power generation.17
“Electric cooking offers the opportunity for rapid scaling through its incorporation into existing electrification programs [that have] the kind of large budgets that dwarf the amounts available for separate clean cooking programs,” Oparaocha said. Another of MECS’s committed partners, ESMAP, collaborated on a key 2020 report26 that found the rate of access to modern energy cooking services for cooking (including electricity, LPG, and other cleaner-burning fuels) stands at only 10% in sub-Saharan Africa, 36% in East Asia, and 56% in Latin America and the Caribbean. The report also estimated that $150 billion will be needed annually to reach universal access to modern energy cooking services by 2030.
To encourage adoption of electric cooking, MECS has developed a series of eCookbooks with recipes and advice tailored to specific countries. For example, the Myanmar eCookbook27 suggests that in off-grid areas or where the electrical grid is weak, lower power devices be used, including electric fryers, rice cookers, and pressure cookers. Images, clockwise from top left: © iStock.com/undefined, © Muhammad Wafa/Shutterstock.com, © Arina P Habich/Shutterstock.com.
Three-photo montage showing fried egg in electric fryer, electric rice cooker, and electric pressure cooker holding soup.
ESMAP is technology-neutral; it evaluates household stoves and cooking appliances for pollutant exposures, energy efficiency, convenience or ease of use, fuel availability, safety, and affordability following guidelines known as the Multi-Tier Framework28 to measure energy access. Both LPG and electric stoves are clean, efficient, convenient, and safe solutions inside the home, says ESMAP senior energy specialist and MECS coordinator Yabei Zhang. But the metrics do not account for carbon emissions; in that respect, electricity from renewable sources is preferable to LPG. Still, Zhang believes that even setting climate impacts aside and focusing on household-level concerns, electric cooking will play a key role in building progress toward the United Nations Sustainable Development Goal for affordable and clean energy.29 “A lot of countries already have very high access to electricity, but their clean cooking access is low,” says Zhang. “So, there is a lot of opportunity to build on the existing infrastructure to switch to electric cooking.”
Electrified households still using solid fuels, particularly in Southeast Asia, are precisely the market targeted by the Australian induction-stove manufacturer ATEC. Of the 60%–70% of households in Cambodia and Bangladesh that cook with wood, nearly all have access to electricity, says ATEC chief operating officer Ben Jeffreys. “[Electricity] solves one of the biggest problems around cooking, which is how to get fuel into the household.”
The company sells directly to consumers and uses targeted advertising on social media to communicate the benefits of replacing solid fuels with electricity for cooking. To help low-income households defray the upfront cost of purchasing a new stove, ATEC offers interest-free payment plans.
As electrification expands across the African continent, ATEC intends to apply its marketing and sales model there. “Our underlying assumption is that over the next 10 years, Africa will go from relatively few households having access to grid electrification to a relatively high number, much like Asia has done,” Jeffreys says. “We will then work with partners there to follow that electrification with electric cooking.”
Speedier Gains through Tailored Solutions
LPG is likely to have a place in global efforts to address the clean cooking crisis for years to come. In some locations, biofuels made from agricultural residues and waste products will probably play a small but important role. Even hydrogen fuel has been proposed as a clean, carbon-free approach to household cooking.
The consensus in the field, it seems, is that the problem of household air pollution is still so huge, so widespread, and so complex, and the health impacts so severe and immediate, that no single approach will suffice. Instead, solutions should be tailored to specific locales to achieve as much improvement in as little time as possible.
Renewable energy sources such as biogas, wind, solar, and micro hydropower are helping bring electricity to rural parts of Nepal, left. The Nepal-based nongovernmental organization People, Energy & Environment Development Association says that women and girls here play vital roles in spreading awareness about the dangers of indoor air pollution and introducing new technologies to mitigate them.30 A Nepalese woman, right, in the district of Rukum, cooks on an electric induction stovetop. Images, left to right: © iStock.com/LSP1982; Courtesy of People, Energy and Environment Development Association (PEEDA), Nepal.
Montage of electric power pole and wires backed by Himalaya range and woman in Nepalese dress by a pot on an electric burner.
Even if the case for carbon-free, renewable household electricity is ultimately unassailable—and even if electric looks to be the way of the future for much of the world—reliable and affordable power cannot yet be deployed at the flip of a switch in many low-resource settings. That leaves room for other, distributed systems using mini-grids or sustainable biofuels to help improve peoples’ lives right away, says Ash Sharma of the Finland-based Nordic Environment Finance Corporation, which invests in clean cooking and off-grid renewable energy through the Modern Cooking Facility for Africa and the Beyond the Grid Fund for Africa, respectively.
There is evidence that people would welcome electric cooking devices, given the opportunity.31,32 “The demand for clean cooking is so enormous that I think there’s space in the market. We’ll see growth across the board, across technologies, really,” says Sharma. “And I think electric cooking has a rosy future.”
Nate Seltenrich covers science and the environment from the San Francisco Bay Area. His work on subjects including energy, ecology, and environmental health has appeared in a wide variety of regional, national, and international publications.
==== Refs
References
1. WHO (World Health Organization). 2022. Household air pollution. https://www.who.int/news-room/fact-sheets/detail/household-air-pollution-and-health [accessed 26 January 2023].
2. World Bank. 2020. The State of Access to Modern Energy Cooking Services. https://www.worldbank.org/en/topic/energy/publication/the-state-of-access-to-modern-energy-cooking-services [accessed 26 January 2023].
3. Stoner O, Lewis J, Martínez IL, Gumy S, Economou T, Adair-Rohani H, et al. 2021. Household cooking fuel estimates at global and country level for 1990 to 2030. Nat Commun 12 (1 ):5793, PMID: , 10.1038/s41467-021-26036-x.34608147
4. Smith KR, Pillarisetti A. 2017. Household Air Pollution from Solid Cookfuels and Its Effects on Health. In: Injury Prevention and Environmental Health. Mock CN, Nugent R, Kobusingye O, Smith KR, eds. 3rd ed. Washington, DC: International Bank for Reconstruction and Development/World Bank.
5. Harvey CA, Guariguata MR. 2021. Raising the profile of woodfuels in the forest landscape restoration agenda. Conserv Sci Pract 3 (2 ):e342, 10.1111/csp2.342.
6. Yongabi KA, Teko GN, Nji C, Endeley JB, Laar A. 2016. Gender assessment of hazard exposure, risk awareness and cultural perceptions associated with firewood usage: a case study of Mendakwe village in Cameroon. Int Res J Public Environ Health 3 (12 ):278–285, 10.15739/irjpeh.16.034.
7. Krishnapriya PP, Chandrasekaran M, Jeuland M, Pattanayak SK. 2021. Do improved cookstoves save time and improve gender outcomes? Evidence from six developing countries. Energy Econ 102 :105456, 10.1016/j.eneco.2021.105456.
8. Bailis R, Drigo R, Ghilardi A, Masera O. 2015. The carbon footprint of traditional woodfuels. Nat Clim Chang 5 (3 ):266–272, 10.1038/nclimate2491.
9. Johnson MA, Steenland K, Piedrahita R, Clark ML, Pillarisetti A, Balakrishnan K, et al. 2020. Air pollutant exposure and stove use assessment methods for the Household Air Pollution Intervention Network (HAPIN) trial. Environ Health Perspect 128 (4 ):47009, PMID: , 10.1289/EHP6422.32347764
10. Johnson M, Pillarisetti A, Piedrahita R, Balakrishnan K, Peel JL, Steenland K, et al. 2022. Exposure contrasts of pregnant women during the Household Air Pollution Intervention Network randomized controlled trial. Environ Health Perspect 130 (9 ):97005, PMID: , 10.1289/EHP10295.36112539
11. Intergovernmental Panel on Climate Change. 2022. The evidence is clear: the time for action is now. We can halve emissions by 2030. [Press release]. https://www.ipcc.ch/2022/04/04/ipcc-ar6-wgiii-pressrelease/ [accessed 26 January 2023].
12. Singer BC, Pass RZ, Delp WW, Lorenzetti DM, Maddalena RL. 2017. Pollutant concentrations and emission rates from natural gas cooking burners without and with range hood exhaust in nine California homes. Build Environ 122 :215–229, 10.1016/j.buildenv.2017.06.021.
13. Dalaba M, Alirigia R, Mesenbring E, Coffey E, Brown Z, Hannigan M, et al. 2018. Liquified petroleum gas (LPG) supply and demand for cooking in northern Ghana. Ecohealth 15 (4 ):716–728, PMID: , 10.1007/s10393-018-1351-4.30109459
14. Trading Economics. 2022. Propane. [Website.] https://tradingeconomics.com/commodity/propane [accessed 26 January 2023].
15. World Bank. 2022. Tracking SDG7: The Energy Progress Report, 2022. https://trackingsdg7.esmap.org/data/files/download-documents/sdg7-report2022-full_report.pdf [accessed 26 January 2023].
16. Cook J. 2019. Cooking up a storm at the Modern Energy Cooking Services Programme Launch. https://mecs.org.uk/cooking-up-a-storm-at-the-modern-energy-cooking-services-programme-launch/ [accessed 26 January 2023].
17. MECS (Modern Energy Cooking Services). 2022. ASHES MECS webinar. https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcolostate-my.sharepoint.com%2F%3Av%3A%2Fg%2Fpersonal%2Fclorange_colostate_edu%2FEZVDVF7xT0xLujwgRQER3pcBrRZz09vn8xMr55tt5BjMbQ%3Fe%3DVsj6KQ&data=05%7C01%7CJohn.Volckens%40ColoState.EDU%7C9a486979aa944853d97508da968a0c13%7Cafb58802ff7a4bb1ab21367ff2ecfc8b%7C0%7C0%7C637987816922423060%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=Aph8ZnxVTgo4qB%2F%2BKQnWlmZqynH%2BuwHRLqdJ8KZgx4A%3D&reserved=0 [accessed 26 January 2023].
18. International Energy Agency. 2017. Energy Access Outlook 2017: From Poverty to Prosperity. https://iea.blob.core.windows.net/assets/9a67c2fc-b605-4994-8eb5-29a0ac219499/WEO2017SpecialReport_EnergyAccessOutlook.pdf [accessed 26 January 2023].
19. Scott N, Batchelor S, Abbo M S, Akankwatsa D. 2019. Discrete Choice Modelling Survey, Uganda. Draft analysis for comment, September 2019. https://mecs.org.uk/wp-content/uploads/2020/12/MECS-Discrete-Choice-Analysis-Uganda.pdf [accessed 26 January 2023].
20. WHO. 2021. Global launch: tracking SDG7: the Energy Progress Report. https://www.who.int/news/item/07-06-2021-global-launch-tracking-sdg7-the-energy-progress-report [accessed 26 January 2023].
21. MECS. 2023. The Kenya eCookbook. https://mecs.org.uk/the-kenya-ecookbook/ [accessed 26 January 2023].
22. Balakrishnan K, Steenland K, Clasen T, Chang H, Johnson M, Pillarisetti A, et al. 2022. Exposure–response relationships for personal exposure to fine particulate matter (PM2·5), carbon monoxide, and black carbon and birthweight: results from the multi-country household air pollution intervention network (HAPIN) trial. medRxiv. Preprint posted online August 8, 2022, 10.1101/2022.08.06.22278373.
23. WHO. 2021. WHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide. https://apps.who.int/iris/bitstream/handle/10665/345329/9789240034228-eng.pdf [accessed 24 February 2023].
24. MECS. 2023. Challenge fund. https://mecs.org.uk/challenge-fund/ [accessed 26 January 2023].
25. MECS. 2021. Global challenge calling on international leaders to achieve 40,60 by 2030 for clean cooking. https://mecs.org.uk/global-challenge-calling-on-international-leaders-to-achieve-4060-by-2030-for-clean-cooking/ [accessed 26 January 2023].
26. ESMAP (Energy Sector Management Assistance Program). 2020. The State of Access to Modern Energy Cooking Services. https://esmap.org/the-state-of-access-to-modern-energy-cooking-services [accessed 26 January 2023].
27. MECS. Myanmar eCookbook. https://mecs.org.uk/wp-content/uploads/2021/09/Myanmar-eCookbook-English.pdf [accessed 26 January 2023].
28. ESMAP. 2022. Multi-Tier Framework. [Website.] https://mtfenergyaccess.esmap.org/ [accessed 26 January 2023].
29. United Nations. 2020. Affordable and Clean Energy: Why It Matters. [Fact sheet.] https://www.un.org/sustainabledevelopment/wp-content/uploads/2016/08/7_Why-It-Matters-2020.pdf [accessed 26 January 2023].
30. People, Energy & Environment Development Association. 2018. Energy Insight. Vol V. http://peeda.net/wp-content/uploads/2018/12/PEEDA-Energy-Insight-Volume-V.pdf [accessed 26 January 2023].
31. Pattanayak SK, Jeuland M, Lewis JJ, Usmani F, Brooks N, Bhojvaid V, et al. 2019. Experimental evidence on promotion of electric and improved biomass cookstoves. Proc Natl Acad Sci USA 116 (27 ):13282–13287, PMID: , 10.1073/pnas.1808827116.31118284
32. Jeuland M, Pattanayak SK, Tan Soo JS, Usmani F. 2020. Preferences and the effectiveness of behavior-change interventions: evidence from adoption of improved cookstoves in India. J Assoc Environ Resour Econ 7 (2 ):305–343, 10.1086/706937.
|
PMC009xxxxxx/PMC9976611.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36856429
EHP11721
10.1289/EHP11721
Research
Association of Lifetime Exposure to Glyphosate and Aminomethylphosphonic Acid (AMPA) with Liver Inflammation and Metabolic Syndrome at Young Adulthood: Findings from the CHAMACOS Study
Eskenazi Brenda 1
Gunier Robert B. 1
Rauch Stephen 1
Kogut Katherine 1
Perito Emily R. 2 3
Mendez Xenia 1
Limbach Charles 4
Holland Nina 1
Bradman Asa 1 5
Harley Kim G. 1
Mills Paul J. 6
https://orcid.org/0000-0002-2008-9714
Mora Ana M. 1
1 Center for Environmental Research and Community Health (CERCH), School of Public Health, University of California, Berkeley, Berkeley, California, USA
2 Department of Pediatrics, University of California, San Francisco, San Francisco, California, USA
3 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA
4 Monterey County Health Department, Salinas, California, USA
5 Department of Public Health, University of California, Merced, Merced, California, USA
6 Department of Family Medicine and Public Health, University of California, San Diego, La Jolla, California, USA
Address correspondence to Brenda Eskenazi, Center for Environmental Research and Community Health (CERCH), School of Public Health, University of California at Berkeley, 1995 University Ave., Suite 265, Berkeley, CA 94704 USA. Telephone: +1 (510) 517-2831. Email: [email protected]
01 3 2023
3 2023
131 3 03700114 6 2022
06 1 2023
17 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
The prevalence of liver disorders and metabolic syndrome has increased among youth. Glyphosate, the most widely used herbicide worldwide, could contribute to the development of these conditions.
Objective:
We aimed to assess whether lifetime exposure to glyphosate and its degradation product, aminomethylphosphonic acid (AMPA), is associated with elevated liver transaminases and metabolic syndrome among young adults.
Methods:
We conducted a prospective cohort study (n=480 mother–child dyads) and a nested case–control study (n=60 cases with elevated liver transaminases and 91 controls) using data from the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS). We measured glyphosate and AMPA concentrations in urine samples collected during pregnancy and at child ages 5, 14, and 18 y from cases and controls. We calculated glyphosate residue concentrations: [glyphosate + (1.5×AMPA)]. We estimated the amount of agricultural-use glyphosate applied within a 1-km radius of every residence from pregnancy to age 5 y for the full cohort using California Pesticide Use Reporting data. We assessed liver transaminases and metabolic syndrome at 18 y of age.
Results:
Urinary AMPA at age 5 y was associated with elevated transaminases [relative risk (RR) per 2-fold increase=1.27, 95% confidence interval (CI): 1.06, 1.53] and metabolic syndrome (RR=2.07, 95% CI: 1.38, 3.11). Urinary AMPA and glyphosate residues at age 14 y were associated with metabolic syndrome [RR=1.80 (95% CI: 1.10, 2.93) and RR=1.88 (95% CI: 1.03, 3.42), respectively]. Overall, a 2-fold increase in urinary AMPA during childhood was associated with a 14% and a 55% increased risk of elevated liver transaminases and metabolic syndrome, respectively. Living near agricultural glyphosate applications during early childhood (birth to 5 y of age) was also associated with metabolic syndrome at age 18 y in the case–control group (RR=1.53, 95% CI: 1.16, 2.02).
Discussion:
Childhood exposure to glyphosate and AMPA may increase risk of liver and cardiometabolic disorders in early adulthood, which could lead to more serious diseases later in life. https://doi.org/10.1289/EHP11721
Supplemental Material is available online (https://doi.org/10.1289/EHP11721).
A.B. is a volunteer member of the Board of Trustees for The Organic Center, a nonprofit organization addressing scientific issues about organic food and agriculture. R.G. consulted with a law firm on possible chlorpyrifos litigation. All other authors declare that they have no conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
The prevalence of childhood obesity and metabolic syndrome has increased at an alarming rate in the United States,1,2 particularly among populations of color.1–3 Accompanying this, has been an increase in nonalcoholic fatty liver disease (NAFLD),4 a condition that can lead to cirrhosis and hepatocellular carcinoma later in life.5 Although diet and physical activity play an important role in cardiometabolic and liver disorders, some hypothesize that exposure to synthetic chemicals may also be involved.6,7
Use of the herbicide glyphosate has markedly increased in the United States in the last two decades and currently is the most commonly used broad-spectrum herbicide worldwide.8,9 It is used to control broadleaf grasses and weeds in agriculture, forestry, and right-of-way clearances, in parks, and in yards as a component in home weed killers (e.g., Round-up®). Exposure to glyphosate and its prime degradation product, aminomethylphosphonic acid (AMPA), can occur through consumption of contaminated food,10,11 air,12 dust,12 and water.13 In food, glyphosate has been detected primarily in grains14 and legumes, including soybeans,15 but it has also been detected in other fruits and vegetables10,16 and in baby formula.17 AMPA is also the degradation product of amino-polyphosphonates, which are extensively used in detergents, fire retardants, and other compounds.18
The potential impact of glyphosate on human health is controversial and widely debated.19–22 Like glyphosate, AMPA raises toxicologic concern.23 In 2015, the International Agency for Research on Cancer (IARC) classified glyphosate as probably carcinogenic to humans (Group 2A),21 but to date the U.S. Environmental Protection Agency (U.S. EPA) has found no evidence of risk to human health.20 Animal24–28 and human29–31 studies have suggested that exposure to glyphosate may be related to liver disease, and some researchers have hypothesized a potential relationship with metabolic disorders.32,33 In the current study, we investigated the association of prenatal and childhood exposure to glyphosate and AMPA—as indicated by urinary concentrations and registry data of nearby agricultural use of glyphosate—with markers of liver inflammation and metabolic syndrome in young adults.
Methods
Study Population
Participants are mother–child dyads enrolled in the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS) longitudinal cohort of children born between 2000 and 2002 in California’s Salinas Valley.34 Briefly, pregnant women receiving prenatal care at community clinics primarily serving farmworker families were eligible for enrollment if they were at least 18 y old, <20 wk gestation, spoke English or Spanish, met income requirements for public health insurance (Medi-Cal), and planned to deliver at the county hospital. Of 601 CHAMACOS women enrolled, 527 remained in the study at delivery (n=537 live-born children, including twins) in the period 2000–2001 (CHAMACOS1 participants). In 2009–2010 we enrolled a second wave of 305 mother–child dyads of children born in 2000–2002 (CHAMACOS2 participants), using similar selection criteria. CHAMACOS1 mothers completed visits during pregnancy and delivery, and children were followed at approximately 1- to 2-y intervals. CHAMACOS2 mother–child dyads completed a baseline data collection visit at 9 y. Thereafter, CHAMACOS1 and CHAMACOS2 families completed the same study visits.
The present analyses included 480 CHAMACOS1 and CHAMACOS2 participants who completed the 18-y follow-up visit prior to our March 2020 COVID-19 closure and a subset of these who were selected for a nested case–control study of liver inflammation at age 18 y (Supplementary Figure 1). Cases (n=60) were defined by having an alanine transaminase (ALT) >44 IU/L or an aspartate aminotransferase (AST) >40 IU/L for males or >32 IU/L for females (cutoffs are based on LabCorp reference ranges); controls (n=91) had normal ALT and AST levels and were randomly selected and frequency-matched to cases on sex. Cases and controls were required to have maternal pregnancy and/or child urine specimens collected and stored from earlier study waves (n=415 were eligible for selection).
Mothers provided written informed consent; children provided verbal assent starting at age 7 y, written assent starting at age 12 y, and full written consent at age 18 y. All study activities were approved by the University of California, Berkeley Office for the Protection of Human Subjects.
Study Procedures
The specific data and sample collection procedures are described below:
Maternal and youth interviews.
Mothers were interviewed in English or Spanish by trained bilingual bicultural research assistants who used structured questionnaires at each study visit, including the prenatal visit at 26 wk gestation, and the child follow-up visits at 5, 14, and 18 y. We collected information on family occupation history, socioeconomic status, medical history, and lifestyle factors. Youth were interviewed in English at ages 14 and 18 y.
Dietary assessments.
Mothers were interviewed using a validated food frequency questionnaire (FFQ) of their diet at 26 wk gestation35–37 and their child’s diet at age 5 y.38,39 Youth completed a validated food screener at 14 y40 and basic diet questions at 18 y.41,42 In the FFQ administered during pregnancy, women were asked how often they ate various food items in the previous 3 months and how much they consumed each time. In the FFQ at 5 y, mothers were asked about the number of times their child ate various foods in the previous 4 wk. In the food screener at 14 y, the youth were asked the number of days in the previous week they ate or drank different food items and how much in 1 day. The validated FFQs and screener underwent proprietary data processing (Pregnancy35–37 and 14-y FFQL Nutritionquest40; 5-y FFQ: Harvard Nutrition Questionnaire Service Center)38,39,43 to convert reported dietary intake into summary variables and individual food items. We selected a priori those dietary variables that included foods commonly treated with glyphosate. We dichotomized continuous summary variables (i.e., total calories, total carbohydrates, whole grains, bran, fruits, and vegetables) into being above or below the median observed in our sample; we also dichotomized reported intake of individual foods (i.e., cold cereal, hot cereal, bread, tortillas, legumes) (e.g., <1 time per day vs. ≥1 time per day). In addition to the administration of the validated FFQs and screener, we asked mothers how frequently their 5-y-old children consumed fast food; we directly asked young adults this question at the 14-y and 18-y visits. We asked the 14- and 18-y-olds to report on their overall alcohol consumption; for 18-y-olds, we also asked about recent binge drinking (≥4 drinks in a row for females, ≥5 for males). At the 5-, 14-, and 18-y visits, we queried the mothers about family food security (U.S. Department of Agriculture Food Security Scale, Short Form).44
Body measurements.
At the 18-y visit, we recorded young adult participants’ height in triplicate using a wall-mounted stadiometer and their weight (a single measurement) using a Tanita bioimpedance scale (Tanita TBF-300A Body Composition Analyzer; Tanita Corporation). We calculated body mass index (BMI) based on the average recorded height and the single weight measurement. We measured waist circumference three times with a measuring tape wrapped around the abdomen, parallel to the floor, at the iliac crest. We present the average of these three measurements as the waist circumference. We measured blood pressure in triplicate using an automated oscillometric monitor (Dinamap Carescape V100). Participants sat and rested for two minutes prior to the first reading and had a 1-min rest between each subsequent reading. We present the average of the second two systolic and diastolic blood pressure readings, respectively, as the measured blood pressure.
Blood collection for clinical chemistries.
At the 18-y visit, a fasting blood sample was collected from the young adults via venipuncture and analyzed for ALT, AST, glucose, high-density lipoprotein (HDL) cholesterol, and serum triglycerides (LabCorp). For those with elevated ALT, we also measured bilirubin and Hepatitis B and C (LabCorp) as well as ceruloplasmin and actin smooth muscle antibodies (LabCorp) to rule out common causes of liver disease other than NAFLD.
Urinary measurements of glyphosate and AMPA.
For the case–control subgroup, we analyzed maternal urine collected at approximately 26 wk gestation and child urine collected at the 5-y, 14-y, and 18-y visits. All urine specimens were aliquoted into clean glass containers with Teflon caps and stored at −80°C at our Salinas research field office until shipment on dry ice to our University of California Berkeley–based biorepository, where they were stored at −80°C. Aliquots were kept in a frozen state until analysis with no intermediate freeze-thaw cycle. For prenatal, 5-y, and 14-y visits, spot urine samples were collected when participants had not fasted. However, at the 18-y visit, 40% (n=48 out of 121) of participants provided a nonfirst morning void spot urine sample under fasting conditions to coincide with a fasting blood draw.
Aliquots of maternal prenatal and child urine samples for our selected cases and controls were shipped on dry ice to the Center de Toxicologie du Québec, Institut National de Santé Publique du Québec (INSPQ). Glyphosate and AMPA were measured in a single extraction by ultraperformance liquid chromatography (UPLC)–mass spectroscopy/mass spectroscopy (MSMS) at INSPQ. This method was performed as previously published.45 External quality control for glyphosate and/or AMPA was ensured by INSPQ’s successful participation in the Quebec External Quality Assessment Scheme for Organic Substances in Urine (OSEQAS), German External Quality Assessment Scheme (G-EQUAS), and Human Biomonitoring for Europe (HBM4EU, reference laboratory) program (see certificates of participation in the Supplementary Material). Limits of detection (LOD) were 0.08μg/L for glyphosate and 0.09μg/L for AMPA. Specific gravity was measured by a refractometer (Atago Company Ltd.).
California Pesticide Use Reporting (PUR) data.
For the full cohort, we recorded families’ residential addresses each time they attended a study visit. In addition, at the 16-y visit, mothers completed a detailed residential history interview in which all residences from the start of their pregnancy through their child’s 16-y visit were obtained.
To characterize potential exposure, we estimated agricultural glyphosate use near each participant’s residence during the prenatal and postnatal (birth to 5-y visit) time periods using PUR data from 1999 to 2007.46–48 PUR data include the amount (kilograms) of active ingredient applied, application date, and location to a 1-square-mile section (1.6km×1.6km) defined by the U.S. Public Land Survey System (PLSS).47–49 We weighted the amount of glyphosate applied in each section by the proportion of land area that was included in a 1-km radius and accounted for the potential downwind transport of glyphosate from the application site using wind direction from the closest meteorological station50 based on the daily proportion of time the wind blew from each of eight directions. We summed all glyphosate agricultural applications to determine estimates of the wind-weighted amount of glyphosate (kg) applied around all residences for each participant during pregnancy and from birth to the 5-y visit.
Data Analysis
We used chi-square tests to compare the detection frequencies of glyphosate and AMPA concentrations measured in urine samples collected from the mother during pregnancy and from the child at 5, 14, and 18 y. We fitted crude and multivariable Poisson regression models using robust standard errors for specific-gravity adjusted urinary glyphosate and AMPA concentrations, as well as total glyphosate residue concentrations, in relationship to a) case–control status defined by liver transaminases, b) metabolic syndrome, and c) other clinical chemistry and anthropometry measures measured at 18 y of age (dichotomized as within or outside normal clinical limits; see more details below). We estimated exposure to total glyphosate residues using the formula [glyphosate + (1.5×AMPA)].29,51,52 This formula, proposed by the Joint Meeting on Pesticide Residues,53 is derived from the ratio of the AMPA molecular weight to the glyphosate molecular weight (∼1.52) and assumes that AMPA and glyphosate have similar human toxicity. We used log2-transformed glyphosate, AMPA, and total glyphosate residue concentrations to reduce the influence of outliers. Values below the LOD were randomly imputed based on a log-normal distribution using maximum likelihood estimation.54 Models were run for time points when at least half of participants had concentrations above the LOD, which included 14- and 18-y glyphosate and glyphosate residue concentrations, and 5-, 14-, and 18-y AMPA concentrations. We fitted models for 18-y nonfasting urinary concentrations and for all 18-y urinary concentrations.
In the models for metabolic syndrome and other clinical chemistry and anthropometry measures collected at 18 y, we corrected for oversampling of individuals with elevated markers of liver inflammation (and by extension, males) using stratum-specific weights for elevated ALT/AST and sex, based on the ratio of the proportions of each group in the case–control subset and full study population (i.e., male controls: 0.887; male cases: 0.373; female controls: 2.668; and female cases: 0.364).55,56 The following clinical cutoffs for adults were used for these models: high-density lipoprotein (HDL) cholesterol <40mg/dL for males or <50mg/dL for females, serum triglycerides ≥150mg/dL, fasting glucose ≥100mg/dL, BMI ≥25 kg/m2, waist circumference ≥40 inches for males or ≥35 inches for females, and systolic blood pressure >130mm Hg and diastolic >85mm Hg.57–59 The presence of metabolic syndrome was indicated by having at least three of the following five factors: a) high systolic blood pressure or diastolic blood pressure; b) large waist circumference; c) elevated fasting serum glucose; d) elevated serum triglycerides; and e) low HDL cholesterol.59
To approximate lifetime glyphosate exposure, we fitted multiple informant models with repeated urinary concentrations at the 5-y, 14-y, and 18-y visits, using mixed-effects Poisson models with a random intercept for each participant60 (we did not include urinary concentrations during pregnancy because the detection frequency was low). For participants who provided a fasting urine sample at 18 y, their 18-y measurement was excluded, but their 5-y and 14-y measurements were retained in the models. To determine whether exposure–outcome associations differed across the visits at which samples were collected and thus the appropriateness of multiple informant models, we also ran models that included exposure × visit interaction terms.
We examined whether BMI (continuous) at 14 y mediated the observed associations of urinary AMPA and glyphosate residue concentrations with the outcomes of interest using Structural Equation Models (SEMs).61 In the case–control study group, we conducted sensitivity analyses excluding eight cases who had high actin (>19 U) or low ceruloplasmin (<16mg/dL male, <19mg/dL female) levels (based on LabCorp adult reference ranges) and/or who reported recent binge drinking in the past 30 d. We also used t-tests to examine the associations of dietary factors (dichotomous summary variables and individual food items) with glyphosate, AMPA, and glyphosate residues, measured concurrently.
We examined the correlation of maternal and child urinary concentrations of glyphosate and AMPA and PUR data from birth to 5 y. We constructed models of prenatal and postnatal PUR data in relationship to dichotomized clinical chemistry measures, anthropometric measures, and metabolic syndrome for the case–control subset (n=151) and the entire 18-y sample (n=415 for clinical chemistries, n=480 for anthropometry). Because only 50.1% of the women lived within 1km of an agricultural glyphosate application during pregnancy, we modeled prenatal exposure to glyphosate as a binary measure (zero vs. nonzero use within 1km). Because 95.2% of children lived near agricultural glyphosate between birth and age 5 y, we modeled postnatal exposure using the sum (log2-transformed) of all agricultural glyphosate used during this period.
Covariates for multivariable models using urinary concentrations and PUR data were selected using a directed acyclic graph (DAG) (Supplemental Figure 2) and included youths’ sex and any alcohol consumption at 18 y (yes vs. no), maternal prepregnancy BMI (continuous), parental work in agriculture during the prenatal period (yes vs. no), as well as household poverty (above the poverty line vs. below)62 and food security (high/marginal food security vs. low/very low food security)44 at the time of sample collection (for models of urinary concentrations) or at the 18-y visit (for models of PUR data).
All statistical analyses were conducted using Stata 15.0 (StataCorp) and ArcMap 10.6.1 (Esri Corp.).
Results
In Table 1, we present demographic information of the study participants. Most mothers were overweight or obese before pregnancy. Seventy percent of the young adult cases were male in comparison with 47.5% in the full 18-y cohort. At the 18-y visit, 42.5% of the families were living at or below the federal poverty line, and 28.2% were at low or very low food security. Among the 18-y young adults, 10.6% fulfilled the criteria for metabolic syndrome, and 57.2% were overweight or obese (Table 2). In the case–control subset, 28.8% of cases vs. 4.4% of controls fulfilled the criteria for metabolic syndrome, and 85.0% of cases were overweight or obese vs. 57.2% of controls.
Table 1 Demographic characteristics of participants in the liver disease nested case–control study and all 18-y-old participants, CHAMACOS study, 1999–2020 [n (%)].
All 18-y-old participantsa
(n=480) Casesb
(n=60) Controlsb
(n=91)
Maternal age at delivery (y)
18–24 200 (41.7) 30 (50.0) 32 (35.2)
25–29 148 (30.8) 15 (25.0) 35 (38.5)
30–34 84 (17.5) 7 (11.7) 11 (12.1)
35–45 48 (10.0) 8 (13.3) 13 (14.3)
Maternal education
≤6th grade 207 (43.1) 25 (41.7) 47 (51.6)
7th–12th grade 162 (33.8) 24 (40.0) 31 (34.1)
High school graduate 111 (23.1) 11 (18.3) 13 (14.3)
Marital status at pregnancy
Not married/living as married 70 (14.6) 8 (13.6) 16 (17.6)
Married/living as married 408 (85.4) 51 (86.4) 75 (82.4)
Missing 2 1 0
Years in U.S. prior to delivery
≤1 y 81 (16.9) 7 (11.7) 22 (24.2)
2–5 y 139 (29.0) 17 (28.3) 22 (24.2)
6–10 y 121 (25.2) 22 (36.7) 26 (28.6)
≥11 y, nonnative 96 (20.0) 12 (20.0) 16 (17.6)
Entire life 43 (9.0) 2 (3.3) 5 (5.5)
Language spoken at home (during pregnancy)
Spanish primarily 438 (91.6) 57 (96.6) 84 (92.3)
Spanish and English equally 17 (3.6) 1 (1.7) 4 (4.4)
English primarily 19 (4.0) 1 (1.7) 2 (2.2)
Other 4 (0.8) 0 (0.0) 1 (1.1)
Missing 2 1 0
Maternal prepregnancy BMI
Normal or underweight (<25.0 kg/m2) 160 (33.4) 14 (23.3) 34 (37.4)
Overweight (25–29.9 kg/m2) 201 (42.0) 22 (36.7) 38 (41.8)
Obese (≥30 kg/m2) 118 (24.6) 24 (40.0) 19 (20.9)
Missing 1 0 0
Parental work in agriculture during pregnancy
Yes 356 (74.5) 41 (69.5) 70 (76.9)
No 122 (25.5) 18 (30.5) 21 (23.1)
Missing 2 1 0
Participant sex
Male 228 (47.5) 42 (70.0) 64 (70.3)
Female 252 (52.5) 18 (30.0) 27 (29.7)
Any alcohol consumption at 18 y
Yes 249 (52.1) 27 (45.8) 48 (52.7)
No 229 (47.9) 32 (54.2) 43 (47.3)
Missing 2 1 0
Household poverty at 18 y
At or below poverty line 196 (42.5) 31 (52.5) 37 (41.1)
Above the poverty line 265 (57.5) 28 (47.5) 53 (58.9)
Missing 19 1 1
Food security at 18 y
High or marginal 338 (71.8) 42 (70.0) 61 (67.0)
Low 97 (20.6) 13 (21.7) 21 (23.1)
Very low 36 (7.6) 5 (8.3) 9 (9.9)
Missing 9 0 0
Note: BMI, body mass index; CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; U.S., United States: .
a Participants with blood draw and clinical chemistry at 18-y visit prior to onset of COVID-19 shelter in place (March 2020).
b Participants with urinary glyphosate measurements.
Table 2 Liver clinical chemistry measures and metabolic outcomes of all 18-y-old CHAMACOS participants and those in the nested case–control subset {GM [GSD] or n (%)}.
Outcome All 18-y-old participants
(n=405–474) Cases
(n=60) Controls
(n=91)
Elevated liver transaminases 61 (14.7) 60 (100.0) 0 (0.0)
ALT (IU/L)a 18.9 [1.9] 57.8 [1.6] 17.1 [1.5]
AST (IU/L)a 19.0 [1.5] 35.6 [1.5] 17.8 [1.3]
BMI category — — —
Normal (<25 kg/m2) 203 (42.8) 9 (15.0) 39 (42.9)
Overweight (25–29.9 kg/m2) 130 (27.4) 13 (21.7) 30 (33.0)
Obese (≥30 kg/m2) 141 (29.8) 38 (63.3) 22 (24.2)
Metabolic syndrome 43 (10.6) 17 (28.8) 4 (4.4)
Blood pressure (systolic ≥130 or diastolic ≥90mm Hg) 52 (11.0) 15 (25.4) 8 (8.8)
Waist circumference (≥40 in for male, ≥35 in for female) 185 (39.2) 42 (70.0) 24 (26.4)
Fasting glucose (≥100mg/dL) 20 (4.8) 8 (13.3) 4 (4.4)
Triglycerides (≥150mg/dL) 50 (12.1) 18 (30.0) 6 (6.6)
HDL cholesterol (<40mg/dL male, <50mg/dL female) 158 (38.1) 29 (48.3) 29 (31.9)
Note: —, no data; ALT, alanine transaminase; AST, aspartate aminotransferase; BMI, body mass index; CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; GM, geometric mean; GSM, geometric standard deviation; HDL, high-density lipoprotein; in, inches: .
a Geometric mean (geometric standard deviation).
Urinary Glyphosate and AMPA Concentrations
Few prenatal samples had detectable concentrations of glyphosate (4.2% >LOD) or AMPA (14.1% >LOD), and these percentages did not differ between cases and controls (Supplemental Tables 1 and 2). Overall, detection frequencies of glyphosate and AMPA were higher for children than for pregnant mothers (Supplemental Table 1).
The detection frequency of glyphosate was low at age 5 y (35.2% >LOD) and differed somewhat between cases and controls (46.9% vs. 28.6%, p=0.08) (Supplemental Table 2), whereas the detection frequency of AMPA was higher than for glyphosate at age 5 y (76.9% >LOD) with a specific gravity-corrected geometric mean (GMsg) of 0.22μg/L and a geometric standard deviation (GSDsg) of 2.53 (Supplemental Table 1) but did not differ between cases and controls (Supplemental Table 2).
At 14 and 18 y, cases and controls did not differ in the proportion of urine samples with detectable levels of glyphosate or AMPA (Supplemental Table 2). Both glyphosate and AMPA had higher detection frequencies and geometric means at age 14 y than at all other ages [78.9% >LOD; GMsg (GSDsg)=0.28μg/L (2.37); and 93.3% >LOD; GMsg (GSDsg)=0.72μg/L (2.27); respectively] (Supplemental Table 1). Specifically, 18-y-olds overall had lower detection frequencies and geometric means of glyphosate [54.6% >LOD; GMsg (GSDsg)=0.16μg/L (2.77)] and AMPA [66.9% >LOD; GMsg (GSDsg)=0.25μg/L (2.46)] than at 14 y, even among those who had not fasted [glyphosate: 65.8% >LOD; GMsg (GSDsg)=0.17μg/L (2.80) and AMPA: 75.3% >LOD; GMsg (GSDsg)=0.27μg/L (2.76)]. In comparison with 18-y-olds who had fasted, those who had not fasted had a greater proportion of samples above the detection limit for glyphosate (65.8% vs. 37.5%) and AMPA (75.3% vs. 54.2%).
Urinary concentrations of glyphosate and AMPA were correlated within each study wave (r=0.35–0.66) (Supplemental Table 3). Dietary factors were only modestly associated with urinary glyphosate and AMPA concentrations (Supplemental Table 4): Higher consumption of cold cereal was associated with somewhat higher AMPA and glyphosate residue concentrations at 5 y; higher total caloric and carbohydrate intake and higher consumption of hot cereal, bread, and fruits and vegetables were associated with higher concentrations of glyphosate at 14 y.
Residential Proximity to Glyphosate Use
Agricultural applications of glyphosate were low during the time of pregnancy (∼Year 2000) and age 5 y visits (∼Year 2005) but higher at ages 14 (∼Year 2014) and 18 (∼Year 2018) (Figure 1; Supplemental Table 5). Glyphosate use near the child’s residence during early childhood (birth to 5 y) was not correlated with urinary glyphosate concentrations at age 5 (r=−0.007) and weakly correlated with urinary AMPA concentrations at this same age (r=0.12) (Supplemental Table 3).
Figure 1. Agricultural use of glyphosate in Monterey County, California, 2000–2018. Note: Sources: Esri, General Bathymetric Chart of the Oceans (GEBCO), National Oceanic and Atmospheric Administration, National Geographic, Garmin, HERE, Geonames.org, and other contributors.
Figure 1 is a set of one bar graph and four maps. The bar graph titled total glyphosate use (kilogram), plotting kilogram, ranging from 0 to 120000 in increments of 20000 (y-axis) across years, ranging from 2000 to 2018 in increments of 2 (x-axis). The four maps of Monterey County, California, depict the agricultural use of glyphosate in the years 2000, 2008, 2015, 2014, and 2018. The range for glyphosate (kilogram) ranges as 0.1 to 50, 51 to 100, 101 to 250, 251 to 500, 501 to 1000, and 1001 to 1420. The areas highlighted are Castroville, Salinas, Gonzales, Soledad, Greenfield, Monterey, and King city.
Associations of Urinary Glyphosate and AMPA Concentrations with Liver and Cardiometabolic Outcomes
We observed associations of urinary AMPA and glyphosate residues with markers of liver inflammation (Table 3; Supplemental Table 6). Higher AMPA concentrations at 5 y were associated with elevated liver transaminases at 18 y [relative risk (RR)=1.27; 95% confidence interval (CI): 1.06, 1.53]. Higher AMPA and glyphosate residue concentrations at 18 y among nonfasting participants were marginally associated with elevated liver enzymes [RR=1.13 (95% CI: 0.98, 1.32) and RR=1.16 (95% CI: 1.00, 1.35), respectively]; these associations were further attenuated when fasting 18-y-olds were included (Supplemental Table 7). RRs remained similar when we excluded the eight cases with high actin, with low ceruloplasmin, and/or who binge-drank alcohol (Supplemental Table 8).
Table 3 Adjusteda RRs and 95% CI for 2-fold increases in child urinary glyphosate, AMPA, and glyphosate residue concentrations (specific gravity-corrected, μg/L) and abnormal markers of liver inflammation and metabolic syndrome (and its components) in CHAMACOS young adults in case–control group.
Outcome Glyphosate AMPA Glyphosate residuesb
14 y
(n=103–104) 18 yc
(n=72–73) 5 y
(n=90–91) 14 y
(n=104–105) 18 yc
(n=72–73) 14 y
(n=103–104) 18 yc
(n=72–73)
Elevated liver transaminases 1.07 (0.89, 1.29) 1.11 (0.97, 1.27) 1.27 (1.06, 1.53) 1.16 (0.96, 1.39) 1.13 (0.98, 1.32) 1.15 (0.93, 1.42) 1.16 (1.00, 1.35)
Metabolic syndrome 1.22 (0.76, 1.96) 1.20 (0.71, 2.02) 2.07 (1.38, 3.11) 1.80 (1.10, 2.93) 1.59 (0.99, 2.54) 1.88 (1.03, 3.42) 1.54 (0.94, 2.52)
High blood pressure 1.30 (0.83, 2.04) 0.98 (0.70, 1.37) 1.29 (0.80, 2.10) 1.50 (0.98, 2.30) 1.28 (0.88, 1.86) 1.55 (0.93, 3.60) 1.23 (0.85, 1.79)
Large waist circumference 0.88 (0.71, 1.10) 1.19 (0.94, 1.50) 1.11 (0.88, 1.40) 1.15 (0.87, 1.53) 1.26 (0.96, 1.65) 1.06 (0.74, 1.52) 1.33 (1.02, 1.73)
High glucose 0.95 (0.57, 1.60) 1.75 (0.20, 15.49) 2.95 (1.70, 5.14) 1.63 (0.92, 2.88) 4.29 (0.31, 59.55) 1.55 (0.75, 3.20) 3.16 (0.30, 33.36)
High triglycerides 0.95 (0.70, 1.31) 1.18 (0.92, 1.52) 1.39 (0.81, 2.37) 1.51 (1.03, 2.22) 1.45 (1.07, 1.96) 1.40 (0.90, 2.18) 1.39 (1.04, 1.88)
Low HDL cholesterol 0.86 (0.70, 1.05) 1.16 (0.94, 1.42) 0.90 (0.72, 1.12) 1.19 (0.96, 1.48) 1.15 (0.89, 1.48) 1.11 (0.84, 1.45) 1.20 (0.93, 1.54)
Note: AMPA, aminomethylphosphonic acid; BMI, body mass index; CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; CI, confidence interval; HDL, high-density lipoprotein; RR, relative risk.
a Models adjusted for sex, any alcohol consumption at 18 y (yes/no), maternal prepregnancy BMI, parental work in agriculture during pregnancy (yes/no), household poverty status at time of visit (above vs. below the poverty threshold), and food security at time of visit (high/marginal security vs. low and very low security).
b Calculated using the formula: [Glyphosate + (1.5×AMPA)].
c Limited to participants with nonfasting urine samples.
We also found associations of urinary AMPA and total glyphosate residue concentrations with metabolic syndrome and related conditions. Two-fold increases in AMPA at 5 y (RR=2.07, 95% CI: 1.38, 3.11) and in AMPA and glyphosate residues at 14 y [RR=1.80 (95% CI: 1.10, 2.93) and RR=1.88 (95% CI: 1.03, 3.42), respectively] and at 18 y (among nonfasters) [RR=1.59 (95% CI: 0.99, 2.54) and RR=1.54 (95% CI: 0.94, 2.52), respectively] were associated with a 50% or greater increased risk of metabolic syndrome at 18 y (Table 3; Supplemental Table 6). In addition, higher AMPA concentrations at age 5 y were associated with elevated glucose levels (RR=2.95, 95% CI: 1.70, 5.14), and higher AMPA concentrations at ages 14 y and 18 y (among nonfasters) were associated with elevated triglycerides [RR=1.51 (95% CI: 1.03, 2.22) and RR=1.45 (95% CI: 1.07, 1.96) respectively]. Higher urinary glyphosate residue concentrations at age 18 y (among nonfasters) were associated with elevated triglycerides (RR=1.39, 95% CI: 1.04, 1.88) and large waist circumference (RR=1.33, 95% CI: 1.02, 1.73).
In multiple informant models, we did not find evidence of interaction by visit in the associations of glyphosate, AMPA, or glyphosate residue concentrations with our outcomes (Supplemental Table 9). Therefore, we fitted models without interaction terms, using repeated measurements to approximate lifetime exposure (Table 4). In these models, a 2-fold increase in childhood urinary concentrations of AMPA was associated with a 14% increased risk of elevated liver transaminases (95% CI: 1.05, 1.23) and a 55% increased risk of metabolic syndrome (95% CI: 1.19, 2.02) at age 18 y. Higher childhood urinary concentrations of AMPA were also associated with elevated blood pressure, glucose, and triglycerides and with larger waist circumference. Higher childhood glyphosate residue concentrations were also associated with increased risks of elevated liver transaminases (RR per 2-fold increase in concentrations=1.13, 95% CI: 1.05, 1.22), metabolic syndrome (RR=1.52, 95% CI: 1.12, 2.06), and elevated triglycerides (RR=1.22, 95% CI: 1.01, 1.46) (Table 4).
Table 4 Multiple informant models (RRs and 95% CI) for repeated child urinary glyphosate, AMPA, and glyphosate residue concentrations (specific gravity-corrected, μg/L) at the 5-y, 14-y, and 18-y visits and abnormal markers of liver inflammation and metabolic syndrome (and its components), using mixed-effects Poisson models with a random intercept for each CHAMACOS participant.a,b
Glyphosate
(n=121–122) AMPA
(n=121–122) Glyphosate residuesc
(n=121–122)
Elevated liver transaminases 1.05 (0.98, 1.13) 1.14 (1.05, 1.23) 1.13 (1.05, 1.22)
Metabolic syndrome 1.14 (0.82, 1.59) 1.55 (1.19, 2.02) 1.52 (1.12, 2.06)
High blood pressure 1.08 (0.85, 1.38) 1.23 (1.01, 1.50) 1.22 (0.99, 1.51)
Large waist circumference 1.02 (0.94, 1.12) 1.06 (0.99, 1.14) 1.06 (0.99, 1.14)
High glucose 1.11 (0.82, 1.49) 1.35 (1.02, 1.77) 1.30 (0.96, 1.77)
High triglycerides 0.99 (0.86, 1.14) 1.27 (1.07, 1.52) 1.22 (1.01, 1.46)
Low HDL cholesterol 0.94 (0.87, 1.02) 1.01 (0.96, 1.07) 0.99 (0.93, 1.06)
Note: AMPA, aminomethylphosphonic acid; BMI, body mass index; CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; CI, confidence interval; HDL, high-density lipoprotein; RR, relative risk.
a Fasting urine samples taken at 18 y are not included in models, but 5-y and 14-y samples are still included for those participants.
b Models adjusted for sex, any alcohol consumption at 18 years (yes/no), maternal prepregnancy BMI, parental work in agriculture during pregnancy (yes/no), household poverty status at 18 y (above vs. below the poverty threshold), and food security at 18 y (high/marginal security vs. low and very low security).
c Calculated using the formula: [Glyphosate + (1.5×AMPA)].
In the SEM models, we found no evidence that associations of urinary AMPA or glyphosate residues with elevated liver transaminases or metabolic syndrome were mediated by young adult BMI (Supplemental Table 10).
Association of Residential Proximity to Glyphosate Use with Liver and Cardiometabolic Outcomes
Any agricultural use of glyphosate near the home during pregnancy was associated with an increased risk of metabolic syndrome in the case–control subset (RR=3.42, 95% CI: 1.12, 10.42) but not in the full sample (RR=1.16, 95% CI: 0.66, 2.05) (Table 5; Supplemental Table 11). A 2-fold increase in nearby glyphosate use during early childhood was also associated with an increased risk of metabolic syndrome in the case–control group (RR=1.53, 95% CI: 1.16, 2.02), and with a somewhat elevated risk in the full sample (RR=1.15, 95% CI: 0.97, 1.35) (Table 5, Supplemental Table 11). In the case–control group, we also observed increased risk of elevated triglyceride levels with any nearby agricultural glyphosate use during pregnancy (RR per 2-fold increase=3.37, 95% CI: 1.36, 8.33) as well as with all agricultural glyphosate use during early childhood (RR=1.45, 95% CI: 1.11, 1.88) (Table 5).
Table 5 Adjusteda RRs and 95% CI for living within 1km of agricultural glyphosate use during maternal pregnancy (any use) and from birth to age 5 y (all use, in kilograms, log2) based on the California Pesticide Use Reporting (PUR) data and presence of elevated markers of liver inflammation or metabolic syndrome (and its components) among all CHAMACOS young adult participants and in case–control subset.
Outcome Any PUR use near home residence during pregnancy (yes/no) Sum of PUR use near home residence from birth to age 5 y (log2)
All 18-y-old participants
(n=397–464) Case–control subset
(n=149–150) All 18-y-old participants
(n=327–373) Case–control subset
(n=129–130)
Elevated liver transaminases 0.83 (0.53, 1.30) 1.14 (0.77, 1.71) 0.93 (0.80, 1.08) 0.98 (0.86, 1.11)
Metabolic syndrome 1.16 (0.66, 2.05) 3.42 (1.12, 10.42) 1.15 (0.97, 1.35) 1.53 (1.16, 2.02)
High blood pressure 1.00 (0.60, 1.67) 0.89 (0.40, 1.99) 0.95 (0.78, 1.16) 1.11 (0.85, 1.46)
Large waist circumference 1.04 (0.84, 1.29) 1.23 (0.74, 2.05) 1.05 (0.97, 1.13) 0.97 (0.82, 1.15)
High glucose 0.78 (0.31, 1.95) 0.39 (0.10, 1.54) 0.87 (0.65, 1.15) 0.79 (0.56, 1.11)
High triglycerides 0.99 (0.58, 1.67) 3.37 (1.36, 8.33) 1.09 (0.90, 1.32) 1.45 (1.11, 1.88)
Low HDL cholesterol 1.04 (0.81, 1.32) 1.07 (0.64, 1.78) 1.00 (0.91, 1.09) 0.91 (0.76, 1.07)
Note: BMI, body mass index; CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; CI, confidence interval; HDL, high-density lipoprotein; PUR, Pesticide Use Reporting; RR, relative risk.
a Models adjusted for sex, alcohol use at 18 y (never/ever), maternal prepregnancy BMI, parental work in agriculture during pregnancy (yes/no), household poverty status at 18 y (above vs. below the poverty threshold), and food security at 18 y (high/marginal security vs. low and very low security).
Discussion
We observed associations of glyphosate or AMPA exposure during childhood with liver inflammation and metabolic syndrome at young adulthood. More specifically, after accounting for multiple potential confounders, we found that higher urinary concentrations of AMPA, a degradation product of glyphosate and amino-polyphosphonates, and glyphosate residues between ages 5 and 18 y were associated with both elevated liver transaminases and metabolic syndrome at age 18 y. This association could not be explained by mediation by body mass. In addition, we found that agricultural glyphosate use during the prenatal period and/or childhood (from birth to age 5 y) was associated with metabolic syndrome at 18 y.
Our findings are consistent with Mills et al.,29 who observed that urinary concentrations of AMPA and total glyphosate residues were elevated in 34 patients with nonalcoholic steatohepatitis (NASH) in comparison with 63 controls and more elevated in cases with more advanced fibrosis than those with less. These findings are also consistent with hepatotoxicity noted at much higher doses in glyphosate poisoning cases and with occupational exposures.30,31 In rodent studies, even low dosages of glyphosate or glyphosate-based herbicide formulations produced signs of NAFLD,24 as evidenced by fibrosis, steatosis, and necrosis of the liver.25 Glyphosate and glyphosate-based herbicide formulations have been found to alter the metabolome, proteome,24 transcriptome,26 epigenome,27 and DNA28 of the liver in rodent studies. Exposure to glyphosate, glyphosate-based herbicides, and AMPA has also induced epigenetic modifications in in vitro studies of human peripheral blood mononuclear cells.27
In a study of male rats, Prasad et al.33 found a dose-related increase in fasting blood glucose and serum insulin in glyphosate-exposed groups in comparison with controls. To our knowledge, the association of glyphosate with insulin resistance and other metabolic disorders has not been previously explored in human populations, although researchers have hypothesized that glyphosate has the potential to induce metabolic disease because of its ability to induce oxidative stress in preadipocytes and in other tissues.32,33,63–65 A second hypothesis for glyphosate’s etiologic role in metabolic disorders is its adverse effect on the gut microbiota, which have been shown in animal studies to be a source of oxidative stress.66 Recent investigations in rats have shown that glyphosate-containing herbicides inhibit the shikimate pathway in the gut microbiome.66 A third hypothesis for the association is through endocrine disruption,67 with evidence that glyphosate and glyphosate-containing herbicides can disrupt endocrine-signaling systems.33,68,69 Although the association of exposure to AMPA with metabolic disorders has been neither explored nor hypothesized, a recent in vitro study based on induced pluripotent stem cells (iPSCs) found changes in glucose metabolism following treatment to glyphosate or AMPA.70
Most of our prenatal urine samples, all collected around year 2000, had nondetectable levels of glyphosate and AMPA, consistent with the low use of glyphosate in agriculture around that time.8 With this exception, detection frequencies and concentrations in urine samples collected during childhood were within the range of those reported in other studies of children.71–74 Adolescents (12- to 19-y-olds) participating in the 2013–2014 National Health and Nutrition Examination Survey (NHANES) had a higher weighted detection frequency (87.2% >LOD) and wet weight geometric mean (GM=0.48μg/L) of urinary glyphosate concentrations than the 14-y-olds included in our study (who provided urine samples collected around the same time) (78.9% >LOD, GM=0.18μg/L, respectively)75; urinary AMPA concentrations were not measured in NHANES adolescents. However, our 14-y-old participants had higher urinary glyphosate and AMPA detection frequencies (78.9 and 93.3% >LOD, respectively) and wet weight geometric means (GM=0.18 and 0.45μg/L, respectively) than 14- to 17-y-old children participating in the 2015–2017 German Environmental Survey for Children and Adolescents (glyphosate: 46% >LOQ, GM <LOQ, respectively; AMPA: 42% >LOQ, GM <LOQ, respectively) (LOQ=0.1μg/L for both glyphosate and AMPA).72 It is likely that diet was a major source of glyphosate and AMPA exposure among our study participants at age 14 y, as indicated by higher urinary glyphosate or AMPA concentrations among those who ate more cereal, fruits, vegetables, bread, and in general, carbohydrates. We observed lower urinary concentrations of glyphosate and AMPA at age 18 y, even among the participants who had not fasted, than at age 14 y, despite increases in agricultural glyphosate use. It is possible that differences in diet may explain the lower concentrations at age 18 y; unfortunately, our 18-y dietary questionnaire was too limited to test this hypothesis. Similar to our findings at age 18 y, the NHANES data revealed lower urinary glyphosate concentrations in those who had fasted more than 8 h in comparison with those who fasted less, supporting the importance of dietary intake in glyphosate exposure.75
A strength of our study was that we could characterize agricultural glyphosate use near homes using California’s unique PUR database. However, our estimates do not reflect the full extent of ambient glyphosate exposure; they do not account for agricultural use near participants’ schools, workplaces, or nonagricultural uses (e.g., homes, roadways, parks).46 We also did not consider use of specific formulations of glyphosate-based pesticides, which may differ in their toxicity.76 Despite these limitations, we observed associations between PUR-assessed agricultural glyphosate exposure and metabolic syndrome that are interesting and merit additional research.77
An important limitation of our study is that a single measure of glyphosate or AMPA concentrations in the urine, and even multiple measures at different developmental periods, might not accurately reflect exposure, given the short half-life of glyphosate and AMPA in humans of between 3.5 and 14.5 h.78 This possibility, along with inaccuracy in dietary recall, could explain the modest associations of urinary AMPA and glyphosate residue concentrations with dietary factors that we observed. The short half-life of glyphosate in the human body78 as well as in the environment79 (unlike AMPA, which has been classified as persistent in soil80 and groundwater81) may have contributed to the weak correlations we observed between glyphosate use near residences and urinary glyphosate or AMPA concentrations. It can also explain the lower detection and urinary concentrations among those 18-y-olds who had fasted (vs. the nonfasters), given that food was likely an important route of exposure and that the maximum concentration of glyphosate and AMPA in urine is 1–3 h and 5–6 h after exposure, respectively.82 Given these short half-lives, the urinary concentrations measured concurrently with the outcomes at 18 y might not accurately reflect the exposure to glyphosate and AMPA preceding the onset of disease, which would be important criterion to establish a causal relationship.
Although we observe some associations of nearby agricultural use of glyphosate during the prenatal period and childhood with metabolic syndrome, urinary glyphosate concentrations were not associated with any health outcomes in the present study. Nevertheless, we observed associations of urinary AMPA and glyphosate residue concentrations (with the latter largely driven by AMPA) with elevated liver transaminases and/or metabolic syndrome. It is likely that urinary AMPA was derived from degradation of glyphosate in the environment. For example, microbial degradation of glyphosate in soil results in the accumulation of AMPA in soil, plants, and animal products.18 AMPA is highly soluble in water, more persistent in the environment than glyphosate,83 and therefore frequently detected at higher concentrations than glyphosate in most hydrological settings, with groundwater and soil water samples having the highest values.81 Tracer studies in Canada have shown that AMPA in groundwater is mainly derived from glyphosate degradation rather than wastewater sources, such as those contaminated with phosphonates.84 Although few studies have measured urinary AMPA concentrations in human populations, it is known that AMPA is poorly metabolized in the human body. For example, in a study of volunteers who ingested glyphosate,82 total dose recovered as unchanged glyphosate was low (1%–6%) but extremely low for AMPA—0.01%–0.04% of the total dose of glyphosate. This low excretion of AMPA in urine after glyphosate exposure has also been demonstrated in other studies.31,85 Thus, urinary AMPA likely results from direct exposure to AMPA from food residues and water, with a lesser extent from the metabolism of glyphosate in vivo. However, additional research is needed to identify the major pathways of AMPA exposure.
Our research suggests that lifetime exposure to glyphosate and AMPA may increase risk of liver and metabolic disease in early adulthood, which could lead to more serious diseases later in life, such as liver cancer,86 diabetes, and cardiovascular disease.87 Longitudinal lifelong studies in humans, such as CHAMACOS, are necessary to connect potential impact of glyphosate and AMPA on organ damage and other intermediate outcomes to chronic illness in adulthood. Future research should include frequent measurements of exposure biomarkers during fetal and child development to determine windows of susceptibility; examine the effects of glyphosate and AMPA in the context of exposure to pesticide mixtures88,89; and explore associations with other outcomes, such as reproductive and endocrine function.67,90 In addition, studies with sufficient sample size should examine differences in susceptibility by sex, as seen in animal studies.91,92
Conclusions
Metabolic and liver diseases are increasing among youth and young adults.93 Our study suggests that glyphosate, the most commonly used herbicide worldwide, and AMPA, a degradation product of glyphosate and amino-phosphonates, may increase risk of liver inflammation and/or cardiometabolic disease in young adulthood. Although previous research on glyphosate in humans has largely focused on its potential carcinogenicity, this study indicates the need for further investigation of its association with metabolic and liver outcomes.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank the Salinas research team for their dedicated work collecting these data and the UC Berkeley biorepository team for preserving and managing biological samples. The authors also thank the CHAMACOS families for their years of participation.
This work was funded by research grant numbers UH3 ES030631, R24 ES028529, R01 ES026994, P01 ES009605, R01 ES017054, and R01 ES021369 from the National Institute of Environmental Health Sciences (NIEHS), National Institutes of Health (NIH); R01 DA035300 from the National Institute on Drug Abuse (NIDA, NIH); and R82670901, RD83171001, and RD83451301 from the U.S. EPA. Additional support for this study was provided by The Solomon Dutka Fund in the New York Community Trust and The Westreich Foundation.
==== Refs
References
1. Sanyaolu A, Okorie C, Qi X, Locke J, Rehman S. 2019. Childhood and adolescent obesity in the United States: a public health concern. Glob Pediatr Health 6 :2333794X19891305, PMID: , 10.1177/2333794X19891305.31832491
2. Wittcopp C, Conroy R. 2016. Metabolic syndrome in children and adolescents. Pediatr Rev 37 (5 ):193–202, PMID: , 10.1542/pir.2014-0095.27139327
3. Jeans MR, Ghaddar R, Vandyousefi S, Landry MJ, Gray MJ, Leidy HJ, et al. 2022. Distinct racial and ethnic metabolic syndrome characteristics: a comparative assessment in low-income children 7–10 years of age. Pediatr Obes 17 (10 ):e12925, PMID: , 10.1111/ijpo.12925.35560860
4. Doycheva I, Watt KD, Alkhouri N. 2017. Nonalcoholic fatty liver disease in adolescents and young adults: the next frontier in the epidemic. Hepatology 65 (6 ):2100–2109, PMID: , 10.1002/hep.29068.28103626
5. Goyal NP, Schwimmer JB. 2016. The progression and natural history of pediatric nonalcoholic fatty liver disease. Clin Liver Dis 20 (2 ):325–338, PMID: , 10.1016/j.cld.2015.10.003.27063272
6. Haverinen E, Fernandez MF, Mustieles V, Tolonen H. 2021. Metabolic syndrome and endocrine disrupting chemicals: an overview of exposure and health effects. Int J Environ Res Public Health 18 (24 ):13047, PMID: , 10.3390/ijerph182413047.34948652
7. Liu Y, Peterson KE. 2015. Maternal exposure to synthetic chemicals and obesity in the offspring: recent findings. Curr Environ Health Rep 2 (4 ):339–347, PMID: , 10.1007/s40572-015-0068-6.26403844
8. Benbrook CM. 2016. Trends in glyphosate herbicide use in the United States and globally. Environ Sci Eur 28 (1 ):3, PMID: , 10.1186/s12302-016-0070-0.27752438
9. Guyton KZ, Loomis D, Grosse Y, El Ghissassi F, Benbrahim-Tallaa L, Guha N, et al. 2015. Carcinogenicity of tetrachlorvinphos, parathion, malathion, diazinon, and glyphosate. Lancet Oncol 16 (5 ):490–491, PMID: , 10.1016/S1470-2045(15)70134-8.25801782
10. Roy DN, Konar SK, Banerjee S, Charles DA, Thompson DG, Prasad R. 1989. Uptake and persistence of the herbicide glyphosate (Vision®) in fruit of wild blueberry and red raspberry. Can J for Res 19 (7 ):842–847, 10.1139/x89-128.
11. Wang YS, Jaw CG, Chen YL. 1994. Accumulation of 2,4-D and glyphosate in fish and water hyacinth. Water Air Soil Pollut 74 (3–4 ):397–403, 10.1007/BF00479802.
12. Bento CPM, Goossens D, Rezaei M, Riksen M, Mol HGJ, Ritsema CJ, et al. 2017. Glyphosate and AMPA distribution in wind-eroded sediment derived from loess soil. Environ Pollut 220 (pt B ):1079–1089, PMID: , 10.1016/j.envpol.2016.11.033.27876225
13. Alferness PL, Iwata Y. 1994. Determination of glyphosate and (aminomethyl)phosphonic acid in soil, plant and animal matrixes, and water by capillary gas chromatography with mass-selective detection. J Agric Food Chem 42 (12 ):2751–2759, 10.1021/jf00048a020.
14. Granby K, Johannesen S, Vahl M. 2003. Analysis of glyphosate residues in cereals using liquid chromatography-mass spectrometry (LC-MS/MS). Food Addit Contam 20 (8 ):692–698, PMID: , 10.1080/0265203031000109477.13129785
15. Çetin E, Şahan S, Ülgen A, Şahin U. 2017. DLLME-spectrophotometric determination of glyphosate residue in legumes. Food Chem 230 :567–571, PMID: , 10.1016/j.foodchem.2017.03.063.28407950
16. Chen MX, Cao ZY, Jiang Y, Zhu ZW. 2013. Direct determination of glyphosate and its major metabolite, aminomethylphosphonic acid, in fruits and vegetables by mixed-mode hydrophilic interaction/weak anion-exchange liquid chromatography coupled with electrospray tandem mass spectrometry. J Chromatogr A 1272 :90–99, PMID: , 10.1016/j.chroma.2012.11.069.23261284
17. Rodrigues NR, Ferreira de Souza AP. 2018. Occurrence of glyphosate and AMPA residues in soy-based infant formula sold in Brazil. Food Addit Contam Part A Chem Anal Control Expo Risk Assess 35 (4 ):723–730, PMID: , 10.1080/19440049.2017.1419286.29267136
18. Grandcoin A, Piel S, Baurès E. 2017. AminoMethylPhosphonic acid (AMPA) in natural waters: its sources, behavior and environmental fate. Water Res 117 :187–197, PMID: , 10.1016/j.watres.2017.03.055.28391123
19. Benbrook CM. 2019. How did the US EPA and IARC reach diametrically opposed conclusions on the genotoxicity of glyphosate-based herbicides? Environ Sci Eur 31 (1 ):2, 10.1186/s12302-018-0184-7.
20. U.S. EPA (U.S. Environmental Protection Agency). 2020. Glyphosate, Interim Registration Review Decision Case Number 0178. https://www.epa.gov/sites/default/files/2020-01/documents/glyphosate-interim-reg-review-decision-case-num-0178.pdf [accessed 8 June 2022].
21. IARC (International Agency for Research on Cancer). 2015. IARC Monographs Volume 112: evaluation of five organophosphate insecticides and herbicides. https://www.iarc.who.int/wp-content/uploads/2018/07/MonographVolume112-1.pdf [accessed 8 June 2022].
22. Van Bruggen AHC, He MM, Shin K, Mai V, Jeong KC, Finckh MR, et al. 2018. Environmental and health effects of the herbicide glyphosate. Sci Total Environ 616–617 :255–268, PMID: , 10.1016/j.scitotenv.2017.10.309.29117584
23. FAO (Food and Agriculture Organization of the United Nations), WHO (World Health Organization). 1998. Pesticide Residues in Food 1997 - Joint FAO/WHO Meeting on Pesticide Residues Part I - Residues. https://www.fao.org/publications/card/en/c/CB2747EN/ [accessed 24 August 2022].
24. Mesnage R, Renney G, Séralini GE, Ward M, Antoniou MN. 2017. Multiomics reveal non-alcoholic fatty liver disease in rats following chronic exposure to an ultra-low dose of roundup herbicide. Sci Rep 7 :39328, PMID: , 10.1038/srep39328.28067231
25. Mesnage R, Arno M, Costanzo M, Malatesta M, Séralini GE, Antoniou MN. 2015. Transcriptome profile analysis reflects rat liver and kidney damage following chronic ultra-low dose Roundup exposure. Environ Health 14 :70, PMID: , 10.1186/s12940-015-0056-1.26302742
26. Mesnage R, Teixeira M, Mandrioli D, Falcioni L, Ibragim M, Ducarmon QR, et al. 2021. Multi-omics phenotyping of the gut–liver axis reveals metabolic perturbations from a low-dose pesticide mixture in rats. Commun Biol 4 (1 ):471, PMID: , 10.1038/s42003-021-01990-w.33854195
27. Rossetti MF, Canesini G, Lorenz V, Milesi MM, Varayoud J, Ramos JG. 2021. Epigenetic changes associated with exposure to glyphosate-based herbicides in mammals. Front Endocrinol (Lausanne) 12 :671991, PMID: , 10.3389/fendo.2021.671991.34093442
28. Mesnage R, Ibragim M, Mandrioli D, Falcioni L, Tibaldi E, Belpoggi F, et al. 2022. Comparative toxicogenomics of glyphosate and roundup herbicides by mammalian stem cell-based genotoxicity assays and molecular profiling in Sprague-Dawley rats. Toxicol Sci 186 (1 ):83–101, PMID: , 10.1093/toxsci/kfab143.34850229
29. Mills PJ, Caussy C, Loomba R. 2020. Glyphosate excretion is associated with steatohepatitis and advanced liver fibrosis in patients with fatty liver disease. Clin Gastroenterol Hepatol 18 (3 ):741–743, PMID: , 10.1016/j.cgh.2019.03.045.30954713
30. Kamijo Y, Takai M, Sakamoto T. 2016. A multicenter retrospective survey of poisoning after ingestion of herbicides containing glyphosate potassium salt or other glyphosate salts in Japan. Clin Toxicol (Phila) 54 (2 ):147–151, PMID: , 10.3109/15563650.2015.1121271.26691886
31. Zouaoui K, Dulaurent S, Gaulier JM, Moesch C, Lachâtre G. 2013. Determination of glyphosate and AMPA in blood and urine from humans: about 13 cases of acute intoxication. Forensic Sci Int 226 (1–3 ):e20–e25, PMID: , 10.1016/j.forsciint.2012.12.010.23291146
32. De Long NE, Holloway AC. 2017. Early-life chemical exposures and risk of metabolic syndrome. Diabetes Metab Syndr Obes 10 :101–109, PMID: , 10.2147/DMSO.S95296.28367067
33. Prasad M, Gatasheh MK, Alshuniaber MA, Krishnamoorthy R, Rajagopal P, Krishnamoorthy K, et al. 2022. Impact of glyphosate on the development of insulin resistance in experimental diabetic rats: role of NFκB signalling pathways. Antioxidants (Basel) 11 (12 ):2436, PMID: , 10.3390/antiox11122436.36552644
34. Eskenazi B, Kogut K, Huen K, Harley KG, Bouchard M, Bradman A, et al. 2014. Organophosphate pesticide exposure, PON1, and neurodevelopment in school-age children from the CHAMACOS study. Environ Res 134 :149–157, PMID: , 10.1016/j.envres.2014.07.001.25171140
35. Block G, Woods M, Potosky A, Clifford C. 1990. Validation of a self-administered diet history questionnaire using multiple diet records. J Clin Epidemiol 43 (12 ):1327–1335, PMID: , 10.1016/0895-4356(90)90099-b.2254769
36. Block G, Thompson FE, Hartman AM, Larkin FA, Guire KE. 1992. Comparison of two dietary questionnaires validated against multiple dietary records collected during a 1-year period. J Am Diet Assoc 92 (6 ):686–693, PMID: , 10.1016/S0002-8223(21)00707-0.1607564
37. Harley K, Eskenazi B, Block G. 2005. The association of time in the US and diet during pregnancy in low-income women of Mexican descent. Paediatr Perinat Epidemiol 19 (2 ):125–134, PMID: , 10.1111/j.1365-3016.2005.00640.x.15787887
38. Rosas LG, Harley K, Fernald LCH, Guendelman S, Mejia F, Neufeld LM, et al. 2009. Dietary associations of household food insecurity among children of Mexican descent: results of a binational study. J Am Diet Assoc 109 (12 ):2001–2009, PMID: , 10.1016/j.jada.2009.09.004.19942017
39. Stein AD, Shea S, Basch CE, Contento IR, Zybert P. 1992. Consistency of the Willett semiquantitative food frequency questionnaire and 24-hour dietary recalls in estimating nutrient intakes of preschool children. Am J Epidemiol 135 (6 ):667–677, PMID: , 10.1093/oxfordjournals.aje.a116346.1580243
40. Hunsberger M, O’Malley J, Block T, Norris JC. 2015. Relative validation of Block Kids Food Screener for dietary assessment in children and adolescents. Matern Child Nutr 11 (2 ):260–270, PMID: , 10.1111/j.1740-8709.2012.00446.x.23006452
41. CDC (U.S. Centers for Disease Control and Prevention). 2022. Behavioral Risk Factor Surveillance System (BRFSS). https://www.cdc.gov/brfss/index.html [accessed 7 June 2022].
42. UCLA Center for Health Policy Research. 2009. Dietary Screener in the 2009 California Health Interview Survey. https://healthpolicy.ucla.edu/chis/data/public-use-data-file/Documents/CHIS2009_Diet_Screener_Intro.pdf [accessed 8 June 2022].
43. Harvard T. H. Chan School of Public Health. 2022. General Documentation. Harvard Nutrition Questionnaire Service Center. https://www.hsph.harvard.edu/nutrition-questionnaire-service-center/general-documentation/ [accessed 24 August 2022].
44. USDA Economic Research Service. 2012. U.S. Household Food Security Survey Module: Six-Item Short Form Economic Research Service, United States Department of Agriculture (USDA). https://www.ers.usda.gov/media/8282/short2012.pdf [accessed 8 June 2022].
45. Bienvenu JF, Bélanger P, Gaudreau É, Provencher G, Fleury N. 2021. Determination of glyphosate, glufosinate and their major metabolites in urine by the UPLC-MS/MS method applicable to biomonitoring and epidemiological studies. Anal Bioanal Chem 413 (8 ):2225–2234, PMID: , 10.1007/s00216-021-03194-x.33547480
46. CDPR (California Department of Pesticide Regulation). 2022. Pesticide Use Reporting (PUR). https://www.cdpr.ca.gov/docs/pur/purmain.htm [accessed 25 May 2022].
47. Gunier RB, Ward MH, Airola M, Bell EM, Colt J, Nishioka M, et al. 2011. Determinants of agricultural pesticide concentrations in carpet dust. Environ Health Perspect 119 (7 ):970–976, PMID: , 10.1289/ehp.1002532.21330232
48. Gunier RB, Jerrett M, Smith DR, Jursa T, Yousefi P, Camacho J, et al. 2014. Determinants of manganese levels in house dust samples from the CHAMACOS cohort. Sci Total Environ 497–498 :360–368, PMID: , 10.1016/j.scitotenv.2014.08.005.25146905
49. Esri (Environmental Systems Research Institute). 2020. World Ocean Base. https://www.arcgis.com/home/item.html?id=1e126e7520f9466c9ca28b8f28b5e500%2F [accessed 4 January 2023].
50. CIMIS (California Irrigation Management Information System). 2014. California Department of Water Resources. https://cimis.water.ca.gov/Default.aspx [accessed 25 May 2022].
51. FAO and WHO. 2019. Pesticide residues in food 2019 - Report 2019 - Extra Joint FAO/WHO Meeting on Pesticide Residues. Rome. https://www.fao.org/3/ca5711en/ca5711en.pdf [accessed 24 August 2022].
52. FAO and WHO. 2005. Pesticide residues in food 2005 - Report 2005 - Joint FAO/WHO Meeting on Pesticide Residues. Series number 183. Rome. https://www.fao.org/3/a0209e/A0209E.pdf [accessed 24 August 2022].
53. van Apeldoorn E, van Hoeven P. 1997. Toxicological and Environmental Evaluations 1994. International Peer Reviewed Chemical Safety Information. https://inchem.org/documents/jmpr/jmpmono/v097pr04.htm [accessed 24 August 2022].
54. Lubin JH, Colt JS, Camann D, Davis S, Cerhan JR, Severson RK, et al. 2004. Epidemiologic evaluation of measurement data in the presence of detection limits. Environ Health Perspect 112 (17 ):1691–1696, PMID: , 10.1289/ehp.7199.15579415
55. Reilly M, Torrång A, Klint A. 2005. Re-use of case-control data for analysis of new outcome variables. Stat Med 24 (24 ):4009–4019, PMID: , 10.1002/sim.2398.16320270
56. Richardson DB, Rzehak P, Klenk J, Weiland SK. 2007. Analyses of case-control data for additional outcomes. Epidemiology 18 (4 ):441–445, PMID: , 10.1097/EDE.0b013e318060d25c.17473707
57. CDC. 2022. About Adult BMI. https://www.cdc.gov/healthyweight/assessing/bmi/adult_bmi/index.html [accessed 8 June 2022].
58. NHLBI (National Heart, Lung, and Blood Institute). 2022. Metabolic Syndrome Diagnosis. https://www.nhlbi.nih.gov/health/metabolic-syndrome/diagnosis [accessed 30 May 2022].
59. Alberti KGMM, Eckel RH, Grundy SM, Zimmet PZ, Cleeman JI, Donato KA, et al. 2009. Harmonizing the metabolic syndrome: a joint interim statement of the International Diabetes Federation Task Force on Epidemiology and Prevention; National Heart, Lung, and Blood Institute; American Heart Association; World Heart Federation; International Atherosclerosis Society; and International Association for the Study of Obesity. Circulation 120 (16 ):1640–1645, PMID: , 10.1161/CIRCULATIONAHA.109.192644.19805654
60. Sánchez BN, Hu H, Litman HJ, Téllez-Rojo MM. 2011. Statistical methods to study timing of vulnerability with sparsely sampled data on environmental toxicants. Environ Health Perspect 119 (3 ):409–415, PMID: , 10.1289/ehp.1002453.21362588
61. Baron RM, Kenny DA. 1986. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Pers Soc Psychol 51 (6 ):1173–1182, PMID: , 10.1037//0022-3514.51.6.1173.3806354
62. U.S. Census Bureau. n.d. Poverty Thresholds: Poverty Thresholds by Size of Family and Number of Children; 2000 & 2018. https://www.census.gov/data/tables/time-series/demo/income-poverty/historical-poverty-thresholds.html [accessed 25 May 2022].
63. Roberts CK, Sindhu KK. 2009. Oxidative stress and metabolic syndrome. Life Sci 84 (21–22 ):705–712, PMID: , 10.1016/j.lfs.2009.02.026.19281826
64. Martini CN, Gabrielli M, Vila MDC. 2012. A commercial formulation of glyphosate inhibits proliferation and differentiation to adipocytes and induces apoptosis in 3T3-L1 fibroblasts. Toxicol In Vitro 26 (6 ):1007–1013, PMID: , 10.1016/j.tiv.2012.04.017.22546541
65. Martini CN, Gabrielli M, Brandani JN, Vila MDC. 2016. Glyphosate inhibits PPAR gamma induction and differentiation of preadipocytes and is able to induce oxidative stress. J Biochem Mol Toxicol 30 (8 ):404–413, PMID: , 10.1002/jbt.21804.27044015
66. Mesnage R, Teixeira M, Mandrioli D, Falcioni L, Ducarmon QR, Zwittink RD, et al. 2021. Use of shotgun metagenomics and metabolomics to evaluate the impact of glyphosate or roundup MON 52276 on the gut microbiota and serum metabolome of Sprague-Dawley rats. Environ Health Perspect 129 (1 ):17005, PMID: , 10.1289/EHP6990.33502259
67. Myers JP, Antoniou MN, Blumberg B, Carroll L, Colborn T, Everett LG, et al. 2016. Concerns over use of glyphosate-based herbicides and risks associated with exposures: a consensus statement. Environ Health 15 :19, PMID: , 10.1186/s12940-016-0117-0.26883814
68. Guerrero Schimpf M, Milesi MM, Zanardi MV, Varayoud J. 2022. Disruption of developmental programming with long-term consequences after exposure to a glyphosate-based herbicide in a rat model. Food Chem Toxicol 159 :112695, PMID: , 10.1016/j.fct.2021.112695.34813928
69. Romano MA, Romano RM, Santos LD, Wisniewski P, Campos DA, de Souza PB, et al. 2012. Glyphosate impairs male offspring reproductive development by disrupting gonadotropin expression. Arch Toxicol 86 (4 ):663–673, PMID: , 10.1007/s00204-011-0788-9.22120950
70. Martinez A, Al-Ahmad AJ. 2019. Effects of glyphosate and aminomethylphosphonic acid on an isogeneic model of the human blood-brain barrier. Toxicol Lett 304 :39–49, PMID: , 10.1016/j.toxlet.2018.12.013.30605748
71. Dereumeaux C, Mercier F, Soulard P, Hulin M, Oleko A, Pecheux M, et al. 2022. Identification of pesticides exposure biomarkers for residents living close to vineyards in France. Environ Int 159 :107013, PMID: , 10.1016/j.envint.2021.107013.34890902
72. Lemke N, Murawski A, Schmied-Tobies MIH, Rucic E, Hoppe H-W, Conrad A, et al. 2021. Glyphosate and aminomethylphosphonic acid (AMPA) in urine of children and adolescents in Germany – human biomonitoring results of the German Environmental Survey 2014–2017 (GerES V). Environ Int 156 :106769, PMID: , 10.1016/j.envint.2021.106769.34274860
73. Stajnko A, Snoj Tratnik J, Kosjek T, Mazej D, Jagodic M, Eržen I, et al. 2020. Seasonal glyphosate and AMPA levels in urine of children and adolescents living in rural regions of northeastern Slovenia. Environ Int 143 :105985, PMID: , 10.1016/j.envint.2020.105985.32731096
74. Sierra-Diaz E, Celis-de la Rosa A, Lozano-Kasten F, Trasande L, Peregrina-Lucano A, Sandoval-Pinto E, et al. 2019. Urinary pesticide levels in children and adolescents residing in two agricultural communities in Mexico. Int J Environ Res Public Health 16 (4 ):562, PMID: , 10.3390/ijerph16040562.30781414
75. Ospina M, Schütze A, Morales-Agudelo P, Vidal M, Wong LY, Calafat AM. 2022. Exposure to glyphosate in the United States: data from the 2013–2014 National Health and Nutrition Examination Survey. Environ Int 170 :107620, PMID: , 10.1016/j.envint.2022.107620.36368224
76. Mesnage R, Defarge N, Spiroux de Vendômois J, Séralini GE. 2015. Potential toxic effects of glyphosate and its commercial formulations below regulatory limits. Food Chem Toxicol 84 :133–153, PMID: , 10.1016/j.fct.2015.08.012.26282372
77. Rani V, Deep G, Singh RK, Palle K, Yadav UCS. 2016. Oxidative stress and metabolic disorders: pathogenesis and therapeutic strategies. Life Sci 148 :183–193, PMID: , 10.1016/j.lfs.2016.02.002.26851532
78. Connolly A, Jones K, Basinas I, Galea KS, Kenny L, McGowan P, et al. 2019. Exploring the half-life of glyphosate in human urine samples. Int J Hyg Environ Health 222 (2 ):205–210, PMID: , 10.1016/j.ijheh.2018.09.004.30293930
79. Jenkins M, Locke M, Reddy K, McChesney DS, Steinriede R. 2018. Glyphosate applications, glyphosate resistant corn, and tillage on nitrification rates and distribution of nitrifying microbial communities. Soil Sci Soc Am J 81 :1371–1380, 10.2136/sssaj2017.02.0063.
80. Bento CPM, Yang X, Gort G, Xue S, van Dam R, Zomer P, et al. 2016. Persistence of glyphosate and aminomethylphosphonic acid in loess soil under different combinations of temperature, soil moisture and light/darkness. Sci Tot Environ 572 :301–311, PMID: , 10.1016/j.scitotenv.2016.07.215.27505263
81. Battaglin WA, Meyer MT, Kuivila KM, Dietze JE. 2014. Glyphosate and its degradation product AMPA occur frequently and widely in U.S. soils, surface water, groundwater, and precipitation. J Am Water Resour Assoc 50 (2 ):275–290, 10.1111/jawr.12159.
82. Faniband MH, Norén E, Littorin M, Lindh CH. 2021. Human experimental exposure to glyphosate and biomonitoring of young Swedish adults. Int J Hyg Environ Health 231 :113657, PMID: , 10.1016/j.ijheh.2020.113657.33130428
83. Carretta L, Masin R, Zanin G. 2022. Review of studies analysing glyphosate and aminomethylphosphonic acid (AMPA) occurrence in groundwater. Environ Rev 30 (1 ):88–109, 10.1139/er-2020-0106.
84. Struger J, Van Stempvoort DR, Brown SJ. 2015. Sources of aminomethylphosphonic acid (AMPA) in urban and rural catchments in Ontario, Canada: glyphosate or phosphonates in wastewater? Environ Pollut 204 :289–297, PMID: , 10.1016/j.envpol.2015.03.038.26187493
85. Zoller O, Rhyn P, Zarn JA, Dudler V. 2020. Urine glyphosate level as a quantitative biomarker of oral exposure. Int J Hyg Environ Health 228 :113526, PMID: , 10.1016/j.ijheh.2020.113526.32305862
86. Massoud O, Charlton M. 2018. Nonalcoholic fatty liver disease/nonalcoholic steatohepatitis and hepatocellular carcinoma. Clin Liver Dis 22 (1 ):201–211, PMID: , 10.1016/j.cld.2017.08.014.29128057
87. Kahn R, Buse J, Ferrannini E, Stern M, American Diabetes Association. 2005. The metabolic syndrome: time for a critical appraisal: joint statement from the American Diabetes Association and the European Association for the Study of Diabetes. Diabetes Care 28 (9 ):2289–2304, PMID: , 10.2337/diacare.28.9.2289.16123508
88. Bonvallot N, Canlet C, Blas-Y-Estrada F, Gautier R, Tremblay-Franco M, Chevolleau S, et al. 2018. Metabolome disruption of pregnant rats and their offspring resulting from repeated exposure to a pesticide mixture representative of environmental contamination in Brittany. PLoS One 13 (6 ):e0198448, PMID: , 10.1371/journal.pone.0198448.29924815
89. Mesnage R, Mahmud N, Mein CA, Antoniou MN. 2021. Alterations in small RNA profiles in liver following a subchronic exposure to a low-dose pesticide mixture in Sprague-Dawley rats. Toxicol Lett 353 :20–26, PMID: , 10.1016/j.toxlet.2021.10.001.34626815
90. Manservisi F, Lesseur C, Panzacchi S, Mandrioli D, Falcioni L, Bua L, et al. 2019. The Ramazzini Institute 13-week pilot study glyphosate-based herbicides administered at human-equivalent dose to Sprague Dawley rats: effects on development and endocrine system. Environ Health 18 (1 ):15, PMID: , 10.1186/s12940-019-0453-y.30857531
91. Giommi C, Ladisa C, Carnevali O, Maradonna F, Habibi HR. 2022. Metabolomic and transcript analysis revealed a sex-specific effect of glyphosate in zebrafish liver. IJMS 23 (5 ):2724, PMID: , 10.3390/ijms23052724.35269866
92. Lozano VL, Defarge N, Rocque L-M, Mesnage R, Hennequin D, Cassier R, et al. 2018. Sex-dependent impact of roundup on the rat gut microbiome. Toxicol Rep 5 :96–107, PMID: , 10.1016/j.toxrep.2017.12.005.29854581
93. Haffner SM. 2006. The metabolic syndrome: inflammation, diabetes mellitus, and cardiovascular disease. Am J Cardiol 97 (2A ):3A–11A, PMID: , 10.1016/j.amjcard.2005.11.010.16675316
|
PMC009xxxxxx/PMC9980311.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36862173
EHP12134
10.1289/EHP12134
Invited Perspective
Invited Perspective: PFAS in Breast Milk and Infant Formula—It’s Time to Start Monitoring
https://orcid.org/0000-0002-0323-9178
LaKind Judy S. 1 2
1 LaKind Associates, LLC, Catonsville, Maryland, USA
2 Department of Epidemiology and Public Health, University of Maryland School of Medicine, Baltimore, Maryland, USA
Address correspondence to Judy S. LaKind, LaKind Associates, LLC, 106 Oakdale Ave., Catonsville, MD 21228 USA. Telephone: (410) 788-8639. Email: [email protected]
2 3 2023
3 2023
131 3 03130111 9 2022
02 11 2022
18 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The author declares she has no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11403
==== Body
pmcThe study by Yao et al.1 in this issue of Environmental Health Perspectives advances our understanding of infant exposures to per- and polyfluoroalkyl substances (PFAS). They note that “it is essential to provide a national baseline of PFAS exposure in human milk … and to assess the risks of emerging and legacy PFAS exposure in exclusively breastfed infants.” It is impossible to argue with that point, and in fact for years my colleagues and I have made similar calls for national breast milk monitoring programs.2 Although several countries have developed national programs to assess human exposures to environmental chemicals using blood and urine (among them the U.S. National Health and Nutrition Examination Survey, the Canadian Health Measures Survey, the European Human Biomonitoring Initiative, the China National Human Biomonitoring study, and the Korean National Environmental Health Survey), similar efforts around breast milk are lacking. Why is this and what does it mean for infants?
Why Are There No National Breast Milk Monitoring Programs?
Some breast milk monitoring programs have been undertaken to assess levels of persistent lipophilic chemicals (e.g., dioxins and furans).3–6 These programs served important purposes such as assessing temporal and geographic variability and informing individual women about their breast milk chemical levels. However, most women did not—and do not—have access to programs such as these.
One obvious reason for not undertaking large-scale breast milk monitoring programs would be the cost and complexity associated with setting up and maintaining such an effort. But there are likely additional reasons. For example, the robust scientific evidence regarding breastfeeding-related benefits to both the mother and infant7,8 makes it difficult to have a conversation about potential risks associated with chemicals in breast milk. The need for thoughtful communications regarding environmental chemicals in breast milk has been described.9 In fact, early in my research in this area I was admonished by lactation consultants and others working to improve breastfeeding rates for unintentionally using language that could scare women away from breastfeeding. I took this feedback seriously and modified how I communicate with mothers, physicians, risk and exposure scientists, and others.
Another factor that may be at play is our general inability to place the measured levels of chemicals into a risk-based context. Breastfeeding presents a unique exposure–risk situation in that exposures are of relatively short duration but occur during an exquisitely sensitive life stage. Available risk-based approaches do not generally offer guidance values specific to this circumstance. I would argue that this combination of factors has limited the will to create either national breast milk monitoring programs or on-request analyses of breast milk samples, at least in the United States.
What Does This Mean for Breastfeeding Infants?
Yao et al.1 state that “further epidemiological studies are needed to demonstrate whether breastfeeding with contaminants has adverse health outcomes on newborns.” It is hard to disagree with the sentiment. For many legacy chemicals, we have several decades’ worth of epidemiologic studies on this subject.10–13
However, in the case of PFAS and infant nutrition, the approach of waiting for robust epidemiological study results is unsatisfactory. Legacy PFAS have been detected in breast milk and drinking water at levels exceeding various guidance values1,14,15; furthermore, emerging PFAS are commonly detected in breast milk.1 The data are too sparse to make general statements about PFAS in infant formula. However, because some infant formulas require reconstitution with water, PFAS intakes above levels of concern cannot be ruled out. Thus, our current risk-based approaches already suggest that action—not waiting—is needed.
Monitoring and Guidance Values Are Needed Now
It has been noted time and time again that infants are our most vulnerable citizens. How can we in good conscience support waiting the many years it will take to complete epidemiological studies while being aware that risk-based approaches indicate that monitoring is needed for today’s generation of infants and children?
A PFAS monitoring program for infant nutrition is long overdue. In addition, we need health-based guidance values for infant PFAS exposures for the myriad of legacy and emerging PFAS—not just the four drinking water screening values that the U.S. Agency for Toxic Substances and Disease Registry has developed. At the very least, people should be aware of the levels of PFAS their children may be exposed to via breastfeeding, tap water, and infant formula to begin to make informed decisions.
We have known about lactational transfer of PFAS for over two decades and about the presence of PFAS in drinking water for just as long. How is it that we still cannot answer the most basic questions regarding PFAS and infant exposure and health? Although scientists—myself included—are trained to be circumspect and to wait for sufficient scientific information to give us confidence in the accumulated evidence before calling for action, we have waited long enough.
==== Refs
References
1. Yao J, Dong Z, Jiang L, Pan Y, Zhao M, Bai X, et al. 2023. Emerging and legacy perfluoroalkyl substances in breastfed Chinese infants: renal clearance, body burden, and implications. Environ Health Perspect 131 (3 ):037003, 10.1289/EHP11403.36862174
2. LaKind JS, Berlin CM, Naiman DQ. 2001. Infant exposure to chemicals in breast milk in the United States: what we need to learn from a breast milk monitoring program. Environ Health Perspect 109 (1 ):75–88, PMID: , 10.1289/ehp.0110975.11171529
3. Fürst P, Fürst C, Wilmers K. 1994. Human milk as a bioindicator for body burden of PCDDs, PCDFs, organochlorine pesticides, and PCBs. Environ Health Perspect 102 (suppl 1 ):187–193, PMID: , 10.1289/ehp.102-1566908.8187707
4. Harden F, Müller J, Toms L. 2004. National Dioxins Program. Technical Report No. 10: dioxins in the Australian Population: Levels in Human Milk. https://www.agriculture.gov.au/sites/default/files/documents/report-10a.pdf [accessed 8 September 2022].
5. Ryan JJ, Rawn DFK. 2014. Polychlorinated dioxins, furans (PCDD/Fs), and polychlorinated biphenyls (PCBs) and their trends in Canadian human milk from 1992 to 2005. Chemosphere 102 :76–86, PMID: , 10.1016/j.chemosphere.2013.12.065.24457050
6. van den Berg M, Kypke K, Kotz A, Tritscher A, Lee SY, Magulova K, et al. 2017. WHO/UNEP global surveys of PCDDs, PCDFs, PCBs and DDTs in human milk and benefit–risk evaluation of breastfeeding. Arch Toxicol 91 (1 ):83–96, PMID: , 10.1007/s00204-016-1802-z.27438348
7. Meek JY, Noble L, Section on Breastfeeding. 2022. Policy statement: breastfeeding and the use of human milk. Pediatrics 150 (1 ):e2022057988, PMID: , 10.1542/peds.2022-057988.35921640
8. WHO (World Health Organization). 2011. Exclusive breastfeeding for six months best for babies everywhere. Statement 15 January 2011. http://www.who.int/mediacentre/news/statements/2011/breastfeeding_20110115/en/ [accessed 14 April 2021].
9. LaKind JS, Brent RL, Dourson ML, Kacew S, Koren G, Sonawane B, et al. 2005. Human milk biomonitoring data: interpretation and risk assessment issues. J Toxicol Environ Health A 68 (20 ):1713–1769, PMID: , 10.1080/15287390500225724.16176917
10. Bonde JP, Flachs EM, Rimborg S, Glazer CH, Giwercman A, Ramlau-Hansen CH, et al. 2016. The epidemiologic evidence linking prenatal and postnatal exposure to endocrine disrupting chemicals with male reproductive disorders: a systematic review and meta-analysis. Hum Reprod Update 23 (1 ):104–125, PMID: , 10.1093/humupd/dmw036.27655588
11. Goodman M, Li J, Flanders WD, Mahood D, Anthony LG, Zhang Q, et al. 2020. Epidemiology of PCBs and neurodevelopment: systematic assessment of multiplicity and reporting. Glob Epidemiol 2 :100040, 10.1016/j.gloepi.2020.100040.
12. LaKind JS, Lehmann GM, Davis MH, Hines EP, Marchitti SA, Alcala C, et al. 2018. Infant dietary exposures to environmental chemicals and infant/child health: a critical assessment of the literature. Environ Health Perspect 126 (9 ):96002, PMID: , 10.1289/EHP1954.30256157
13. Sharma BM, Sáňka O, Kalina J, Scheringer M. 2019. An overview of worldwide and regional time trends in total mercury levels in human blood and breast milk from 1966 to 2015 and their associations with health effects. Environ Int 125 :300–319, PMID: , 10.1016/j.envint.2018.12.016.30735961
14. Andrews DQ, Naidenko OV. 2020. Population-wide exposure to per-and polyfluoroalkyl substances from drinking water in the United States. Environ Sci Technol Lett 7 (12 ):931–936, 10.1021/acs.estlett.0c00713.
15. LaKind JS, Verner MA, Rogers RD, Goeden H, Naiman DQ, Marchitti SA, et al. 2022. Current breast milk PFAS levels in the United States and Canada: after all this time why don’t we know more? Environ Health Perspect 130 (2 ):25002, PMID: , 10.1289/EHP10359.35195447
|
PMC009xxxxxx/PMC9980344.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36862174
EHP11403
10.1289/EHP11403
Research
Emerging and Legacy Perfluoroalkyl Substances in Breastfed Chinese Infants: Renal Clearance, Body Burden, and Implications
https://orcid.org/0000-0001-9162-5425
Yao Jingzhi 1 2
Dong Zhaomin 3
Jiang Lulin 1
Pan Yitao 1 4
Zhao Meirong 5
Bai Xiaoxia 6
https://orcid.org/0000-0003-4908-5597
Dai Jiayin 1 4
1 State Environmental Protection Key Laboratory of Environmental Health Impact Assessment of Emerging Contaminants, School of Environmental Science and Engineering, Shanghai Jiao Tong University, Shanghai, China
2 Key Laboratory of Animal Ecology and Conservation Biology, Institute of Zoology, Chinese Academy of Sciences, Beijing, China
3 School of Space and Environment and Beijing Advanced Innovation Center for Big Data-Based Precision Medicine, Beihang University, Beijing, China
4 State Key Laboratory of Reproductive Medicine, Center for Global Health, School of Public Health, Nanjing Medical University, Nanjing, China
5 College of Environment, Zhejiang University of Technology, Hangzhou, Zhejiang, China
6 Department of Obstetrics, Women’s Hospital, School of Medicine, Zhejiang University, Hangzhou, Zhejiang, China
Address correspondence to Jiayin Dai, State Environmental Protection Key Laboratory of Environmental Health Impact Assessment of Emerging Contaminants, School of Environmental Science and Engineering, Shanghai Jiao Tong University, 800 Dongchuan Rd., Shanghai 200240, China. Telephone: 86-21-54741065. Email: [email protected]. And, Xiaoxia Bai, Women’s Hospital, School of Medicine, Zhejiang University, Hangzhou, Zhejiang 310006, China. Telephone: 86-21-54741065. Email: [email protected]
2 3 2023
3 2023
131 3 03700312 4 2022
12 12 2022
18 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Human breast milk is a primary route of exposure to perfluoroalkyl substances (PFAS) in infants. To understand the associated risks, the occurrence of PFAS in human milk and the toxicokinetics of PFAS in infants need to be addressed.
Objectives:
We determined levels of emerging and legacy PFAS in human milk and urine samples from Chinese breastfed infants, estimated renal clearance, and predicted infant serum PFAS levels.
Methods:
In total, human milk samples were collected from 1,151 lactating mothers in 21 cities in China. In addition, 80 paired infant cord blood and urine samples were obtained from two cities. Nine emerging PFAS and 13 legacy PFAS were analyzed in the samples using ultra high-performance liquid chromatography tandem mass spectrometry. Renal clearance rates (CLrenals) of PFAS were estimated in the paired samples. PFAS serum concentrations in infants (<1 year of age) were predicted using a first-order pharmacokinetic model.
Results:
All nine emerging PFAS were detected in human milk, with the detection rates of 6:2 Cl-PFESA, PFMOAA, and PFO5DoDA all exceeding 70%. The level of 6:2 Cl-PFESA in human milk (median concentration=13.6 ng/L) ranked third after PFOA (336 ng/L) and PFOS (49.7 ng/L). The estimated daily intake (EDI) values of PFOA and PFOS exceeded the reference dose (RfD) of 20 ng/kg BW per day recommended by the U.S. Environmental Protection Agency in 78% and 17% of breastfed infant samples, respectively. 6:2 Cl-PFESA had the lowest infant CLrenal (0.009mL/kg BW per day), corresponding to the longest estimated half-life of 49 y. The average half-lives of PFMOAA, PFO2HxA, and PFO3OA were 0.221, 0.075, and 0.304 y, respectively. The CLrenals of PFOA, PFNA, and PFDA were slower in infants than in adults.
Conclusions:
Our results demonstrate the widespread occurrence of emerging PFAS in human milk in China. The relatively high EDIs and half-lives of emerging PFAS suggest potential health risks of postnatal exposure in newborns. https://doi.org/10.1289/EHP11403
Supplemental Material is available online (https://doi.org/10.1289/EHP11403).
The authors declare no conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Per- and polyfluoroalkyl substances (PFAS) are a broad range of synthetic chemicals widely used as water- and oil-repellents in textiles, food packaging, and leather; processing aids in the fluoropolymer production; anti-mist agents in chrome plating; and surfactants in film-forming foams.1,2 Over the past several decades, long-chain PFAS, including perfluoroalkane sulfonic acids (PFSAs; CnF2n+1SO3H, n≥six carbon atoms) and perfluoroalkyl carboxylic acids (PFCAs; CnF2n+1COOH, n≥seven carbon atoms), have attracted increasing scientific and regulatory attention owing to their environmental persistence, bioaccumulation, and potential toxicity in wildlife3,4 and humans.5,6 As such, various regulations have been implemented to restrict the manufacture and use of specific PFAS,7–10 especially perfluorooctanesulfonate (PFOS)11 and perfluorooctanoate (PFOA).12 To replace these restricted PFAS, many manufacturers have introduced short-chain and partially fluorinated PFAS alternatives, characterized by the substitution of hydrogen or insertion of ether-oxygen in their backbones,2,13,14 for example, per- and polyfluoroalkyl ether carboxylic acids and sulfonic acids (PFECAs and PFESAs).15–17
However, there is growing evidence that these novel compounds are also environmentally persistent and exhibit similar or even higher toxicity than PFOS and PFOA. In recent years, hexafluoropropylene oxide dimer acid (HFPO-DA; commercial name: GenX), hexafluoropropylene oxide trimer acid (HFPO-TA), and 6:2 chlorinated polyfluorinated ether sulfonate (6:2 Cl-PFESA; commercial name: F-53B) have been increasingly detected in the environment,18,19 even in polar regions,20,21 and in wildlife.16,22 In addition, various emerging PFAS [i.e., HFPO-TA, perfluoro(3,5,7,9,11-pentaoxadodecanoic) acid (PFO5DoDA), and 6:2 Cl-PFESA] are reported to impact human liver function23 and bioaccumulate in the estuarine food web.24 Recent toxicological evidence also suggests that perfluoro-(3,5,7,9-tetraoxadecanoic) acid (PFO4DA) and PFO5DoDA may exert higher developmental toxicity to zebrafish than PFOA.25 Of concern, however, the potentially harmful effects of many novel PFAS have not yet been verified in humans, including vulnerable populations, such as infants.
Given their immature metabolic and immune systems, infants are more susceptible than adults to risks associated with hazardous chemicals.26 Human milk provides excellent nutrition and antibodies for the healthy development of infants27,28 but is also considered a source of exposure to external pollutants,29 potentially accounting for most PFAS intake in infants, including >94% and 83% of the exposure to PFOS and PFOA, respectively.30 PFAS content in human milk has been reported globally, although previous studies have focused on legacy PFAS.31–33 For example, studies have found that PFOS and PFOA are the most common PFAS found in samples of human milk from mothers in parts of the United States (median concentration=30 ng/L of PFOS and 14 ng/L of PFOA),31 Czech Republic (20 ng/L and 23 ng/L),34 Sweden (72 ng/L and 89 ng/L),35 and Korea (47 ng/L and 39 ng/L).29
Nevertheless, few studies have reported on PFAS concentrations in human milk from Chinese mothers, except for small-scale local investigations in single cities, namely, Shanghai35 and Hangzhou.36 Animal model studies have shown that early postnatal PFAS exposure may lead to thyroid dysfunction,37 decreased immunity,38 and diseases in later life, such as obesity and diabetes.39 Additional epidemiological evidence suggests that prenatal and postnatal PFAS exposure may increase the risk of poor growth in early childhood during the most sensitive stages of life.40–42 Thus, it is essential to provide a national baseline of PFAS exposure in human milk in China and to assess the risks of emerging and legacy PFAS exposure in exclusively breastfed infants.
Studies on PFAS clearance and elimination half-life have been reported in highly exposed adult populations (e.g., fluorochemical production workers, airport employees) by biomonitoring serum samples across multiple longitudinal time points43–45 or by analyzing paired serum and urine samples at a single time point,46,47 with research on infants remaining scarce. In occupationally exposed workers, average renal clearance rates (CLrenals) of PFOA, perfluorohexane sulfonate (PFHxS), and PFOS are 0.067, 0.023, and 0.010mL/kg body weight (BW) per day, respectively.46 Furthermore, PFOA (at 0.674mL/kg BW per day) and PFOS (at 0.034mL/kg BW per day) are eliminated more rapidly in the general population compared with occupationally exposed populations.48 Thus, PFAS clearance estimates show considerable variation among different populations.
To evaluate the health risks of breast milk to infants, the accumulation and elimination behaviors of PFAS must be comprehensively considered given that infant organs are more sensitive to contaminants. To date, however, studies on renal clearance of PFAS in infants are limited. Despite the detection of PFAS in the serum, placenta, and breast milk of mothers, as well as in the cord blood of neonates, exploration of their associations in various paired biological matrices remains poor. One of the main reasons is that biological sample selection in birth cohorts is restricted, particularly body fluids from neonates (i.e., serum). Thus, cord blood is considered a good biological matrix reflecting internal PFAS exposure in infants at birth.33 In addition, the difficulty in collecting urine samples from neonates has also hampered investigations on the elimination of hazardous chemicals from the body. To date, only a few studies have analyzed phthalate excretion in infants based on urine samples collected from disposable gel absorbent diapers.49 To the best of our knowledge, however, studies have not reported on the occurrence and distribution of PFAS in multiple matched mother–infant samples. Therefore, a comprehensive assessment of postnatal PFAS exposure and accumulation via daily human milk intake and urinary excretion is lacking. At the same time, given the increase in emerging PFAS exposure, it is critical to understand the renal clearance of PFAS and predict their serum concentrations at key developmental stages in human infants.
In the present study, we aimed to a) determine the concentrations of 22 PFAS (including nine emerging PFAS) in human milk from lactating women obtained in the first 2 weeks postpartum in 21 cities in China, b) estimate the daily intake of PFAS and net accumulation exposure to PFAS in nursing infants, and c) calculate the renal clearance of PFAS in three paired biological samples (cord blood, human milk, and urine) and predict PFAS serum concentrations using a single compartment and first-order pharmacokinetic model. This study should not only provide health protection thresholds for breastfeeding infants but also provide information on the elimination and half-life of emerging PFAS, which can be used to compare the potential health risks of emerging and legacy PFAS.
Methods
Study Population and Sample Collection
Expectant mothers were recruited prior to delivery in the local hospital. The mothers were informed regarding the breast milk study plan and purpose in the hospital before delivery. Those willing to participate in the study provided fully informed consent for our use of their samples and study aim. The criteria for recruitment were minimal being able to a) produce breast milk after delivery (including both vaginal and cesarean section) and b) complete a basic questionnaire. Mothers suffering miscarriage, stillbirth, or delivery complications/injury, such as amniotic fluid embolism and placental abruption, were excluded from recruitment. From October 2020 to June 2021, a total of 1,151 postpartum milk samples were provided by participants from 21 cities in China (Table S1 and Figure S1). Breast milk samples were collected from two hospitals in Shanghai and one hospital in each of the other study cities. Lactating women within the 2-week postpartum period were instructed to express breast milk manually into a centrifuge tube. During the collection period, breast milk samples were stored at −80°C at the local hospital. Once the collection process was completed, the samples were transported on dry ice to our laboratory at Shanghai Jiao Tong University for PFAS measurement. Of the 1,151 participants, those from Huantai County (n=15) and Hangzhou City (n=65) provided additional paired cord blood and infant urine samples. Cord blood samples were collected during delivery. Disposable diapers containing urine were collected in the hospital obstetrics department during the first postnatal week and sealed in polypropylene bags. All samples were stored at −20°C within 24 h of collection and transferred to −80°C until further analysis.
All mothers provided fully informed consent for our usage of their samples and completed a questionnaire with the assistance of well-trained nurses. The questionnaire included information on maternal age (in years; continuous), postnatal weight (in kilograms; continuous), height (in meters; continuous), education (high school, bachelor’s degree, advanced degree; categorical), annual family income [Chinese Yuan Renminbi (CNY); low: <50,000, middle: 50,000–100,000, upper middle: >100,000; categorical)], hyperglycemia (yes or no; categorical), hypertension (yes or no; categorical), type of delivery (cesarean section or vaginal delivery; categorical), gravidity (primiparous or multiparous; categorical), infant sex (boy or girl; categorical), birth weight (in grams; continuous), and birth length (in centimeters; continuous). All research protocols and ethics applications were approved by the participating hospitals. The ethics committee of Shanghai Jiao Tong University provided approval for this study.
Standards and Reagents
A total of 22 target PFAS, including 9 emerging PFAS [i.e., perfluoro-2-methoxyacetic acid (PFMOAA), perfluoro(3,5-dioxahexanoic) acid (PFO2HxA), perfluoro(3,5,7-trioxaoctanoic) acid (PFO3OA), PFO4DA, PFO5DoDA, HFPO-DA, HFPO-TA, and chlorinated polyfluorinated ether sulfonates (6:2 and 8:2 Cl-PFESA)] and 13 legacy PFAS (C4–C12 PFCAs and C4, C6–C8 PFSAs), were analyzed in this study. Except for PFMOAA, PFO2HxA, PFO3OA, PFO4DA, PFO5DoDA, and HFPO-TA, all native and mass-labeled internal standards (purity >99%; listed in Table S3) were purchased from Wellington Laboratories. The remaining native standards (purity >98%) were synthesized using previously reported methods.23
Ammonium acetate (≥99.9%) was purchased from Sigma. Methanol [liquid chromatography–mass spectrometry (LC-MS) grade], acetonitrile [ACN; high-pressure LC (HPLC) grade], water (LC-MS grade), formic acid, acetic acid, and ammonium hydroxide were obtained from Fisher Scientific. N-propylethylenediamine (PSA) adsorbent was purchased from Agilent Technologies. Calcium chloride (CaCl2; ≥96%) was purchased from Shanghai Macklin Biochemical Co., Ltd. Oasis weak anion exchange (WAX) cartridges (6 cc/150mg) were purchased from the Waters Corporation.
PFAS Measurement
The target compounds were measured in human milk, cord blood, and urine based on previous extraction methods, with minor modifications.23,31,50 In brief, 2mL of breast milk was extracted using 2mL of acidified ACN (containing 1% formic acid), with the process repeated twice. The supernatants were combined and concentrated to near dryness (<0.5mL) under nitrogen (N2). The remaining 0.5mL of solution was diluted with 10mL of 1% formic acid water and further purified using a solid phase extraction (SPE) cartridge (Oasis WAX 6 cc/150mg; Waters). The cartridge was preconditioned with 8mL of 0.5% ammonium hydroxide in methanol, 4mL of methanol, and 4mL of ultrapure water. After loading the samples, the cartridges were washed with 4mL of buffer solution (25 mM acetic acid/ammonium acetate, pH = 4) and 4mL of methanol. The target compounds were eluted using 4mL of 0.5% ammonium hydroxide in methanol. The eluent was evaporated under N2 and reconstituted with 200μL of methanol/water (vol/vol=50:50) for instrumental analysis.
Cord blood was extracted using ACN for protein precipitation.23 Briefly, 200μL of blood was spiked with 5 ng of internal standard and extracted with 1mL of ACN. The mixture was vortex-mixed and sonicated for 10 min. The supernatant was transferred into a new 15-mL tube after vortexing for 10 min at 900 rpm and centrifuged for 15 min at 148,000 rpm and 4°C. Another 1mL of ACN was added to the remaining tube, and the extraction procedure was repeated as described above. The extract solution (2mL) was evaporated to dryness under N2 at 40°C and reconstituted with 200μL of methanol/water (vol/vol=50:50).
Urine samples were extracted from the gel diaper by mixing the gel absorbent with CaCl2, as per previously reported methods,50 with a customized glass cup used to separate urine from the dehydrated gel absorbent. In detail, the wet diaper was cut open after thawing at room temperature. The gel absorbent particles were carefully transferred from the diaper to a glass cup with a filter at the bottom. An amount of CaCl2 powder was added to the glass cup and mixed thoroughly with the gel absorbent (ratio of CaCl2 to gel absorbent, 1:50 wt/wt). The mixture was then incubated for 15 min at room temperature to release the urine from the gel absorbent. The tube was centrifuged at 3,000 rpm for 3 min and a “clear” urine sample was collected for further preparation. The urine sample (10mL) was concentrated using an Oasis WAX cartridge following the same procedure as for breast milk. Additional cleanup was performed by adding 50mg of PSA absorbents before evaporation. The supernatant was evaporated under N2 and reconstituted with 200μL of methanol/water (vol/vol=50:50) for instrumental analysis. Considering further computational convenience for PFAS elimination and half-life estimation, as well as limitations of urine sampling, PFAS concentrations were not adjusted for creatinine concentration or specific gravity.
The extractants were finally injected into an ultra-high-performance liquid chromatograph (ExionLC AD; Applied Biosystems/SCIEX Inc.) coupled to an electrospray ionization tandem mass spectrometer (QTRAP 6500 plus; AB SCIEX) for quantification of PFAS. Owing to the fragmentation of PFECAs in the ion source, another quantification method was used exclusively for PFECAs. Details on chromatographic columns, instrumental parameters, and tandem mass spectrometry parameters are presented in Tables S2 and S3.
Quality Assurance and Quality Control
Potential laboratory background, field blank, and solvent blank contaminations were checked before processing formal samples. One procedural blank (Milli-Q water) and one quality control sample (Standard Reference Material, SRM 1957, used for cord blood and mixed and spiked human milk and urine samples used for human milk and infant urine, respectively) were used every 20 samples to monitor background contamination and ensure the accuracy of data from each batch. PFAS were not detected in the procedural blanks (n=70), except for HFPO-DA and PFBA (means=2.23 ng/L and 1.68 ng/L, respectively, in the blanks). Thus, the levels of HFPO-DA and PFBA were blank-corrected by subtracting the average procedural blank concentrations from the sample concentrations. An 11-point internal calibration curve (0.01, 0.02, 0.05, 0.1, 0.2, 0.5, 1, 2, 5, 10, and 20 ng/mL) was prepared for all analytes in matrix-free solvent, which exhibited excellent linearity (R2>0.99), except for PFMOAA (1–20 ng/mL) in the urine analysis matrix calibration curve.
Instrumental drift was monitored by injecting a calibration standard every 10 sample injections, and a new calibration curve was constructed if a ±20% deviation from its initial value was observed. The limits of detection (LODs) were set as the PFAS concentration resulting in a matrix-specific signal-to-noise (S/N) ratio of 3. The method limits of quantification (LOQs) were established based on the concentration resulting in a matrix-specific S/N ratio of 10 or the lowest concentration in the calibration curve with measured concentrations within ±20% of its theoretical value, followed by adjustment with the concentration or dilution factor in extraction. Details on LODs and LOQs in different matrices are provided in Table S4. Matrix-spiked recoveries (n=4) were assessed by spiking 0.2 ng of mixed native standards (final concentration 1.0 ng/mL) into a blank matrix (pooled human milk, infant cord blood, and infant urine), followed by the same analytical methods above. The recoveries of PFAS in human milk, cord blood, and urine samples ranged from 78% to 114% (Table S5).
Data Analysis
Descriptive statistics were assessed for participant demographics, neonatal birth information, and PFAS concentrations in three biomatrices (human milk, cord blood, and urine). In calculations of average proportions among all PFAS in the biomatrices, PFAS concentrations below LOQ were set to zero, and the proportions were calculated via two steps: a) contribution percentage of PFAS in each sample, and b) average percentage of individual PFAS across all human milk samples.
Considering the nonnormality of data, the Mann–Whitney U-test and Kruskal–Wallis test were used to compare differences in PFAS concentrations in human milk samples between demographic variables. Maternal age (continuous variable) was divided to four groups (<25, 25–29, 30–34, ≥35; categorized variable). Maternal weight (continuous variable) and height (continuous variable) were calculated to postnatal body mass index (BMI; categorized variable) based on Chinese BMI standards (underweight: <18.5 kg/m2, normal: 18.5–23.9 kg/m2, overweight: ≥24 kg/m2). For missing data on demographic characteristics, PFAS concentrations in these samples were excluded, and comparisons of PFAS concentrations were analyzed in other subgroups. Concentrations below the LOQs were substituted by the LOQ divided by 2 for analytes with detection rates of >50% in human milk samples for analysis of differences. Other PFAS with low detection rates in human milk [i.e., HFPO-DA, HFPO-TA, PFO2HxA, PFO3OA, PFO4DA, 8:2 Cl-PFESA, perfluoropentanoate (PFPeA), perfluorohexanoate (PFHxA), perfluorobutane sulfonate (PFBS), and perfluoroheptane sulfonate (PFHpS)] were not further explored.
Hierarchical cluster analysis was used to assess distribution similarities or differences in PFAS concentrations between various sampling cities. The complete-linkage approach was used for clustering to avoid the generation of links in single linkages.51 The distance between clusters was defined as the longest distance between clusters. Hierarchical cluster analysis was conducted using TBtools (open source software; https://github.com/CJ-Chen/TBtools/releases). Based on the clustering results, the various sampling cities were divided into different regions. Significant differences in individual PFAS within clusters were verified using the Kruskal–Wallis test. Statistical analyses were performed using SPSS (version 24.0; SPSS Inc.), and the significance threshold was set to p<0.05.
The estimated daily intake (EDI; in nanograms per kilogram BW per day) of each PFAS was calculated for breastfed infants using the following equation: EDIs=CBM×VBMBW,
where CBM represents PFAS concentration measured in breast milk (in nanograms per liter); VBM is the average daily consumption volume of breast milk (510mL/d for infants <1 month of age)52; and BW is the body weight of the infant at birth (in grams). BW was assumed as 3,400g for infants missing birth weight data, based on average food ingestion rate (FIR; 150mL/kg BW per day) and VBM (510mL/d) published in the Exposure Factors Handbook.52 In total, 78 infants were missing birth weight data, including 41 in Dalian, 3 in Huantai, 2 in Shouguang, 6 in Shanghai, and 26 in Guangzhou. Sensitivity analysis was conducted by repeating EDI calculations after excluding those with missing birth weight. Independent-samples t-tests were used to assess differences in PFAS EDIs based on infant sex.
The daily absolute intake/excretion mass of PFAS (AIM/AEM; in nanograms per day) and initial body burden of PFAS exposure (EIB; in nanograms) for infants at birth were calculated using the following equations: AIM=CBM×FIR×BW1,000×1,000,
AEM=CUR×UER×BW1,000×1,000,
EIB=CCB×VB×BW1,000×1,000,
where CUR is the concentration of PFAS in urine (in nanograms per liter); UER (urine excretion rate) is the average daily excretion via urine, set to 48mL/kg BW per day53; CCB is the concentration of PFAS in cord blood (in nanograms per liter); and VB is the blood volume in the body, estimated as 85mL/kg BW for infants <3 months of age.54 The two factors (1,000) in the denominator of the equations were used for unit conversion, representing the conversion of milliliters per kilogram BW per day (or milliliters per kilogram BW) to liters per kilogram BW per day (or liters per kilogram BW) for intake/excretion rate (or blood volume) and grams to kilograms for BW. Here, we assumed that PFAS levels in cord blood represent internal exposure of infants at birth, given that cord blood is present on the fetal side of the vascular organ.55,56 The correlation analysis of PFAS concentrations was explored between breast milk and cord blood samples and between breast milk and urine samples using regression analyses.
We calculated the daily CLrenal (in milliliters per kilogram BW per day) of individual PFAS on the basis of paired blood and urine concentrations, and we estimated the elimination half-lives (T1/2, in days) using the following equations.48 As in Zhang et al. (2013),48 we assumed renal clearance to be the major pathway for PFAS elimination to simplify half-life estimations. CLrenal=CUR×UER2×CCB,
T1/2=ln(2)×VdCLrenal,
where the factor of 2 in the denominator of the equation represents the PFAS concentrations in serum-to-whole blood ratio57,58 and Vd is the apparent volume of distribution (in liters per kilogram). The Vd values of PFOS and PFOS were set to 230 and 170mL/kg, respectively.59 Comparable Vd values of PFOA have been reported in different mammals [mean±standard deviation (SD),191±67mL/kg].60 Owing to the lack of Vd values for PFAS in humans, we tentatively used the toxicokinetic parameters of rats and mice for the calculation of half-life. We set 130, 199, 265, and 394mL/kg as the Vd values for PFBA, PFHpA, PFNA, and PFDA, respectively.61,62 The Vd values of PFO3OA, PFO4DA, and PFO5DoDA were set to 103, 154, and 296mL/kg based on toxicokinetic data of mice.63 The Vd values of other PFECAs and PFCAs were estimated based on the Vd of substances with similar structures, such as 103mL/kg for PFMOAA, PFO2HxA, PFO3OA, and HFPO-DA; 296mL/kg for HFPO-TA; and 130mL/kg for PFPeA. Except for the Vd value of PFBS (277mL/kg) estimated in CD-1 mice,64 we assumed a Vd value of 230mL/kg for other PFSAs to maintain consistency with other studies.46 We further compared the estimated CLrenals in the present work to published data in prior studies. Available data on renal PFAS clearance in humans were identified by searching keywords “PFAS” and “RENAL CLEARANCE” in PubMed and Web of Science for publications appearing between 1 January 2000 and 31 December 2021.
Given that the one-compartment model has been previously adopted to estimate time-dependent serum concentrations and PFAS compounds are not highly lipophilic,59,65 a first-order model based on the one-compartment model was employed to address temporal trends in PFAS in infants in days after birth (Ct) via breast milk intake66: dCtdt=−k×Ct+EDIsVd,
where k is the elimination rate calculated using the half-life with the following equation: k=ln(2)T1/2.
The solution for Ct can be expressed as follows: Ct=EDIsk×Vd(1−e−k×t)+C0×e−k×t,
where Ct is the serum concentration in infants after birthday t, and C0 is the initial serum concentration in infants at birth. As above, PFAS levels in cord blood were regarded as the initial blood concentration in infants. Thus, PFAS concentrations in whole blood were multiplied by a factor of 2 to convert to a serum concentration before substituting the equation.54 A factor of 2 was used for the ratio of PFAS concentrations in serum to whole blood based on previous studies of the general adult population.57,58 The prediction models for PFAS concentrations in serum of infants <1 year of age were generated using R (version 4.0.2; R Development Core Team).
Results
Population Characteristics
A summary of the demographic characteristics and basic information of the 1,151 mothers and their infants is provided in Table 1. In total, 98% of participants provided their maternal age, which ranged from 16 to 47 years of age (mean±SD, 30.3±4.7 y). Nearly half of the participants had a bachelor’s degree or above, with an upper middle-class income (>1 00,000 CNY). More than 50% of mothers were primiparous and chose to have a cesarean section. The average postpartum BMI value was 25.6 kg/m2, and 65% of participants were overweight. Most mothers (>80%) did not suffer from hyperglycemia or hypertension during pregnancy. Of the neonates with known sex, males accounted for 55% and females accounted for 45%. The mean±SD of birth weight was 3,277±708g, with 82% of infants within the normal weight range.
Table 1 Summary of demographic characteristics of study participants (n=1,151).
Parameter n Percentage (%)
Maternal age (y)
<25 107 9.5
25–29 414 36.7
30–34 398 35.3
≥35 209 18.5
Missing 23 —
Education
High school 657 60.2
Bachelor’s degree 371 34.0
Advanced degree 63 5.8
Missing 60 —
Annual family income (CNY)
Low (<50,000) 74 8.4
Middle (50,000–100,000) 336 38.1
Upper middle (>100,000) 471 53.5
Missing 270 —
Type of delivery
Cesarean section 644 56.3
Vaginal delivery 500 43.7
Missing 7 —
Gravidity
Primiparous 636 55.9
Multipara 501 44.1
Missing 14 —
Postnatal BMI (kg/m2)
Underweight (<18.5) 26 2.4
Normal (18.5–24.0) 354 32.5
Overweight (>24.0) 710 65.1
Missing 61 —
Hyperglycemia
Yes 180 15.8
No 959 84.2
Missing 12 —
Hypertension
Yes 177 15.9
No 933 84.1
Missing 41 —
Infant sex
Girl 493 45.0
Boy 603 55.0
Missing 55 —
Infant weight (g)
<2,500 106 9.9
2,500–4,000 874 81.5
>4,000 93 8.7
Missing 78 —
Note: The number of missing was removed from the calculation of percentages. —, not applicable; BMI, body mass index; CNY, Chinese Yuan Renminbi.
PFAS Concentrations in Breast Milk
The detection rates and descriptive statistics of PFAS concentrations in human milk are summarized in Table 2. Among the 22 PFAS detected in milk, 3 emerging PFAS (PFMOAA, PFO5DoDA, and 6:2 Cl-PFESA) and 9 legacy PFAS were detected in more than half of the samples (Figure 1, Table 2, and Excel Table S1). The other PFAS analytes were detected in 8.5%–42.7% of samples and hence were not further analyzed. Among all PFAS analytes, PFOA was dominant (median=336 ng/L), accounting for 63% of total PFAS (on average), followed by PFOS (49.7 ng/L, 13%) and 6:2 Cl-PFESA (13.6 ng/L, 4.5%). To the best of our knowledge, 2 emerging PFECAs (PFMOAA and PFO5DoDA) were found in human milk for the first time, with median concentrations comparable to those of PFBA and PFHpA. We examined differences in PFAS concentrations and distribution among demographic subgroups (Table S6). The concentrations of several long-chain PFAS, such as PFO5DoDA, 6:2 Cl-PFESA, and PFOS, were significantly higher in older and upper-middle income groups. Higher PFAS levels, except for PFOA, were also observed in human milk from mothers with a bachelor’s degree. In contrast, the highest PFOA concentrations were found in mothers with a high school diploma and low annual family income. Notably, milk concentrations of four long-chain PFAS (PFO5DoDA, 6:2 Cl-PFESA, PFOA, and PFOS) were significantly higher in mothers with gestational hypertension. The median concentrations of PFO5DoDA, 6:2 Cl-PFESA, and PFOA were about 1.4 times higher in postpartum overweight females than in the normal-weight group.
Table 2 Concentrations of PFAS (ng/L) in human milk (n=1,151).
Analyte n DR (%) GM SD Min Percentile Max Proportion (%)a
5th 25th 50th 75th 95th
Emerging PFAS
HFPO-DA 98 8.5 3.056 10.09 <LOQ <LOQ <LOQ <LOQ <LOQ 5.527 199.1 0.2
HFPO-TA 212 18.4 1.783 22.52 <LOQ <LOQ <LOQ <LOQ <LOQ 13.41 524.3 0.2
PFMOAA 1,099 95.5 4.816 75.34 <LOQ 1.042 1.946 3.416 7.411 107.0 1,086 2.4
PFO2HxA 208 18.1 0.733 7.887 <LOQ <LOQ <LOQ <LOQ <LOQ 8.604 125.5 0.2
PFO3OA 492 42.7 1.275 34.27 <LOQ <LOQ <LOQ <LOQ 2.600 21.24 744.4 0.8
PFO4DA 343 29.8 1.763 11.03 <LOQ <LOQ <LOQ <LOQ 2.659 15.82 144.9 0.4
PFO5DoDA 820 71.2 2.380 9.002 <LOQ <LOQ <LOQ 2.633 5.731 17.78 117.5 0.7
6:2 Cl-PFESA 1,147 99.7 15.24 75.70 <LOQ 2.886 6.923 13.59 32.02 104.3 1,183 4.5
8:2 Cl-PFESA 266 23.1 0.696 5.224 <LOQ <LOQ <LOQ <LOQ <LOQ 3.561 163.3 0.1
Legacy PFAS
PFBA 959 83.3 4.872 15.62 <LOQ <LOQ 2.716 4.549 7.784 24.93 258.5 1.6
PFPeA 168 14.6 0.640 2.395 <LOQ <LOQ <LOQ <LOQ <LOQ 3.609 38.52 0.1
PFHxA 373 32.4 0.867 9.713 <LOQ <LOQ <LOQ <LOQ 1.317 6.789 238.0 0.4
PFHpA 925 80.4 2.095 4.647 <LOQ <LOQ 1.173 2.093 3.862 11.81 58.29 0.5
PFOA 1,151 100 325.7 1,129 12.04 58.07 144.9 335.9 668.9 1,933 2,059 63.0
PFNA 1,151 100 12.71 21.42 1.161 3.799 7.396 12.28 20.51 50.37 307.0 3.1
PFDA 1,142 99.2 9.343 38.88 <LOQ 1.893 4.379 8.721 17.92 59.24 798.3 2.6
PFUnDA 1,144 99.4 9.244 22.09 <LOQ 2.459 5.025 8.598 15.81 43.87 438.2 2.4
PFDoDA 744 64.6 1.523 6.510 <LOQ <LOQ <LOQ 1.471 2.949 10.25 135.2 0.4
PFBS 291 25.3 0.696 1.492 <LOQ <LOQ <LOQ <LOQ <LOQ 2.788 30.42 0.1
PFHxS 1,092 94.9 7.619 38.23 <LOQ <LOQ 3.514 7.286 16.86 54.93 533.8 2.6
PFHpS 324 28.1 0.712 1.437 <LOQ <LOQ <LOQ <LOQ 1.063 2.531 29.29 0.1
PFOS 1,151 100 52.83 154.6 2.621 13.81 29.38 49.72 87.79 257.4 2,503 13.7
Note: Cl-PFESA, chlorinated polyfluorinated ether sulfonates; DR, detection rate; GM, geometric mean; HFPO-DA, hexafluoropropylene oxide dimer acid; HFPO-TA, perfluoro-2,5-dimethyl-3,6-dioxanonanoic acid; LOQ, limit of quantitation; Max, maximum; Min, minimum; PFAS, per- and polyfluoroalkyl substances; PFBA, perfluorobutanoate; PFBS, perfluorobutane sulfonate; PFDA, perfluorodecanoate; PFDoDA, perfluorododecanoate; PFHpA, perfluoroheptanoate; PFHpS, perfluoroheptane sulfonate; PFHxA, perfluorohexanoate; PFHxS, perfluorohexane sulfonate; PFMOAA, perfluoro-2-methoxyacetic acid; PFNA, perfluorononanoate; PFO2HxA, perfluoro(3,5-dioxahexanoic) acid; PFO3OA, perfluoro(3,5,7-trioxaoctanoic) acid; PFO4DA, perfluoro(3,5,7,9-tetraoxadecanoic) acid; PFO5DoDA, perfluoro(3,5,7,9,11-pentaoxadodecanoic) acid; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFPeA, perfluoropentanoate; PFUnDA, perfluoroundecanoate; SD, standard deviation.
a Average proportion among all PFAS in human milk samples. Proportions were calculated by two steps: namely, determined contribution percentage of PFAS in each sample, then averaged percentages of individual PFAS across all human milk samples. PFAS concentrations below LOQ were set to zero.
Figure 1. Concentrations (ng/L) of PFAS detected in >50% of human milk samples (n=1,151). Boxes display 25th, 50th, and 75th percentiles for PFAS concentrations, and whiskers represent 10th and 90th percentiles. Values above box represent median concentration. Corresponding raw data are provided in Excel Table S1. Boxes are ranked from highest to lowest median level for each PFAS. Blue boxes (left of dashed line) represent legacy PFAS, pink boxes (right of dashed line) represent emerging PFAS. Note: Cl-PFESA, chlorinated polyfluorinated ether sulfonates; PFAS, per- and polyfluoroalkyl substances; PFBA, perfluorobutanoate; PFDA, perfluorodecanoate; PFDoDA, perfluorododecanoate; PFHpA, perfluoroheptanoate; PFHxS, perfluorohexane sulfonate; PFMOAA, perfluoro-2-methoxyacetic acid; PFNA, perfluorononanoate; PFO5DoDA, perfluoro(3,5,7,9,11-pentaoxadodecanoic) acid; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFUnDA, perfluoroundecanoate.
Figure 1 is a box and whiskers plot, plotting concentration (nanograms per liter), ranging from 0.1 to 1 in increments of 0.9, 1 to 10 in increments of 9, 10 to 100 in increments of 90, 100 to 1,000 in increments of 900, 1,000 to 10,000 in increments of 9,000 (y-axis) across perfluorooctanoate, perfluorooctane sulfonate, perfluorononanoate, perfluorodecanoate, perfluoroundecanoate, perfluorohexane sulfonate, perfluorobutanoate, perfluoroheptanoate, perfluorododecanoate, 6 to 2 chlorinated polyfluorinated ether sulfonates, perfluoro-2-methoxyacetic acid, and perfluoro(3,5,7,9,11-pentaoxadodecanoic) acid (x-axis) for Legacy per- and polyfluoroalkyl substances and Emerging per- and polyfluoroalkyl substances.
PFAS Profiles and Clustering in Various Cities
Figure 2A shows the similarities in PFAS composition in human milk samples across all sampled cities (corresponding numeric results are provided in the Excel Table S2). Median concentrations (in nanograms per liter) of PFAS (detection rate >50%) in breast milk obtained in individual cities are summarized in Table S7. Although levels varied in the different cities, PFOA accounted for the largest proportion of total PFAS in all areas, ranging from 37% (Taizhou) to 87% (Guangzhou). PFOS was the second-most common compound, ranging from 2.2% (Huantai County) to 32% (Wuhan), although a higher proportion of PFECAs was detected in several specific areas (e.g., Sanming, Huantai, Shouguang). The percentage contributions of several long-chain PFCAs (PFDA, PFUnDA, and PFDoDA) and PFHxS were higher in eastern China, including Shanghai, Nanjing, Hangzhou, Huzhou, Taizhou, Jiaxing, Hefei, Anqing, Wenzhou, and Ningbo, than in other cities, whereas the proportions of PFBA and PFHpA were higher in Xiamen, Sanming, and Wuhan. The remaining PFAS (PFPeA, PFHxA, PFBS, and PFHpS) were found at negligible percentages in the studied locations. The cities were divided into four clusters based on PFAS composition using complete-linkage cluster analysis. As shown in Figure 2B, the median concentrations of PFO5DoDA in human milk differed significantly in the four clusters. PFMOAA levels were highest in Region 1, whereas PFOA levels were significantly higher in Regions 1 and 3. Highly elevated concentrations of 6:2 Cl-PFESA, PFOS, and PFNA were observed in Region 4. The raw data are provided in the Excel Table S1 and p-values from different comparisons of the four regions are reported in Table S8.
Figure 2. Composition profiles and concentrations of PFAS in human milk samples from different cities and regions. PFAS median concentrations were used to calculate composition profiles in each city. (A) Hierarchical cluster analysis was performed on profiles of 21 cities using complete-linkage cluster analysis. Cities were divided into four clusters. (B) Box and whisker plots for PFMOAA, PFO5DoDA, 6:2 Cl-PFESA, PFOA, PFNA, and PFOS concentrations in regions based on clustering results. Region 1 (R1, orange boxes, n=169) included Sanming, Huantai, and Shouguang. Region 2 (R2, red boxes, n=172) included Shenyang, Chongqing, Wuhan, Dalian, and Xiamen. Region 3 (R3, blue boxes, n=259) included Guangzhou and Zigong. Region 4 (R4, green boxes, n=551) included Huzhou, Taizhou, Nanjing, Shanghai, Hefei, Anqing, Ningbo, Wenzhou, Quzhou, Jiaxing, and Hangzhou. Different letters represent significant differences between groups at p<0.05 by Kruskal–Wallis test. All specific p-values from comparisons are reported in Table S8. Corresponding raw data are provided in Excel Tables S1 and S2. ∑PFECAs represent sum of HFPO-DA, HFPO-TA, PFMOAA, PFO2HxA, PFO3OA, PFO4DA, and PFO5DoDA; ∑PFESAs represent sum of 6:2 Cl-PFESA and 8:2 Cl-PFESA. Note: Cl-PFESA, chlorinated polyfluorinated ether sulfonates; HFPO-DA, hexafluoropropylene oxide dimer acid; HFPO-TA, perfluoro-2,5-dimethyl-3,6-dioxanonanoic acid; PFAS, per- and polyfluoroalkyl substances; PFBA, perfluorobutanoate; PFBS, perfluorobutane sulfonate; PFDA, perfluorodecanoate; PFDoDA, perfluorododecanoate; PFECAs, per- and polyfluoroalkyl ether carboxylic acids; PFESAs, per- and polyfluoroalkyl ether sulfonic acids; PFHpA, perfluoroheptanoate; PFHpS, perfluoroheptane sulfonate; PFHxA, perfluorohexanoate; PFHxS, perfluorohexane sulfonate; PFMOAA, perfluoro-2-methoxyacetic acid; PFNA, perfluorononanoate; PFO2HxA, perfluoro(3,5-dioxahexanoic) acid; PFO3OA, perfluoro(3,5,7-trioxaoctanoic) acid; PFO4DA, perfluoro(3,5,7,9-tetraoxadecanoic) acid; PFO5DoDA, perfluoro(3,5,7,9,11-pentaoxadodecanoic) acid; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFPeA, perfluoropentanoate; PFUnDA, perfluoroundecanoate; R, region.
Figure 2A is a horizontal stacked bar graph, plotting Hangzhou, Jiaxing, Quzhou, Wenzhou, Ningbo, Anqing, Hefei, Shanghai, Nanjing, Taizhou, Huzhou, Zigong, Guangzhou, Xiamen, Dalian, Wuhan, Chongqing, Shenyang, Shouguang, Huantai, and Sanming (y-axis) across Composition profiles of per- and polyfluoroalkyl substances, ranging from 0 to 100 percent in increments of 20 (x-axis) for uppercase sigma perfluoroalkyl ether carboxylic, uppercase sigma polyfluoroalkyl ether sulfonic acid, perfluorobutanoate, perfluoropentanoate, perfluorohexanoate, perfluoroheptanoate, perfluorooctanoate, perfluorononanoate, perfluorodecanoate, perfluoroundecanoate, perfluorododecanoate, perfluorobutane sulfonate, perfluorohexane sulfonate, perfluoroheptane sulfonate, and perfluorooctane sulfonate. Figure 2B is a set of six box and whisker plots titled perfluoro-2-methoxyacetic acid, perfluoro(3,5,7,9,11-pentaoxadodecanoic) acid, 6 to 2 chlorinated polyfluorinated ether sulfonates, perfluorooctanoate, perfluorononanoate, and perfluorooctane sulfonate, plotting Concentration (nanograms per liter), ranging from 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript; 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript; 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript; 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, 10 begin superscript 4 end superscript, and 10 begin superscript 5 end superscript; 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript; and 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, and 10 begin superscript 4 end superscript (y-axis) across Region 1, Region 2, Region 3, and Region 4 (x-axis) for Region 1 (R 1), Region 2 (R 2), Region 3 (R 3), and Region 4 (R 4), respectively.
Daily Intake of PFAS via Breast Milk
The PFOA and PFOS EDIs in individual cities are shown in Figure 3 and Figures S2, respectively (corresponding numeric data are provided in the Excel Tables S3 and S4). Descriptive statistics of EDIs for PFOA and PFOS and their differences by neonatal sex are summarized in Table S9. We found that the EDIs of PFOA and PFOS exceeded the reference dose (RfD) of 20 ng/kg BW per day recommended by the U.S. Environmental Protection Agency (EPA) in 78% and 17% of breastfed infant samples in all cities, respectively. Furthermore, more than half of the PFOA EDIs were above the threshold in 14 cities, including Huantai, Shouguang, Guangzhou, and Quzhou, in which the EDIs exceeded the recommended threshold (20 ng/kg BW per day) in all tested infants.67 In comparison, in half the cities, only 20% of the PFOS EDIs exceeded the RfD (20 ng/kg BW per day),68 whereas in other cities, levels were below the reference threshold in all tested samples. Of concern, when compared with stricter reference values, average PFOA intake via human breast milk was more than two orders of magnitude higher than the tolerable daily intake (0.857 ng/kg BW per day) established by the EFSA (European Food Safety Authority).69 Of note, all PFOA EDIs exceeded this reference value in all cities. In addition, almost all PFOS EDIs (>90% samples) were higher than the EFSA reference dose (1.857 ng/kg BW per day).69 Furthermore, 100% of PFOS EDIs exceeded 1.857 ng/kg BW per day in one-third of the cities (including Dalian, Xiamen, Huzhou, Nanjing, Anqing, Quzhou, and Jiaxing).
Figure 3. Lactational estimated daily intake (EDI, ng/kg BW per day) of PFOA in <1-month-old infants in different cities, compared with reference values. The red dashed line represents the threshold (20 ng/kg BW per day) from the U.S. Environmental Protection Agency.67 The green dashed line represents the threshold (0.857 ng/kg BW per day) from European Food Safety Authority.69 Percentage values are ratio of number of samples exceeding 20 ng/kg BW per day to total sample size in city. All EDIs in each city exceeded 0.857 ng/kg BW per day. All calculated EDI values are provided in Excel Table S3. Note: BW, body weight; PFOA, perfluorooctanoate.
Figure 3 is a graph, plotting estimated daily intake (nanograms per kilogram body weighted per day), ranging from 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, 10 begin superscript 4 end superscript (y-axis) across Sanming, Huantai, Shouguang, Shenyang, Chongqing, Wuhan, Dalian, Xiamen, Guangzhou, Zigong, Huzhou, Taizhou, Nanjing, Shanghai, Hefei, Anqing, Ningbo, Wenzhou, Quzhou, Jiaxing, and Hangzhou (x-axis).
PFAS Mass Balance via Human Milk and Urine
The distribution of PFAS in paired cord blood and neonate urine samples (n=80) is shown in Tables S10 and S11 and Excel Table S5. Except for HFPO-DA, PFO2HxA, 8:2 Cl-PFESA, and PFBA, the detection rates of all target analytes were >75% in cord blood, and the median concentrations ranged from 15.65 ng/L (in PFPeA) to 5,185 ng/L (in PFOA). The dominant PFAS in cord blood were the same as in human milk (PFOA>PFOS>6:2 Cl-PFESA). The PFAS levels in neonatal urine were much lower than the levels in cord blood. Several long-chain PFAS, such as PFO4DA, PFO5DoDA, 8:2 Cl-PFESA, PFUnDA, PFDoDA, PFHpS, and PFOS, were not detected in any urine sample.
In contrast, short-chain PFBA and PFMOAA were detected in 100% of urine samples at concentrations 3–9 times higher than that of PFOA. The different PFAS proportions in the three biomatrices and the PFAS mass balance estimates are shown in Figure 4 and Figures S3–S4, respectively (corresponding numeric data are provided in the Excel Tables S6 and S7). Similar proportions of PFAS were detected in human milk and cord blood, but not in urine. For example, PFOA, PFOS, and 6:2 Cl-PFESA accounted for >70% of total PFAS in human milk and cord blood, whereas perfluorobutanoate (PFBA) and PFMOAA accounted for >70% of total PFAS in urine.
Figure 4. PFAS absolute mass (ng) of daily intake by human milk and excretion by urine in infants. Absolute masses of four PFAS (PFMOAA, PFBA, PFOA, and PFNA) with detection rates >50% in both human milk and urine were calculated. Actual number of paired samples in calculation differed (i.e., n=80, 72, 78, and 59 for PFMOAA, PFBA, PFOA, and PFNA, respectively). Absolute intake was calculated by PFAS concentration (ng/L) in human milk and daily intake rate (510mL/d).52 Absolute excretion was calculated by PFAS concentration (ng/L) in urine, clearance rate (48mL/kg BW per day),53 and infant weight (kg). Differences between PFAS absolute mass of human milk and urine represent net daily intake for infants. Corresponding raw data are provided in Excel Table S6. Note: PFAS, per- and polyfluoroalkyl substances; PFBA, perfluorobutanoate; PFMOAA, perfluoro-2-methoxyacetic acid; PFNA, perfluorononanoate; PFOA, perfluorooctanoate.
Figure 4 is a bar graph, per- and polyfluoroalkyl substances absolute mass (nanograms), ranging from 10 begin superscript negative 2 end superscript, 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, and 10 begin superscript 4 end superscript (y-axis) across perfluoro-2-methoxyacetic acid, perfluorobutanoate, perfluorooctanoate, perfluorononanoate (x-axis) for intake and excretion.
In the mass balance estimation model, the absolute daily mass of most PFAS detected in human milk and urine, except for PFMOAA, PFBA, and PFBS, showed a significant net accumulation in neonates (Figure 4 and Figure S3). Correlation analysis of the dominant legacy PFAS (i.e., PFOA, PFNA, and PFOS) and emerging PFAS (i.e., PFMOAA, PFO5DoDA, and 6:2 Cl-PFESA) in different matrices revealed significant positive correlations (0.24≤r2≤0.78) between individual PFAS levels in cord blood and human milk (Figure S5 and Table S12). Given that PFO5DoDA and PFOS were not detected in urine, we were unable to explore their associations between human milk and urine. The four other PFAS (0.01≤r2≤0.29) showed no significant correlations between human milk and urine.
Estimation of Renal Clearance and Half-Life
Table 3 shows the daily CLrenal (in milliliters per kilogram BW per day) of individual PFAS based on paired cord blood and urine concentrations. Among the PFCAs with C4 to C10 chain lengths, the median CLrenal decreased from 31.80 to 0.022mL/kg BW per day as the perfluorinated carbon chain length increased. PFMOAA, PFO2HxA, and PFO3OA exhibited moderate CLrenal, comparable to that of PFBS (median=3.210mL/kg BW per day). Overall, PFCAs were preferentially excreted via urine compared with PFSAs with the same perfluoroalkyl chain length, especially 6:2 Cl-PFESA, with the lowest CLrenal (0.009mL/kg BW per day).
Table 3 Summary of PFAS renal clearance (mL/kg per day) in infants (present study) and adults (previous studies).
Compound n a Infants in present study Adults in published studies
Mean Min 25th P Median 75th P Max n=9 b n=39 c n=7 d n=72 e n=20 f n=66 g n=207 h n=61 i n=20 j
Location Hangzhou/Huantai Wuhan Wuhan/Yantai Shijiazhuang/Handan Yingcheng Kyoto
HFPO-TA 24 0.408 0.048 0.108 0.369 0.581 1.206 — — — — — — — — —
PFMOAA 79 24.49 0.088 0.978 3.851 17.67 677.5 — — — — — — — — —
PFO2HxA 7 7.217 1.084 2.095 2.605 9.348 23.95 — — — — — — — — —
PFO3OA 26 3.409 0.054 1.338 2.349 4.324 11.54 — — — — — — — — —
PFBA 21 34.61 4.076 15.86 31.80 46.40 93.89 28.70 3.820 10.30 — — — — — —
PFPeA 63 25.79 3.305 13.47 21.43 32.95 95.42 — — — — — — — — —
PFHpA 33 1.071 0.105 0.540 0.722 1.176 7.679 3.260 17.70 7.980 0.170 0.410 — — —
PFOA 78 0.064 0.002 0.031 0.049 0.076 0.259 0.121 0.061 0.079 — 0.140 0.180 0.067k 0.070 1.365
PFNA 60 0.082 0.016 0.028 0.047 0.100 0.543 — 0.055 0.074 — 0.200 0.094 — — —
PFDA 13 0.037 0.008 0.013 0.022 0.055 0.118 — 0.015 0.031 — 0.047 0.035 — — —
PFBS 28 5.537 0.700 2.156 3.210 7.423 27.53 22.20 8.210 174.0 — — — — — —
PFHxS 20 0.956 0.052 0.108 0.270 0.781 11.71 6.470 0.012 0.006 — 0.033 0.015 0.023k 0.030 —
6:2 Cl-PFESA 33 0.012 0.003 0.006 0.009 0.014 0.053 — — — 0.0016 — — — — —
Note: —, not applicable; Cl-PFESA, chlorinated polyfluorinated ether sulfonates; HFPO-TA, perfluoro-2,5-dimethyl-3,6-dioxanonanoic acid; P, percentile; PFBA, perfluorobutanoate; PFBS, perfluorobutane sulfonate; PFDA, perfluorodecanoate; PFHpA, perfluoroheptanoate; PFHxS, perfluorohexane sulfonate; PFMOAA, perfluoro-2-methoxyacetic acid; PFNA, perfluorononanoate; PFO2HxA, perfluoro(3,5-dioxahexanoic) acid; PFO3OA, perfluoro(3,5,7-trioxaoctanoic) acid; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; PFPeA, perfluoropentanoate.
a Number of paired cord blood–urine samples available for estimating renal clearance.
bBackground-exposed person (n=9).70
cFishery employee (n=39).70
dFishery family (n=7).70
ePredominantly male (58 males/14 females) population (n=72).47
fYoung female group (n=20).48
gMale and older female groups (n=66).48
hOccupational workers (n=207).46
iOccupational workers (n=61).71
jAdults (n=20).72
kValues represent average renal clearance in the studied population.
We assumed that renal clearance was the only body elimination pathway for PFAS in infants. Thus, we estimated the half-lives of PFAS in all infants, regardless of sex (Table S13). Although the estimated half-lives of emerging PFECAs were 4–130 times shorter than that of PFOA (median=6.6y), 6:2 Cl-PFESA displayed a median half-life of 49 y. Considering continual PFAS intake, we estimated PFAS serum concentrations (in nanograms per milliliter) in infants (<1 year of age) based on measurements in human milk and calculated the CLrenal (estimated values are provided in the Excel Table S8). As shown in Figure 5, the model-predicted PFOA level in serum exceeded 100 ng/mL at 6 months of age. The predicted HFPO-TA concentration was consistent with the reported value (Table S14). However, the predicted 6:2 Cl-PFESA level (geometric mean: 6.410 ng/mL) was approximately eight times higher than the measured value (0.847 ng/mL), and the predicted PFMOAA value was approximately one 1/100th that of the measured value in serum (Table S14). The inconsistencies in model prediction for different PFAS are difficult to explain, but may be due, at least in part, to the immaturity of the current prediction model regarding emerging PFAS.
Figure 5. Simulated PFAS serum concentrations (ng/mL) in <1-y-old infants. Simulated values are provided in Excel Table S8. Fitted curves from top to bottom are labeled by ordinal numbers (1–13), responding to the order of PFAS in the right legend. Note: Cl-PFESA, chlorinated polyfluorinated ether sulfonates; HFPO-TA, perfluoro-2,5-dimethyl-3,6-dioxanonanoic acid; PFAS, per- and polyfluoroalkyl substances; PFBA, perfluorobutanoate; PFBS, perfluorobutane sulfonate; PFDA, perfluorodecanoate; PFHpA, perfluoroheptanoate; PFHxS, perfluorohexane sulfonate; PFMOAA, perfluoro-2-methoxyacetic acid; PFNA, perfluorononanoate; PFO2HxA, perfluoro(3,5-dioxahexanoic) acid; PFO3OA, perfluoro(3,5,7-trioxaoctanoic) acid; PFOA, perfluorooctanoate; PFPeA, perfluoropentanoate.
Figure 5 is a set of line graphs, plotting Serum concentration (nanograms per milliliter), ranging from 10 begin superscript negative 3 end superscript, 10 begin superscript negative 2 end superscript, 10 begin superscript negative 1 end superscript, 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript (y-axis) across Time (year), ranging from 0.0 to 1.0 in increments of 0.2 (x-axis) across perfluoropentanoate, perfluorobutane sulfonate, perfluorobutanoate, perfluoro-2-methoxyacetic acid, perfluoro(3,5,7-trioxaoctanoic) acid, perfluoroheptanoate, perfluoro(3,5-dioxahexanoic) acid, perfluoro-2,5-dimethyl-3,6-dioxanonanoic acid, perfluorobutanoate, perfluorononanoate, perfluorohexane sulfonate, 6 to 2 chlorinated polyfluorinated ether sulfonates, and perfluorooctanoate.
Discussion
The present study was conducted to assess background exposure to PFAS in human milk in various cities in China. Overall, our results showed that PFAS exposure is still widespread in Chinese mothers and that the health implications to neonates are not negligible considering the high detection and abundance of these compounds in human milk, even under increasingly strict worldwide regulations.7,73 Notably, the concentrations of legacy PFAS, especially PFOA, in breast milk were ∼4- to 24-fold higher in the study population (median=336 ng/L , sample size: n=1,151, sampling year: 2020–2021) than in previously studied populations from other countries, such as the United States (median=13.9 ng/L, n=50, 2019),31 Czech Republic (median=44 ng/L, n=50, 2010),74 Japan (median=89 ng/L, n=30, 2010),75 Spain (median=57.5 ng/L, n=10, 2012),76 Italy (mean=60 ng/L, n=37, 2010),77 and France (mean=41 ng/L, n=61, 2010–2013),33 which could be attributed to the ongoing use of PFOA-related products in China.
With the phasing out of PFOA production and usage in North America and Europe, China has become one of the largest fluorochemical manufacturers and consumers in the world,78 with many fluorochemical plants found across China, such as Shandong Dongyue Chemical Co., Ltd., with a polytetrafluoroethylene (PTFE) production capacity of 45,000 tons/y,79 and Fujian Sannong Chemical Co., Ltd., with a production capacity of 6,000 tons/y,79 resulting in an increased risk of exposure to PFOA from environmental discharge into drinking water.79 Various emerging PFAS used as raw material or intended products in the manufacturing process, including PFMOAA, PFO5DoDA, and 6:2 Cl-PFESA,2 have been detected in Wilmington residents in North Carolina living downstream of a fluorochemical manufacturing facility.80 Similar to previous surveys of residents living near fluorochemical plants, PFMOAA, PFO5DoDA, and 6:2 Cl-PFESA were the dominant emerging PFAS detected in the human milk samples (median concentrations=13.6, 3.42, and 2.63 ng/mL, respectively), accounting for a similar proportion (7.6% in human milk in this study) to that found in a highly exposed population (10.8% in serum).23 To the best of our knowledge, this study is the first to report on these emerging PFAS in human milk and the first to conduct a comprehensive nationwide (21 cities) survey of baseline PFAS exposure in breastfed infants.
In the present study, there was wide variation in human milk PFAS concentrations in the different cities. The highest median Σ22PFAS concentration was found in Huantai (1,482 ng/L), followed by Shouguang (869.1 ng/L), Quzhou (829.9 ng/L), Guangzhou (824.9 ng/L), and Hangzhou (802.6 ng/L), which were 2–11 times higher than levels in other cities (Table S7). The lack of relevant prenatal data on maternal dietary intake, occupation, and lifestyle habits during pregnancy makes it difficult to explain the discrepancies in the cities studied, although potential pollution sources may offer a partial explanation in some cases where exposure levels are particularly high. Thus, we assessed the composition of PFAS between cities and divided the cities into four clusters (Figure 2A). Cluster 1 consisted of Sanming, Huantai, and Shouguang, all of which contain typical fluorochemical industrial plants.16,81 This cluster showed significant differences from other cities, especially in the use of PFMOAA and PFO5DoDA (Figure 2B). Cluster 2, consisting of Shenyang, Chongqing, Wuhan, Dalian, and Xiamen, showed PFOA and PFOS as the most dominant PFAS. Cluster 3, consisting of Guangzhou and Zigong, showed a dominant PFOA exposure burden but almost no quantifiable PFO5DoDA in human milk, which may be related to the high PFOA concentration found in drinking water (Zigong: 3,165 ng/L, Guangzhou: 53.4 ng/L).79 Cluster 4, consisting of 11 cities located in eastern China, showed the highest levels of PFOS and 6:2 Cl-PFESA, similar to the high levels reported in surface water and municipal sewage sludge in the region,82,83 which may be related to the local electroplating industry.17,84
Whether compared with the U.S. EPA-recommended threshold (20 ng/kg BW per day) or the EFSA reference values (0.857 ng/kg BW per day for PFOA and 1.857 ng/kg BW per day for PFOS), the lactational PFOA and PFOS EDI values obtained in this study are of concern, especially regarding their potential health effects, both alone and in combination, on infants. Our estimates for all breastfed infants far exceeded the latest combined tolerable daily intake rates (TDI; 0.63 ng/kg BW per day) for PFOA, PFNA, PFHxS, and PFOS established by the EFSA,85 and even the minimum estimate was close to 10 times the threshold. Given that birth weight data were missing for five cities, additional sensitivity analysis was conducted for those locations, excluding those with missing birth weight data (Figure S6 and Excel Table S9). The percentages of EDIs exceeding the U.S. EPA-recommended threshold (20 ng/kg BW per day) or EFSA reference values (0.857 ng/kg BW per day for PFOA and 1.857 ng/kg BW per day for PFOS) in each city remained unchanged.
At present, there are insufficient studies to prove that the potential risks of PFAS exposure in infants via breast milk intake reduce the benefits of breastfeeding in terms of growth outcomes. However, based on previous experiments in rodents, exposure to PFAS in early life can have adverse developmental effects, such as growth restriction, obesity, and endocrine disruption.86,87 Thus, our results suggest that breastfeeding may be a critical contributor to adverse effects later in childhood. However, we must be careful in interpreting these results because, as infants gain body weight and reduce human milk intake, EDI values decrease with infant age. Although average EDI in male infants was slightly lower than that in female infants (Table S7), the difference was not significant, suggesting that sex did not impact exposure risk during infancy. To date, only trace amounts of PFAS have been reported in infant formula,74,75,88 in contrast to the high detection and abundance of PFAS in human milk. This suggests that infant formula feeding may be safer than breastfeeding from the perspective of PFAS exposure. Nevertheless, considering the irreplaceable role of human milk in strengthening the immune system and providing balanced nutrients to infants, further epidemiological studies are needed to demonstrate whether breastfeeding with contaminants has adverse health outcomes on newborns.89
In the paired samples, the different PFAS composition in human milk and urine could be attributed to their polarity. For example, short-chain PFAS can be readily transferred in the aqueous phase because of their relatively high polarity and water solubility. Human milk, as a complex mixture of endogenous compounds, contains a high lipid content, multiple proteins, vitamins, and antibodies.88 Therefore, the percentage contribution of PFMOAA and PFBA increased significantly from 2.4% and 1.6% in human milk to 30% and 49% in urine (Figure S4A). In contrast, the proportions of long-chain PFAS, such PFOA, PFOS, and 6:2 Cl-PFESA, were basically constant, accounting for 70%, 9%, and 4% in human milk and 58%, 10%, and 8% in cord blood, respectively. The similar PFAS composition profiles in human milk and cord blood were primarily ascribed to their protein-binding affinity.90 Comparing daily breast milk intake and urinary excretion, only short-chain PFAS (i.e., PFMOAA and PFBA) were underaccumulated in the body. This result may be interpreted as the original infant body load transferred from cord blood at birth (Figure S4B). However, in the simulated serum concentrations derived from the PFAS clearance rate and cord blood level, PFMOAA was not completely excreted over time (Figure 5). In contrast, previous research has shown that serum PFMOAA is positively correlated with age,23 suggesting other nonnegligible exposure routes for infants in addition to human milk, such as ingestion of dust and dermal absorption.30 However, given the difficulty of direct blood collection in infants, we regarded PFAS levels in cord blood to be the same as levels in neonate blood. Although measurements of xenobiotics in cord blood are common, such as estimation of total body burden in newborns,56 comparison of gestation and lactation exposure,54 and calculation of placental transport efficiency, simultaneous evaluation of multiple neonatal matrices is recommended to avoid bias from inappropriate sample types.
In this study, we explored PFAS CLrenals and half-lives in infants, especially emerging PFECAs, which have not yet been estimated in humans based on current literature. Overall, the median CLrenal values of long-chain PFCAs were 0.049mL/kg BW per day for PFOA, 0.047mL/kg BW per day for PFNA, and 0.022mL/kg BW per day for PFDA, which are much lower than those reported in prior studies (Table 3). Results showed that the estimated median values of PFHxA and PFHxS (0.722 and 0.270mL/kg BW per day, respectively) were in the range of those reported in adults in prior studies (0.170–17.70 and 0.006–6.470mL/kg BW per day, respectively).46,70,71 However, the median renal clearance efficiencies of PFBA and 6:2 Cl-PFESA were slightly higher than those reported previously in adults (Table 3). The reason for these differences is unclear (although participant age, location, diet, and sampling size may be a factor), but in all cases, renal PFAS clearance was consistently ranked in the following order: PFBA>PFHpA>PFOA>PFNA>PFDA. In addition, the corresponding half-life of PFOA was 6.6 y in infants, compared with 2.9 y in low-exposure adults and 8.5 years in high-exposure adults estimated in prior studies.44 It should be noted that our study used single time-paired whole blood and urine samples and assumed renal clearance as total clearance in the half-life calculation, which may cause an overestimation of half-life. Further, deviation caused by conversion of PFAS levels in whole blood to serum should also be considered. Although previous studies have reported serum-to-whole blood ratios of 1.9–2.3 for most legacy PFAS (e.g., PFOA, PFNA, PFDA, PFHxS, and PFOS),58 we still found slight differences in these ratios for different PFAS due to their different protein-binding affinities and erythrocyte-binding capacities.91
Based on the higher estimated half-lives (3- to 8-fold higher in infants than in adults), our results showed greater accumulation of long-chain PFCAs (>8 carbons) in infants compared with adults,48 which is unsurprising given the immature metabolic system of newborns. Therefore, the slower excretion of PFAS in infants than in adults provides an opportunity to address the renal clearance and half-lives of these emerging PFECAs, in which oxygen atoms inserted between carbon chains lead to higher water solubility than in legacy PFAS and to faster excretion via urine.92,93 One of the strengths of our study is that neonatal urine samples were successfully collected with disposable gel absorbent diapers, which made it possible to evaluate PFAS elimination.
We used a one-compartment model to estimate renal clearance and temporal trends of PFAS concentrations in infants. Although this model is simple and easy to implement, it has several limitations: For example, a) it cannot fully describe the distribution, metabolism, and excretion processes; and b) it does not consider physiological life stage characteristics. Various physiologically based pharmacokinetic (PBPK) models for different PFAS have been published for different age groups,87,94 including a recent model for sensitive populations (e.g., pregnant women, fetuses, lactating mothers, neonates).87 However, for some emerging PFAS, model parameters are often lacking. Our data not only provide validation for the PBPK model in infants, but also provide reliable estimates of crucial parameters, such as absorption and excretion factors.
The CLrenals of emerging short-chain PFECAs (PFMOAA, PFO2HxA, and PFO3OA) were lower than those of PFBA and PFPeA but comparable to that of PFBS (3.210mL/kg BW per day), indicating that the PFECAs were not rapidly eliminated. Also unexpectedly, HFPO-TA was excreted faster than PFOA in infants, in contrast to the higher bioaccumulation potential of HFPO-TA than PFOA reported in studies on wild common carp16 and animal models.95 Some bias may exist in the results owing to the large variation in sample size in single PFAS elimination estimates. However, in our study, no other reference data were available to compare PFECA estimates. Although PFECAs showed lower accumulation and faster excretion than PFOA in the studied infants, ongoing exposure from daily human milk intake deserves attention owing to the potential effects on neonatal health. For example, significantly high levels of PFECAs (i.e., HFPO-TA, PFMOAA, PFO4DA, and PFO5DoDA) have been detected in children <1 to 18 years of age.23 Studies have also found elevated PFECA concentrations in vegetables, eggs, and seafood,24,96 suggesting exposure risk for consumers. Therefore, epidemiological studies on infant exposure to emerging PFAS are required to determine their associations with growth and developmental outcomes.
Our study showed a marked decrease in clearance rates with increasing carbon chain length for both legacy PFAS and emerging PFECAs, which may be related to their binding affinities to serum albumin97 and the transport proteins governing reabsorption.98 For example, PFOS shows strong interactions with human serum albumin71 and long-chain PFAS and organic anion transporting protein (Oatp 1a1) show strong interactions in the rat kidney, resulting in lower urinary excretion.99 However, the above theory fails to explain the presence of 6:2 Cl-PFESA but not PFOS in the urine samples in our study, even though 6:2 Cl-PFESA is longer. This finding is also inconsistent with previous research showing slower urinary elimination of 6:2 Cl-PFESA than PFOS in highly exposed metal plate workers.47 Therefore, molecular chain length–dependent mechanisms cannot fully explain the discrepancies found in PFAS elimination in different studied populations, and other transport mechanisms are likely acting on PFAS to cross fluid barriers in humans. In addition to the discrepancy in 6:2 Cl-PFESA and PFOS elimination, we also found a 6-fold difference in the estimated half-life of 6:2 Cl-PFESA in infants compared with highly exposed adult populations.47 Although this difference may be attributable to differences in study population (e.g., age, occupation, and sex composition), dose-dependent renal clearance estimation and ongoing exposure cannot be ruled out, which may result in slower elimination in highly exposed populations.45 However, despite the uncertainty surrounding the urinary elimination of 6:2 Cl-PFESA and PFOS, their potential health risks to infant development cannot be ignored. Notably, previous toxicological studies have shown that 6:2 Cl-PFESA induces developmental and cardiac toxicity in zebrafish embryos100 and causes serious liver injury and lipid metabolism disturbance in rodents.101
To the best of our knowledge, this study is the first to report on the clearance rate of PFECAs in humans. However, our study has several limitations regarding CLrenal and half-life estimates owing to the lack of Vd data for most PFAS. We tentatively used Vd values in rats and mice for the calculation of half-life to ensure estimations were as accurate as possible. However, there were some inevitable uncertainties in the current model based on the differences in toxicokinetic parameters for PFAS among various species. Taking PFOA as an example, previous Vd values have been estimated at 346 and 211mL/kg BW in male and female rats,102 226 and 135mL/kg BW in male and female mice,103 181 and 198mL/kg BW in male and female cynomolgus monkeys,104 and 170mL/kg BW in humans.59 Thus, we should express caution in discussing the half-lives of PFAS when extrapolating Vd values from animals to humans. As such, additional data on distribution volumes of emerging PFAS are needed to improve the reliability of future CLrenal and half-life estimates. In addition, the relatively small number of study subjects decreased the reliability of the PFAS excretion estimates. Furthermore, although we succeeded in obtaining neonatal urine samples, long-chain PFAS levels in infant urine were too low to estimate half-lives. Given the limitations in infant urine collection, we could not obtain fresh urine samples and needed to add CaCl2 salt to release urine from the gel absorbents. The PFAS concentrations in urine were not adjusted for specific gravity, which may interfere with the dilution or concentration of urine. The health risks of PFAS may also be underestimated given that the half-life estimates were taken at a single time point, despite infants being continuously exposed. Therefore, sample collection over an extended period should be considered in future studies to calculate PFAS elimination and determine the correlations between PFAS exposure and infant growth outcomes.
Conclusions
We determined the occurrence and distribution of PFAS in human milk collected from 1,151 lactating women in 21 cities in China and provided baseline data on the potential exposure risk for breastfeeding infants. Our results showed that PFOA, PFOS, and 6:2 Cl-PFESA were the dominant PFAS in breast milk. To the best of our knowledge, this study is the first report on emerging PFECAs in breast milk samples, demonstrating wide distribution in China. Several PFECAs found with high detection frequencies (such as PFMOAA and PFO5DoDA) deserve greater attention. Of concern, the EDIs for PFOA in more than half of the breastfed infants in most studied cities were higher than the recommended threshold, although this should be interpreted with caution given the rapid changes in infant weight and breast milk ingestion. Based on measurements of PFAS in matched cord blood and urine samples, albeit with a small sample size, we determined the PFAS CLrenals and half-lives in infants and predicted PFAS levels in serum during the lactation period, which broadens our understanding of the elimination and accumulation of emerging PFECAs and potential health risks of both legacy and emerging PFAS in infants. However, further studies are needed to explore the association between PFAS exposure and health outcomes and to elucidate the key factors contributing to adverse health effects.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by the National Natural Science Foundation of China [U22A20618 and 22276124 (both to J.D.)] and Shanghai Science and Technology Committee [21DZ1202101 (to J.D.)].
==== Refs
References
1. Buck RC, Franklin J, Berger U, Conder JM, Cousins IT, de Voogt P, et al. 2011. Perfluoroalkyl and polyfluoroalkyl substances in the environment: terminology, classification, and origins. Integr Environ Assess Manag 7 (4 ):513–541, PMID: , 10.1002/ieam.258.21793199
2. Wang Z, DeWitt JC, Higgins CP, Cousins IT. 2017. A never-ending story of per- and polyfluoroalkyl substances (PFASs)? Environ Sci Technol 51 (5 ):2508–2518, PMID: , 10.1021/acs.est.6b04806.28224793
3. Wang Y, Yao J, Dai J, Ma L, Liu D, Xu H, et al. 2020. Per- and polyfluoroalkyl substances (PFASs) in blood of captive Siberian tigers in China: occurrence and associations with biochemical parameters. Environ Pollut 265 (pt B ):114805, PMID: , 10.1016/j.envpol.2020.114805.32480004
4. Cui Q, Shi F, Pan Y, Zhang H, Dai J. 2019. Per- and polyfluoroalkyl substances (PFASs) in the blood of two colobine monkey species from China: occurrence and exposure pathways. Sci Total Environ 674 :524–531, PMID: , 10.1016/j.scitotenv.2019.04.118.31022542
5. Luo D, Wu W, Pan Y, Du B, Shen M, Zeng L. 2021. Associations of prenatal exposure to per- and polyfluoroalkyl substances with the neonatal birth size and hormones in the growth hormone/insulin-like growth factor axis. Environ Sci Technol 55 (17 ):11859–11873, PMID: , 10.1021/acs.est.1c02670.34378915
6. Liu M, Zhang G, Meng L, Han X, Li Y, Shi Y, et al. 2022. Associations between novel and legacy per- and polyfluoroalkyl substances in human serum and thyroid cancer: a case and healthy population in Shandong Province, East China. Environ Sci Technol 56 (10 ):6144–6151, PMID: , 10.1021/acs.est.1c02850.34618433
7. Bălan SA, Mathrani VC, Guo DF, Algazi AM. 2021. Regulating PFAS as a chemical class under the California Safer Consumer Products program. Environ Health Perspect 129 (2 ):25001, PMID: , 10.1289/EHP7431.33595352
8. OECD (Organisation for Economic Cooperation and Development). 2015. Risk Reduction Approaches for PFASs—A Cross Country Analysis. https://www.oecd.org/chemicalsafety/risk-management/Risk_Reduction_Approaches%20for%20PFASS.pdf [accessed 9 June 2020].
9. U.S. EPA (U.S. Environmental Protection Agency). 2020. Significant new use rule: long-chain perfluoroalkyl carboxylate and perfluoroalkyl sulfonate chemical substances. https://www.regulations.gov/document/EPA-HQ-OPPT-2013-0225-0232 [accessed 17 September 2021].
10. Dong Z, Bahar MM, Jit J, Kennedy B, Priestly B, Ng J, et al. 2017. Issues raised by the reference doses for perfluorooctane sulfonate and perfluorooctanoic acid. Environ Int 105 :86–94, PMID: , 10.1016/j.envint.2017.05.006.28521193
11. UNEP (United Nations Environment Programme). 2009. Chemicals listed in Annex B. http://chm.pops.int/Implementation/Alternatives/AlternativestoPOPs/ChemicalslistedinAnnexB/tabid/5850/Default.aspx [accessed 5 October 2021].
12. UNEP. 2019. Chemicals listed in Annex A. http://chm.pops.int/Implementation/Alternatives/AlternativestoPOPs/ChemicalslistedinAnnexA/tabid/5837/Default.aspx [accessed 5 October 2020].
13. Glüge J, London R, Cousins IT, DeWitt J, Goldenman G, Herzke D, et al. 2022. Information requirements under the essential-use concept: PFAS case studies. Environ Sci Technol 56 (10 ):6232–6242, PMID: , 10.1021/acs.est.1c03732.34608797
14. Wang Z, Cousins IT, Scheringer M, Buck RC, Hungerbühler K. 2014. Global emission inventories for C4-C14 perfluoroalkyl carboxylic acid (PFCA) homologues from 1951 to 2030, part I: production and emissions from quantifiable sources. Environ Int 70 :62–75, PMID: , 10.1016/j.envint.2014.04.013.24932785
15. Wang Z, Cousins IT, Scheringer M, Hungerbühler K. 2013. Fluorinated alternatives to long-chain perfluoroalkyl carboxylic acids (PFCAs), perfluoroalkane sulfonic acids (PFSAs) and their potential precursors. Environ Int 60 :242–248, PMID: , 10.1016/j.envint.2013.08.021.24660230
16. Pan Y, Zhang H, Cui Q, Sheng N, Yeung LWY, Guo Y, et al. 2017. First report on the occurrence and bioaccumulation of hexafluoropropylene oxide trimer acid: an emerging concern. Environ Sci Technol 51 (17 ):9553–9560, PMID: , 10.1021/acs.est.7b02259.28780851
17. Wang S, Huang J, Yang Y, Hui Y, Ge Y, Larssen T, et al. 2013. First report of a Chinese PFOS alternative overlooked for 30 years: its toxicity, persistence, and presence in the environment. Environ Sci Technol 47 (18 ):10163–10170, PMID: , 10.1021/es401525n.23952109
18. Pan Y, Zhang H, Cui Q, Sheng N, Yeung LWY, Sun Y, et al. 2018. Worldwide distribution of novel perfluoroether carboxylic and sulfonic acids in surface water. Environ Sci Technol 52 (14 ):7621–7629, PMID: , 10.1021/acs.est.8b00829.29749740
19. Zhou J, Li S, Liang X, Feng X, Wang T, Li Z, et al. 2021. First report on the sources, vertical distribution and human health risks of legacy and novel per- and polyfluoroalkyl substances in groundwater from the Loess Plateau, China. J Hazard Mater 404 (pt A ):124134, PMID: , 10.1016/j.jhazmat.2020.124134.33022527
20. Sebastiano M, Angelier F, Blévin P, Ribout C, Sagerup K, Descamps S, et al. 2020. Exposure to PFAS is associated with telomere length dynamics and demographic responses of an Arctic top predator. Environ Sci Technol 54 (16 ):10217–10226, PMID: , 10.1021/acs.est.0c03099.32696640
21. Gao K, Miao X, Fu J, Chen Y, Li H, Pan W, et al. 2020. Occurrence and trophic transfer of per- and polyfluoroalkyl substances in an Antarctic ecosystem. Environ Pollut 257 :113383, PMID: , 10.1016/j.envpol.2019.113383.31727419
22. Robuck AR, Cantwell MG, McCord JP, Addison LM, Pfohl M, Strynar MJ, et al. 2020. Legacy and novel per- and polyfluoroalkyl substances in juvenile seabirds from the U.S. Atlantic coast. Environ Sci Technol 54 (20 ):12938–12948, PMID: , 10.1021/acs.est.0c01951.32894676
23. Yao J, Pan Y, Sheng N, Su Z, Guo Y, Wang J, et al. 2020. Novel perfluoroalkyl ether carboxylic acids (PFECAs) and sulfonic acids (PFESAs): occurrence and association with serum biochemical parameters in residents living near a fluorochemical plant in China. Environ Sci Technol 54 (21 ):13389–13398, PMID: , 10.1021/acs.est.0c02888.33047597
24. Li Y, Yao J, Zhang J, Pan Y, Dai J, Ji C, et al. 2022. First report on the bioaccumulation and trophic transfer of perfluoroalkyl ether carboxylic acids in estuarine food web. Environ Sci Technol 56 (10 ):6046–6055, PMID: , 10.1021/acs.est.1c00965.34296857
25. Wang J, Shi G, Yao J, Sheng N, Cui R, Su Z, et al. 2020. Perfluoropolyether carboxylic acids (novel alternatives to PFOA) impair zebrafish posterior swim bladder development via thyroid hormone disruption. Environ Int 134 :105317, PMID: , 10.1016/j.envint.2019.105317.31733528
26. Landrigan PJ, Claudio L, Markowitz SB, Berkowitz GS, Brenner BL, Romero H, et al. 1999. Pesticides and inner-city children: exposures, risks, and prevention. Environ Health Perspect 107 (suppl 3 ):431–437, PMID: , 10.1289/ehp.99107s3431.10346991
27. WHO (World Health Organization). 2012. Breastfeeding. https://www.who.int/news-room/facts-in-pictures/detail/breastfeeding [accessed 23 June 2021].
28. Andreas NJ, Kampmann B, Mehring Le-Doare K. 2015. Human breast milk: a review on its composition and bioactivity. Early Hum Dev 91 (11 ):629–635, PMID: , 10.1016/j.earlhumdev.2015.08.013.26375355
29. Lee S, Kim S, Park J, Kim HJ, Choi G, Choi S, et al. 2018. Perfluoroalkyl substances (PFASs) in breast milk from Korea: time-course trends, influencing factors, and infant exposure. Sci Total Environ 612 :286–292, PMID: , 10.1016/j.scitotenv.2017.08.094.28865262
30. Haug LS, Huber S, Becher G, Thomsen C. 2011. Characterisation of human exposure pathways to perfluorinated compounds—comparing exposure estimates with biomarkers of exposure. Environ Int 37 (4 ):687–693, PMID: , 10.1016/j.envint.2011.01.011.21334069
31. Zheng G, Schreder E, Dempsey JC, Uding N, Chu V, Andres G, et al. 2021. Per- and polyfluoroalkyl substances (PFAS) in breast milk: concerning trends for current-use PFAS. Environ Sci Technol 55 (11 ):7510–7520, PMID: , 10.1021/acs.est.0c06978.33982557
32. Fiedler H, Sadia M. 2021. Regional occurrence of perfluoroalkane substances in human milk for the global monitoring plan under the Stockholm Convention on Persistent Organic Pollutants during 2016–2019. Chemosphere 277 :130287, PMID: , 10.1016/j.chemosphere.2021.130287.33774256
33. Cariou R, Veyrand B, Yamada A, Berrebi A, Zalko D, Durand S, et al. 2015. Perfluoroalkyl acid (PFAA) levels and profiles in breast milk, maternal and cord serum of French women and their newborns. Environ Int 84 :71–81, PMID: , 10.1016/j.envint.2015.07.014.26232143
34. Černá M, Grafnetterová AP, Dvořáková D, Pulkrabová J, Malý M, Janoš T, et al. 2020. Biomonitoring of PFOA, PFOS and PFNA in human milk from Czech Republic, time trends and estimation of infant’s daily intake. Environ Res 188 :109763, PMID: , 10.1016/j.envres.2020.109763.32540571
35. Awad R, Zhou Y, Nyberg E, Namazkar S, Yongning W, Xiao Q, et al. 2020. Emerging per- and polyfluoroalkyl substances (PFAS) in human milk from Sweden and China. Environ Sci Process Impacts 22 (10 ):2023–2030, PMID: , 10.1039/d0em00077a.32940316
36. Jin H, Mao L, Xie J, Zhao M, Bai X, Wen J, et al. 2020. Poly- and perfluoroalkyl substance concentrations in human breast milk and their associations with postnatal infant growth. Sci Total Environ 713 :136417, PMID: , 10.1016/j.scitotenv.2019.136417.31955077
37. Aimuzi R, Luo K, Chen Q, Wang H, Feng L, Ouyang F, et al. 2019. Perfluoroalkyl and polyfluoroalkyl substances and fetal thyroid hormone levels in umbilical cord blood among newborns by prelabor caesarean delivery. Environ Int 130 :104929, PMID: , 10.1016/j.envint.2019.104929.31228788
38. Rappazzo KM, Coffman E, Hines EP. 2017. Exposure to perfluorinated alkyl substances and health outcomes in children: a systematic review of the epidemiologic literature. Int J Environ Res Public Health 14 (7 ):691, PMID: , 10.3390/ijerph14070691.28654008
39. Jain RB. 2020. Impact of the co-occurrence of obesity with diabetes, anemia, hypertension, and albuminuria on concentrations of selected perfluoroalkyl acids. Environ Pollut 266 (pt 2 ):115207, PMID: , 10.1016/j.envpol.2020.115207.32698119
40. Gao Y, Luo J, Zhang Y, Pan C, Ren Y, Zhang J, et al. 2022. Prenatal exposure to per- and polyfluoroalkyl substances and child growth trajectories in the first two years. Environ Health Perspect 130 (3 ):37006, PMID: , 10.1289/EHP9875.35285689
41. Luo J, Ramlau-Hansen CH, Kesmodel US, Xiao J, Vasiliou V, Deziel NC, et al. 2022. Prenatal exposure to per- and polyfluoroalkyl substances and facial features at 5 years of age: a study from the Danish National Birth Cohort. Environ Health Perspect 130 (1 ):17006, PMID: , 10.1289/EHP9478.35080464
42. Vuong AM, Yolton K, Xie C, Dietrich KN, Braun JM, Webster GM, et al. 2021. Childhood exposure to per- and polyfluoroalkyl substances (PFAS) and neurobehavioral domains in children at age 8 years. Neurotoxicol Teratol 88 :107022, PMID: , 10.1016/j.ntt.2021.107022.34438039
43. Olsen GW, Burris JM, Ehresman DJ, Froehlich JW, Seacat AM, Butenhoff JL, et al. 2007. Half-life of serum elimination of perfluorooctanesulfonate, perfluorohexanesulfonate, and perfluorooctanoate in retired fluorochemical production workers. Environ Health Perspect 115 (9 ):1298–1305, PMID: , 10.1289/ehp.10009.17805419
44. Seals R, Bartell SM, Steenland K. 2011. Accumulation and clearance of perfluorooctanoic acid (PFOA) in current and former residents of an exposed community. Environ Health Perspect 119 (1 ):119–124, PMID: , 10.1289/ehp.1002346.20870569
45. Xu Y, Fletcher T, Pineda D, Lindh CH, Nilsson C, Glynn A, et al. 2020. Serum half-lives for short- and long-chain perfluoroalkyl acids after ceasing exposure from drinking water contaminated by firefighting foam. Environ Health Perspect 128 (7 ):77004, PMID: , 10.1289/EHP6785.32648786
46. Fu J, Gao Y, Cui L, Wang T, Liang Y, Qu G, et al. 2016. Occurrence, temporal trends, and half-lives of perfluoroalkyl acids (PFAAs) in occupational workers in China. Sci Rep 6 :38039, PMID: , 10.1038/srep38039.27905562
47. Shi Y, Vestergren R, Xu L, Zhou Z, Li C, Liang Y, et al. 2016. Human exposure and elimination kinetics of chlorinated polyfluoroalkyl ether sulfonic acids (Cl-PFESAs). Environ Sci Technol 50 (5 ):2396–2404, PMID: , 10.1021/acs.est.5b05849.26866980
48. Zhang Y, Beesoon S, Zhu L, Martin JW. 2013. Biomonitoring of perfluoroalkyl acids in human urine and estimates of biological half-life. Environ Sci Technol 47 (18 ):10619–10627, PMID: , 10.1021/es401905e.23980546
49. Li X, Liu L, Wang H, Zhang X, Xiao T, Shen H. 2019. Phthalate exposure and cumulative risk in a Chinese newborn population. Environ Sci Pollut Res Int 26 (8 ):7763–7771, PMID: , 10.1007/s11356-019-04216-z.30673950
50. Liu L, Xia T, Guo L, Cao L, Zhao B, Zhang J, et al. 2012. Expressing urine from a gel disposable diaper for biomonitoring using phthalates as an example. J Expo Sci Environ Epidemiol 22 (6 ):625–631, PMID: , 10.1038/jes.2012.51.22643239
51. Bu J, Liu W, Pan Z, Ling K. 2020. Comparative study of hydrochemical classification based on different hierarchical cluster analysis methods. Int J Environ Res Public Health 17 (24 ):9515, PMID: , 10.3390/ijerph17249515.33353090
52. U.S. EPA. 2011. Exposure Factors Handbook: 2011 Edition (Final Report). https://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=236252 [accessed 26 June 2021].
53. Hazinski MF. 2012. Nursing Care of the Critically Ill Child. 3rd edition. St. Louis, MO: Elsevier/Mosby.
54. Liu J, Li J, Liu Y, Chan HM, Zhao Y, Cai Z, et al. 2011. Comparison on gestation and lactation exposure of perfluorinated compounds for newborns. Environ Int 37 (7 ):1206–1212, PMID: , 10.1016/j.envint.2011.05.001.21620474
55. Wright TE. 2015. Biochemical screening for in utero drug exposure. Drug Metab Lett 9 (2 ):65–71, PMID: , 10.2174/1872312809666150904182334.26411469
56. Lee J, Choi K, Park J, Moon HB, Choi G, Lee JJ, et al. 2018. Bisphenol A distribution in serum, urine, placenta, breast milk, and umbilical cord serum in a birth panel of mother–neonate pairs. Sci Total Environ 626 :1494–1501, PMID: , 10.1016/j.scitotenv.2017.10.042.29146078
57. Ehresman DJ, Froehlich JW, Olsen GW, Chang SC, Butenhoff JL. 2007. Comparison of human whole blood, plasma, and serum matrices for the determination of perfluorooctanesulfonate (PFOS), perfluorooctanoate (PFOA), and other fluorochemicals. Environ Res 103 (2 ):176–184, PMID: , 10.1016/j.envres.2006.06.008.16893538
58. Poothong S, Thomsen C, Padilla-Sanchez JA, Papadopoulou E, Haug LS. 2017. Distribution of novel and well-known poly- and perfluoroalkyl substances (PFASs) in human serum, plasma, and whole blood. Environ Sci Technol 51 (22 ):13388–13396, PMID: , 10.1021/acs.est.7b03299.29056041
59. Thompson J, Lorber M, Toms LML, Kato K, Calafat AM, Mueller JF. 2010. Use of simple pharmacokinetic modeling to characterize exposure of Australians to perfluorooctanoic acid and perfluorooctane sulfonic acid. Environ Int 36 (4 ):390–397, PMID: , 10.1016/j.envint.2010.02.008.20236705
60. Han X, Nabb DL, Russell MH, Kennedy GL, Rickard RW. 2012. Renal elimination of perfluorocarboxylates (PFCAs). Chem Res Toxicol 25 (1 ):35–46, PMID: , 10.1021/tx200363w.21985250
61. Chang SC, Das K, Ehresman DJ, Ellefson ME, Gorman GS, Hart JA, et al. 2008. Comparative pharmacokinetics of perfluorobutyrate in rats, mice, monkeys, and humans and relevance to human exposure via drinking water. Toxicol Sci 104 (1 ):40–53, PMID: , 10.1093/toxsci/kfn057.18353799
62. Ohmori K, Kudo N, Katayama K, Kawashima Y. 2003. Comparison of the toxicokinetics between perfluorocarboxylic acids with different carbon chain length. Toxicology 184 (2–3 ):135–140, PMID: , 10.1016/S0300-483X(02)00573-5.12499116
63. Chen J, Li H, Yao J, Guo H, Zhang H, Guo Y, et al. 2021. Chronic exposure to PFO4DA and PFO5DoDA, two perfluoroalkyl ether carboxylic acids (PFECAs), suppresses hepatic stress signals and disturbs glucose and lipid metabolism in male mice. J Hazard Mater 411 :124963, PMID: , 10.1016/j.jhazmat.2020.124963.33440278
64. Lau C, Rumpler J, Das KP, Wood CR, Schmid JE, Strynar MJ, et al. 2020. Pharmacokinetic profile of perfluorobutane sulfonate and activation of hepatic nuclear receptor target genes in mice. Toxicology 441 :152522, PMID: , 10.1016/j.tox.2020.152522.32534104
65. Andersen ME, Mallick P, Clewell HJ III, Yoon M, Olsen GW, Longnecker MP. 2021. Using quantitative modeling tools to assess pharmacokinetic bias in epidemiological studies showing associations between biomarkers and health outcomes at low exposures. Environ Res 197 :111183, PMID: , 10.1016/j.envres.2021.111183.33887277
66. Fan X, Tang S, Wang Y, Fan W, Ben Y, Naidu R, et al. 2022. Global exposure to per- and polyfluoroalkyl substances and associated burden of low birthweight. Environ Sci Technol 56 (7 ):4282–4294, PMID: , 10.1021/acs.est.1c08669.35293723
67. U.S. EPA. 2016. Drinking Water Health Advisory for Perfluorooctanoic Acid (PFOA). EPA 822-R-16-005. https://hero.epa.gov/hero/index.cfm/reference/ details/reference_id/3982042 [accessed 23 February 2023].
68. U.S. EPA. 2016. Drinking Water Health Advisory for Perfluorooctane Sulfonate (PFOS). EPA 822-R-16-004. https://www.epa.gov/sites/production/files/2016-05/documents/pfos_health_advisory_final-plain.pdf [accessed 1 September 2021].
69. EFSA CONTAM Panel (EFSA Panel on Contaminants in the Food Chain), Knutsen HK, Alexander J, Barregård L, Bignami M, Brüschweiler B, et al. 2018. Risk to human health related to the presence of perfluorooctane sulfonic acid and perfluorooctanoic acid in food. EFSA J 16 (12 ):e05194, PMID: , 10.2903/j.efsa.2018.5194.32625773
70. Zhou Z, Shi Y, Vestergren R, Wang T, Liang Y, Cai Y. 2014. Highly elevated serum concentrations of perfluoroalkyl substances in fishery employees from Tangxun Lake, China. Environ Sci Technol 48 (7 ):3864–3874, PMID: , 10.1021/es4057467.24588690
71. Gao Y, Fu J, Cao H, Wang Y, Zhang A, Liang Y, et al. 2015. Differential accumulation and elimination behavior of perfluoroalkyl acid isomers in occupational workers in a manufactory in China. Environ Sci Technol 49 (11 ):6953–6962, PMID: , 10.1021/acs.est.5b00778.25927957
72. Harada K, Inoue K, Morikawa A, Yoshinaga T, Saito N, Koizumi A. 2005. Renal clearance of perfluorooctane sulfonate and perfluorooctanoate in humans and their species-specific excretion. Environ Res 99 (2 ):253–261, PMID: , 10.1016/j.envres.2004.12.003.16194675
73. Lindstrom AB, Strynar MJ, Libelo EL. 2011. Polyfluorinated compounds: past, present, and future. Environ Sci Technol 45 (19 ):7954–7961, PMID: , 10.1021/es2011622.21866930
74. Lankova D, Lacina O, Pulkrabova J, Hajslova J. 2013. The determination of perfluoroalkyl substances, brominated flame retardants and their metabolites in human breast milk and infant formula. Talanta 117 :318–325, PMID: , 10.1016/j.talanta.2013.08.040.24209347
75. Fujii Y, Yan J, Harada KH, Hitomi T, Yang H, Wang P, et al. 2012. Levels and profiles of long-chain perfluorinated carboxylic acids in human breast milk and infant formulas in East Asia. Chemosphere 86 (3 ):315–321, PMID: , 10.1016/j.chemosphere.2011.10.035.22113060
76. Lorenzo M, Farré M, Blasco C, Onghena M, Picó Y, Barceló D. 2016. Perfluoroalkyl substances in breast milk, infant formula and baby food from Valencian Community (Spain). Environ Nanotechnol Monit Manag 6 :108–115, 10.1016/j.enmm.2016.09.001.
77. Barbarossa A, Masetti R, Gazzotti T, Zama D, Astolfi A, Veyrand B, et al. 2013. Perfluoroalkyl substances in human milk: a first survey in Italy. Environ Int 51 :27–30, PMID: , 10.1016/j.envint.2012.10.001.23138018
78. Chen C, Lu Y, Zhang X, Geng J, Wang T, Shi Y, et al. 2009. A review of spatial and temporal assessment of PFOS and PFOA contamination in China. Chem Ecol 25 (3 ):163–177, 10.1080/02757540902918321.
79. Liu L, Qu Y, Huang J, Weber R. 2021. Per- and polyfluoroalkyl substances (PFASs) in Chinese drinking water: risk assessment and geographical distribution. Environ Sci Eur 33 :6, 10.1186/s12302-020-00425-3.
80. Kotlarz N, McCord J, Collier D, Lea CS, Strynar M, Lindstrom AB, et al. 2020. Measurement of novel, drinking water-associated PFAS in blood from adults and children in Wilmington, North Carolina. Environ Health Perspect 128 (7 ):77005, PMID: , 10.1289/EHP6837.32697103
81. Song X, Vestergren R, Shi Y, Huang J, Cai Y. 2018. Emissions, transport, and fate of emerging per- and polyfluoroalkyl substances from one of the major fluoropolymer manufacturing facilities in China. Environ Sci Technol 52 (17 ):9694–9703, PMID: , 10.1021/acs.est.7b06657.30118215
82. Wang T, Vestergren R, Herzke D, Yu J, Cousins IT. 2016. Levels, isomer profiles, and estimated riverine mass discharges of perfluoroalkyl acids and fluorinated alternatives at the mouths of Chinese rivers. Environ Sci Technol 50 (21 ):11584–11592, PMID: , 10.1021/acs.est.6b03752.27689437
83. Ruan T, Lin Y, Wang T, Liu R, Jiang G. 2015. Identification of novel polyfluorinated ether sulfonates as PFOS alternatives in municipal sewage sludge in China. Environ Sci Technol 49 (11 ):6519–6527, PMID: , 10.1021/acs.est.5b01010.25961764
84. Zhang L, Liu J, Hu J, Liu C, Guo W, Wang Q, et al. 2012. The inventory of sources, environmental releases and risk assessment for perfluorooctane sulfonate in China. Environ Pollut 165 :193–198, PMID: , 10.1016/j.envpol.2011.09.001.21995873
85. EFSA CONTAM Panel (EFSA Panel on Contaminants in the Food Chain), Schrenk D, Bignami M, Bodin L, Chipman JK, Del Mazo J, et al. 2020. Risk to human health related to the presence of perfluoroalkyl substances in food. EFSA J 18 (9 ):e06223, PMID: , 10.2903/j.efsa.2020.6223.32994824
86. Blake BE, Cope HA, Hall SM, Keys RD, Mahler BW, McCord J, et al. 2020. Evaluation of maternal, embryo, and placental effects in CD-1 mice following gestational exposure to perfluorooctanoic acid (PFOA) or hexafluoropropylene oxide dimer acid (HFPO-DA or GenX). Environ Health Perspect 128 (2 ):27006, PMID: , 10.1289/EHP6233.32074459
87. Chou WC, Lin Z. 2021. Development of a gestational and lactational physiologically based pharmacokinetic (PBPK) model for perfluorooctane sulfonate (PFOS) in rats and humans and its implications in the derivation of health-based toxicity values. Environ Health Perspect 129 (3 ):37004, PMID: , 10.1289/EHP7671.33730865
88. Lehmann GM, LaKind JS, Davis MH, Hines EP, Marchitti SA, Alcala C, et al. 2018. Environmental chemicals in breast milk and formula: exposure and risk assessment implications. Environ Health Perspect 126 (9 ):96001, PMID: , 10.1289/EHP1953.30187772
89. Arnold C. 2019. Baby steps forward: recommendations for better understanding environmental chemicals in breast milk and infant formula. Environ Health Perspect 127 (6 ):64001, PMID: , 10.1289/EHP4804.31162980
90. Macheka-Tendenguwo LR, Olowoyo JO, Mugivhisa LL, Abafe OA. 2018. Per- and polyfluoroalkyl substances in human breast milk and current analytical methods. Environ Sci Pollut Res Int 25 (36 ):36064–36086, PMID: , 10.1007/s11356-018-3483-z.30382519
91. Jackson TW, Scheibly CM, Polera ME, Belcher SM. 2021. Rapid characterization of human serum albumin binding for per- and polyfluoroalkyl substances using differential scanning fluorimetry. Environ Sci Technol 55 (18 ):12291–12301, PMID: , 10.1021/acs.est.1c01200.34495656
92. Zhao S, Guo Y, Su Z, Cao W, Wu C, Chen QY. 2020. A series of deoxyfluorination reagents featuring OCF2 functional groups. Org Lett 22 (21 ):8634–8637, PMID: , 10.1021/acs.orglett.0c03238.33049136
93. Bhhatarai B, Gramatica P. 2011. Prediction of aqueous solubility, vapor pressure and critical micelle concentration for aquatic partitioning of perfluorinated chemicals. Environ Sci Technol 45 (19 ):8120–8128, PMID: , 10.1021/es101181g.20958003
94. Loccisano AE, Campbell JL Jr, Andersen ME, Clewell HJ III. 2011. Evaluation and prediction of pharmacokinetics of PFOA and PFOS in the monkey and human using a PBPK model. Regul Toxicol Pharmacol 59 (1 ):157–175, PMID: , 10.1016/j.yrtph.2010.12.004.21168463
95. Sheng N, Pan Y, Guo Y, Sun Y, Dai J. 2018. Hepatotoxic effects of hexafluoropropylene oxide trimer acid (HFPO-TA), a novel perfluorooctanoic acid (PFOA) alternative, on mice. Environ Sci Technol 52 (14 ):8005–8015, PMID: , 10.1021/acs.est.8b01714.29927593
96. Feng X, Chen X, Yang Y, Yang L, Zhu Y, Shan G, et al. 2021. External and internal human exposure to PFOA and HFPOs around a mega fluorochemical industrial park, China: differences and implications. Environ Int 157 :106824, PMID: , 10.1016/j.envint.2021.106824.34411760
97. Bischel HN, Macmanus-Spencer LA, Zhang C, Luthy RG. 2011. Strong associations of short-chain perfluoroalkyl acids with serum albumin and investigation of binding mechanisms. Environ Toxicol Chem 30 (11 ):2423–2430, PMID: , 10.1002/etc.647.21842491
98. Weaver YM, Ehresman DJ, Butenhoff JL, Hagenbuch B. 2010. Roles of rat renal organic anion transporters in transporting perfluorinated carboxylates with different chain lengths. Toxicol Sci 113 (2 ):305–314, PMID: , 10.1093/toxsci/kfp275.19915082
99. Yang CH, Glover KP, Han X. 2009. Organic anion transporting polypeptide (Oatp) 1a1-mediated perfluorooctanoate transport and evidence for a renal reabsorption mechanism of Oatp1a1 in renal elimination of perfluorocarboxylates in rats. Toxicol Lett 190 (2 ):163–171, PMID: , 10.1016/j.toxlet.2009.07.011.19616083
100. Shi G, Cui Q, Pan Y, Sheng N, Sun S, Guo Y, et al. 2017. 6:2 Chlorinated polyfluorinated ether sulfonate, a PFOS alternative, induces embryotoxicity and disrupts cardiac development in zebrafish embryos. Aquat Toxicol 185 :67–75, PMID: , 10.1016/j.aquatox.2017.02.002.28187362
101. Zhang H, Zhou X, Sheng N, Cui R, Cui Q, Guo H, et al. 2018. Subchronic hepatotoxicity effects of 6:2 chlorinated polyfluorinated ether sulfonate (6:2 Cl-PFESA), a novel perfluorooctanesulfonate (PFOS) alternative, on adult male mice. Environ Sci Technol 52 (21 ):12809–12818, PMID: , 10.1021/acs.est.8b04368.30256107
102. Kudo N, Katakura M, Sato Y, Kawashima Y. 2002. Sex hormone-regulated renal transport of perfluorooctanoic acid. Chem Biol Interact 139 (3 ):301–316, PMID: , 10.1016/s0009-2797(02)00006-6.11879818
103. Lou I, Wambaugh JF, Lau C, Hanson RG, Lindstrom AB, Strynar MJ, et al. 2009. Modeling single and repeated dose pharmacokinetics of PFOA in mice. Toxicol Sci 107 (2 ):331–341, PMID: , 10.1093/toxsci/kfn234.19005225
104. Butenhoff JL, Kennedy GL Jr, Hinderliter PM, Lieder PH, Jung R, Hansen KJ, et al. 2004. Pharmacokinetics of perfluorooctanoate in cynomolgus monkeys. Toxicol Sci 82 (2 ):394–406, PMID: , 10.1093/toxsci/kfh302.15470233
|
PMC009xxxxxx/PMC9994178.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36883823
EHP11112
10.1289/EHP11112
Research
Short-Term Association between Sulfur Dioxide and Mortality: A Multicountry Analysis in 399 Cities
O’Brien Edward 1
Masselot Pierre 2 3
Sera Francesco 4
Roye Dominic 5 6
Breitner Susanne 7 8
Ng Chris Fook Sheng 9 10
de Sousa Zanotti Stagliorio Coelho Micheline 11
Madureira Joana 12 13 14
Tobias Aurelio 9 15
Vicedo-Cabrera Ana Maria 16 17
Bell Michelle L. 18
Lavigne Eric 19 20
Kan Haidong 21 *
https://orcid.org/0000-0002-2271-3568
Gasparrini Antonio 2 3 22 *
MCC Collaborative Research Network †
1 London School of Hygiene & Tropical Medicine, London, UK
2 Department of Public Health Environments and Society, London School of Hygiene & Tropical Medicine, London, UK
3 Centre on Climate Change & Planetary Health, London School of Hygiene & Tropical Medicine, London, UK
4 Department of Statistics, Computer Science and Applications “G. Parenti,” University of Florence, Florence, Italy
5 Department of Geography, University of Santiago de Compostela, Santiago de Compostela, Spain
6 CIBER Epidemiología y Salud Pública, Madrid, Spain
7 Institute for Medical Information Processing, Biometry, and Epidemiology, Ludwig Maximilian University of Munich, Munich, Germany
8 Institute of Epidemiology, Helmholtz Zentrum München–German Research Center for Environmental Health, Neuherberg, Germany
9 Department of Global Health Policy, Graduate School of Medicine, University of Tokyo, Tokyo, Japan
10 School of Tropical Medicine and Global Health, Nagasaki University, Nagasaki, Japan
11 Department of Pathology, Faculty of Medicine, University of São Paulo, São Paulo, Brazil
12 Department of Environmental Health, Instituto Nacional de Saúde Dr Ricardo Jorge, Porto, Portugal
13 EPIUnit, Instituto de Saúde Publica, Universidade do Porto, Porto, Portugal
14 Laboratório para a Investigação Integrativa e Translacional em Saúde Populacional, Porto, Portugal
15 Institute of Environmental Assessment and Water Research, Spanish Council for Scientific Research, Barcelona, Spain
16 Institute of Social and Preventive Medicine, University of Bern, Bern, Switzerland
17 Oeschger Center for Climate Change Research, University of Bern, Bern, Switzerland
18 School of the Environment, Yale University, New Haven, Connecticut, USA
19 School of Epidemiology & Public Health, Faculty of Medicine, University of Ottawa, Ottawa, Ontario, Canada
20 Air Health Science Division, Health Canada, Ottawa, Ontario, Canada
21 Department of Environmental Health, School of Public Health, Fudan University, Shanghai, China
22 Centre for Statistical Methodology, London School of Hygiene & Tropical Medicine, London, UK
Address correspondence to Antonio Gasparrini, London School of Hygiene & Tropical Medicine, 15–17 Tavistock Place, WC1H 9SH, London, UK. Telephone: +44 (0)20 7927 2406. Email: [email protected]
08 3 2023
3 2023
131 3 03700216 2 2022
22 1 2023
30 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Epidemiological evidence on the health risks of sulfur dioxide (SO2) is more limited compared with other pollutants, and doubts remain on several aspects, such as the form of the exposure–response relationship, the potential role of copollutants, as well as the actual risk at low concentrations and possible temporal variation in risks.
Objectives:
Our aim was to assess the short-term association between exposure to SO2 and daily mortality in a large multilocation data set, using advanced study designs and statistical techniques.
Methods:
The analysis included 43,729,018 deaths that occurred in 399 cities within 23 countries between 1980 and 2018. A two-stage design was applied to assess the association between the daily concentration of SO2 and mortality counts, including first-stage time-series regressions and second-stage multilevel random-effect meta-analyses. Secondary analyses assessed the exposure–response shape and the lag structure using spline terms and distributed lag models, respectively, and temporal variations in risk using a longitudinal meta-regression. Bi-pollutant models were applied to examine confounding effects of particulate matter with an aerodynamic diameter of ≤10μm (PM10) and 2.5μm (PM2.5), ozone, nitrogen dioxide, and carbon monoxide. Associations were reported as relative risks (RRs) and fractions of excess deaths.
Results:
The average daily concentration of SO2 across the 399 cities was 11.7 μg/m3, with 4.7% of days above the World Health Organization (WHO) guideline limit (40 μg/m3, 24-h average), although the exceedances occurred predominantly in specific locations. Exposure levels decreased considerably during the study period, from an average concentration of 19.0 μg/m3 in 1980–1989 to 6.3 μg/m3 in 2010–2018. For all locations combined, a 10-μg/m3 increase in daily SO2 was associated with an RR of mortality of 1.0045 [95% confidence interval (CI): 1.0019, 1.0070], with the risk being stable over time but with substantial between-country heterogeneity. Short-term exposure to SO2 was associated with an excess mortality fraction of 0.50% [95% empirical CI (eCI): 0.42%, 0.57%] in the 399 cities, although decreasing from 0.74% (0.61%, 0.85%) in 1980–1989 to 0.37% (0.27%, 0.47%) in 2010–2018. There was some evidence of nonlinearity, with a steep exposure–response relationship at low concentrations and the risk attenuating at higher levels. The relevant lag window was 0–3 d. Significant positive associations remained after controlling for other pollutants.
Discussion:
The analysis revealed independent mortality risks associated with short-term exposure to SO2, with no evidence of a threshold. Levels below the current WHO guidelines for 24-h averages were still associated with substantial excess mortality, indicating the potential benefits of stricter air quality standards. https://doi.org/10.1289/EHP11112
Supplemental Material is available online (https://doi.org/10.1289/EHP11112).
* These authors contributed equally to this work.
† A full list of MCC Network authors is displayed in the “Acknowledgments” section.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Sulfur dioxide (SO2) is an important air pollutant linked with increased health risks.1,2 It originates largely from the combustion of fossil fuels to generate electricity and transportation.1,3,4 SO2 is also released during industrial processes, mainly the production of metals, such as copper, and other chemical plants, or from emissions from fuel combustion in shipping, whereas a very small percentage occurs naturally from volcanoes and fissures. In many developed nations, the desulfurization of cars and power plants has dramatically reduced emissions and the related population exposure, but the same cannot be said of many developing countries, where levels remain relatively very high.2 In 2005, the World Health Organization (WHO) published their recommendation for air quality standards, stating that the average SO2 concentration over a 24-h period should not exceed 20 μg/m3 or 500 μg/m3 over 10 min.5 The WHO guidelines were revised in 2021, with the 24-h limit increased to 40 μg/m3 following a new criterion based on the distribution of daily SO2 concentrations and the corresponding limit of annual averages.6
Short-term mortality risks of SO2 have been assessed in several ecological studies primarily based on time-series data. Early multicity studies were conducted in Europe,7,8 the United States,9 and East Asia.10 More recently, large investigations were performed in China, where the SO2 concentrations far exceed those of most high-income countries.11,12 A recent meta-analysis of 67 eligible studies systematically reviewed the evidence and provided pooled estimates of the association.13 When restricting the unit of analysis to 24-h averages of SO2, a 10-μg/m3 increment of SO2 was associated with an increase of 0.59% (95% CI: 0.46%, 0.71%) in all-cause mortality. The association remained when controlling for particulate matter (PM), but not for nitrogen dioxide (NO2) or ozone (O3). Moreover, there was no evidence of an exposure threshold below which no risk can be assumed.13
However, several gaps in knowledge still exist, as discussed in a comprehensive report from the U.S. Environmental Protection Agency (EPA).14 In addition to the uncertainty regarding the potential confounding effects from copollutants mentioned above, limited evidence is available on other aspects of the short-term association between exposure to SO2 and mortality risks. For instance, there is no conclusive evidence on the shape of the exposure–response relationships and possible nonlinearities or about the presence of more complex temporal dependencies and lagged associations. Further, it is still unclear whether results from studies from China are generalizable elsewhere, or if the risk shows geographical heterogeneity. More importantly, the published analyses assessed the association at relatively high exposure ranges. It is unclear to what extent the risk can be extrapolated at lower concentrations, for instance, below current air quality guidelines. This information is critical for revising air quality limits using evidence-based processes.
In this contribution, we address these limitations and present results from an investigation of the short-term mortality risks associated with SO2 exposure using data from 399 cities in 23 countries across the globe. The analysis used advanced study designs and statistical methods to characterize the associations of interest, whereas the large database and broad exposure contrasts offered enough statistical power to assess geographical variations and risks at low exposure levels.
Methods
Data Collection
The data were collected within the Multi-Country Multi-City (MCC) Collaborative Research Network, an international collaboration investigating environmental stressors and their impacts on human health.15 The MCC database has been used in previous publications that evaluated associations between air pollutants and mortality.15–20 Mortality data were gathered from the local health authorities and were represented as daily counts of all causes [International Classification of Diseases, Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death (ICD-9),21 codes 0–799], if available, or nonexternal [International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10),22 codes A00–R99] deaths. Nonexternal causes of death exclude intentional and unintentional injury, poisoning (including drug overdose), and complications of medical or surgical care.
The data set also contains 24-h average city-level concentrations of SO2, NO2, O3, carbon monoxide (CO), and PM with an aerodynamic diameter of ≤10μm (PM10) and 2.5μm (PM2.5), in addition to daily temperature. All pollutants were harmonized using the unit of micrograms per meter cubed, except CO, which was harmonized with milligrams per meter cubed. The country-specific data sets generally include all the major cities; however, this analysis was restricted to the 399 locations within 23 countries with SO2 data available and at least 365 d of measurement. The United States of America (USA) was divided into nine regions to account for the large heterogeneity in SO2 values, for a total of 31 areas. The geographical location of each city and the mean SO2 concentrations are displayed in Figure 1. Detailed information on data collection is reported in the Supplemental Material, “Information on country-specific datasets.”
Figure 1. Geographical location of the 399 urban areas and related average annual concentrations of SO2 (in μg/m3) within the study period 1980–2018. Data can be found in Table S1. Note: SO2, sulfur dioxide.
Figure 1 is a world map, depicting the 399 urban areas and related average annual concentrations of Sulfur dioxide within the study period from 1980 to 2018.
Statistical Methods
Main Model
We applied a two-stage procedure to analyze the short-term association between SO2 and mortality. In the first stage, we performed city-specific time-series analyses using a quasi-Poisson generalized linear model with distributed lag terms.23,24 The city-specific model included an indicator for the day of the week to account for within-week variation and a natural spline function with 7 degrees of freedom per year to control for long-term trends and seasonal variations. Air temperature was modeled with a distributed lag nonlinear model (DLNM), composed of a quadratic B-spline with three knots placed at the 10th, 75th, and 90th percentiles for the exposure–response and a step function with strata lags of 0 and 1–3 d for the lag–response. In the main model, SO2 was modeled assuming a linear exposure–response relationship of the moving average computed over lag 0–3 d.
In the second stage, we combined the city-specific estimates using a multilevel random-effect meta-analysis fitted with restricted maximum likelihood (REML) and nested random effects defined by city and country.25 The pooled estimate represents the global average SO2-mortality association, whereas city and country-specific estimates were derived as best linear unbiased predictions (BLUPs) at the corresponding aggregation level. The BLUPs use information from pooled associations to make more accurate location-specific estimates by borrowing information from the whole sample, especially for cities/countries with higher uncertainty, while at the same time accounting for heterogeneity in risks.25,26 All estimates are reported as the relative risk (RR) for a 10-μg/m3 increase in SO2, with corresponding 95% confidence intervals (CIs). Heterogeneity was reported as I2 statistics and tested with the Cochran’s Q test.25–27
Secondary Analyses
We performed a series of secondary analyses. First, we explored potential nonlinear exposure–response shapes and more complex lag structures of the SO2–mortality relationship, extending the model first using a quintic polynomial, and then using a distributed lag model (DLM) with a natural spline with knots at lag 1 and 3 plus an intercept over lag 0–7 d. The polynomial parameterization was adopted to decrease the sensitivity of the estimates to different ranges of SO2 concentrations because polynomial terms are not local and are, instead, defined across the whole variable range. In both extensions, city-specific estimates of the multiparameter associations were pooled using a multivariate multilevel meta-analysis.28
Second, we evaluated possible changes in risk over time by subsetting the city-specific data and performing the first-stage model in multiple subperiods, splitting the time series into ∼5-y intervals. City and period-specific estimates were pooled in a longitudinal multilevel meta-regression using time (defined as the midyear of each subperiod) as a continuous fixed-effect term.28
Third, we assessed the potential confounding effect of other pollutants in bi-pollutant models, where PM10, PM2.5, O3 (8-h daily maximum), NO2, and CO entered the model linearly, using a moving average of lag 0–3 d. These choices were informed by previous studies.12,29,30 Only one other pollutant was controlled for at a time because of the high correlation between pollutants.14,31 These models were fitted both with and without adjustment in the subset of cities, providing measurements for both pollutants. Information on the levels of copollutants across cities and countries has been provided in previous articles.16–20
Computation of Excess Mortality
Finally, using the main model, we computed the excess mortality associated in the short term with exposure to SO2 in each city, using a previously described method.32 Briefly, the cumulative RR within lag 0–3 was applied to compute the excess daily deaths, adopting a forward perspective using the standard formula (1−exp(−βj(xjt−c)+))×djt for continuous exposure, as in previous analyses.17 In the formula, βj represents the log-RR for a unit increase in SO2, defined as the country-specific BLUP for city j, and xjt and djt are the corresponding SO2 levels at day t and the average daily mortality in the same and next 3 d, respectively. The term (xjt−c)+ represents the exceedance in SO2 concentration above a limit c. We used c=0 and c=40 to compute the burden attributable to short-term exposure to SO2 in total and above the WHO guideline, respectively. The results are reported as fraction of excess deaths, both in total and for levels below the WHO guideline, together with 95% empirical CIs (eCIs).
All analyses were conducted in R (version 4.2.2; R Development Core Team), using the dlnm and mixmeta packages. The R code for the original analysis and for performing a reproducible example using simulated data is available in a GitHub repository (https://github.com/gasparrini/MCC-SO2).
Results
Descriptive Analysis
The analysis included 43,729,018 deaths across 399 cities, in 23 countries (separating the USA into nine regions), with an average period of 14.5 y. Table 1 shows the total deaths, number of cities, period of analysis, and levels of SO2 across cities in each country. The list of cities, together with basic information on study periods and SO2 exposure levels, is provided in Table S1. The SO2 exposure was widely heterogeneous both within and between countries, with an average across cities of 11.7 μg/m3 and an average 5th–95th percentile range of 2.6–29.3 μg/m3 (see the map in Figure 1 for the geographical distribution). The highest mean concentrations were found in cities of China and Iran (50.1 and 98.9 μg/m3, respectively), and the lowest concentrations were in Portugal and Estonia (2.6 and 3.5 μg/m3, respectively). Across the whole set of cities and periods, 4.7% of days registered concentrations higher than the WHO limits of 40 μg/m3.6
Table 1 Descriptive statistics reported by country (with the United States separated into nine regions): number of cities, total days, and deaths in the time series, distribution of sulfur dioxide (SO2) across cities, and percentage of days with SO2 concentrations >40 μg/m3.
Country Cities (n) Period City-days (n) Deaths (n) Average mean and 5th–95th percentile range of SO2 (μg/m3) Percentage of days >40 μg/m3 (%)a
Canada 24 2000–2015 109,953 1,741,439 5.6 (0.3–17.7) 1.1
USA-Central 30 1985–2006 216,457 3,724,610 19.6 (3.5–49.0) 9.9
USA-NE Central 14 1985–2006 76,137 1,518,119 10.2 (1.0–29.2) 2.4
USA-Northeast 42 1985–2006 291,202 6,912,465 19.8 (3.6–49.8) 10.2
USA-Northwest 5 1985–2006 19,994 394,004 14.6 (3.8–32.3) 2.3
USA-NW Central 2 1985–2006 10,894 69,007 2.8 (0.2–9.7) 0.4
USA-South 17 1985–2006 105,338 1,513,772 8.7 (0.9–26.1) 2.7
USA-Southeast 29 1985–2006 173,040 2,945,570 9.5 (1.5–25.9) 2.2
USA-Southwest 9 1985–2006 50,766 794,800 8.0 (0.8–22.6) 1.6
USA-West 15 1985–2006 81,604 3,367,343 5.0 (0.2–16.9) 0.2
Puerto Rico 1 2009–2016 2,877 26,161 4.6 (0.0–9.4) 0.7
Brazil 1 1997–2011 5,076 909,305 12.3 (4.0–26.3) 0.6
Colombia 1 1998–2013 5,800 423,344 20.9 (5.3–41.6) 6.4
Ecuador 1 2014–2018 1,819 44,369 16.7 (8.0–26.0) 0.2
Peru 1 2010–2014 1,489 148,775 12.8 (3.4–27.8) 0.1
Estonia 4 2002–2018 18,239 96,455 2.6 (0.2–10.1) 0.6
Finland 1 1994–2014 7,663 153,166 9.1 (2.1–25.9) 1.0
UK 31 1990–2016 157,489 3,823,644 8.6 (0.5–23.6) 1.9
Czech Republic 1 1994–2009 5,835 213,706 14.6 (2.0–55.7) 8.6
Germany 12 1993–2015 70,542 2,098,705 7.9 (1.9–23.8) 2.0
Romania 8 2008–2016 14,964 221,816 8.4 (3.2–17.7) 0.4
Switzerland 4 1995–2013 24,229 151,898 5.6 (0.7–16.9) 0.2
Portugal 6 1995–2018 28,860 847,377 3.5 (0.5–9.9) 0.3
Spain 48 2002–2014 197,008 1,480,869 5.5 (2.6–10.4) 0.1
Iran 1 2002–2015 5,025 683,739 98.9 (17.0–279.1) 70.7
China 15 1996–2015 23,139 1,181,405 50.1 (14.8–124.2) 45.8
Japan 46 1980–2015 133,940 4,382,591 6.0 (2.2–12.6) 0.3
South Korea 7 1999–2015 42,979 1,658,788 15.5 (7.5–29.1) 1.3
Thailand 18 1999–2008 52,980 722,911 10.1 (3.0–20.7) 2.3
Taiwan 3 1994–2014 22,981 1,208,118 15.4 (6.1–32.1) 3.6
Australia 2 2000–2009 6,609 270,751 5.1 (0.8–18.8) 0.8
All MCC countriesb 399 1980–2018 1,964,928 43,729,018 11.7 (2.6–29.3) 4.7
Note: The analysis includes data from 399 cities within the study period 1980–2018 from the Multi-Country Multi-City (MCC) Collaborative Research Network.
a Current limit of daily concentration of SO2 in the World Health Organization guidelines.6
b Data shown represents the statistics for all the areas included above.
Figure 2 displays the distribution of city-specific average SO2 concentrations over the years, revealing a strong attenuation in exposure levels within the study period, although the comparison should account for the different sample of cities/countries contributing to each interval. Actual figures of SO2 levels and percentage exceedances of WHO limits aggregated by decade are reported in Table S2. Results show that the average daily concentration decreased from 19.0 μg/m3 in 1980–1989 to 6.3 μg/m3 in 2010–2018, and the corresponding percentage of days >40 μg/m3 from 11.7% to 1.3%.
Figure 2. Box plot of the distribution of the average concentration of SO2 (in μg/m3) across cities for each year. The horizontal line identifies the current limit of daily concentration of SO2 in the WHO guidelines (40 μg/m3). The y-axis is represented in a logarithmic scale. The analysis includes data from 399 cities within the study period 1980–2018. Note that a different set of countries contributes to each study period (see Table 1 for details). Note: SO2, sulfur dioxide; WHO, World Health Organization.
Figure 2 is a box and whiskers plot, plotting average sulfur dioxide concentration (micrograms per meter cubed), ranging from 1 to 2 in unit increment, 2 to 5 in increments of 3, 5 to 10 in increments of 5, 10 to 20 in increments of 10, 20 to 50 in increments of 30, 50 to 100 in increments of 50, and 100 to 200 in increments of 100 (y-axis) across year, ranging from 1980 to 2016 in increments of 4 years (x-axis).
Association between SO2 and Mortality
The risk associations estimated from the main model using a linear exposure–response relationship and a moving average of lag 0–3 are illustrated in Figure 3, with the pooled RR and related country-specific BLUPs (see Table S3 for the numeric data). On average across all cities and countries, each 10-μg/m3 increase of SO2 was associated with an RR of mortality of 1.0045 (95% CI: 1.0019, 1.0070). Although country-specific estimates were less precise, there is evidence of substantial heterogeneity (I2=45.8%, Cochran Q p<0.001), with the RRs ranging from 0.9968 (95% CI: 0.9883, 1.0054) in Finland to 1.0234 (95% CI: 1.0168, 1.0300) in Brazil, and a few nonsignificant negative estimates.
Figure 3. Country-specific and pooled relative risks (RRs, with 95% CIs) for mortality corresponding to a 10-μg/m3 increase in SO2 over lag 0–3 d. The analysis includes data from 399 cities within the study period 1980–2018. Data can be found in Table S3. Note: CI, confidence interval; SO2, sulfur dioxide.
Figure 3 is a forest plot, plotting (bottom to top) Pooled; Australia (Australia); Taiwan, Thailand (South-East Asia); South Korea, Japan, China (East Asia); Iran (Middle East Asia); Spain, Portugal (South Europe); Switzerland, Romania, Germany, Czech Republic (Central Europe); United Kingdom, Finland, Estonia (North Europe); Peru, Ecuador, Colombia, Brazil (South America); Puerto Rico (Central America); United States of America West, United States of America southeast, United States of America southwest, United States of America south, United States of America northwest central, United States of America northwest, United States of America northeast, United States of America northeast central, United States of America central, and Canada (North America) (y-axis) across Relative risks for 10 micrograms per meter cubed increase in sulfur dioxide, ranging from 0.99 to 1.03 in increments of 0.01 (x-axis).
Exposure and Lag–Response Relationships
Results from secondary analyses are displayed in Figure 4. Figure 4A shows the estimate of the pooled SO2-mortality relationship from the flexible model allowing nonlinear exposure–response associations. The graph is augmented with the estimated log-linear relationship from the main model and a bar representing the number of cities with SO2 measurements within the related exposure range. The figure indicates some evidence of nonlinearity, with a steep increase in risk and an attenuation at high concentrations. It is unclear if this supralinear shape resulted from lower risks at high SO2 exposures or if it is attributable to a different sample of countries contributing at various ranges, given that only 23.3% of cities were exposed to levels >150 μg/m3. In any case, the nonlinear parameterization confirms the evidence of mortality risks for exposures below the WHO limit of 40 μg/m3. A comparison of country-specific exposure–response relationships (as BLUPs) estimated using the linear and nonlinear models is presented in Figure S1. The results of the second modeling extension with the application of a DLM over lag 0–7 to assess the lag structure are displayed in Figure 4B. The graph suggests no risk of same-day exposure to SO2, with the risk then increasing in the next 3–4 d. This analysis indicated that the main model with the moving average of lag 0–3 can capture these lagged associations and can provide valid estimates of the association.
Figure 4. Secondary analysis on the short-term association between SO2 (per 10-μg/m3 increase) and mortality. (A) Pooled exposure–response curve obtained using a linear term (dashed line) and a quintic polynomial (continuous line, with 95% confidence intervals), with a bar representing the percentage of studies contributing to the specific exposure range. (B) Pooled lag–response curve obtained using a natural spline with knots at lags 1 and 3, plus intercept. (C) Analysis of temporal variation of the pooled relative risk (RR) associated with a 10-μg/m3 increase in SO2 over lag 0–3 d. Shaded areas represent the 95% confidence intervals. The analysis includes data from 399 cities within the study period 1980–2018. Summary data on city-specific exposure distributions can be found in Table S1. Note: SO2, sulfur dioxide.
Figures 4A, 4B, and 4C are ribbon plus line graphs, plotting relative risks, ranging from 0.98 to 1.08 in increments of 0.02, relative risks for 10 micrograms per meter cubed increase in sulfur dioxide, ranging from 0.999 to 1.002 in increments of 0.001, and relative risks for 10 micrograms per meter cubed increase in sulfur dioxide, ranging from 0.998 to 1.010 in increments of 0.002 (y-axis) across sulfur dioxide (micrograms per meter cubed), ranging from 0 to 200 in increments of 50, Lag (days), ranging from 0 to 7 in unit increments, and year, ranging from 1980 to 2020 in increments of 10 years (x-axis) for linear and nonlinear. A color scale depicts percentage of contributing locations ranges from 0 to 100 percent in increments of 20 for Figure 4A.
Analysis of Temporal Variation in Risk
The extension of the two-stage design for assessing potential temporal changes in risk involved the analysis by subperiod and the pooling of estimates using time as a continuous meta-predictor. The results are reported in Figure 4C, which shows the pooled mortality RR for a 10-μg/m3 increase in SO2 along the years predicted from the longitudinal meta-analytical model. The graph suggests little evidence of variation in time in the short-term association, with a p-value of the time term equal to 0.67.
Bi-Pollutant Analysis
Results from bi-pollutant analyses are provided in Table 2, with information on the number of cities contributing to the analysis of each copollutant provided in Table S4. The comparison suggests that an independent risk associated with SO2 remained even after adjustment for each of the five other pollutants, although with some variations indicative of partial confounding. Specifically, the estimated risks seemed to attenuate after controlling for PM10 and NO2, whereas they increased when including PM2.5 in the model. Results were only negligibly affected by control for O3 and CO.
Table 2 Relative risk with 95% confidence interval) [RR (95% CI)] associated with a 10-μg/m3 increase in sulfur dioxide (SO2, in μg/m3) over lag 0–3 d with and without adjustment for each copollutant in selected cities with both measurements within the study period 1980–2018.
Copollutant Cities (n) RR±95% CI
Without adjustment With adjustment
PM10 265 1.0045 (1.0017, 1.0074) 1.0028 (1.0004, 1.0052)
PM2.5 217 1.0028 (1.0005, 1.0050) 1.0056 (1.0028, 1.0084)
O3 8h 309 1.0044 (1.0023, 1.0065) 1.0040 (1.0017, 1.0062)
NO2 358 1.0046 (1.0019, 1.0073) 1.0032 (1.0012, 1.0052)
CO 302 1.0038 (1.0016, 1.0060) 1.0036 (1.0023, 1.0048)
Note: Number of cities included in the bi-pollutant models are reported in Table S4. CO, carbon monoxide; NO2, nitrogen dioxide; O3, ozone; PM2.5, particulate matter with an aerodynamic diameter of 2.5μm; PM10, particulate matter with an aerodynamic diameter of ≤10μm.
Excess Mortality
Finally, Table 3 depicts the excess mortality fraction associated with SO2 exposure computed from the main model, assuming a linear relationship both by decade and within the whole study period of each area. Overall across the 399 cities, SO2 was found to be associated with an excess increase of 0.50% (95% eCI: 0.42%, 0.57%) of the total deaths. This fraction showed a strong decrease in time and variation across regions, consistent with the reduction in SO2 levels and the geographical variation in risk described above. Specifically, the excess mortality overall decreased from 0.74% (0.61%, 0.85%) in 1980–1989 to 0.37% (0.27%, 0.47%) in 2010–2018. The corresponding quota of excess deaths attributable to levels <40 μg/m3 is shown in Table S5, indicating that 93.8% of the excess was due to exposures below the WHO limits on average across the study period.
Table 3 Excess mortality fraction with 95% empirical confidence interval [% (95% eCI)] attributable to short-term exposure to sulfur dioxide (SO2, per 10 μg/m3) by country (separating the USA into nine regions) and decade.
Country/region 1980–1989 1990–1999 2000–2009 2010–2019 Full period
Canada — — 0.24 (0.12, 0.37) 0.11 (0.05, 0.17) 0.19 (0.10, 0.29)
USA-Central 0.82 (0.56, 1.05) 0.57 (0.40, 0.73) 0.40 (0.28, 0.51) — 0.58 (0.40, 0.73)
USA-NE Central 0.71 (0.28, 1.16) 0.50 (0.19, 0.82) 0.35 (0.12, 0.58) — 0.50 (0.19, 0.82)
USA-Northeast 1.09 (0.79, 1.36) 0.77 (0.56, 0.97) 0.55 (0.40, 0.70) — 0.78 (0.57, 0.98)
USA-Northwest 0.15 (−0.52, 0.80) 0.15 (−0.48, 0.73) 0.06 (−0.32, 0.39) — 0.13 (−0.46, 0.67)
USA-NW Central 0.20 (−0.28, 0.65) 0.22 (−0.31, 0.72) 0.08 (−0.11, 0.25) — 0.16 (−0.22, 0.53)
USA-South 0.63 (0.45, 0.83) 0.49 (0.35, 0.64) 0.33 (0.22, 0.44) — 0.47 (0.33, 0.61)
USA-Southeast 0.11 (−0.06, 0.26) 0.07 (−0.02, 0.17) 0.05 (−0.01, 0.11) — 0.07 (−0.02, 0.15)
USA-Southwest 0.55 (0.21, 0.88) 0.47 (0.18, 0.74) 0.26 (0.09, 0.43) — 0.40 (0.16, 0.63)
USA-West 1.07 (0.86, 1.29) 0.80 (0.66, 0.94) 0.76 (0.64, 0.88) — 0.84 (0.70, 0.99)
Puerto Rico — — 0.04 (−0.32, 0.40) 0.05 (−0.41, 0.50) 0.05 (−0.40, 0.49)
Brazil — 3.75 (2.69, 4.70) 2.85 (2.04, 3.57) 1.69 (1.21, 2.12) 2.87 (2.05, 3.60)
Colombia — 0.14 (−2.02, 2.26) 0.10 (−1.55, 1.73) 0.03 (−0.49, 0.56) 0.09 (−1.32, 1.47)
Ecuador — — — 0.76 (−1.07, 2.55) 0.76 (−1.07, 2.55)
Peru — — — 0.83 (−0.41, 2.17) 0.83 (−0.41, 2.17)
Estonia — — 0.08 (−0.12, 0.28) 0.05 (−0.05, 0.14) 0.06 (−0.07, 0.19)
Finland — −0.35 (−1.25, 0.59) −0.28 (−1.00, 0.47) −0.27 (−0.98, 0.46) −0.29 (−1.07, 0.50)
UK — −0.46 (−0.79, −0.14) −0.14 (−0.21, −0.08) −0.06 (−0.10, −0.02) −0.24 (−0.39, −0.09)
Czech Republic — −0.03 (−1.65, 1.69) −0.01 (−0.36, 0.38) — −0.01 (−0.88, 0.91)
Germany — 0.38 (0.16, 0.59) 0.15 (0.07, 0.22) 0.09 (0.02, 0.15) 0.24 (0.11, 0.35)
Romania — — 0.54 (−0.27, 1.31) 0.39 (−0.13, 0.89) 0.42 (−0.16, 0.98)
Switzerland — 0.82 (0.15, 1.57) 0.41 (0.14, 0.71) 0.21 (0.07, 0.36) 0.48 (0.13, 0.86)
Portugal — 0.42 (0.04, 0.81) 0.27 (0.07, 0.46) 0.11 (0.04, 0.18) 0.24 (0.06, 0.40)
Spain — — 0.82 (0.60, 1.04) 0.60 (0.46, 0.74) 0.73 (0.55, 0.92)
Iran — — −0.37 (−4.30, 3.31) −0.13 (−1.45, 1.16) −0.26 (−3.02, 2.35)
China — 0.58 (0.11, 1.07) 1.55 (1.12, 1.98) 1.68 (0.79, 2.55) 1.48 (1.04, 1.95)
Japan 0.17 (−0.18, 0.51) 0.11 (−0.10, 0.31) 0.06 (−0.04, 0.16) 0.11 (0.02, 0.19) 0.11 (−0.05, 0.25)
South Korea — 2.32 (1.72, 2.91) 1.49 (1.09, 1.88) 1.37 (1.00, 1.74) 1.49 (1.09, 1.88)
Thailand — 1.21 (0.80, 1.61) 1.07 (0.74, 1.40) — 1.08 (0.75, 1.41)
Taiwan — −0.13 (−0.87, 0.53) −0.08 (−0.54, 0.35) −0.06 (−0.42, 0.27) −0.09 (−0.59, 0.37)
Australia — — 0.20 (−0.10, 0.50) — 0.20 (−0.10, 0.50)
All MCC countries 0.74 (0.61, 0.85) 0.46 (0.38, 0.53) 0.50 (0.40, 0.60) 0.37 (0.27, 0.47) 0.50 (0.42, 0.57)
Note: The analysis includes data from the Multi-Country Multi-City (MCC) Collaborative Research Network for 399 cities within the study period 1980–2018. Estimates based on the main model assuming a linear exposure–response relationship and a moving average of lag 0–3 d. —, not applicable.
Discussion
To our knowledge, this study represents the most extensive epidemiological assessment of the short-term mortality risks associated with exposure to SO2, investigating the relationship using a large database that includes almost 44 million deaths from 399 cities in 23 countries across 5 continents. We found an overall increased short-term risk, with an RR=1.0045 (95% CI: 1.0019, 1.0070) per 10-μg/m3 increase in SO2, although with evidence of heterogeneity across countries and cities. This translated to an annual excess corresponding to 0.5% of the total mortality on average across the study period. Stratified by decades, the analysis indicates a strong decrease in the health impacts, in line with the reduction in the concentration levels of SO2 although this result should be interpreted with caution given the differential temporal coverage across countries. The application of more flexible models suggested some evidence of nonlinearity, with steeper RR estimates at low exposure levels, and a complex lag structure with the maximal risk arising 1–3 d after exposure. Independent associations were still measurable even after controlling for copollutants, and there was no evidence that the risk associated with a given exposure level had changed over time.
This study provides an essential contribution to the literature on the mortality risks associated with short-term SO2 exposure. The pooled RR estimates are consistent with previous epidemiological studies, although slightly lower when compared with figures published in the multicity studies and the meta-analysis described in the introduction.7–13 The difference can be due to the selection of locations, which in our study represent a broader sample and a wider geographical area. Our results strengthen the evidence on the association of short-term SO2 exposure with total mortality, which was determined as suggestive but not sufficient to infer a causal relationship in the U.S. EPA report mentioned above.14 One of the main reasons described in the report for such a conclusion was the limited knowledge about the potential confounding by other pollutants. Our analysis confirms that risk estimates were somewhat sensitive to control for copollutants but that there was still strong evidence of the association when PM10, PM2.5, O3, NO2, and CO were each included in the model. In addition, our contribution addresses other knowledge gaps on the association, such as the shape of the exposure–response relationship, the lag patterns, and possible temporal variations in risks.
There are several biological pathways and mechanisms through which SO2 can lead to higher mortality risks. The body of evidence from animal, experimental, and epidemiological studies is deemed sufficient to suggest a causal relationship between exposure and effects on the respiratory system.13 Specifically, short-term exposure to SO2 is known to induce neural reflex responses, release of inflammatory mediators, and modulation of allergic inflammation, leading to several end points that include bronchoconstriction and increased airway responsiveness.33 The impacts are known to be more severe in susceptible individuals, such as those with asthma.34 Previous studies have also assessed potential risks for cardiovascular outcomes. SO2 was shown to cause drops in measures of baroreflex sensitivity and cardiac vagal control, as well as increases in plasma fibrinogen, oxidative stress, and blood viscosity in young adults.35 However, the body of evidence is not sufficient and consistent enough to establish a causal relationship.14
Another important result of this study is the evidence of the potential public health benefits achievable with more stringent air quality policies. As stated above, we found that short-term exposure to SO2 was associated with an excess risk of mortality. Although the impact has decreased in time consistent with the reduction in concentration levels, exposure to SO2 was still linked with a considerable number of excess deaths in recent years. More importantly, the majority of the additional deaths were linked with exposures sustained on days in which SO2 levels were at or below the current daily WHO threshold, which was recently increased from 20 to 40 μg/m3. These results, in line with previous multicountry studies on other pollutants,16–20 indicate the presence of a considerable risk even at low exposure ranges that is also associated with a substantial health burden in countries that comply with the current WHO guidelines. These results therefore support the efforts to enforce national and international policy guidelines and to consider the opportunity to revise the limits downward.
An important aspect of this assessment is the application of state-of-the-art study design and statistical methods on a large multicountry database. Given the large sample of locations covering such a wide geographic area, we were able to obtain consistent evidence using a uniform model of a short-term association between exposure to SO2 and all-cause mortality. The use of advanced analytical techniques applied in the first stage, including time-series regression and distributed lag models, allows a nuanced characterization of complex exposure–lag–response relationships. Similarly, the analysis of hundreds of locations and the use of extended meta-analytical techniques provide both pooled and city- and country-specific estimates and offer a comprehensive geographical and temporal comparison across various regions of the globe.
Some limitations must be acknowledged. First, although we were able to provide risk summaries across four inhabited continents, our results should not be considered truly global estimates given that some areas, such as Africa, South America, and the Middle East, were underrepresented or not assessed. Moreover, the study was restricted to urban populations, with several countries represented by a small number of cities, and therefore it cannot be entirely representative of the risks across whole populations. Notably, although the assessment offers a consistent comparison of risks across locations and found important differences in risks and impacts across countries, we did not attempt to characterize such heterogeneity. Part of this variability can be due to systemic differences concerning measurements from atmospheric monitors (type of station, proximity to the study area), study area boundaries, temporal coverage, and data processing, whereas the other part can be related to actual differences in susceptibility. This will be a topic for future research. In addition, the extension using polynomial terms indicates a degree of nonlinearity in the exposure–response relationship. However, it is challenging to disentangle to what extent this is due to heterogeneous risks in areas with high exposure ranges, and the main results are therefore reported from a model assuming a linear association. Further research on the shape of the exposure–response relationship and the risk at high SO2 levels is needed. Furthermore, although we assessed potential confounding effects in bi-pollutant models, we did not extend the analysis to evaluate possible synergistic effects between SO2 and other pollutants. Finally, the study was conducted using aggregated time-series data, preventing a more refined analysis of potential biological mechanisms and differential susceptibility patterns at the individual level.
Conclusions
This large multicountry study provides evidence of an independent short-term association between exposure to SO2 and all-cause mortality. The assessment indicates that even if current air quality guidelines for SO2 were enforced, many deaths would still occur, and additional health benefits could be attained by further lowering existing limits. These findings have important implications for the design of future health and environmental policy actions. More generally, they can contribute to the design and implementation of mitigation strategies to reduce the environmental risks and impacts on health in the context of climate change.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This study was supported by the Medical Research Council UK [MR/R013349/1 (to A.G.)] and the European Union’s Horizon 2020 Project Exhaustion [820655 (to A.G.)]. Additional grants supported individual researchers: H.K. was supported by the National Natural Science Foundation of China (92043301), J.M. was supported by a fellowship of Fundação para a Ciência e a Tecnlogia (SFRH/BPD/115112/2016), and A.T. was supported by Ministerio de Ciencia e Innovacion/Agencia Estatal de Investigación (MCIN/AEI/10.13039/501100011033; CEX2018-000794-S).
The Multi-City Multi-Country Collaborative Research Network group authorship includes G. Carrasco, B.-Y. Chen, A. Entezari, Y. Guo, Y.L. Guo, M. Hashizume, I.-H. Holobaca, C. Íñiguez, J.J.K. Jaakkola, H. Kim, W. Lee, S. Li, M. Maasikmets, F. Mayvaneh, B. Nunes, H. Orru, S. Osorio, M.S. Ragettli, N. Ryti, P.H.N. Saldiva, A. Schneider, J. Schwartz, S. Tong, and A. Zanobetti.
==== Refs
References
1. Guttikunda SK, Tang Y, Carmichael GR, Kurata G, Pan L, Streets DG, et al. 2005. Impacts of Asian megacity emissions on regional air quality during spring 2001. J Geophys Res 110 (D20 ):D20301, 10.1029/2004JD004921.
2. Greenstone M. 2004. Did the Clean Air Act cause the remarkable decline in sulfur dioxide concentrations? J Environ Econ Manage. 47 (3 ):585–611, 10.1016/j.jeem.2003.12.001.
3. WHO (World Health Organization). Air pollution. https://www.who.int/health-topics/air-pollution [accessed 8 August 2019].
4. Bell ML, Samet JM, Dominici F. 2004. Time-series studies of particulate matter. Annu Rev Public Health 25 (1 ):247–280, PMID: , 10.1146/annurev.publhealth.25.102802.124329.15015920
5. WHO. 2006. WHO Air Quality Guidelines for Particulate Matter, Ozone, Nitrogen Dioxide and Sulfur Dioxide: Global Update 2005: Summary of Risk Assessment. https://apps.who.int/iris/handle/10665/69477 [accessed 25 November 2021].
6. WHO. 2021. WHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide. https://apps.who.int/iris/handle/10665/345329 [accessed 25 November 2021].
7. Katsouyanni K, Touloumi G, Spix C, Schwartz J, Balducci F, Medina S, et al. 1997. Short term effects of ambient sulphur dioxide and particulate matter on mortality in 12 European cities: results from time series data from the APHEA project. BMJ 314 (7095 ):1658–1663, PMID: , 10.1136/bmj.314.7095.1658.9180068
8. Biggeri A, Baccini M, Bellini P, Terracini B. 2005. Meta-analysis of the Italian studies of short-term effects of air pollution (MISA), 1990–1999. Int J Occup Environ Health 11 (1 ):107–122, PMID: , 10.1179/oeh.2005.11.1.107.15859200
9. Moolgavkar SH, McClellan RO, Dewanji A, Turim J, Luebeck EG, Edwards M. 2013. Time-series analyses of air pollution and mortality in the United States: a subsampling approach. Environ Health Perspect 121 (1 ):73–78, PMID: , 10.1289/ehp.1104507.23108284
10. Kan H, Wong CM, Vichit-Vadakan N, Qian Z, PAPA Project Teams. 2010. Short-term association between sulfur dioxide and daily mortality: the Public Health and Air Pollution in Asia (PAPA) study. Environ Res 110 (3 ):258–264, PMID: , 10.1016/j.envres.2010.01.006.20122685
11. Chen R, Huang W, Wong CM, Wang Z, Thach TQ, Chen B, et al. 2012. Short-term exposure to sulfur dioxide and daily mortality in 17 Chinese cities: the China Air Pollution and Health Effects Study (CAPES). Environ Res 118 :101–106, PMID: , 10.1016/j.envres.2012.07.003.22831556
12. Wang L, Liu C, Meng X, Niu Y, Lin Z, Liu Y, et al. 2018. Associations between short-term exposure to ambient sulfur dioxide and increased cause-specific mortality in 272 Chinese cities. Environ Int 117 :33–39, PMID: , 10.1016/j.envint.2018.04.019.29715611
13. Orellano P, Reynoso J, Quaranta N. 2021. Short-term exposure to sulphur dioxide (SO2) and all-cause and respiratory mortality: a systematic review and meta-analysis. Environ Int 150 :106434, PMID: , 10.1016/j.envint.2021.106434.33601225
14. U.S. EPA (U.S. Environmental Protection Agency). 2017. Integrated Science Assessment (ISA) for Sulfur Oxides—Health Criteria (Final Report, Dec 2017). EPA/600/R-17/451. Washington, DC: U.S. EPA.
15. Multi-Country Multi-City (MCC) Collaborative Research Network. https://mccstudy.lshtm.ac.uk/ [accessed 28 August 2022].
16. Liu C, Chen R, Sera F, Vicedo-Cabrera AM, Guo Y, Tong S, et al. 2019. Ambient particulate air pollution and daily mortality in 652 cities. N Engl J Med 381 (8 ):705–715, PMID: , 10.1056/NEJMoa1817364.31433918
17. Vicedo-Cabrera AM, Sera F, Liu C, Armstrong B, Milojevic A, Guo Y, et al. 2020. Short term association between ozone and mortality: global two stage time series study in 406 locations in 20 countries. BMJ 368 :m108, PMID: , 10.1136/bmj.m108.32041707
18. Meng X, Liu C, Chen R, Sera F, Vicedo-Cabrera AM, Milojevic A, et al. 2021. Short term associations of ambient nitrogen dioxide with daily total, cardiovascular, and respiratory mortality: multilocation analysis in 398 cities. BMJ 372 :n534, PMID: , 10.1136/bmj.n534.33762259
19. Zhao Q, Guo Y, Ye T, Gasparrini A, Tong S, Overcenco A, et al. 2021. Global, regional, and national burden of mortality associated with non-optimal ambient temperatures from 2000 to 2019: a three-stage modelling study. Lancet Planet Health 5 (7 ):e415–e425, PMID: , 10.1016/S2542-5196(21)00081-4.34245712
20. Chen K, Breitner S, Wolf K, Stafoggia M, Sera F, Vicedo-Cabrera AM, et al. 2021. Ambient carbon monoxide and daily mortality: a global time-series study in 337 cities. Lancet Planet Health 5 (4 ):e191–e199, PMID: , 10.1016/S2542-5196(21)00026-7.33838734
21. WHO. 1997. International Classification of Diseases, Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death. Geneva, Switzerland: World Health Organization.
22. WHO. 2016. International Statistical Classification of Diseases and Related Health Problems, 10th Revision. http://apps.who.int/classifications/icd10/browse/2016/en [accessed 25 November 2021].
23. Bhaskaran K, Gasparrini A, Hajat S, Smeeth L, Armstrong B. 2013. Time series regression studies in environmental epidemiology. Int J Epidemiol 42 (4 ):1187–1195, PMID: , 10.1093/ije/dyt092.23760528
24. Gasparrini A, Armstrong B, Kenward MG. 2010. Distributed lag non-linear models. Stat Med 29 (21 ):2224–2234, PMID: , 10.1002/sim.3940.20812303
25. Sera F, Armstrong B, Blangiardo M, Gasparrini A. 2019. An extended mixed‐effects framework for meta‐analysis. Stat Med 38 (29 ):5429–5444, PMID: , 10.1002/sim.8362.31647135
26. Gasparrini A, Armstrong B, Kenward MG. 2012. Multivariate meta-analysis for non-linear and other multi-parameter associations. Stat Med 31 (29 ):3821–3839, PMID: , 10.1002/sim.5471.22807043
27. Higgins JPT, Thompson SG. 2002. Quantifying heterogeneity in a meta-analysis. Stat Med 21 (11 ):1539–1558, PMID: , 10.1002/sim.1186.12111919
28. Sera F, Gasparrini A. 2022. Extended two-stage designs for environmental research. Environ Health 21 (1 ):41, PMID: , 10.1186/s12940-022-00853-z.35436963
29. Samoli E, Analitis A, Touloumi G, Schwartz J, Anderson HR, Sunyer J, et al. 2005. Estimating the exposure–response relationships between particulate matter and mortality within the APHEA multicity project. Environ Health Perspect 113 (1 ):88–95, PMID: , 10.1289/ehp.7387.15626653
30. Chen R, Yin P, Meng X, Liu C, Wang L, Xu X, et al. 2017. Fine particulate air pollution and daily mortality. A nationwide analysis in 272 Chinese cities. Am J Respir Crit Care Med 196 (1 ):73–81, PMID: , 10.1164/rccm.201609-1862OC.28248546
31. Schwartz J, Spix C, Touloumi G, Bachárová L, Barumamdzadeh T, le Tertre A, et al. 1996. Methodological issues in studies of air pollution and daily counts of deaths or hospital admissions. J Epidemiol Community Health 50 (suppl 1 ):S3–S11, PMID: , 10.1136/jech.50.suppl_1.s3.8758217
32. Gasparrini A, Leone M. 2014. Attributable risk from distributed lag models. BMC Med Res Methodol 14 (1 ):55, PMID: , 10.1186/1471-2288-14-55.24758509
33. Johns DO, Svendsgaard D, Linn WS. 2010. Analysis of the concentration–respiratory response among asthmatics following controlled short-term exposures to sulfur dioxide. Inhal Toxicol 22 (14 ):1184–1193, PMID: , 10.3109/08958378.2010.535220.21126154
34. Zheng XY, Ding H, Jiang LN, Chen SW, Zheng JP, Qiu M, et al. 2015. Association between air pollutants and asthma emergency room visits and hospital admissions in time series studies: a systematic review and meta-analysis. PLoS One 10 (9 ):e0138146, PMID: , 10.1371/journal.pone.0138146.26382947
35. Chuang KJ, Chan CC, Su TC, Lee CT, Tang CS. 2007. The effect of urban air pollution on inflammation, oxidative stress, coagulation, and autonomic dysfunction in young adults. Am J Respir Crit Care Med 176 (4 ):370–376, PMID: , 10.1164/rccm.200611-1627OC.17463411
|
PMC009xxxxxx/PMC9994181.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36883836
EHP11391
10.1289/EHP11391
Research
Long-Term Exposure to Nitrate and Trihalomethanes in Drinking Water and Prostate Cancer: A Multicase–Control Study in Spain (MCC-Spain)
https://orcid.org/0000-0002-4523-4148
Donat-Vargas Carolina 1 2 3 4
Kogevinas Manolis 1 2 3 5
Castaño-Vinyals Gemma 1 2 3 5
Pérez-Gómez Beatriz 3 6
Llorca Javier 3 7
Vanaclocha-Espí Mercedes 8
Fernandez-Tardon Guillermo 3 9
Costas Laura 3 10
Aragonés Nuria 3 11
Gómez-Acebo Inés 3 7 12
Moreno Victor 3 13 14 15
Pollan Marina 3 6
Villanueva Cristina M. 1 2 3 5
1 Instituto de Salud Global de Barcelona (ISGlobal), Barcelona, Spain
2 Universitat Pompeu Fabra (UPF), Barcelona, Spain
3 CIBER Epidemiología y Salud Pública (CIBERESP), Madrid, Spain
4 Unit of Cardiovascular and Nutritional Epidemiology, Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden
5 Hospital del Mar Medical Research Institute (IMIM), Barcelona, Spain
6 Cancer and Environmental Epidemiology Unit, National Centre for Epidemiology, Carlos III Institute of Health, Madrid, Spain
7 Faculty of Medicine, University of Cantabria, Spain
8 Cancer and Public Health Area, Foundation for the Promotion of Health and Biomedical Research-Public Health Research (FISABIO), Valencia, Spain
9 Health Research Institute of Asturias (ISPA), Oviedo, Spain
10 Cancer Epidemiology Research Programme, Catalan Institute of Oncology, Bellvitge Biomedical Research Institute (IDIBELL), Barcelona, Spain
11 Epidemiology Section, Public Health Division, Department of Health of Madrid, Madrid, Spain
12 Instituto de Investigación Sanitaria Valdecilla (IDIVAL), Santander, Spain
13 Unit of Biomarkers and Susceptibility, Cancer Prevention and Control Program, Catalan Institute of Oncology (ICO), Hospitalet de Llobregat, Spain
14 Colorectal Cancer Group, IDIBELL, Hospitalet de Llobregat, Spain
15 Department of Clinical Sciences, Faculty of Medicine, University of Barcelona, Barcelona, Spain
Address correspondence to Cristina M. Villanueva Belmonte, ISGlobal – Institut de Salut Global de Barcelona, Campus MAR, Barcelona Biomedical Research Park (PRBB), Barcelona, Spain. Telephone: +34 93 214 73 44. Email: [email protected]
08 3 2023
3 2023
131 3 03700409 4 2022
20 1 2023
26 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Nitrate and trihalomethanes (THMs) in drinking water are widespread and are potential human carcinogens.
Objective:
We evaluated the association between drinking-water exposure to nitrate and THMs and prostate cancer.
Methods:
During the period 2008–2013, 697 hospital-based incident prostate cancer cases (97 aggressive tumors) and 927 population-based controls were recruited in Spain, providing information on residential histories and type of water consumed. Average nitrate and THMs levels in drinking water were linked with lifetime water consumption to calculate waterborne ingestion. Odds ratios (OR) and 95% confidence intervals (CI) were estimated using mixed models with recruitment area as random effect. Effect modification by tumor grade (Gleason score), age, education, lifestyle, and dietary factors was explored.
Results:
Mean (±standard deviation) adult lifetime waterborne ingested nitrate (milligrams per day), brominated (Br)-THMs (micrograms per day), and chloroform (micrograms per day) were 11.5 (±9.0), 20.7 (±32.4), and 15.1 (±14.7) in controls. Waterborne ingested nitrate >13.8 vs. <5.5mg/d was associated with an OR of 1.74 (95% CI: 1.19, 2.54) overall, and 2.78 (95% CI: 1.23, 6.27) for tumors with Gleason scores ≥8. Associations were higher in the youngest and those with lower intakes of fiber, fruit/vegetables, and vitamin C. Waterborne ingested THMs were not associated with prostate cancer. Residential tap water levels of Br-THMs and chloroform showed, respectively, inverse and positive associations with prostate cancer.
Conclusions:
Findings suggest long-term waterborne ingested nitrate could be a risk factor of prostate cancer, particularly for aggressive tumors. High intakes of fiber, fruit/vegetables and vitamin C may lower this risk. Association with residential levels but not ingested chloroform/Br-THM may suggest inhalation and dermal routes could be relevant for prostate cancer. https://doi.org/10.1289/EHP11391
Supplemental Material is available online (https://doi.org/10.1289/EHP11391).
The authors declare that they have no competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Prostate cancer has become widespread worldwide,1,2 with 1,414,259 estimated new cases in 2020 (7.3% of all cancer sites),2 and the prostrate is the leading incident cancer site in Spanish men (22% of all cancers sites).3 However, the etiology of prostate cancer remains largely unknown, and it is one of the few types of cancer for which the International Agency for Research on Cancer (IARC) has not identified a clear carcinogenic agent.4 Currently recognized risk factors are nonmodifiable, including age, ethnicity, and family history of cancer (including genetic heritage).1,5,6 Aggressive and fatal prostate cancers have been suggested to have different underlying causes in comparison with slow-growing tumors with an indolent course.7–9 Other suggested risk factors, particularly for advanced-stage and aggressive prostate cancer, are lifestyle/behaviors such as smoking, unhealthy diet, overweight status, and lack of exercise,9–11 as well as exposure to endocrine-disrupting chemicals12,13 such as Agent Orange (i.e., dioxins)8 and pesticides.14 An association between nitrite and nitrate from food additives and prostate cancer has also been recently reported.15 Cancer mortality maps reporting spatial and temporal distribution within countries and globally suggest that environmental exposures may contribute to prostate cancer development and could partly explain increasing incidence rates.16–18
Nitrate occurrence in the water cycle is rising worldwide because of growing use of nitrogen fertilizers and intensive farming.19 Human exposure to nitrate mainly occurs through ingestion of food and drinking water.20 Ingested nitrate is reduced to nitrite, which can react with amines and amides under acidic conditions in the stomach to form N-nitroso compounds. Ingested nitrate or nitrite under conditions that result in endogenous nitrosation is probably carcinogenic to humans.21,22 There are limited epidemiological studies seeking to disclose the relationship between nitrate exposure from drinking water and cancer, and, to date, consistent evidence has only been established with colorectal cancer.19 Disinfectants added to raw water to inactivate microbial pathogens result in the formation of several disinfection by-products (DBPs). DBPs constitute a complex mixture of chemicals formed as by-products of the reactions of the disinfectants applied to drinking-water.23 Chlorine is the most widespread disinfectant used worldwide, and trihalomethanes (THMs) and haloacetic acids are the DBPs formed at the highest concentrations after chlorination. When water containing ammonia is chlorinated, chloramines are formed, which, in turn, lead to the formation of nitrogenated by-products such as the carcinogenic N-nitrosamines.24 Several DBPs are genotoxic in vitro and carcinogenic in animal experiments,25,26 and the IARC has classified some DBPs as possible human carcinogens.27–29
This study was designed to evaluate the association between prostate cancer and long-term exposure to nitrate and THMs in drinking water. Because risk factors for advanced-stage and aggressive prostate cancer may differ from slow-growing tumors7,18 we also evaluated the associations by Gleason score (<8 vs. ≥8 as aggressive prostate cancer). We further investigated the effect modification by age, dietary factors, education, and adherence to a healthy lifestyle.
Methods
Study Design and Population
The MCC-Spain study (http://www.mccspain.org) is a multicase–control study conducted in different provinces in Spain between 2008 and 2013. MCC-Spain included breast, colorectal, prostate, and gastroesophageal cancer, as well as chronic lymphocytic leukemia cases, along with a common pool of population-based controls.30 Cases were histologically confirmed incident prostate cancer patients [International Classification of Diseases, Tenth Revision (ICD)-10 C61 and D07.5], identified through active searches that included periodic visits to hospital departments, and were interviewed closely after diagnosis (median of 58 d). Controls were selected from the general population, identified from the lists of randomly selected family practitioners in primary health centers, and were frequency matched to cases by age for each region (12 recruitment areas).30 Inclusion criteria required participants to be 20–85 y old, to have the ability to understand and answer the questionnaire, and to have lived for at least 6 months in the study area. The study protocol was approved by the ethics committee at all collaborating institutions, and each participant signed an informed consent form prior to enrollment. The overall response rate (subjects interviewed divided by subjects interviewed plus refusals) was 72% for prostate cancer cases and 53% for controls, leading to 996 prostate cancer cases and 1,281 controls recruited in the areas included in the present analysis (Asturias, Barcelona, Cantabria, Madrid, and Valencia).
Data Collection
Those who agreed to participate answered a structured, computerized questionnaire administered by trained personnel in a face-to-face interview to gather information on anthropometrics (self-reported), sociodemographics, lifestyle factors, and personal and family medical history. Participants provided full address, year started and stopped living in all the residences where they lived for at least 12 months since age 18 y until the time of the interview, and the type of water consumed in each residence (municipal, bottled, well, other). Amount (glasses per day) of water ingested on average lifetime at home, work, and other places was ascertained. A final section evaluating the reliability of the interview was completed by the interviewer. Dietary habits the year before the interview were collected through a self-administered semiquantitative food frequency questionnaire, including a total of 140 food items, previously validated in Spain.31 Questionnaires used are available online (http://mccspain.org).
The Gleason score was collected from the pathological records. Two prostate cancer grading categories were constructed: low- to medium-grade prostate cancer (Gleason score <8) and high-grade/aggressive prostate cancer (Gleason score ≥8).32,33
Nitrate and THM Levels in Municipal Drinking Water
We designed a structured questionnaire aimed at water utilities, local authorities, and/or health authorities to collect drinking water source (surface or ground water proportion) and treatment in the study areas back to 1940. In addition, available data from routine monitoring in the drinking water treatment plants and the distribution network were collected for nitrate and THMs (chloroform, bromodichloromethane, dibromochloromethane, and bromoform). We targeted data collection among study municipalities that contributed up to 80% of person-years.
For the years 2004–2010, centralized routine monitoring data was provided by the SINAC (Spanish National Information System on Water for Consumption), that includes information at the water-zone level introduced by water supply operators from public or private companies or municipalities, as well as from public or private laboratories. The water zone, which mostly corresponds to municipality, was defined as a geographical area supplied by water with a homogeneous source and treatment and whose quality in the water distributed in the networks can be considered homogeneous. We linked each postal code from the residence to the corresponding water zone.
The distribution of the sampling points and the sampling frequency varied greatly, depending on the population served, extension of the water zone, and the year, and could be more than once a day (e.g., Madrid), up to once every 3 months, or once a year in less-populated areas. Measurements below the analytical limit of quantification (QL) were substituted with half the QL (QL/2).34 If the QL was missing, we imputed half of the most frequently reported.
Nitrate and THM Levels in Nonmunicipal Drinking Water
We measured nitrate in the 9 most-consumed bottled water brands in Spain using ultraviolet spectrophotometry, with 0.5/0.1mg/L detection/quantification limit. Nitrate concentrations were in the range of 2.3–15.6mg/L.35 THMs were previously measured in 15 popular bottled water brands in Spain through purge-and-trap and gas chromatography. Mean concentrations for chloroform, bromodichloromethane, dibromochloromethane, and bromoform were ≤0.1μg/L,36 and limits of detection were, respectively, 0.015, 0.004, 0.005, and 0.011μg/L. We used THM data from 56 measurements in different Spanish areas that were supplied by chlorinated ground water. Average concentrations were 0.3, 0.3, 0.8, and 1.8μg/L for chloroform, bromodichloromethane, dibromochloromethane, and bromoform, respectively. Nitrate data in private wells were not available.
Estimation of Long-Term Levels in Municipal Drinking Water
We calculated the annual average levels of nitrate and THMs at the water zone level. Years without measurements were assigned the average of all available measurements in the water zone if the water source and treatment did not change over the years. In the case of changes in the water source and/or treatment, procedures to back-extrapolate were applied.
For THMs, because their concentrations in surface water are generally higher than in ground sources,37 we used surface water percentage as a weight to back extrapolate individual THM concentrations when water source changed through linear interpolation, assuming that concentrations increased proportionately to the percentage of surface water. Likewise, water zones with changes in treatment over the years and THMs measurements were used to estimate the change percentage of THMs concentrations after introducing such treatments. These percentages were applied as a weight to back-extrapolate THM concentrations in areas with changes in these specific treatments when measurements were unavailable. Before chlorination started, THMs concentrations were assumed to be zero. Total THMs (TTHM) levels were calculated by adding up chloroform, bromodichloromethane, dibromochloromethane, and bromoform concentrations.
For years without nitrate measurements in water zones where water source changed over the years, the groundwater percentage was used as a weight to back-extrapolate concentrations using linear interpolation, assuming that nitrate levels were higher with increasing groundwater proportion.19 In municipalities without any nitrate measurement (covering ∼0.5% of the total person-years), we imputed the levels of neighboring municipalities supplied with similar ground water proportion plus or minus 10%.
Individual Exposures in the Study Population
Average nitrate and THMs concentrations in residential tap water.
We used municipality and year to link municipal levels in drinking water with residential histories of study participants from age 18 y to 2 y before the interview. We estimated the average concentration of nitrate (milligrams per liter) and THMs (micrograms per liter) for this period, henceforth referred to as “lifetime” or “long-term exposure.” Generally, because participants lived in three residences on average during the exposure window period, they were assigned to the water zone where they had lived the longest as of the date of the interview (≈30 y).
Average ingested nitrate and THMs.
To calculate waterborne ingested nitrate (milligrams per day) and THMs (micrograms per day), we assigned levels in drinking water by year according to the water type consumed at home, including municipal (tap), bottled, and private well/other water. Nitrate and THMs levels in municipal water were assigned for tap-water consumption. Nitrate levels in the sampled bottled waters (range 2.3–15.6mg/L)35 were averaged using the sales frequency of each brand as a weight, leading to 6.1mg/L of NO3−, which was assigned to study participants consuming bottled water. Because nitrate levels in well water were not available, waterborne ingested nitrate was considered missing for years when well-water consumption was reported (≈2%). A zero THM level was assigned to bottled-water consumers, according to a previous study.36 THM values assigned for well-water consumers were 0.3, 0.3, 0.8, and 1.8μg/L for chloroform, bromodichloromethane, dibromochloromethane, and bromoform, respectively. The annual nitrate and THMs estimates were averaged from age 18 y to 2 y before the interview and multiplied by the average daily water intake at the residence. Total amount of ingested water was ascertained as the number of water glasses per day consumed on average by the participant at home (liters per day, assuming 200mL/glass). Water intakes that equaled zero and those above the 99th percentile (4L/d), considered implausible, were treated as missing values in the analyses.
Covariables
Age (years), education (less than primary school, primary school, secondary school, university), self-reported weight and height 1 y before the interview to compute body mass index (BMI; kilograms per square meter), family history of prostate cancer (yes, no), smoking (never, former, current), and physical activity were considered. Smokers were defined as those smoking at least one cigarette per day for ≥6 months. Former smokers were defined as those who quit smoking ≥1 y before the interview. Physical activity was ascertained through open questions on any type of physical activity practiced in life, years, and frequency (hours/week), to calculate metabolic equivalents (METs) from age 16 y to 2 y before the interview. We estimated a cancer prevention score based on The World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) cancer prevention recommendations38 based on six items: BMI, physical activity, consumption of foods and drinks that promote weight gain, plant foods, animal foods, and alcohol. Briefly, the method of estimating the score according to the standardized scoring system for 2018 WCRF/AICR cancer prevention recommendations39 is based on the following criteria: 1 point was assigned when the recommendation was met, 0.5 point when it was partially met, and 0 points when it was not met. The score of each recommendation was added to obtain the total score, which ranged from a minimum value of 0 to a maximum value of 6 points, with higher values indicating high compliance with the cancer prevention recommendations.40 We collected information on self-reported family history of prostate cancer (i.e., malignant tumors in first-degree relatives). Based on food composition tables, frequencies in servings per day of red and processed meat were converted to grams per day and total dietary fiber intake and vitamin intake.
Statistical Analyses
The initial sample of prostate cancer cases and controls was 2,277 (996 cases, 1,281 controls). Number of controls was greater than the number of cases because the controls were matched to different cancer sites. We excluded subjects with interviews qualified as unreliable by the trained interviewers (n=3 cases); those with nitrate or THMs estimates covering fewer than 70% of the years between age 18 y to 2 y before the interview (n=376) and those reporting no water consumption or implausible values were also excluded (n=162). Finally, to have a similar geographical distribution of cases and controls, only municipalities with at least one case and one control were included (n=112 excluded). The final sample included 1,624 subjects, 697 cases (590 low- to medium-grade, tumors, 97 high-grade tumors, and 10 without this information) and 927 controls with ages between 38 and 85 y. (Figure 1). Characteristics of cases and controls excluded from the study (n=653) are presented in Supplemental Table 1.
Figure 1. Flowchart showing exclusions of study participants from the Multicase–Control Study in Spain (MCC-Spain). The main exposure periods were from 18 y of age to 2 y before the interview. The interviewers rated the quality of the interview, and those unreliable or inconsistent were excluded. A total of 653 participants were excluded from the study.
Figure 1 is a flowchart with five steps. Step 1: The initial sample size is 2277. The case samples are 996 and the control samples are 1281, excluding 3 samples of cases with unreliable interviews. Step 2: The initial sample size is 2274, excluding 376 samples, which include 173 samples of cases and 203 samples of controls with nitrate and THM exposures covering less than 70 percent of the main exposure periods. Step 3: The initial sample size is 1898, excluding 162 samples, which include 67 samples of cases and 95 samples of controls with water consumption not reported or implausible values (0 liters or greater liters). Step 4: The initial sample size is 1736, excluding 112 samples, which include 56 samples of cases and 56 samples of controls from municipalities without any case or any control. Step 5: The final sample size is 1624, including 697 samples of cases and 927 samples of controls.
Table 1 Characteristics and drinking-water contaminant exposures of the study population from the Multicase–Control Study in Spain (MCC-Spain): 697 cases, 927 controls (N=1,624).
Controls n (%) or mean (±SD) Cases n (%) or mean (±SD)
Total Low- to medium-grade tumors (Gleason score <8) High-grade tumors (Gleason score ≥8)
Number of participants 927 697 590 97
Characteristics
Age (y) 66.6 (8.3) 66.0 (7.3) 65.6 (7.2) 68.7 (7.7)
Educational level (%)
Less than primary 157 (16.9) 156 (22.4) 128 (21.7) 26 (26.8)
Primary school 300 (32.4) 276 (39.6) 234 (39.7) 39 (40.2)
Secondary school 262 (28.3) 157 (22.5) 139 (22.0) 23 (23.7)
University 208 (22.4) 108 (15.5) 98 (16.6) 9 (9.3)
Family history of prostate cancer (first degree) (%) 108 (11.7) 144 (20.7) 122 (20.7) 20 (20.6)
Smoking status (%)
Never 237 (25.6) 218 (31.3) 180 (30.5) 34 (35. 1)
Former 480 (51.8) 349 (50.1) 296 (50.2) 47 (48.5)
Current smoker 210 (22.7) 130 (18.7) 114 (19.3) 16 (16.5)
WCRF/AICR cancer prevention score 3.4 (1.0) 3.3 (0.9) 3.3 (0.9) 3.4 (1.0)
Intake of red and processed meat (g/d) 73.3 (38.0) 77 (40.6) 77.1 (40.9) 75.4 (40.2)
Intake of total fiber (g/d) 11.3 (3.8) 11.1 (3.6) 11.1 (3.6) 11.3 (3.7)
Intake of fruit and vegetables (g/d) 486 (277) 501 (245) 499 (239) 524 (275)
Intake of vitamin C (mg/d) 148 (91) 150 (80) 150 (78) 156 (94)
Intake of vitamin E (mg/d) 10.4 (5.45) 10.6 (5.2) 10.6 (5.1) 11.1 (5.9)
Recruitment area (%)
Asturias 47 (5.1) 8 (1.2) 8 (1.36) 0 (0)
Barcelona 421 (45.4) 301 (43.2) 252 (42.7) 49 (50.5)
Cantabria 120 (12.9) 101 (14.5) 84 (14.2) 14 (14.4)
Madrid 276 (29.8) 239 (34.3) 206 (34.9) 27 (27.8)
Valencia 63 (6.8) 48 (6.9) 40 (6.8) 7 (7.2)
Drinking-water contaminant exposures
Average concentrations in residential tap water
Nitrate (mg/L) 7.2 (4.0) 7.1 (4.2) 7.1 (4.14) 7.8 (4.1)
Brominated trihalomethanes (μg/L) 34.7 (33.0) 28.3 (27.1) 28.1 (27.2) 30.6 (26.7)
Chloroform (μg/L) 20.7 (8.0) 21.4 (8.4) 21.6 (8.4) 20.4 (8.1)
Average waterborne ingestion
Nitrate (mg/d) 11.5 (9.0) 12.8 (10.8) 12.8 (11.1) 13.2 (9.2)
Brominated trihalomethanes (μg/d) 20.7 (32.4) 19.2 (29.2) 19.2 (28.1) 20.6 (36.1)
Chloroform (μg/d) 15.1 (14.7) 15.4 (14.0) 15.8 (14.2) 12.9 (13.4)
Note: Continuous variables are presented as means and standard deviation and categorical variables as percentage (%) and number of subjects (n). A total of 171 subjects had missing data for WCRF/AICR cancer prevention score and dietary variables (intake of red and processed meat, total fiber, fruit and vegetables, vitamin C, and vitamin E). The mismatch between total cases and the sum of early-stage (low- to medium-grade tumors) and aggressive (high-grade tumors) prostate cancer is because there are 10 subjects without information on the grade of the tumor. Brominated trihalomethanes include bromodichloromethane, dibromochloromethane, and bromoform. WCRF/AICR cancer prevention score is based on the WCRF/AICR cancer prevention recommendations. SD, standard deviation; WCRF/AICR, World Cancer Research Fund/American Institute for Cancer Research.
Spearman correlations between tap water residential concentrations and waterborne ingested nitrate, Br-THMs (sum of bromodichloromethane, dibromochloromethane, and bromoform), and chloroform were examined. Drinking-water exposures were categorized into tertiles defined according to the distribution among controls. The main models estimated the association between prostate cancer and lifetime waterborne ingested nitrate, TTHMs, chloroform and Br-THMs, expressed in tertiles and continuous (per 5-unit increment). We estimated odds ratios (OR) and 95% confidence intervals (CI) of prostate cancer using mixed models with recruitment area (Asturias, Barcelona, Cantabria, Madrid, Valencia) as random effect. To test for linear trends (p-trend) across increasing categories of exposure, the median concentration within each category was treated as a continuous variable in the model.
To test whether the associations varied across tumor grade and aggressiveness, we used multinomial logistic regression models with Gleason scores <8 and ≥8 for the two categories, with control group as the reference (base outcome), and OR [also referred to as relative risk ratios (RRRs)] and their 95% CI were estimated. Heterogeneity of effects for the two grades of tumor severity was tested using the Wald statistic. Smoothed spline with three degrees of freedom from general additive models (GAM) were used to visually display the exposure–response relationships on continuous variables. We further explored the associations using concentrations in residential tap water as exposure, because it might be a better indicator of an exposure though multiple routes (not just ingestion), which is especially relevant for THMs.
All models were adjusted for recruitment area, age, and education. Further adjustment included first-degree family history of prostate cancer, smoking status, and WCRF/AICR cancer prevention score. An additional model was reported with mutual adjustments between nitrate, chloroform, and Br-THMs levels. Multicollinearity was explored using the variance inflation factor (VIF). A mean VIF of 1.43 was obtained, and all variable categories had a VIF <2. We used stochastic regression (which adds a random error term that appropriately reproduces the correlation between X and Y) to impute 171 missing values in the WCRF/AICR cancer prevention score. There were no missing data for the other covariates of adjustment.
Additionally, subgroup analyses were performed for waterborne nitrate ingested, stratifying the sample (above or below the median among controls) by the following suspected effect modifiers: age (≤66 y vs. >66 y), education (≤primary school vs. >primary school), WCRF/AICR score (≤3.5 vs. >3.5), intakes of red and processed meat (≤67g/d vs. >67g/d), total fiber (≤11g/d vs. >11g/d), total fruit and vegetables (≤473g/d vs. >473g/d), vitamin C (≤130mg/d vs. >130mg/d), and vitamin E (≤9.2mg/d vs. >9.2mg/d). Interaction p-value was obtained using the likelihood ratio test of the models with and without the multiplicative interaction term.
All p-values presented are two-tailed; <0.05 was considered statistically significant. Analyses were performed using STATA (version 16.0; Stata Corp.).
Results
Approximately 330 water zones with data on nitrate and THM levels were involved during the study exposure window (Supplemental Figure 1; Excel Table S1). Mean (plus or minus SD) values for average lifetime waterborne ingested nitrate (milligrams per day), Br-THMs (micrograms per day) and chloroform (micrograms per day) were 11.5 (9.0), 20.7 (32.4), and 15.1 (14.7), respectively, in controls; and 12.8 (10.8), 19.2 (29.2), and 15.4 (14.0) in cases. The average age was 66.6 (8.3) y old for controls and 66.0 (7.3) for cases. On average, cases had lower education, had twice as frequent family history of prostate cancer (first degree), and had consumed slightly more red and processed meat in comparison with controls. The recruitment area contributing with the largest population was Barcelona (44% of all subjects) followed by Madrid (30%), whereas Asturias was the province with the greatest difference in percentage of cases (1.6%) in comparison with controls (5.5%) (Table 1).
The proportion (in person-years) of municipal, bottled- and well-water consumption was approximately 78%, 20%, and 2%, respectively, during the exposure window. The average water intake was 1.9L/d for cases and 1.8L/d for controls. Spearman correlations between tap-water residential concentrations and waterborne ingested contaminants were moderate, rho 0.67 for nitrate and ∼0.50 for Br-THM and chloroform (Supplemental Figure 2).
Waterborne Ingested Nitrate
Considering the mutually adjusted models, lifetime average waterborne ingested nitrate was positively associated with prostate cancer when comparing the highest with the lowest exposure category (OR=1.74, 95% CI: 1.19, 2.54; p-trend 0.002). For each 5mg/d increase of waterborne ingested nitrate, the overall OR of prostate cancer increased by 22% (OR=1.22; 95% CI: 1.12, 1.33) (Table 2). When examining prostate cancer by tumor severity, comparing extreme categories, the OR for low- to medium-grade prostate cancer (Gleason score <8) was 1.59, 95% CI: 1.05, 2.39; p-trend 0.014, and for high-grade (Gleason score ≥8, i.e., aggressive prostate cancer) was 2.78, 95% CI: 1.23, 6.27; p-trend 0.002 (Wald test p=0.189) (Table 3).
Table 2 Association between prostate cancer and lifetime average waterborne ingested nitrate and trihalomethanes (THMs). Multicase–Control Study in Spain (MCC-Spain): 697 cases, 927 controls (N=1,624).
Exposure Controls Cases OR (95% CI) Multivariable adjusteda OR (95% CI) Multivariable adjustedb OR (95% CI) Multivariable adjustedc
Nitrate (mg/d)
Tertile 1 (<5.5) 309 221 1.00 1.00 1.00
Tertile 2 (5.5–13.8) 309 209 1.11 (0.82, 1.49) 1.09 (0.81, 1.48) 1.16 (0.85, 1.57)
Tertile 3 (>13.8) 309 267 1.58 (1.12, 2.23) 1.54 (1.08, 2.19) 1.74 (1.19, 2.54)
p-Trend 927 697 0.004 0.007 0.002
Per 5mg/d 927 697 1.13 (1.06, 1.21) 1.13 (1.05, 1.21) 1.22 (1.12, 1.33)
TTHMs (μg/d)
Tertile 1 (<13.7) 309 243 1.00 1.00 1.00
Tertile 2 (13.7–37.5) 309 221 0.88 (0.68, 1.13) 0.87 (0.67, 1.13) 0.87 (0.67, 1.13)
Tertile 3 (>37.5) 309 233 0.95 (0.73, 1.23) 0.91 (0.70, 1.18) 0.90 (0.69, 1.17)
p-Trend 927 697 0.783 0.535 0.508
Per 5mg/d 927 697 1.00 (0.98, 1.01) 0.99 (0.98, 1.01) 0.99 (0.98, 1.01)
Brominated THMs (μg/d)
Tertile 1 (<3.2) 311 254 1.00 1.00 1.00
Tertile 2 (3.2–16.8) 307 225 0.91 (0.71, 1.16) 0.89 (0.69, 1.14) 0.82 (0.61, 1.08)
Tertile 3 (>16.8) 309 218 0.90 (0.67, 1.20) 0.86 (0.64, 1.16) 0.65 (0.42, 1.00)
p-Trend 927 697 0.575 0.447 0.109
Per 5μg/d 927 697 0.99 (0.98, 1.01) 0.99 (0.97, 1.01) 0.99 (0.97, 1.01)
Chloroform (μg/d)
Tertile 1 (<5.4) 309 212 1.00 1.00 1.00
Tertile 2 (5.4–19.1) 309 255 1.17 (0.90, 1.51) 1.16 (0.89, 1.51) 1.35 (0.98, 1.87)
Tertile 3 (>19.1) 309 230 1.01 (0.76, 1.33) 0.98 (0.74, 1.31) 1.19 (0.80, 1.77)
p-Trend 927 697 0.795 0.671 0.925
Per 5μg/d 927 697 0.99 (0.95, 1.03) 0.99 (0.95, 1.02) 0.98 (0.93, 1.02)
Note: ORs and 95% CI were calculated using mixed models with recruitment area (Asturias, Barcelona, Cantabria, Madrid, Valencia) as random effect. Brominated THMs includes bromodichloromethane, dibromochloromethane, and bromoform. TTHMs includes chloroform, bromodichloromethane, dibromochloromethane, and bromoform. CI, confidence interval; OR, odds ratio; THMs, trihalomethanes; TTHMs, total trihalomethanes; WCRF/AICR, World Cancer Research Fund/American Institute for Cancer Research.
a Model adjusted for age (years) and educational level (less than primary, primary school, secondary school, university).
b Model further adjusted for first-degree family history of prostate cancer (yes, no), smoking status (never, former, current smoker), and the WCRF/AICR cancer prevention score.
c Model 2 mutually adjusted for the other corresponding components, i.e., total THMs (nitrate model), nitrate (THMs model), chloroform and nitrate (brominated THMs model), brominated THMs and nitrate (chloroform model).
Table 3 Association between prostate cancer and lifetime average waterborne ingested nitrate by tumor grade (according to Gleason score). Multicase–Control Study in Spain (MCC-Spain): 687 cases, 927 controls (N=1,614).
Exposure Controls Cases OR (95% CI) Multivariable adjusteda OR (95% CI) Multivariable adjustedb OR (95% CI) Multivariable adjustedc
Nitrate waterborne ingestion (mg/d)
Low- to medium-grade
Gleason score <8
Tertile 1 (<5.5) 309 190 1.00 1.00 1.00
Tertile 2 (5.5–13.8) 309 179 1.10 (0.80, 1.52) 1.09 (0.79, 1.50) 1.13 (0.81, 1.56)
Tertile 3 (>13.8) 309 221 1.52 (1.04, 2.21) 1.47 (1.01, 2.15) 1.59 (1.05, 2.39)
p-Trend 927 590 0.015 0.024 0.014
Per 5mg/d 927 590 1.14 (1.06, 1.23) 1.14 (1.06, 1.22) 1.23 (1.13, 1.35)
High-grade
Gleason score ≥8
Tertile 1 (<5.5) 309 27 1.00 1.00 1.00
Tertile 2 (5.5–13.8) 309 25 0.92 (0.46, 1.83) 0.90 (0.45, 1.81) 1.12 (0.55, 2.29)
Tertile 3 (>13.8) 309 45 1.80 (0.84, 3.85) 1.77 (0.82, 3.82) 2.78 (1.23, 6.27)
p-Trend 927 97 0.035 0.039 0.002
Per 5mg/d 927 97 1.07 (0.94, 1.23) 1.07 (0.93, 1.23) 1.18 (1.01, 1.39)
Note: The number of these analyses is 1,614 (10 cases had no information on Gleason score). Multinomial logistic regression models with the two categories of tumor grade Gleason score < and ≥8, with control group as the reference (base outcome). ORs and 95% CI were calculated using multinomial (polytomous) logistic regression. Wald test p-values for heterogeneity of effects were 0.1894. CI, confidence interval; OR, odds ratio; THMs, trihalomethanes; WCRF/AICR, World Cancer Research Fund/American Institute for Cancer Research.
a Model adjusted for recruitment area (Asturias, Barcelona, Cantabria, Madrid, Valencia), age (years) and educational level (less than primary, primary school, secondary school, university).
b Model further adjusted for first-degree family history of prostate cancer (yes, no), smoking status (never, former, current smoker) and the WCRF/AICR cancer prevention score.
c Model 2 mutually adjusted for total THMs.
We also explored whether other factors (Supplemental Table 2) modified the association between waterborne ingested nitrate and prostate cancer. In the analyses stratified by several factors (Table 4), those with greatest impact on the estimates were age and dietary fiber intake, followed by intakes of fruit and vegetables and vitamin C, although interaction p-values were not statistically significant. Comparing highest vs. lowest categories, waterborne ingested nitrate was significantly associated with a higher odds of prostate cancer in the youngest (1.75, 95% CI: 1.07, 2.89; p-trend 0.007) but not in the oldest [1.23, 95% CI: 0.75, 2.01; p-trend 0.505 (p for interaction 0.095)]. Likewise, the OR of prostate cancer for those with the lowest intake of total fiber (≤11g/d) was 2.34 (95% CI: 1.39, 3.94; p-trend 0.001), whereas for those with the highest intake (>11g/d), the OR was 0.95 (95% CI: 0.52, 1.73; p-trend 0.709). Interaction p-value was 0.102. Comparable results were observed when comparing prostate cancer likelihood in those with higher and lower intakes of fruit and vegetables and vitamin C. Finally, waterborne ingested nitrate was associated with prostate cancer only in those participants with highest education attained, without significant interaction (Table 4).
Table 4 Subgroup analysis. Association between waterborne ingested nitrate and prostate cancer by age, education, WCRF score adherence, and intakes of red and processed meat, fiber, total fruit and vegetables, vitamin C and vitamin E (above and below the median among controls). Multicase–Control Study in Spain (MCC-Spain): 629 cases, 824 controls (N=1,453).
By age ≤66 y (n=848) p Interaction >66 y (n=776)
Ingested nitrate (mg/d) Controls/Cases OR 0.095 Controls/Cases OR
Tertile 1 (<5.5) 156/124 1.00 — 153/97 1.00
Tertile 2 (5.5–13.8) 165/98 1.02 (0.67, 1.54) — 144/111 118 (0.76, 1.84)
Tertile 3 (>13.8) 154/151 1.75 (1.07, 2.89) — 155/116 1.23 (0.75, 2.01)
p-Trend 475/373 0.007 — 452/324 0.505
Per 5mg/d 475/373 1.15 (1.04, 1.28) — 452/324 1.08 (0.98, 1.19)
By education Primary school or less (n=889) — Secondary school or higher (n=735)
Ingested nitrate (mg/d) Controls/Cases OR 0.495 Controls/Cases OR
Tertile 1 (<5.5) 140/124 1.00 — 169/97 1.00
Tertile 2 (5.5–13.8) 146/119 0.87 (0.53, 1.44) — 163/90 1.40 (0.90, 2.16)
Tertile 3 (>13.8) 171/189 1.17 (0.54, 2.54) — 138/78 1.89 (1.09, 3.29)
p-Trend 457/432 0.215 — 470/265 0.032
Per 5mg/d 457/432 1.07 (0.98, 1.16) — 470/265 1.22 (1.08, 1.38)
By WCRF/AICR score Score ≤3.5 (n=854) p Interaction Score >3.5 (n=599)
Ingested nitrate (mg/day) Controls/Cases OR 0.844 Controls/Cases OR
Tertile 1 (<5.5) 156/113 1.00 — 121/79 1.00
Tertile 2 (5.5–13.8) 160/115 1.26 (0.82, 1.92) — 113/71 0.96 (0.60, 1.54)
Tertile 3 (>13.8) 165/145 1.74 (1.07, 2.82) — 109/106 1.43 (0.84, 2.44)
p-Trend 481/373 0.021 — 343/256 0.110
Per 5mg/d 481/373 1.12 (1.02, 1.23) — 343/256 1.13 (1.01, 1.26)
By intake of red and processed meat Intake ≤66g/d (n=676) p Interaction Intake >66g/d (n=777)
Ingested nitrate (mg/d) Controls/Cases OR 0.705 Controls/Cases OR
Tertile 1 (<5.5) 150/78 1.00 — 127/114 1.00
Tertile 2 (5.5–13.8) 140/86 1.14 (0.73, 1.78) — 133/100 1.05 (0.67, 1.64)
Tertile 3 (>13.8) 122/100 1.66 (0.99, 1.78) — 152/151 1.39 (0.84, 2.29)
p-Trend 412/264 0.043 — 412/365 0.128
Per 5mg/d 412/264 1.12 (1.01, 1.24) — 412/365 1.11 (1.01, 1.23)
By intake of total fiber Intake ≤11g/d (n=764) p Interaction Intake >11g/d (n=689)
Ingested nitrate (mg/d) Controls/Cases OR 0.102 Controls/Cases OR
Tertile 1 (<5.5) 145/97 1.00 — 132/95 1.00
Tertile 2 (5.5–13.8) 136/113 1.48 (0.94, 2.32) — 137/73 0.75 (0.45, 1.23)
Tertile 3 (>13.8) 131/142 2.34 (1.39, 3.94) — 143/109 0.95 (0.52, 1.73)
p-Trend 412/352 0.001 — 412/277 0.709
Per 5mg/d 412/352 1.17 (1.05, 1.30) — 412/277 1.06 (0.97, 1.17)
By intake of fruit and vegetables Intake ≤460g/d (n=710) p Interaction Intake >473 g/d (n=743)
Ingested nitrate (mg/d) Controls/Cases OR 0.783 Controls/Cases OR
Tertile 1 (<5.5) 135/84 1.00 — 142/108 1.00
Tertile 2 (5.5–13.8) 149/98 1.26 (0.81, 1.97) — 124/88 0.88 (0.60, 1.28)
Tertile 3 (>13.8) 128/116 2.07 (1.24, 3.45) — 146/135 1.01 (0.71, 1.45)
p-Trend 412/298 0.003 — 412/331 0.799
Per 5mg/d 412/298 1.20 (1.08, 1.34) — 412/331 1.02 (0.95, 1.10)
By intake of vitamin C Intake ≤130mg/d (n=700) p Interaction Intake >130mg/d (n=753)
Ingested nitrate (mg/d) Controls/Cases OR 0.987 Controls/Cases OR
Tertile 1 (<5.5) 131/82 1.00 — 146/110 1.00
Tertile 2 (5.5–13.8) 148/95 1.29 (0.81, 2.07) — 125/91 0.99 (0.65, 1.50)
Tertile 3 (>13.8) 133/111 1.78 (1.04, 3.02) — 141/140 1.31 (0.81, 2.13)
p-Trend 412/288 0.034 — 412/341 0.198
Per 5mg/d 412/288 1.17 (1.05, 1.31) — 412/341 1.07 (0.97, 1.17)
By intake of vitamin E Intake ≤9.2mg/d (n=695) p Interaction Intake >9.2mg/d (n=758)
Ingested nitrate (mg/d) Controls/Cases OR 0.764 Controls/Cases OR
Tertile 1 (<5.5) 140/95 1.00 — 137/97 1.00
Tertile 2 (5.5–13.8) 147/87 1.05 (0.68, 1.62) — 126/99 1.20 (0.77, 1.88)
Tertile 3 (>13.8) 125/101 1.59 (0.95, 2.65) — 149/150 1.54 (0.93, 2.54)
p-Trend 412/283 0.047 — 412/346 0.084
Per 5mg/d 412/283 1.13 (1.02, 1.26) — 412/346 1.10 (0.99, 1.21)
Note: These analyses are performed excluding 171 subjects with missing data in the dietary variables. WCRF/AICR cancer prevention score is based on the WCRF/AICR cancer prevention recommendations. OR and 95% CI were calculated using mixed models with recruitment area (Asturias, Barcelona, Cantabria, Madrid, Valencia) as random effect. Model adjusted for age (years), educational level (less than primary, primary school, secondary school, university), first-degree family history of prostate cancer (yes, no), smoking status (never, former, current smoker) and the WCRF/AICR cancer prevention score. —, no data; CI, confidence interval; OR, odds ratio; WCRF/AICR, World Cancer Research Fund/American Institute for Cancer Research.
Waterborne Ingested THMs
The relationship between waterborne ingested Br-THMs and prostate cancer showed an inverse pattern (Table 2; Figure 2). However, neither waterborne ingested TTHM (OR 0.90, 95% CI: 0.69, 1.17; p-trend 0.508), Br-THMs (OR 0.65, 95% CI: 0.42, 1.00; p-trend 0.109), nor chloroform (OR 1.19, 95% CI: 0.80, 1.77; p-trend 0.925) were significantly associated with prostate cancer (Table 2).
Figure 2. Exposure–response relationship between prostate cancer and waterborne ingested nitrate, brominated THMs, and chloroform (expressed as ORs). Multicase–Control Study in Spain (MCC-Spain): 697 cases, 927 controls (N=1,624). Smoothed spline with three degrees of freedom from general additive models adjusted for recruitment area (Asturias, Barcelona, Cantabria, Madrid, Valencia), age (years), educational level (less than primary, primary school, secondary school, university), first-degree family history of prostate cancer (yes, no), smoking status (never, former, current smoker) and the WCRF/AICR cancer prevention score. The dashed lines represent the 95% CIs. Upper CI red and lower green color. Note: CI, confidence interval; OR, odds ratio; THMs, trihalomethanes; WCRF/AICR, World Cancer Research Fund/American Institute for Cancer Research.
Figure 2A is a set of three line graphs, plotting odds ratio (95 percent confidence intervals), ranging from 1 to 4 in unit increments; negative 5 to 1 in increments of 5, 1 to 2 in increments of 0.5; and negative 5 to 1 in increments of 5, 1 to 2 in increments of 0.5 (y-axis) across Waterborne ingested nitrate (milligram per day), ranging from 0 to 100 in increments of 20, waterborne ingested brominated T H MS (microgram per day), ranging from 0 to 250 in increments of 50, and Waterborne ingested chloroform (microgram per day), ranging from 0 to 100 in increments of 20 (x-axis). Figure 2B is a set of three line graphs, plotting odds ratio (95 percent confidence intervals), ranging from 1 to 5 in unit increments; negative 5 to 1 in increments of 5, 1 to 2 in increments of 0.5; and 1 to 4 in unit increments (y-axis) across nitrate in residential tap water (milligram per day), ranging from 0 to 40 in increments of 10, brominated T H Ms in residential tap water (microgram per day), ranging from 0 to 150 in increments of 50, chloroform in residential tap water (microgram per day), ranging from 0 to 40 in increments of 10 (x-axis).
Residential Nitrate and THM Concentrations
Residential nitrate levels were associated with prostate cancer; OR for 5mg/d increase of residential nitrate level was 1.59 (95% CI: 1.13, 2.25) (Table 5). Br-THMs were inversely associated with prostate cancer both comparing extreme tertiles and the continuous exposure, OR 0.92 (95% CI: 0.89, 0.95) for 5μg/L increase (Table 5). Although the relationship between residential chloroform levels and prostate cancer was not linear (Figure 2), those with the highest residential chloroform exposure were more likely to develop prostate cancer than those with lowest residential exposure (OR 2.61, 95% CI: 1.55, 4.39; p-trend <0.001) (Table 5).
Table 5 Association between prostate cancer and lifetime average nitrate and trihalomethanes concentrations in residential tap water. Multicase–Control Study in Spain (MCC-Spain): 697 cases, 927 controls (N=1,624).
Exposure Controls Cases OR (95% CI) Multivariable adjusteda OR (95% CI) Multivariable adjustedb OR (95% CI) Multivariable adjustedc
Nitrate (mg/L)
Tertile 1 (<2.8) 309 237 1.00 1.00 1.00
Tertile 2 (2.8–10.0) 309 242 1.32 (0.73, 2.38) 1.33 (0.73, 2.44) 1.37 (0.75, 2.51)
Tertile 3 (>10.0) 309 218 1.30 (0.69, 2.43) 1.33 (0.70, 2.52) 1.52 (0.78, 2.97)
p-trend 927 697 0.473 0.441 0.214
Per 5mg/L 927 697 1.18 (0.88, 1.59) 1.18 (0.87, 1.60) 1.59 (1.13, 2.25)
TTHMs (μg/L)
Tertile 1 (<32.5) 309 231 1.00 1.00 1.00
Tertile 2 (32.5–64.4) 309 278 1.14 (0.81, 1.61) 1.11 (0.79, 1.58) 1.06 (0.75, 1.51)
Tertile 3 (>64.4) 309 188 0.88 (0.57, 1.36) 0.86 (0.55, 1.33) 0.78 (0.49, 1.24)
p-Trend 927 697 0.322 0.274 0.172
Per 5mg/L 927 697 0.93 (0.90, 0.96) 0.92 (0.89, 0.95) 0.92 (0.89, 0.95)
Brominated THMs (μg/L)
Tertile 1 (<8.9) 309 259 1.00 1.00 1.00
Tertile 2 (8.9–44.6) 309 261 0.86 (0.53, 1.40) 0.87 (0.53, 1.42) 0.68 (0.39, 1.20)
Tertile 3 (>44.6) 309 177 0.57 (0.32, 0.99) 0.57 (0.32, 1.01) 0.48 (0.25, 0.93)
p-Trend 927 697 0.007 0.007 0.017
Per 5μg/L 927 697 0.93 (0.90, 0.96) 0.93 (0.90, 0.96) 0.92 (0.89, 0.95)
Chloroform (μg/L)
Tertile 1 (<18.7) 309 183 1.00 1.00 1.00
Tertile 2 (18.4–25.5) 311 228 2.02 (1.36, 3.02) 2.19 (1.46, 3.29) 2.18 (1.43, 3.34)
Tertile 3 (>25.5) 307 286 2.37 (1.52, 3.71) 2.55 (1.61, 4.02) 2.61 (1.55, 4.39)
p-Trend 927 697 <0.001 <0.001 <0.001
Per 5μg/L 927 697 1.20 (1.02, 1.42) 1.21 (1.02, 1.43) 1.07 (0.94, 1.23)
Note: OR and 95% CI were calculated using mixed models with recruitment area (Asturias, Barcelona, Cantabria, Madrid, Valencia) as random effect. Brominated THMs includes bromodichloromethane, dibromochloromethane, and bromoform. TTHMs includes chloroform, bromodichloromethane, dibromochloromethane, and bromoform. CI, confidence interval; OR, odds ratio; THMs, trihalomethanes; TTHMs, total trihalomethanes; WCRF/AICR, World Cancer Research Fund/American Institute for Cancer Research.
a Model adjusted for age (years) and educational level (less than primary, primary school, secondary school, university).
b Model further adjusted for first degree family history of prostate cancer (yes, no), smoking status (never, former, current smoker) and the WCRF/AICR cancer prevention score.
c Model 2 mutually adjusted for the other corresponding components, i.e., total THMs (nitrate model), nitrate (THMs model), chloroform and nitrate (brominated THMs model), brominated THMs and nitrate (chloroform model).
Discussion
This is, to our knowledge, the first study to explore the relationship between exposure to nitrate and THMs through drinking water and prostate cancer at the individual level. In this case–control study, long-term waterborne ingested nitrate was associated with an increased OR of prostate cancer, especially with aggressive tumors. Residential chloroform levels showed a nonlinear positive association, whereas brominated THM were negatively associated with prostate cancer.
We examined two distinct exposure estimates, i.e., residential levels at the tap and waterborne ingested exposure. Residential levels provide a rough estimate of exposure through multiple routes (ingestion, inhalation, dermal), which is particularly relevant for volatile and skin permeable THMs.41–43 Given that nitrate is only ingested, the waterborne ingested estimates provide a more relevant exposure in comparison with residential nitrate levels. This finding is consistent with the differences we observed between exposure metrics for nitrate and THMs, i.e., higher associations for the ingested vs. residential nitrate exposure estimates and for residential vs. ingested THMs exposure estimates.
Prostate cancer is increasingly studied as two distinct phenotypes with suggested different etiologies: one slow-growing, indolent form and an aggressive form that can be fatal. Although age is more related to early-stage and indolent prostate tumors, lifestyle factors such as obesity, cigarette smoking, Western diets,44 or exposures such as pesticides14 have been linked to advanced-stage and more aggressive prostate cancer.9 Although our study had limited power to conduct stratified analyses for the most aggressive tumors, we observed a higher effect size for aggressive in comparison with early-stage prostate cancer (based on Gleason score <8 vs. ≥8). This observation might suggest that nitrate could have a greater influence on prostate cancer progression than initiation. Future work on the eventual mechanisms of nitrate on prostate carcinogenesis considering the role of grade and stage are warranted.
Our findings suggested that men ≤66 y old might be more susceptible to the carcinogenic effect of drinking-water nitrate on the prostate. In Spain, prostate cancer incidence at age ≤65 y is increasing at a higher rate than in older men45 and follows different spatial patterns,18 supporting the hypothesis that environmental factors may be involved. Prostate tumors in younger men differ from tumors diagnosed at an older age in terms of biological features and clinical entity.46,47 On the other hand, an effect modification has been identified by education level, suggesting that unaccounted factors related to socioeconomic status may play a role in the association between nitrate and prostate cancer. This effect modification is probably independent from that of the exposure to water contaminants, because previous studies have not been able to identify a clear link between socioeconomic status and exposure to drinking water contaminants.48,49 In light of this evidence, whether subgroups within the population may respond differently to the toxicity of drinking-water nitrate needs to be further examined.50
We also performed subgroup analysis by dietary factors involved in endogenous nitrosation.50 Men with high intake of total fruit and vegetables or vitamin C did not show an association between waterborne ingested nitrate and prostate cancer, whereas in those with low intakes the associations were strong. Antioxidants, vitamins, and polyphenols present in fruits and vegetables are inhibitors of endogenous nitrosation,51–54 and epidemiological evidence suggests the role of vitamins on prostate cancer prevention.55–57 Vitamin C (ascorbate) has shown significant antitumor activity, and high-dose vitamin C has been investigated as a treatment for cancer patients.58,59 Similarly, the association between waterborne ingested nitrate and prostate cancer was only found among people who consumed low amounts of fiber. These findings are consistent with the recognized benefit of dietary fiber to the gut microbiome, with protective capacity against food-derived toxicants, including N-nitrosamine.60 Furthermore, men with aggressive prostate cancer share a specific gut microbial profile, and recent studies have associated gut microbiome–related metabolites such as choline, betaine, and PAGln61 with prostate cancer, particularly its lethal form.62 Overall, these findings may suggest microbiome disruption as a possible biological mechanism of nitrate.
We did not find significant associations between waterborne ingested THMs and prostate cancer. However, residential levels of brominated THMs showed an inverse association with prostate cancer, whereas chloroform showed a nonlinear positive association. Current evidence is limited and methodology of previous studies is not comparable with our study. Previous cohort studies have evaluated prostate cancer in relation to residential-based exposure estimates, not specifically THMs, and showed null associations.63,64 The higher associations we observed for residential levels in comparison with ingestion suggest the relevance of inhalation and dermal exposure routes. These observations are supported by experimental evidence showing a higher internal dose of THMs and longer duration in the bloodstream when exposure is through activities involving inhalation and dermal absorption in comparison with ingestion.41,65 Different effects between chloroform and brominated THMs are expected, given the different genotoxicities.66 However, the inverse association with brominated THMs was unexpected.
Although THMs have been used in epidemiological studies as surrogates of total DBPs in drinking water, they are not the most toxic,67 and correlations among specific DBPs are variable and strongly depend on raw water quality and the type of treatment (including the disinfection processes).68 During the recruitment for the present study, DBPs other than THMs were analyzed in drinking water of a representative sample across study areas.68 The median (range) concentration of 3-Chloro-4-(dichloromethyl)-5-hydroxy-2(5H)-furanone (MX), which is a major mutagenic constituent of DBP,69 was 16.7 (0.8–54.1) ng/L. Chloroform concentrations were positively correlated to MX, whereas Br-THMs concentrations were negatively correlated to MX.68 This correlation might explain the inverse association we observed between residential Br-THMs and prostate cancer and the positive association with chloroform.
Exposure measurement error is the main concern in this study, particularly because the exposure difference between cases and controls is small. The limited historical measurements (particularly before 1980) and the assumptions used to model historical concentrations could reduce the accuracy of exposure estimates. To minimize exposure measurement error, we included only subjects with known exposures for at least 70% of the exposure window. Inability to account for exposures outside home and use of domestic filtration systems may have introduced nondifferential misclassification in the waterborne ingested estimates. However, the reported amount of water consumed at work (mean±SD: 0.2±0.3L/d) and other places (0.01±0.05L/d) was smaller than that consumed at home (1.2±0.7L/d), and minor bias was expected. As for the use of domestic filters, a reduction of THMs levels has been reported.70 Although there were no statistics on the use of domestic filters in Spain for the study exposure window, expert knowledge suggests that the use of domestic water filters during the exposure window was most likely infrequent. Overall, the expected effect on the associations from exposure measurement error is attenuation toward the null,71 as has been shown for other residence-based exposures.72 This could partly explain the lack of association between waterborne ingested chloroform and prostate cancer, whereas residential levels were associated.
Frequency of routine monitoring by water zone is determined by the population served; thus the number of measurements and accuracy of exposure estimates are expected to be higher in large municipalities (or cities), which in turn concentrate most of the study participants. Although the number of measurements below the QL was small (e.g., ≈5% for nitrate), the imputation of values below the QL may have introduced nondifferential error at the low-exposure range that may have attenuated associations for the continuous variable. However, this approach likely did not affect the results based on exposure categories, because values <QL remain in the referent category. The use of the average instead of geometric mean to calculate long-term levels was due to the constraints of the data provided by water operators, which mostly reported averages and did not provide the raw database. We speculate that this use of the average has probably reduced the accuracy of the exposure estimates, which in turn has probably led to attenuated associations, although bias away from the null cannot be excluded.
Personal information was collected retrospectively after diagnosis, and differential recall between cases and controls may not be totally ruled out. However, the questionnaire was administered by trained personnel in a face-to-face interview, and there is no obvious link between the water questions and prostate cancer that could motivate different responses between cases and controls. Other questions more prone to differential recall bias (e.g., occupational history) are not included in the present analysis. Moreover, the interviewers rated the quality of the interview, and unreliable or inconsistent interviews were excluded from our analyses. Thus, minor differential recall bias is expected to affect the results on waterborne ingested nitrate/THMs. Selection bias arising from control sampling might be of concern. Response rates were moderate, especially among controls, and is partly explained by the population-based source as opposed to hospital-based cases. Controls had a slightly higher educational level in comparison with cases, and all logistic regression models were adjusted for education. In addition, we conducted stratified analysis to identify eventual effect modification on the associations, suggesting that unaccounted factors related to socioeconomic status may be relevant in the association with nitrate. Finally, the probability of participation can be assumed to be independent from the exposure, and nondifferential bias, if any, is expected. Residual confounding by unmeasured factors including environmental exposures with geographical distribution, such as air pollution, green spaces, neighborhoods, or other drinking-water contaminants, cannot be ruled out. It was not possible to perform analyses by recruitment area due to the limited within area variability, and we conducted mixed models with area as random effect. This approach indirectly adjusted for environmental factors geographically distributed, and expected effect on results is minimal because correlation of these factors with our main exposures is unlikely.
Strengths of this study are the relatively large sample size, the long-term exposure approach (from 18 y of age to 2 y before the study interview), and detailed individual information on a range of covariables. These elements facilitated the assessment of several potential confounders and effect modifiers and assessment of coexposure to two main water contaminants.
Conclusions
Findings suggest long-term waterborne ingested nitrate could be a risk factor of prostate cancer, particularly for aggressive tumors and in men <66 y old. A high dietary intake of fiber, fruits and vegetables, or vitamin C may reduce this negative effect of drinking-water nitrate. Association with residential levels but not ingested chloroform/Br-THM may suggest inhalation and dermal routes could be relevant for prostate cancer. Further studies are warranted to draw firm conclusions.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors would like to acknowledge all the research assistants and interviewers in the study centers, the staff of all participating hospitals, and, most of all, the study participants.
The study was partially funded by the “Accion Transversal del Cancer,” approved on the Spanish Ministry Council on 11 October 2007, by the Instituto de Salud Carlos III-FEDER (PI08/1770, PI08/0533, PI08/1359, PS09/00773, PS09/01286, PS09/01903, PS09/02078, PS09/01662, PI11/01889, PI11/00226, PI12/01270, PI12/00715, PI14/0613, PI15/00914, PI17CIII/00034), by the Fundación Marqués de Valdecilla (API 10/09), by the Conselleria de Sanitat of the Generalitat Valenciana (AP_061/10), by the European Commission grants FOOD-CT-2006-036224-HIWATE, by the Spanish Association Against Cancer (AECC) Scientific Foundation, by the Catalan Government-Agency for Management of University and Research Grants (AGAUR) grants 2017SGR723 and 2014SGR850, by the Fundación Caja de Ahorros de Asturias, and by the University of Oviedo. ISGlobal acknowledges support from the Spanish Ministry of Science and Innovation through the “Centro de Excelencia Severo Ochoa 2019–2023” Program (CEX2018-000806-S) and support from the Generalitat de Catalunya through the CERCA Program.
==== Refs
References
1. Rawla P. 2019. Epidemiology of prostate cancer. World J Oncol 10 (2 ):63–89, PMID: , 10.14740/wjon1191.31068988
2. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. 2021. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 71 (3 ):209–249, PMID: , 10.3322/caac.21660.33538338
3. Galceran J, Ameijide A, Carulla M, Mateos A, Quirós JR, Rojas D, et al. REDECAN Working Group. 2017. Cancer incidence in Spain, 2015. Clin Transl Oncol 19 (7 ):799–825, PMID: , 10.1007/s12094-016-1607-9.28093701
4. IARC (International Agency for Research on Cancer). 2018. Global Cancer Observatory: Cancer Today. Ferlay J, Ervik M, Lam F, Colombet M, Mery L, Piñeros M, et al. , eds. Lyon, France: IARC.
5. Eeles RA, Olama AAA, Benlloch S, Saunders EJ, Leongamornlert DA, Tymrakiewicz M, et al. 2013. Identification of 23 new prostate cancer susceptibility loci using the iCOGS custom genotyping array. Nat Genet 45 (4 ):385–391, PMID: , 10.1038/ng.2560.23535732
6. Tan SH, Petrovics G, Srivastava S. 2018. Prostate cancer genomics: recent advances and the prevailing underrepresentation from racial and ethnic minorities. Int J Mol Sci 19 (4 ):1255, PMID: , 10.3390/ijms19041255.29690565
7. Giovannucci E, Liu Y, Platz EA, Stampfer MJ, Willett WC. 2007. Risk factors for prostate cancer incidence and progression in the Health Professionals Follow-Up Study. Int J Cancer 121 (7 ):1571–1578, PMID: , 10.1002/ijc.22788.17450530
8. Chamie K, DeVere White RW, Lee D, Ok JH, Ellison LM. 2008. Agent Orange exposure, Vietnam War veterans, and the risk of prostate cancer. Cancer 113 (9 ):2464–2470, PMID: , 10.1002/cncr.23695.18666213
9. Pernar CH, Ebot EM, Wilson KM, Mucci LA. 2018. The epidemiology of prostate cancer. Cold Spring Harb Perspect Med 8 (12 ):a030361, PMID: , 10.1101/cshperspect.a030361.29311132
10. Gandaglia G, Leni R, Bray F, Fleshner N, Freedland SJ, Kibel A, et al. 2021. Epidemiology and prevention of prostate cancer. Eur Urol Oncol 4 (6 ):877–892, PMID: , 10.1016/j.euo.2021.09.006.34716119
11. Perez-Cornago A, Key TJ, Allen NE, Fensom GK, Bradbury KE, Martin RM, et al. 2017. Prospective investigation of risk factors for prostate cancer in the UK Biobank cohort study. Br J Cancer 117 (10 ):1562–1571, PMID: , 10.1038/bjc.2017.312.28910820
12. Gore AC, Chappell VA, Fenton SE, Flaws JA, Nadal A, Prins GS, et al. 2015. EDC-2: the Endocrine Society’s second scientific statement on endocrine-disrupting chemicals. Endocr Rev 36 (6 ):E1–E150, PMID: , 10.1210/er.2015-1010.26544531
13. De Falco M, Laforgia V. 2021. Combined effects of different endocrine-disrupting chemicals (EDCs) on prostate gland. Int J Environ Res Public Health 18 (18 ):9772, PMID: , 10.3390/ijerph18189772.34574693
14. Koutros S, Beane Freeman LE, Lubin JH, Heltshe SL, Andreotti G, Barry KH, et al. 2013. Risk of total and aggressive prostate cancer and pesticide use in the Agricultural Health Study. Am J Epidemiol 177 (1 ):59–74, PMID: , 10.1093/aje/kws225.23171882
15. Chazelas E, Pierre F, Druesne-Pecollo N, Esseddik Y, Szabo de Edelenyi F, Agaesse C, et al. 2022. Nitrites and nitrates from food additives and natural sources and cancer risk: results from the NutriNet-Santé cohort. Int J Epidemiol 51 (4 ):1106–1119, PMID: , 10.1093/ije/dyac046.35303088
16. Quinn M, Babb P. 2002. Patterns and trends in prostate cancer incidence, survival, prevalence and mortality. Part I: international comparisons. BJU Int 90 (2 ):162–173, PMID: , 10.1046/j.1464-410x.2002.2822.x.12081758
17. Dasgupta P, Baade PD, Aitken JF, Ralph N, Chambers SK, Dunn J. 2019. Geographical variations in prostate cancer outcomes: a systematic review of international evidence. Front Oncol 9 :238, PMID: , 10.3389/fonc.2019.00238.31024842
18. Rodriguez-Sanchez L, Fernández-Navarro P, López-Abente G, Nuñez O, Fernández de Larrea-Baz N, Jimenez-Moleón JJ, et al. 2019. Different spatial pattern of municipal prostate cancer mortality in younger men in Spain. PLoS One 14 (1 ):e0210980, PMID: , 10.1371/journal.pone.0210980.30682085
19. Ward MH, Jones RR, Brender JD, de Kok TM, Weyer PJ, Nolan BT, et al. 2018. Drinking water nitrate and human health: an updated review. Int J Environ Res Public Health 15 (7 ):1557, PMID: , 10.3390/ijerph15071557.30041450
20. Espejo-Herrera N, Gracia-Lavedan E, Pollan M, Aragonés N, Boldo E, Perez-Gomez B, et al. 2016. Ingested nitrate and breast cancer in the Spanish Multicase-Control Study on Cancer (MCC-Spain). Environ Health Perspect 124 (7 ):1042–1049, PMID: , 10.1289/ehp.1510334.26942716
21. Grosse Y, Baan R, Straif K, Secretan B, El Ghissassi F, Cogliano V, et al. 2006. Carcinogenicity of nitrate, nitrite, and cyanobacterial peptide toxins. Lancet Oncol 7 (8 ):628–629, PMID: , 10.1016/s1470-2045(06)70789-6.16900606
22. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. 2010. IARC monographs on the evaluation of carcinogenic risks to humans. Ingested nitrate and nitrite, and cyanobacterial peptide toxins. IARC Monogr Eval Carcinog Risks Hum 94 :v–vii, 1–412, PMID: .21141240
23. Richardson SD, Thruston Jr AD, Caughran TV, Chen PH, Collette TW, Schenck KM, et al. 2000. Identification of new drinking water disinfection by-products from ozone, chlorine dioxide, chloramine, and chlorine. Water Air Soil Pollut 123 :95–102, 10.1023/A:1005265509813.
24. Kristiana I, Tan J, Joll CA, Heitz A, von Gunten U, Charrois JW. 2013. Formation of N-nitrosamines from chlorination and chloramination of molecular weight fractions of natural organic matter. Water Res 47 (2 ):535–546, PMID: , 10.1016/j.watres.2012.10.014.23164216
25. Richardson SD, Plewa MJ, Wagner ED, Schoeny R, Demarini DM. 2007. Occurrence, genotoxicity, and carcinogenicity of regulated and emerging disinfection by-products in drinking water: a review and roadmap for research. Mutat Res 636 (1–3 ):178–242, PMID: , 10.1016/j.mrrev.2007.09.001.17980649
26. Wilbourn JD, Partensky C, Rice JM. 1999. Agents that induce epithelial neoplasms of the urinary bladder, renal cortex and thyroid follicular lining in experimental animals and humans: summary of data from IARC monographs volumes 1–69. IARC Sci Publ 147 :191–209, PMID: .10457918
27. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. 2013. Some Chemicals Present in Industrial and Consumer Products, Food and Drinking-Water. IARC Monogr Eval Carcinog Risks Hum 101 :9–549. https://publications.iarc.fr/_publications/media/download/5700/5febe9a9b9aae2224f6c0cc3db1c430149ec5593.pdf [accessed 24 February 2023].24772663
28. Chang SL. 1982. The safety of water disinfection. Annu Rev Public Health 3 :393–418, PMID: , 10.1146/annurev.pu.03.050182.002141.6756434
29. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. 1999. Some Chemicals that Cause Tumours of the Kidney or Urinary Bladder in Rodents and Some Other Substances. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, No. 73. Lyon, France: IARC.
30. Castano-Vinyals G, Aragones N, Perez-Gomez B, Martén V, Llorca J, Moreno V, et al. 2015. Population-based multicase–control study in common tumors in Spain (MCC-Spain): rationale and study design. Gac Sanit 29 (4 ):308–315, PMID: , 10.1016/j.gaceta.2014.12.003.25613680
31. Martin-Moreno JM, Boyle P, Gorgojo L, Maisonneuve P, Fernandez-Rodriguez JC, Salvini S, et al. 1993. Development and validation of a food frequency questionnaire in Spain. Int J Epidemiol 22 (3 ):512–519, PMID: , 10.1093/ije/22.3.512.8359969
32. Epstein JI, Egevad L, Amin MB, Delahunt B, Srigley JR, Humphrey PA, Grading Committee. 2016. The 2014 International Society of Urological Pathology (ISUP) consensus conference on Gleason grading of prostatic carcinoma: definition of grading patterns and proposal for a new grading system. Am J Surg Pathol 40 (2 ):244–252, PMID: , 10.1097/PAS.0000000000000530.26492179
33. Hurwitz LM, Agalliu I, Albanes D, Barry KH, Berndt SI, Cai Q, et al. 2021. Recommended definitions of aggressive prostate cancer for etiologic epidemiologic research. J Natl Cancer Inst 113 (6 ):727–734, PMID: , 10.1093/jnci/djaa154.33010161
34. Croghan C, Egeghy PP. 2003. Methods of dealing with values below the limit of detection using SAS. Southern SAS User Group 22 :24. St. Petersburg, FL, September 22-24, 2003. https://cfpub.epa.gov/si/si_public_record_report.cfm?Lab=NERL&dirEntryId=64046.
35. Espejo-Herrera N, Kogevinas M, Castaño-Vinyals G, Aragonés N, Boldo E, Ardanaz E, et al. 2013. Nitrate and trace elements in municipal and bottled water in Spain. Gac Sanit 27 (2 ):156–160, PMID: , 10.1016/j.gaceta.2012.02.002.22444517
36. Font-Ribera L, Kogevinas M, Nieuwenhuijsen MJ, Grimalt JO, Villanueva CM. 2010. Patterns of water use and exposure to trihalomethanes among children in Spain. Environ Res 110 (6 ):571–579, PMID: , 10.1016/j.envres.2010.05.008.20554274
37. Villanueva CM, Kogevinas M, Grimalt JO. 2003. Haloacetic acids and trihalomethanes in finished drinking waters from heterogeneous sources. Water Res 37 (4 ):953–958, PMID: , 10.1016/s0043-1354(02)00411-6.12531279
38. Olmedo-Requena R, Lozano-Lorca M, Salcedo-Bellido I, Jiménez-Pacheco A, Vázquez-Alonso F, García-Caballos M, et al. 2020. Compliance with the 2018 World Cancer Research Fund/American Institute for Cancer Research cancer prevention recommendations and prostate cancer. Nutrients 12 (3 ):768, PMID: , 10.3390/nu12030768.32183345
39. Shams-White MM, Brockton NT, Mitrou P, Romaguera D, Brown S, Bender A, et al. 2019. Operationalizing the 2018 World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) cancer prevention recommendations: a standardized scoring system. Nutrients 11 (7 ):1572, PMID: , 10.3390/nu11071572.31336836
40. Romaguera D, Gracia-Lavedan E, Molinuevo A, de Batlle J, Mendez M, Moreno V, et al. 2017. Adherence to nutrition-based cancer prevention guidelines and breast, prostate and colorectal cancer risk in the MCC-Spain case-control study. Int J Cancer 141 (1 ):83–93, PMID: , 10.1002/ijc.30722.28380695
41. Ashley DL, Blount BC, Singer PC, Depaz E, Wilkes C, Gordon S, et al. 2005. Changes in blood trihalomethane concentrations resulting from differences in water quality and water use activities. Arch Environ Occup Health 60 (1 ):7–15, PMID: , 10.3200/AEOH.60.1.7-15.16961003
42. Gordon SM, Brinkman MC, Ashley DL, Blount BC, Lyu C, Masters J, et al. 2006. Changes in breath trihalomethane levels resulting from household water-use activities. Environ Health Perspect 114 (4 ):514–521, PMID: , 10.1289/ehp.8171.16581538
43. Costet N, Villanueva CM, Jaakkola JJK, Kogevinas M, Cantor KP, King WD, et al. 2011. Water disinfection by-products and bladder cancer: is there a European specificity? A pooled and meta-analysis of European case-control studies. Occup Environ Med 68 (5 ):379–385, PMID: , 10.1136/oem.2010.062703.21389011
44. Chan JM, Gann PH, Giovannucci EL. 2005. Role of diet in prostate cancer development and progression. J Clin Oncol 23 (32 ):8152–8160, PMID: , 10.1200/JCO.2005.03.1492.16278466
45. Zhou CK, Check DP, Lortet-Tieulent J, Laversanne M, Jemal A, Ferlay J, et al. 2016. Prostate cancer incidence in 43 populations worldwide: an analysis of time trends overall and by age group. Int J Cancer 138 (6 ):1388–1400, PMID: , 10.1002/ijc.29894.26488767
46. Ding Y, Wu H, Warden C, Steele L, Liu X, Iterson MV, et al. 2016. Gene expression differences in prostate cancers between young and old men. PLoS Genet 12 (12 ):e1006477, PMID: , 10.1371/journal.pgen.1006477.28027300
47. Salinas CA, Tsodikov A, Ishak-Howard M, Cooney KA. 2014. Prostate cancer in young men: an important clinical entity. Nat Rev Urol 11 (6 ):317–323, PMID: , 10.1038/nrurol.2014.91.24818853
48. Castaño-Vinyals G, Cantor KP, Villanueva CM, Tardon A, Garcia-Closas R, Serra C, et al. 2011. Socioeconomic status and exposure to disinfection by-products in drinking water in Spain. Environ Health 10 :18, PMID: , 10.1186/1476-069X-10-18.21410938
49. Vrijheid M, Martinez D, Aguilera I, Ballester F, Basterrechea M, Esplugues A, et al. 2012. Socioeconomic status and exposure to multiple environmental pollutants during pregnancy: evidence for environmental inequity? J Epidemiol Community Health 66 (2 ):106–113, PMID: , 10.1136/jech.2010.117408.20974841
50. Bryan NS, Alexander DD, Coughlin JR, Milkowski AL, Boffetta P. 2012. Ingested nitrate and nitrite and stomach cancer risk: an updated review. Food Chem Toxicol 50 (10 ):3646–3665, PMID: , 10.1016/j.fct.2012.07.062.22889895
51. Bartsch H, Ohshima H, Pignatelli B. 1988. Inhibitors of endogenous nitrosation. Mechanisms and implications in human cancer prevention. Mutat Res 202 (2 ):307–324, PMID: , 10.1016/0027-5107(88)90194-7.3057363
52. Mergens WJ, Kamm JJ, Newmark HL, Fiddler W, Pensabene J. 1978. Alpha-tocopherol: uses in preventing nitrosamine formation. IARC Sci Publ 19 :199–212, PMID: .28277
53. Mirvish SS. 1975. Blocking the formation of N-nitroso compounds with ascorbic acid in vitro and in vivo. Ann NY Acad Sci 258 :175–180, PMID: , 10.1111/j.1749-6632.1975.tb29277.x.1106296
54. Mirvish SS. 1996. Inhibition by vitamins C and E of in vivo nitrosation and vitamin C occurrence in the stomach. Eur J Cancer Prev 5 (suppl 1 ):131–136, PMID: , 10.1097/00008469-199609000-00027.8972308
55. Bidoli E, Talamini R, Zucchetto A, Bosetti C, Negri E, Lenardon O, et al. 2009. Dietary vitamins E and C and prostate cancer risk. Acta Oncol 48 (6 ):890–894, PMID: , 10.1080/02841860902946546.19452333
56. Helzlsouer KJ, Huang HY, Alberg AJ, Hoffman S, Burke A, Norkus EP, et al. 2000. Association between alpha-tocopherol, gamma-tocopherol, selenium, and subsequent prostate cancer. J Natl Cancer Inst 92 (24 ):2018–2023, PMID: , 10.1093/jnci/92.24.2018.11121464
57. McCann SE, Ambrosone CB, Moysich KB, Brasure J, Marshall JR, Freudenheim JL, et al. 2005. Intakes of selected nutrients, foods, and phytochemicals and prostate cancer risk in western New York. Nutr Cancer 53 (1 ):33–41, PMID: , 10.1207/s15327914nc5301_4.16351504
58. Mastrangelo D, Pelosi E, Castelli G, Lo-Coco F, Testa U. 2018. Mechanisms of anti-cancer effects of ascorbate: cytotoxic activity and epigenetic modulation. Blood Cells Mol Dis 69 :57–64, PMID: , 10.1016/j.bcmd.2017.09.005.28954710
59. Vissers MCM, Das AB. 2018. Potential mechanisms of action for vitamin C in cancer: reviewing the evidence. Front Physiol 9 :809, PMID: , 10.3389/fphys.2018.00809.30018566
60. Farag MA, Shakour ZTA, Elmassry MM, Donia MS. 2022. Metabolites profiling reveals gut microbiome-mediated biotransformation of green tea polyphenols in the presence of N-nitrosamine as pro-oxidant. Food Chem 371 :131147, PMID: , 10.1016/j.foodchem.2021.131147.34808759
61. Reichard CA, Naelitz BD, Wang Z, Jia X, Li J, Stampfer MJ, et al. 2022. Gut microbiome-dependent metabolic pathways and risk of lethal prostate cancer: prospective analysis of a PLCO cancer screening trial cohort. Cancer Epidemiol Biomarkers Prev 31 (1 ):192–199, PMID: , 10.1158/1055-9965.EPI-21-0766.34711629
62. Fujita K, Matsushita M, Banno E, De Velasco MA, Hatano K, Nonomura N, et al. 2022. Gut microbiome and prostate cancer. Int J Urol 29 (8 ):793–798, PMID: , 10.1111/iju.14894.35388531
63. Wilkins JR 3rd, Comstock GW. 1981. Source of drinking water at home and site-specific cancer incidence in Washington County, Maryland. Am J Epidemiol 114 (2 ):178–190, PMID: , 10.1093/oxfordjournals.aje.a113181.7304553
64. Koivusalo M, Pukkala E, Vartiainen T, Jaakkola JJ, Hakulinen T. 1997. Drinking water chlorination and cancer–a historical cohort study in Finland. Cancer Causes Control 8 (2 ):192–200, PMID: , 10.1023/a:1018420229802.9134243
65. Leavens TL, Blount BC, DeMarini DM, Madden MC, Valentine JL, Case MW, et al. 2007. Disposition of bromodichloromethane in humans following oral and dermal exposure. Toxicol Sci 99 (2 ):432–445, PMID: , 10.1093/toxsci/kfm190.17656487
66. Landi S, Hanley NM, Warren SH, Pegram RA, DeMarini DM. 1999. Induction of genetic damage in human lymphocytes and mutations in Salmonella by trihalomethanes: role of red blood cells and GSTT1-1 polymorphism. Mutagenesis 14 (5 ):479–482, PMID: , 10.1093/mutage/14.5.479.10473651
67. Plewa MJ, Muellner MG, Richardson SD, Fasano F, Buettner KM, Woo Y-T, et al. 2008. Occurrence, synthesis, and mammalian cell cytotoxicity and genotoxicity of haloacetamides: an emerging class of nitrogenous drinking water disinfection byproducts. Environ Sci Technol 42 (3 ):955–561, PMID: , 10.1021/es071754h.18323128
68. Villanueva CM, Castaño-Vinyals G, Moreno V, Carrasco-Turigas G, Aragonés N, Boldo E, et al. 2012. Concentrations and correlations of disinfection by-products in municipal drinking water from an exposure assessment perspective. Environ Res 114 :1–11, PMID: , 10.1016/j.envres.2012.02.002.22436294
69. Komulainen H, Kosma VM, Vaittinen SL, Vartiainen T, Kaliste-Korhonen E, Lötjönen S, et al. 1997. Carcinogenicity of the drinking water mutagen 3-chloro-4-(dichloromethyl)-5-hydroxy-2(5H)-furanone in the rat. J Natl Cancer Inst 89 (12 ):848–856, PMID: , 10.1093/jnci/89.12.848.9196250
70. Egorov AI, Tereschenko AA, Altshul LM, Vartiainen T, Samsonov D, LaBrecque B, et al. 2003. Exposures to drinking water chlorination by-products in a Russian city. Int J Hyg Environ Health 206 (6 ):539–551, PMID: , 10.1078/1438-4639-00244.14626901
71. Rothman KJ, Greenland S, Lash TL. 2008. Modern Epidemiology, Vol. 3. Philadelphia, PA: Wolters Kluwer Health/Lippincott Williams & Wilkins.
72. Wei Y, Qiu X, Yazdi MD, Shtein A, Shi L, Yang J, et al. 2022. The impact of exposure measurement error on the estimated concentration-response relationship between long-term exposure to PM2.5 and mortality. Environ Health Perspect 130 (7 ):77006, PMID: , 10.1289/EHP10389.35904519
|
PMC009xxxxxx/PMC9994489.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36884004
EHP12503
10.1289/EHP12503
Invited Perspective
Invited Perspective: Leveraging Air Quality Research Advancements to Assess and Address Exposure Inequities
https://orcid.org/0000-0003-1787-9851
Goldman Gretchen T. 1
1 Union of Concerned Scientists, Washington, District of Columbia, USA
Address correspondence to Gretchen T. Goldman, 1825 K St. NW, Suite 800, Washington, DC 20006 USA. Email: [email protected]
8 3 2023
3 2023
131 3 03130228 11 2022
19 12 2022
27 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The author declares she has no potential or actual competing interests to disclose.
This perspective was written in the author’s personal capacity. All opinions expressed by the author are solely based on her knowledge and experience and are not representative of her professional affiliations, the White House Office of Science and Technology Policy, or the U.S. Government.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org/10.1289/EHP11605
==== Body
pmcReductions in air pollution emissions, whether via policy, economics, or other factors, can benefit public health and welfare. However, additional scrutiny can reveal that the distribution and extent of such benefits vary widely across geography, time, and populations.
Illuminating a case of such variability in emissions reductions, Henneman et al., in this issue of Environmental Health Perspectives, quantified nationwide long-term decreases in exposure to fine particulate matter (PM2.5, particulate matter with aerodynamic diameter ≤2.5μm) associated with sulfur dioxide (SO2) emissions from coal-fired power plants. The authors attributed this decrease to scrubber installations (especially between 2007 and 2010), reduced operations, and plant unit retirements (especially after 2010).1 They further assessed how emissions changes in different locations affected inequities in exposure to associated PM2.5 across demographic groups. They found that nationwide population-weighted exposures to coal-related PM2.5 declined from 1.96μg/m3 in 1999 to 0.06μg/m3 in 2020.
Emissions from coal-fired power plants and fossil fuel combustion emissions at large are a significant contributor to the global health burden from air pollution.2,3 The findings reported by Henneman et al. underscore both the potential for policy measures to guide improved health outcomes and the role of individual plants in driving down exposures. The latter leads to their second major finding: Although emissions reductions had the effect of decreasing inequities overall, Black populations continue to be disproportionately exposed to coal-related PM2.5 from facilities in states across the North Central United States, and Native populations continue to be disproportionately exposed by facilities in the West.
This finding aligns with other work that has demonstrated inequities in how emissions reductions and exposures are distributed across regions and demographic groups. Jbaily et al.,4 for example, found that disparities in exposure relative to safety standards set by the U.S. Environmental Protection Agency (U.S. EPA) and the World Health Organization have increased over time, informing the kinds of PM2.5 emissions-reductions strategies that might be most effective (or least effective) at reducing harmful emissions in communities overburdened with multiple air pollution emissions sources. Further, Wang et al.5 assessed how different emissions-reduction strategies would affect “national inequality,” defined by the authors as the difference between the average exposure of the most exposed racial/ethnic group vs. that of the overall population. They estimated that national inequalities in exposure can be eliminated with relatively minor emissions reductions if specific locations are strategically targeted to maximally reduce disparities for the most exposed group relative to the overall population average. Wang et al. found that accomplishing a similar outcome relying on existing U.S. policy mechanisms—i.e., emissions-based standards for economic sectors and ambient concentration standards—would either be impossible or require elimination of all emissions.
Such inequities have persisted despite significant reductions in ambient air pollution over recent decades. The U.S. Clean Air Act, for example, has been wildly successful. In 2011, the U.S. EPA estimated the 1990 Clean Air Act Amendments would prevent more than 230,000 early deaths by 2020.6 Although concentrations of many ambient air pollutants have fallen in recent decades nationally, benefits have not been enjoyed equally across the population.4 Improvements in air quality have been slower in communities of color and low-income communities; in some instances, this effect has actually worsened exposure disparities.4
As science and technology advance, we have a critical opportunity for researchers to push the bounds of our understanding of these exposure inequities and how to address them. Researchers are using better models, more precise measurements, additional remote sensing products, and other innovations to assess air pollution exposure inequities at unprecedented scales over time and space. In doing so, researchers are providing new opportunities to inform decision makers on the efficacy of policy measures and other interventions.
To reduce pollution exposure inequities and associated health effects, which emissions-reduction strategies would be most effective? Where should cities, states, tribal nations, and countries put resources to make the greatest air quality improvements? What are the exposure and health consequences of specific policy choices now and in the future? These critical policy questions must be answered if our society seeks to tackle environmental injustices, and researchers have a central role to play in generating useful evidence to inform these answers.
In my career-spanning research, policy, and environmental justice community partnerships, I have often observed disconnects between the focus of researchers, the scope of policy makers, and the interests of affected communities.7,8 When we bridge these gaps, we advance not only policy-relevant research and evidence-based policy decisions but better outcomes in the real world—a world in which fewer communities are disproportionately burdened with air pollution and in which cleaner air affords everyone the opportunity to live healthier lives.
==== Refs
References
1. Henneman LRF, Rasel MM, Choirat C, Anenberg S, Zigler C. 2023. Inequitable exposures to U.S. coal power plant–related PM2.5: 22 years and counting. Environ Health Perspect 131 (3 ):037005, 10.1289/EHP11605.36884005
2. McDuffie E, Martin R, Yin H, Brauer M. 2021. Global burden of disease from major air pollution sources (GBD MAPS): a global approach. Res Rep Health Eff Inst 210 :1–45, PMID: .36148817
3. Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K, et al. 2017. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 389 (10082 ):1907–1918, PMID: , 10.1016/S0140-6736(17)30505-6.28408086
4. Jbaily A, Zhou X, Liu J, Lee TH, Kamareddine L, Verguet S, et al. 2022. Air pollution exposure disparities across US population and income groups. Nature 601 (7892 ):228–233, PMID: , 10.1038/s41586-021-04190-y.35022594
5. Wang Y, Apte JS, Hill JD, Ivey CE, Patterson RF, Robinson AL, et al. 2022. Location-specific strategies for eliminating US national racial-ethnic PM2.5 exposure inequality. Proc Natl Acad Sci USA 119 (44 ):e2205548119, PMID: , 10.1073/pnas.2205548119.36279443
6. U.S. Environmental Protection Agency. 2011. The Benefits and Costs of the Clean Air Act from 1990 to 2020. Washington, DC: U.S. Environmental Protection Agency. https://www.epa.gov/sites/default/files/2015-07/documents/fullreport_rev_a.pdf [accessed 25 February 2023].
7. Goldman GT, Desikan A, Morse R, Kalman C, MacKinney T, Cohan DS, et al. 2021. Assessment of air pollution impacts and monitoring data limitations of a spring 2019 chemical facility fire. Environ Justice 15 (6 ):362–372, 10.1089/env.2021.0030.
8. Declet-Barreto J, Goldman GT, Desikan A, Berman E, Goldman J, Johnson C, et al. 2020. Hazardous air pollutant emissions implications under 2018 guidance on U.S. Clean Air Act requirements for major sources. J Air Waste Manage Assoc 70 (5 ):481–490, PMID: , 10.1080/10962247.2020.1735575.32101104
|
PMC009xxxxxx/PMC9994529.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36884005
EHP11605
10.1289/EHP11605
Research
Inequitable Exposures to U.S. Coal Power Plant–Related PM2.5: 22 Years and Counting
https://orcid.org/0000-0003-1957-8761
Henneman Lucas R.F. 1
Rasel Munshi Md 1
Choirat Christine 2
Anenberg Susan C. 3
Zigler Corwin 4
1 Department of Civil, Environmental, and Infrastructure Engineering; George Mason University, Fairfax, Virginia, USA
2 Swiss Data Science Center, ETH Zürich and EPFL, Lausanne, Switzerland
3 Department of Environmental and Occupational Health, George Washington University, Washington, District of Columbia, USA
4 Department of Statistics and Data Sciences, University of Texas, Austin, USA
Address correspondence to Lucas R.F. Henneman, 4400 University Dr., MS-6C1, Fairfax, VA 22030 USA. Email: [email protected]
8 3 2023
3 2023
131 3 03700523 5 2022
23 1 2023
27 1 2023
15 5 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Emissions from coal power plants have decreased over recent decades due to regulations and economics affecting costs of providing electricity generated by coal vis-à-vis its alternatives. These changes have improved regional air quality, but questions remain about whether benefits have accrued equitably across population groups.
Objectives:
We aimed to quantify nationwide long-term changes in exposure to particulate matter (PM) with an aerodynamic diameter ≤2.5μm (PM2.5) associated with coal power plant SO2 emissions. We linked exposure reductions with three specific actions taken at individual power plants: scrubber installations, reduced operations, and retirements. We assessed how emissions changes in different locations have influenced exposure inequities, extending previous source-specific environmental justice analyses by accounting for location-specific differences in racial/ethnic population distributions.
Methods:
We developed a data set of annual PM2.5 source impacts (“coal PM2.5”) associated with SO2 emissions at each of 1,237 U.S. coal-fired power plants across 1999–2020. We linked population-weighted exposure with information about each coal unit’s operational and emissions-control status. We calculate changes in both relative and absolute exposure differences across demographic groups.
Results:
Nationwide population-weighted coal PM2.5 declined from 1.96μg/m3 in 1999 to 0.06 μg/m3 in 2020. Between 2007 and 2010, most of the exposure reduction is attributable to SO2 scrubber installations, and after 2010 most of the decrease is attributable to retirements. Black populations in the South and North Central United States and Native American populations in the western United States were inequitably exposed early in the study period. Although inequities decreased with falling emissions, facilities in states across the North Central United States continue to inequitably expose Black populations, and Native populations are inequitably exposed to emissions from facilities in the West.
Discussion:
We show that air quality controls, operational adjustments, and retirements since 1999 led to reduced exposure to coal power plant related PM2.5. Reduced exposure improved equity overall, but some populations continue to be inequitably exposed to PM2.5 associated with facilities in the North Central and western United States. https://doi.org/10.1289/EHP11605
Supplemental Material is available online (https://doi.org/10.1289/EHP11605).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Electricity generation from U.S. coal power plants was historically a major contributor to poor air quality—notably in the form of particulate matter (PM) and ozone from sulfur and nitrogen emissions.1–3 To curb pollution from coal power plants, the U.S. Congress enacted coal power plant emissions regulations under the 1990 Clean Air Act Amendments, and the U.S. Environmental Protection Agency (U.S. EPA) has further tightened emissions standards under the 1990 Amendments with the Acid Rain Program, NOx Budget Trading Program and State Implementation Plan (SIP) Call, Clean Air Interstate Rule, and Cross-State Air Pollution Rule.4–7 National Ambient Air Quality Standards (NAAQS), tightened periodically by the U.S. EPA and attained through state actions, have led to further emissions reductions.8 Since the early 2010s, much of the new electricity generation capacity in the U.S. electricity generation sector has come from natural gas; no new coal power plants have been built in the contiguous United States since 2015.9
These programs resulted in regional improvements in air quality and health.10–15 In addition, interventions on individual facilities have been shown to improve air quality and population health in local populations.16–18 It is less clear, however, whether the overall benefits of these regulations have accrued equally across different populations; researchers have identified exposure inequities related to U.S. power plant air pollution impacts,19,20 but evidence is lacking on how major emissions reductions over the last two decades have reduced disparities. Although the Clean Air Act Amendments did not address inequity directly, multiple executive orders21,22 have directed the U.S. EPA to address environmental justice (EJ). Notably, the Biden administration’s Justice40 initiative establishes the goal that 40% of certain federal investments will go toward “disadvantaged communities that are marginalized, underserved, and overburdened by pollution.” Harper et al. describe how quantitative comparisons between and within population health status and exposure are important for designing regulatory actions that address EJ.23 For regulations targeting specific air pollution source categories (e.g., electricity generation units), the U.S. EPA’s definition of EJ suggests the need to quantify health and/or exposure inequities associated with the regulated source and its change in response to regulations.
Because the coal power plant sector is a combination of individual point sources at which operational decisions and control installations attributable to regulatory action happen semi-independently, thorough understanding of evolving exposure associated with dramatic sector-wide emissions reductions is best achieved by assessing each power plant’s historical contribution to exposure and inequities.19,20,24 For example, in crafting its recent climate-focused Clean Power Plan (later replaced by the now-vacated Affordable Clean Energy Act25) the U.S. EPA identified many facilities located in neighborhoods with higher environmental burdens and vulnerable populations.26 The U.S. EPA continues to provide information on EJ issues for populations surrounding power plants through its Power Plants and Neighboring Communities online tool and communicates that these issues will continue to be accounted for in future power plant regulations.27
The U.S. EPA’s tool, however, only accounts for populations living within 3 mi of the facilities (thereby missing populations impacted by pollution transport), and this tool and previous work in this area have focused on a single year’s exposure. The desire to expand this previous work to investigate trends led to the development of the HYSPLIT with Average Dispersion (HyADS) model, which is computationally nimble enough to simulate individual unit spatial impacts over long periods. Although other reduced complexity models such as the Intervention Model for Air Pollution (InMAP), the Estimating Air pollution Social Impact Using Regression model (EASIUR), and the Air Pollution Emission Experiments and Policy analysis model (APEEP) have been applied to estimate population damages and power generation source exposure inequities,20,28–34 the HyADS model is appropriate for long-term simulations because it accounts for both meteorological variability and emissions changes over time.
We sought to quantify nationwide long-term changes in exposure and inequities to PM2.5 associated with coal power plant SO2 emissions because particulate sulfate, an atmospheric product of SO2, has historically been associated with most of the PM2.5-related mortalities from coal (one estimate attributed 75% of annual mortalities from electricity generation to SO2 emissions in 2005, in comparison with 7% from NOx and 14% from primary PM2.5).35 Using HyADS, we estimated annual spatial-temporal PM2.5 source impacts related to SO2 emissions and other correlated precursor emitted species (termed “coal PM2.5” here) from each of over 1,200 coal electricity-generating units (EGUs) in the United States across 1999–2020, the most extensive data set of its kind to our knowledge. We report here the extent that SO2 emissions changes from each EGU influenced national PM2.5 concentrations, and we used facility-level information to quantify the influence of three interventions—retirements, reduced operations, and emissions controls installations—on population exposure. Finally, we performed a detailed disparities assessment that accounts for location-specific differences in racial/ethnic population distributions relative to locations of power plants. The accompanying publicly available data set has further potential for application in long-term nationwide exposure, epidemiological, and regulatory accountability studies.
Methods
This work applied the HyADS model to estimate PM2.5 associated with SO2 emissions from all U.S. coal power plants (coal PM2.5). Attributing changes in coal PM2.5 exposure and exposure disparities to actions taken at each of the over 1,200 coal electricity-generation units necessitated the choice of a reduced complexity model such as HyADS over traditional full-complexity photochemical grid models such as CMAQ, CAMx, and GEOS-Chem.
Emissions and Facility Attributes Data
We employed monthly data from the U.S. EPA Clean Air Markets Division’s Air Markets Program Data (AMPD) database, which houses facility attributes (e.g., location and fuel type) and emissions (SO2 in U.S. tons) information for 480 coal electricity-generation facilities operating in the United States since at least 1999.10 Most facilities comprise multiple units—the base denomination for emissions and emissions controls—and the entire database contains 1,237 units. We supplemented AMPD data with stack height information obtained from the U.S. EPA’s 2014 National Emissions Inventory. Of the 1,237 coal units, 69% have stack height information; for the others, we assigned the mean stack height of all available units (182m).
The HyADS Model
HyADS has been applied previously to address multiple topics relevant for exposure and health impacts studies, including ranking the influence of individual coal plants on specific locations16,36,37 quantifying long-term changes in coal PM2.5 source impacts18; and estimating the relative influence of meteorological and emissions variability on coal PM2.5 source impacts.38 In addition, HyADS has been used to address questions relevant for population health, including investigating the effect of long-term nationwide coal emissions changes on hospitalizations of adults ≥65 y of age18; the influence of short-term interventions at power plants with high local impacts on urban asthma outcomes16; and differential toxicity of coal PM2.5 source impacts relative to bulk PM2.5.18 A key feature of HyADS is its ability to resolve individual source contributions to population exposures to coal PM2.5. Although the overall model framework of HyADS has been described in previous applications, we have made updates reflected in the model description below and in the Supplemental Information (“Updates to the HyADS Model,” Figures S1–S3).
HyADS acts as a wrapper to initiate and process many runs of the HYSPLIT air transport and dispersion model,39,40 and the output is power plant–specific PM2.5 concentrations associated with SO2 emissions. The modeling approach was based on emissions events, for each of which we tracked 100 air parcels. HYSPLIT tracks the transport and dispersion of these air parcels for 7 d using wind fields from the NCEP/NCAR Reanalysis meteorological product, an assimilation of meteorological observations from multiple platforms.41 We repeated emissions events at 6-h intervals each day [2400 hours (12:00 A.M.), 0600 hours (6:00 A.M.), 1200 hours (12:00 P.M.), and 1800 hours (6:00 P.M.)], resulting in [365 days]×[4 emissions events]=1,460 HYSPLIT runs per year for each facility. Air parcel locations were summed by month over a 36-km grid; this resolution was selected because of the regional nature of annual PM2.5 related to SO2 emissions, which is discussed in detail below. At this stage in the modeling process, the intermediate output consisted of monthly gridded air parcel counts for each emissions source.
Next, we multiplied the gridded air parcel counts by each unit’s monthly SO2 emissions. We averaged the monthly spatial emissions-weighted HyADS impacts across the year, accounting for the number of days in each month.
The resulting unitless concentrations at this stage were not directly relatable to measured air pollution species because they corresponded to only a portion of total ambient pollution. To post-process the unitless concentrations to coal PM2.5, we regressed the unitless concentrations against coal PM2.5 source impacts estimated with the Hybrid CMAQ-DDM model.2,42 Hybrid CMAQ-DDM employs the Community Multiscale Air Quality (CMAQ) model with the Direct-Decoupled Method (DDM) to simulate PM2.5 source impact sensitivities from all U.S. coal sources and corrects them using an optimization-based adjustment technique to more closely match observed PM2.5 observations (the approach for estimating coal source impacts is described in detail by Ivey et al.2). We used PM2.5 source impacts for all U.S. coal sources from the Hybrid CMAQ-DDM model in 2005 to post-process the unitless HyADS exposures to PM2.5 source impacts. As described in detail in Henneman et al.37 and updated as detailed in the Supplemental Information (“Updates to the HyADS Model,” Figures S1–S3), the postprocessing was based on the following regression (the model was trained independently for a given period; therefore, we dropped temporal terms from the notation above): (1) log(PM2.5,iCMAQ−DDM)=β0+βHyADS∑u=1UHyADSi,u+βHyADS2(∑u=1UHyADSi,u)2+(∑u=1UHyADSi,u)X⇀iTβX⇀,HyADS+ϵi,
where PM2.5,iCMAQ−DDM is the Hybrid CMAQ-DDM PM2.5 coal source impacts for location i, X⇀iT is the vector of meteorological variables at location i, HyADSi,u for u=1, 2,…, U are the contributions of unitless HyADS concentrations from each of U units to that location, and ϵi is assumed iid normal. β0 is the intercept, which permits some constant amount of PM2.5 predicted by Hybrid CMAQ-DDM coal PM2.5 source impacts that are not explained by HyADS (e.g., PM2.5 related to NOx or primary PM2.5 emissions not otherwise correlated with SO2 emissions). βHyADS, βHyADS2, and βX⇀,HyADS are parameters governing the relationships between each of the corresponding variables with Hybrid CMAQ-DDM PM2.5 coal source impacts, (βX⇀,HyADS corresponds to interactions between HyADS and each meteorological variable). We employed meteorological data from the NCEP/NCAR North American Regional Reanalysis project, including average temperature, accumulated precipitation, relative humidity, and x and y wind vectors. The meteorological data (originally on a ∼32-km grid) and CMAQ-DDM Hybrid PM2.5 source impacts (36km) were spatially projected to the HyADS 36-km grid. A log link was used because Hybrid CMAQ-DDM PM2.5 coal source impacts has an approximately log normal distribution.
CMAQ-DDM Hybrid coal source impacts were available in 2005 and 2006, with year 2006 results used to evaluate the model trained in 2005 (more details and model evaluations are discussed in detail in the Supplemental Information).37 HyADS employs the model trained on 2005 Hybrid CMAQ-DDM results to predict total coal PM2.5 and unit-specific coal PM2.5. We predicted total coal PM2.5 using each year’s ∑u=1UHyADSi,u and meteorology in Model 1, and we subtracted the intercept from the predictions to eliminate the portion of CMAQ-DDM Hybrid coal source impacts not accounted for by raw HyADS. To predict each unit’s annual spatial coal PM2.5, we distributed the coal PM2.5 from all sources to each unit based on its fractional contribution to the total ∑u=1UHyADSi,u at each location.
The model relies on the assumption that SO2 emissions are an important contributor to PM2.5 source impacts through their atmospheric processing to particulate sulfate. To the extent that emissions of other atmospheric constituents such as NOx and trace metals are correlated with SO2, their influence may be reflected in coal PM2.5. A detailed evaluation of unit-specific HyADS coal PM2.5 against GEOS-Chem adjoint sensitivities from Dedoussi et al.43 shows that population-weighted PM2.5 exposure from the two models by state and across the United States compare at levels consistent with other reduced complexity model evaluations44 (details in the Supplemental Information section “HyADS model evaluations and applications,” Tables S1–S3 and Figures S4–S6). In the comparison with GEOS-Chem, HyADS showed similar agreement with both PM2.5 from only SO2 emissions and PM2.5 from SO2 + NOx emissions, suggesting some potential that the HyADS coal PM2.5 metric captures at least some PM2.5 impacts from NOx as well as those attributable to SO2. The finding that coal PM2.5 agrees with directly comparable quantities from a more complex model provides evidence that the model was sufficient for the applications described below and highlights potential influence on the findings by uncertainties related to the simplified approach to atmospheric chemistry taken by HyADS. We mitigated these uncertainties by aggregating to regional and nationwide exposure averages and exploring long-term trends instead of year-specific results.
Population-Weighted Exposure (PWE)
We employed PWE to reduce multidimensional spatial exposure fields to a single dimension by emphasizing portions of the exposure field that impact areas with high population. For gridded coal PM2.5 source impacts, we calculated population-weighted exposure PWEu,y,d from each unit or group of units u in each year y for each demographic group d as: (2) PWEu,y,d=∑i=1I[Coal PM2.5]u,y,i×(Py,d,iPy,d,total),
where i=1, 2,…, I denotes grid cell locations, Py,d,i is the population of a given demographic in each grid cell, and Py,d,total is the total population in the domain of the demographic group.
Annual grid cell population was calculated by spatially apportioning U.S. Census county populations to the 36-km HyADS grid. We used annual intercensal population estimates from 2000 to 2020.45 We assigned 1 April 2000 population estimates for year 1999 population. We calculated PWu,y,d for the following racial/ethnic groups (census names in parentheses): White (White alone), Black (Black or African-American population alone), Native (American Indian and Alaska Native population alone), Asian (Asian population alone), Pacific (Native Hawaiian and Other Pacific Islander population alone), and Hispanic (population of Hispanic origin and any race). Although county data and the HyADS 36-km grid are relatively coarse measures of air pollution and demographic spatial distribution, they are justified by the regional nature of coal power plant pollution.20,43,46,47 Recent findings, for example, have shown that cross-state EGU source impacts accounted for around half of total fatalities attributable to EGU emissions,20,43 likely because local contributions to sulfate PM2.5 from SO2 emissions are less likely due to elevated stack heights and the delay introduced by atmospheric processing of SO2 to sulfate.48 HyADS uses a 36-km spatial grid; the average land area of counties in the contiguous United States is 1,555 km2,49 which is of similar order as the 1,296-km2 area of a 36-km grid cell. Daouda et al. previously used a county-level analysis and a portion of the data set described here to quantify racial disparities in preterm birth outcomes attributable to EGU SO2 emissions.50
Exposure Contributed or Avoided by Regulatory or Operational Activity
Population exposure to coal PM2.5 was reduced through various actions taken on individual coal EGUs across the study period, including reduced operations, emissions controls (“scrubbers”; control technologies identified by the following labels in the AMPD data set: Dry Lime FGD, Dry Sorbent Injection, Dual Alkali, Fluidized Bed Limestone Injection, Magnesium Oxide, Sodium Based, Wet Lime FGD, Wet Limestone, and Other), and retirements. Using PWE from HyADS and data from EPA AMPD, we calculated PWE contributed by operational facilities and PWE avoided through each of these three interventions.
We used dates of unit retirements and scrubber installations listed in the AMPD data set to designate each unit’s operational or emissions control status. Additionally, we employ each unit’s annual heat input—also available in the AMPD data set—to characterize units as operating at high capacity (annual heat input above each unit’s median annual heat input reported in operational years from 1999 to 2020) or low capacity (annual heat input below median heat input). This characterization of high vs. low operational capacity allows for the quantification of exposure avoided by reduced operations. Using this information, we characterize each unit into one of six categories: a) operating at high capacity without a scrubber, b) operating at low capacity without a scrubber, c) operating at high capacity with a scrubber, d) operating at low capacity with a scrubber, e) retired without previously installing a scrubber, and f) retired after operating with a scrubber. These six operational/control categories led to seven contributed and avoided exposure designations that could be calculated using modeled PWE across subsets of years for each unit (Table 1).
Table 1 Contributed and avoided PWE designations and calculation approaches.
Naming convention PWE calculation Applied to each unit in years that meet these criteria
Contributed: uncontrolled Average PWE in years that meet these criteria: Operating
Heat input above median
Before scrubber installation
Operating
Heat input above median
Before scrubber installation
Contributed: scrubber Average PWE in years that meet these criteria: Operating
Heat input above median
After scrubber installation
Operating
Heat input above median
After scrubber installation
Avoided: reduced operation Difference between average PWE in two sets of years: High-operation years
○ Operating
○ Heat input above median
○ Before scrubber installation
Low-operation years
○ Operating
○ Heat input below median
○ Before scrubber installation
Operating
Heat input below median
Before scrubber installation
Avoided: reduced operation after scrubber Difference between average PWE in two sets of years: High-operation years
○ Operating
○ Heat input above median
○ After scrubber installation
Low-operation years
○ Operating
○ Heat input below median
○ After scrubber installation
Operating
Heat input below median
After scrubber installation
Avoided: scrubber Difference between average PWE in two sets of years: No scrubber years
○ Operating
○ Heat input above median
○ Before scrubber installation
Scrubber years
○ Operating
○ Heat input above median
○ After scrubber installation
Operating
After scrubber installation
Avoided: retirement Average PWE in years that meet these criteria: Operating
Before scrubber installation
Retired
Avoided: retirement after scrubber Average PWE in years that meet these criteria: Operating
After scrubber installation
Retired
After scrubber installation
Note: PWE, population-weighted exposure.
We calculated each quantity listed in Table 1 for each unit in years that met the corresponding criteria and presented the sum of each exposure class across units. We did not include the years of scrubber installation or retirement in the PWE averaging to avoid transition years. Each unit’s potential PWE designation among these five categories remained constant across any given range of years for which its scrubber and operational status did not change. We presented the annual results as a percentage of total potential exposure in each year.
The approach was designed to explore trends across years, and the calculated values were somewhat sensitive to the criteria listed in Table 1. Therefore, the results were not precise enough to diagnose a given year’s exposure distribution across the seven categories, and we focused on overarching trends in the discussion. Sensitivity of the results to the selection of the heat input value cutoff used to define high/low operating capacity is presented in Figure S7.
PWE Inequities Accounting for Spatial Distributions of Sources and Populations
We explored power plant exposure inequities through two lenses: a) comparisons of demographic-specific PWE relative to the total population PWE and b) comparisons of demographic-specific PWE relative to an “expected” exposure disparity based on regional demographic distributions. The second comparison acknowledges that, before high voltage transmission lines made long-distance electricity transport possible in recent decades, power plants were cited nearby the population that would use the electricity and were operated mostly independently across regions.51 Even as regional electricity transport has become more viable, historical citing policies continue to influence exposure to power plants. For example, in 2010 50% of operating coal plants had been in service for 38 y,52 and power plant operations are still managed somewhat independently across regions.20 The differences between the two types of exposure inequality measures highlights the extent that regional demographic differences influence interpretation of exposure inequality from coal power plants. We adapted multiple literature-based exposure metrics to calculate source-specific PWE inequities for groups of EGUs, and we developed novel region-specific “relative expected PWE” to account for location-specific demographic makeups.
PWE Inequities
To quantify exposure inequities in region- and state-specific coal PM2.5, we presented relative and absolute comparisons of each demographic’s PWEu,y,d to the total population PWEu,y,d=all as described previously.23 Absolute PWE differences contributed by coal facilities were calculated by subtracting the PW exposure for the total population from the PW exposure from a given group: (3) PWEu,y,dabsolute=PWEu,y,d−PWEu,y,d=all,
where PWEu,y,d=all is the population-weighted exposure across the entire population. PWEu,y,dabsolute provides the difference in exposure contributed by a unit or group of units u in year y on population group d relative to the total population. The units are in concentration units (micrograms per cubic meter here); a value of zero signifies that the coal units in question do not inequitably expose the group relative to the population average. Relatedly, relative PW exposure was calculated as the ratio between the two terms in Equation 3: (4) PWEu,y,drelative=PWEu,y,dPWEu,y,d=all.
PWEu,y,drelative is unitless; values represent the fractional difference of a group’s exposure relative to the total population.
The absolute comparison (PWEu,y,dabsolute) quantifies inequity in concentration (micrograms per cubic meter) units and is useful for showing how health-relevant exposures differ in space and have evolved over time. The relative comparison (PWEu,y,drelative) is in fractional units, which are useful for comparing across years and regions with large absolute exposure differences (e.g., inequities may exist in the western United States, even though absolute exposure is low because of a paucity of coal power plants). A PWEu,y,drelative value of 1 signifies that group’s exposure is equal to that of the total population. We defined all disparity measures relative to the average member of the total population, which revealed different interpretation than what may be found, for example, by comparing to the worst- or best-off member of the population.23 Such a decision was justified by coal pollution’s regional nature.
Regional differences in average population demographics mean that a given facility located in an arbitrary location is expected to contribute to exposure disparities that simply reflect that area’s demographics. For example, the southeastern United States’ population is characterized by a higher percentage of Black or African-American people than any other region; therefore, a facility located anywhere in the Southeast would seem to inequitably impact the Black or African-American population relative to other regions. Assuming facilities must be located in a specific region (as is often the case to provide nearby power sources53) the most equitably sited facilities would impact demographic groups at a level equal to their expected exposure in that region. We calculated the PWE enhancement for a group of units in region (or state) r as the ratio between the relative PW exposure and the relative expected PW exposure: (5) PWEu,r,y,denhancement=PWEu,y,drelativePWer,y,drelative,expected.
PWer,y,drelative,expected is expected relative population-weighted exposure for facilities located in an average location in region (or state) r in year y on demographic group d. The lowercase “e” signifies that the expected exposure calculated is not in micrograms per cubic meter, as clarified below. (6) PWer,y,drelative,expected=PWer,y,dPWer,y,d=total.
To calculate regional control population-weighted exposure PWer,y,d, we started by assuming that a source could feasibly be located at any overland location, here implemented as the centroids of the 36-km HyADS grid cells. From each location, we calculated exposure as the inverse of distance from that location in all 36-km grid cells across the domain (distance for the containing grid cell is set at 18km), and we used the same population-weighting approach as above: (7) PWec,y,d=∑i=1I[1distc,i]u,y,i×(Py,d,iPy,d,total),
where c denotes grid cell centroids, and distc,i references the distance between centroid c and grid cell i. Expected PWE for region/state r in year y on demographic group d (PWer,y,d) is calculated as the sum of PWec,y,d for each centroid location in region r.
This approach controlled for population group spatial distributions in locations impacted by coal units and population demographic changes over time (PWer,y,drelative,expected is plotted in Figure S8). Changes in wind patterns and changes in emissions at coal facilities are not captured in this expected exposure, meaning the PWEu,r,y,denhancement measures the extent that these two variables increase or decrease exposure disparities. Step changes surrounding census years in PWer,y,drelative,expected reflect discontinuities in intercensal years; such step corrections signify some potential for bias surrounding 2010, but note that the steps are smaller in all cases than the overall trend across 1999–2020.
Results
Coal Power Plant SO2 Emissions, 1999–2020
Annual coal power plant SO2 emissions decreased from 11.8 million tons in 1999 to 0.8 million tons in 2020 (93%; Figure 1). Individual units’ emissions are mostly below 10,000 tons SO2 per year, but some units emitted more than an order of magnitude above that in some years. Monthly emissions have two peaks in most years: one each in the summer and winter. All units emitted <20,000 tons SO2 per year since 2018.
Figure 1. Line plot of monthly total coal electricity generating unit SO2 emissions from 1,237 units in the U.S. EPA Clean Air Markets database.10 Data for this plot are provided in Supplemental Excel Table S1.
Figure 1 is a line graph, plotting total sulfur dioxide emissions, 10 begin superscript 6 end superscript tons, ranging from 0.0 to 1.0 in increments of 0.5 (y-axis) across years, ranging from 2000 to 2020 in increments of 5 years (x-axis).
SO2 emissions control installations (noted generically as scrubbers here) and coal unit retirements drove the large sectorwide emissions reduction seen in Figure 1. The total number of operational units in the AMPD database peaked in 2004 at 1,256 (Figure S9). The number of units reporting SO2 emissions each year peaked at 1,090 in 2009 (units listed as operational do not always report emissions greater than zero). After 2004, both the number of units in operation with a scrubber and the number of retired units increased. The number of units operating with a scrubber increased from 316 (25% of operating units) in 2006 to 503 (42%) in 2010, and peaked in 2016 at 545 (55%), and declined thereafter as units with and without scrubbers retired. After 2010, new scrubber installations slowed, and emissions changes were driven by coal unit retirements—an average of 70 units retired per year after 2010 in the contiguous United States. In 2020, 438 (80%) of operational units had scrubbers installed.
National Coal PM2.5 Source Impacts and PWE, 1999–2020
Annual average overland coal PM2.5 over the contiguous United States decreased by more than an order of magnitude from 1.17 μg/m3 (interquartile range across spatial grid cells=0.10–1.75) in 1999 to 0.05μg/m3 (0.01–0.07) in 2020 (Figure 2). Average population-weighted coal PM2.5 decreased from 1.96 to 0.06μg/m3 across the same period (Figure 3). Higher population-weighted coal PM2.5 exposure relative to average exposure is attributable to the power plants’ proximity to large population centers in the eastern United States.
Figure 2. Gridded coal PM2.5 from all coal power plants in the United States between 1999 and 2020. The top panel provides boxplots displaying annual median with first and third quartile and outlier coal PM2.5 from coal power plants in the U.S. EPA Clean Air Markets database.10 Select years’ spatial distribution maps are provided as examples in the bottom panel. United States maps are from the USAboundaries R package.42 Data for this plot are provided in Supplemental Excel Tables S2 and S3.
Figure 2 is a set of one box and whiskers plot and four maps. The box and whiskers plot, plotting coal particulate matter begin subscript 2.5 end subscript microgram per meter begin superscript negative 3 end superscript, ranging from 0 to 6 in increments of 2 (y-axis) across years, ranging from 1999 to 2020 (x-axis). A scale depicts coal particulate matter begin subscript 2.5 end subscript microgram per meter begin superscript negative 3 end superscript range from 0.0 to greater than or equal to 2.0 in increments of 1.0. From 1999 to 2020, the four maps of the United States of America depict gridded coal particulate matter begin superscript 2.5 end superscript from coal power plants in various regions.
Figure 3. Line plot of annual average population-weighted coal PM2.5 by demographic group averaged across all locations in the contiguous United States. Points denote population-weighted exposure for the total population. U.S. Census racial/ethnic designations are denoted by colors and line type and grouped as follows: White (White population alone), Black (Black or African-American population alone), Native (American Indian and Alaska Native population alone), Asian (Asian population alone), Pacific (Native Hawaiian and Other Pacific Islander population alone and any race), and Hispanic (population of Hispanic origin). Data for this plot are provided in Supplemental Excel Table S4.
Figure 3 is a line graph, plotting population-weighted coal particulate matter begin subscript 2.5 end subscript microgram per meter begin superscript negative 3 end superscript, ranging from 0.0 to 2.0 in increments of 0.5 y-axis) across year, ranging from 2000 to 2020 in increments of 5 years (x-axis) for population range, including Asian, Black, Hispanic, Native, Pacific, and White.
In the early 2000s, the eastern United States saw elevated regional exposure, and the western United States saw hot spots near coal facilities. The highest exposures in earlier years are seen along the Ohio River Valley, which houses the highest density of large coal facilities in the country. In recent years, elevated concentrations across the eastern United States have been reduced, but hot spots remain, especially in a band stretching from eastern Texas to the northeast and along the Ohio River Valley.
Modeled coal PM2.5 is correlated (R2>0.5) with measured sulfate PM2.5 at rural IMPROVE monitors in the Northeast, North Central, and South in most years before 2014 (correlation falls off in the South after these years; Figure S4 and Figure S5). The comparison is imperfect, because ambient sulfate can originate from many sources, but the U.S. EPA estimates that electric utilities contributed ∼70% of total U.S. SO2 emissions from 1999 to 2014, thereafter falling to under 50% in 2020,54 suggesting that much of the ambient sulfate in the United States should be related to electricity generation. We restrict the evaluation to rural IMPROVE sites55 to limit interference from urban sources. Coal PM2.5 follows a similar annual regional trend as observed sulfate PM2.5 at monitoring sites, although the average predicted coal PM2.5 is about 1μg/ m3 lower than observed sulfate PM2.5 concentrations at monitor locations (Figure S4). Coal PM2.5 changed at a similar rate in the Northeast, North Central, and South until around 2010, after which coal PM2.5 decreased faster than observed sulfate. Our data suggest little relationship between coal PM2.5 and observed sulfate in the West, where coal PM2.5 exposure is low. Although this evaluation suggests a potential negative bias of coal PM2.5, some negative bias is expected due to the focus on coal emissions specifically, and comparisons with more complex models do not show a similarly systematic negative bias (Figure S6).
National Exposure Contributed or Avoided by Regulatory or Operational Activity
Between 1999 and 2007, at least 85% of the contributed exposure—and 70% of the total (contributed + avoided) exposure—came from units without SO2 scrubbers (such units represented between 60% and 70% of annual input heat capacity in these years). Units with installed scrubbers contributed about 15% of total exposure between 1999 and 2020 (Figure 4). Between 2006 and 2013, both scrubber installations and reduced operations contributed to a large reduction in PWE. Avoided exposure from scrubbers increased from 5% of the total (avoided + emitted) PWE in 2006 to 50% in 2011, where it held steady until 2020. Avoided exposure from reduced operations in units without scrubbers was <10% of total PWE before 2008 and between 10% and 20% from 2009 to 2014, coinciding with decreased operations simultaneous with the economic recession. After 2010, reductions in PWE were driven further by retirements of facilities with and without scrubbers. Although the percentage magnitudes are somewhat sensitive to the high/low operational cutoff (larger avoided exposure is found for higher cutoffs; Figure S7), the overall trends are consistent across applied cutoffs.
Figure 4. Stacked area chart of PWE contributed and avoided relative to each unit’s (n=1,237) baseline PWE (methodological approach details are shown in in Table 1). Baseline is taken as the average exposure contributed before an intervention (either scrubber installation or retirement) in years the unit operated at least at its median capacity. “Avoided: retirement after scrubber” considers only PWE avoided by shutting down; exposure avoided by scrubber installment continues to contribute to “Avoided: scrubber” even after a unit is retired. The top-to-bottom order of categories in the legend is consistent with the order of the plotted data. Data for this plot are provided in Supplemental Excel Table S5. Note: PWE, population-weighted exposure.
Figure 4 is an area graph, plotting percentage baseline P W E contributed or avoided, ranging from 0 to 100 percent in increments of 25 (y-axis) across years, ranging from 2000 to 2020 in increments of 5 years (x-axis) for Avoided: retirement after scrubber, Avoided: retirement, Avoided: scrubber, Avoided: reduced operation after scrubber, Avoided: reduced operation before scrubber, Contributed: scrubber, and Contributed: uncontrolled.
Together, the facility retirement/control equipment counts and exposure related to both interventions tell a consistent story: Nationwide, reductions in exposure were primarily driven by scrubber installations from 2006 to 2010, reduced operations from 2008 to 2012, and facility retirements after 2010. Since 2015, avoided exposure has made up over 95% of total (combined + avoided) exposure. Of the small remaining contributed nationwide exposure since 2015, nonscrubbed facilities contributed between 25% and 50%, suggesting similar importance of scrubbed and nonscrubbed facilities for the relatively small remaining PWE.
PWE Inequities Accounting for Spatial Distributions of Sources and Populations
Nationwide population-weighted coal PM2.5 exposure has decreased for all demographic groups since 1999 (Figure 3). Overall, White population exposure tracked the population average because White population represents the largest population group across most of the country (Figure S10). Population average exposure changed little before 2005 and decreased approximately linearly thereafter until 2012. Annual exposure decreased at a slower pace after 2012. Exposure in the Black population was higher than the population average in 1999 and similarly remained relatively constant until 2005. Starting in 2005, Black population exposure decreased fastest among all population groups and by 2016 was at a level nearly identical to the population average. Other demographic groups were exposed at much lower levels than the average population, but these national results obscure regional differences.
The Black population is inequitably exposed in comparison with the population average by facilities located in the Northeast, South, and North Central regions (Figure 5; regions defined in Figure S11). In the South, the Black population was exposed to >0.2μg/m3 more coal PM2.5 most years between 1999 and 2010, and the absolute difference fell thereafter to close to zero by 2017. The relative exposure for the Black population in the South, however, is above 1.3 (30% above the population average) and declines only slightly across the period even as the absolute difference approaches zero. Other demographic groups are exposed at levels similar to (as in the White population) and less than the population average in the Northeast, South, and North Central.
Figure 5. Line plots of absolute and relative annual average population-weighted exposure differences (PWu,y,dabsolute and PWu,y,drelative) attributable to coal facilities located in each of four regions on demographic groups in all locations. U.S. Census demographic variables are denoted by colors and line type and grouped as follows: White (White population alone), Black (Black or African-American population alone), Native (American Indian and Alaska Native population alone), Asian (Asian population alone), Pacific (Native Hawaiian and Other Pacific Islander population alone), and Hispanic (population of Hispanic origin and any race). Data for this plot are provided in Supplemental Excel Table S6.
Figure 5 is a set of eight line graphs. On the left, the first set of two line graphs are titled Northeast, plotting Absolute P W E difference, microgram per meter begin superscript negative 3 end superscript, ranging from negative 0.50 to 0.25 in increments of 0.25 and Relative P W E difference, ranging from 0.5 to 2.0 in increments of 0.5 (y-axis) across years, ranging from 2000 to 2020 in increments of 5 years (x-axis) for Asian, Black, Hispanic, Native, Pacific, and White. The second set of two line graphs are titled South, Absolute P W E difference, microgram per meter begin superscript negative 3 end superscript, ranging from negative 0.50 to 0.25 in increments of 0.25 and Relative P W E difference, ranging from 0.5 to 2.0 in increments of 0.5 (y-axis) across years, ranging from 2000 to 2020 in increments of 5 years (x-axis) for Asian, Black, Hispanic, Native, Pacific, and White. The third set of two line graphs are titled North Central, Absolute P W E difference, microgram per meter begin superscript negative 3 end superscript, ranging from negative 0.50 to 0.25 in increments of 0.25 and Relative P W E difference, ranging from 0.5 to 2.0 in increments of 0.5 (y-axis) across years, ranging from 2000 to 2020 in increments of 5 years (x-axis) for Asian, Black, Hispanic, Native, Pacific, and White. The fourth set of two line graphs are titled West, plotting Absolute P W E difference, microgram per meter begin superscript negative 3 end superscript, ranging from negative 0.50 to 0.25 in increments of 0.25 and Relative P W E difference, ranging from 0.5 to 2.0 in increments of 0.5 (y-axis) across years, ranging from 2000 to 2020 in increments of 5 years (x-axis) for Asian, Black, Hispanic, Nat ive, Pacific, and White.
Multiple populations—notably the Native American, Pacific, and Hispanic populations—were exposed to emissions from facilities located in the West at levels higher than the population average. The relative exposure difference is pronounced in the Native American population, with values over 1.6 (160% of the population average; recall, however, that absolute differences in the West were very low after 2010). Exposure inequities are more pronounced for these populations in the relative exposure difference than the absolute difference because of the lower overall exposure in the western United States (there were many fewer coal plants in the West than in other regions during the study period).
Accounting for regional population demographic distributions and changes over time, many of the observed exposure inequities shown in Figure 5 in the Black population in the Northeast and South are reduced to below the population average (Figure 6). Although the Black population experienced slightly enhanced exposure before 2015, thereafter the population has been exposed at levels less than the population average. An opposite pattern is observed in the Black population exposure to emissions from facilities in the North Central region; accounting for the regional population distribution enhances the Black population’s average exposure inequity across the period from 7% over the average to 13% over the average (absolute differences in the North Central were very low after 2010).
Figure 6. Line plots of PWE enhancement from facilities in each region (top) and state (bottom) that account for regional demographics and change over time (PWEr,y,denhancement; denoted as Relative PWE enhancement and Rel. PWE enhmnt in the figure). U.S. Census demographic variables are denoted by colors and line type in the top plots and grouped as follows: White (White population alone), Black (Black or African-American population alone), Native (American Indian and Alaska Native population alone), Asian (Asian population alone), Pacific (Native Hawaiian and Other Pacific Islander population alone), and Hispanic (population of Hispanic origin and any race). Only states with coal power plants are shown in the lower plots (California and Idaho are omitted); blue shading and hashing denote PWEr,y,denhancement below one, and red shading and dots denote PWEr,y,denhancement above one. United States maps are from the USAboundaries R package.42 Data for this plot are provided in Supplemental Excel Tables S7 and S8. Note: PWE, population-weighted exposure.
Figure 6 is a set of four line graphs and twelve maps. The four line graphs are titled Northeast, South, North Central, and West, plotting Relative P W E enhancement, ranging from 0.6 to 1.4 in increments of 0.2 (y-axis) across year, ranging from 2000 to 2020 in increments of 5 years (x-axis) for Asian, Native, Black, Pacific, Hispanic, and White, respectively. Six maps in the year 1999 and six maps in the year 2020 of the United States of America, titled Asian, Black, Hispanic, Native, Pacific, and White, depict the regional demographics and change over time into two categories, namely, yes and no, respectively. A scale that depicts the relative P W E enhancement ranges from 0.5 to 1.5 in increments of 0.5.
In the West, exposure inequity in the Native American population persists (but is slightly mitigated) by controlling for regional population distribution. Other populations that see higher exposures relative to the population average without controlling for the regional population distribution (Pacific and Hispanic populations) are exposed to smaller levels of coal PM2.5 than would be expected for a facility in an arbitrary location throughout the region.
Regional averages of exposure enhancements obscure underlying heterogeneity: Facilities in some states contributed more exposure inequities than others (Figure 6, bottom). Facilities in small groups of states stand out as contributing inequitably to exposure in each of the demographic groups: Washington State (Asian population); Nevada and Louisiana (Hispanic population); Oregon, Nevada, Arizona, and New Mexico (Native population); and Washington State, Nevada, and Utah (Pacific population). Facilities in a small number of states stand out (30 in 1999, 25 in 2020) contributed enhanced exposure to the Black population.
Uncertainty and Limitations
The HyADS model is a reduced complexity model in that it uses the assumptions detailed above to approximate atmospheric transport and deposition. Further, the chemical conversion of SO2 to PM2.5 is approximated in each year using a statistical model trained on CMAQ-DDM output from a single year (2005). This approach, necessitated by limited availability of Hybrid CMAQ-DDM output, makes it difficult to quantify the uncertainty associated with relationships that changed over the time scale of this study. For example, although model (1) allows for varied relationships between meteorological influence on coal PM2.5, it does not account for long-term changes in the nitrate and ammonium emissions, which exert influence over PM2.5 mass formed by sulfur emissions.56,57 Our evaluations have shown, however, that although uncertainties in individual-unit coal PM2.5 are nonnegligible, high correlations in source impacts from individual units with complementary estimates from more sophisticated models suggest that the model is able to provide useful information for the applications above, particularly in the regional and national relative exposure long-term trends. Uncertainty in the exposure modeling with HyADS is not propagated into reported air quality or disparities metrics presented here, which motivated our approach to focus on temporal trends and relative exposure disparities that would be robust to modeling uncertainties that are not systematically related to time or the spatial distribution of different populations. There is the potential for future studies to explore reduced complexity approaches that more fully capture the atmospheric dynamics that are not accounted for in our approach.
The study is limited by its geographic resolution at the county level, which follows from the HyADS grid resolution (36km). As in any model based on a grid, there is potential for within-grid variation to introduce exposure misclassification and the possibility that this coarse resolution averages out intragrid exposure and demographic variability as addressed by Spiller et al.,58 but such potential is mitigated by annual averages used here and the regional nature of SO2-related coal PM2.5 impacts. Using PM2.5-emitting facility locations and a 2.5-mi (4-km) buffer, for example, Mikati et al. found higher relative disparities in year 2011 emissions burden than the values we report, albeit for a different group of facilities.59
Calculations of PWE enhancement begin with the assumption that coal power plants could be located at any overland location in a region, which is not actually the case. Locations are restricted by water availability, access to fuel sources, and feasibility to deliver electricity demand.53 Although the method uses inverse-distance weighting and therefore does not control for meteorological variability, such a control could be added in future work. One approach to undertaking this would be to run the HyADS model from each grid centroid location and using the output to calculate PWec,y,d. Given the high agreement between inverse-distance weighted exposure and HyADS output found in Henneman et al., however, we concluded the inverse distance approach provided sufficient exposure control.37
The analysis focuses on impacts from SO2 emissions reported at coal power plants. Coal combustion leads to emissions of other pollution species, including NOx, primary PM2.5, and mercury.35 Although emissions of some of these other pollution species may be captured in coal PM2.5 owing to their correlation with SO2 emissions and the reduced-complexity nature of the model, we have not quantified the extent that they have contributed to exposure inequities insofar as they are not coincidentally captured by HyADS and its conversion to coal PM2.5. Additionally, it is possible that locations of now-retired coal power plants were repurposed to house natural gas facilities; inequities contributed by such situations are not captured in this analysis.
Implications
Regulatory policies, technological improvements, and economics had wide influence on decreasing exposure to PM2.5 from coal power plant SO2 emissions since 1999. Three unit-level interventions taken at many individual coal power plants throughout the country resulted in decreasing exposure and exposure inequities— SO2 scrubber installations, unit retirements, and decreased operations. In recent years, more nationwide PWE is attributable to units with scrubbers installed than to uncontrolled emissions, and exposure inequities persist throughout the country. This finding suggests that reducing emissions from the 505 units still in operation in 2020, which contribute similar magnitudes of exposure from scrubbed and nonscrubbed units, offers potential benefits for overall exposure and exposure inequities reduction.
Although air pollution from coal power plants is generally regional in nature, this work reveals exposure inequities attributable to regional and state facility siting (Figure 6). These inequities persist even as emissions have been reduced from the entire source sector, for example in Black populations impacted by facilities in the North Central region and in individual states across the center of the country and in Native American populations impacted by facilities in the West. Exposure in both groups remained inequitably high in 2020, although average exposure after 2015 has been very low in the West.
We find that accounting for population demographics near sources in exposure-inequity calculations reduces estimated inequities relative to calculations that do not account for underlying population demographics in many cases. Results from the analyses may be of interest to policymakers at varying levels of government; for example, the U.S. EPA may be interested in the national inequities, whereas local decision makers may use results that account for regional population makeup.
In the United States, coal is being rapidly phased out as an energy source. This analysis, therefore, serves primarily as a detailed post hoc assessment of coal’s impacts across 1999–2020 and a period of dramatic improvements due to air quality regulations and the economics of coal price relative to other electricity sources. We find that coal electricity-generation SO2 emissions contributions to overall PM2.5 are diminishing to the point of having minimal effect on most locations’ ability to meet national ambient air quality standards.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by research funding from the National Institutes of Health (NIH) NIHR01ES026217 and the U.S. EPA 835872. S.C.A. acknowledges support from NASA grant no. 80NSSC21K0511. Its contents are solely the responsibility of the grantee and do not necessarily represent the official views of the U.S. EPA. Further, the U.S. EPA does not endorse the purchase of any commercial products or services mentioned in the publication.
Code and data needed to reproduce the figures in this manuscript are available on Github (https://github.com/lhenneman/coal_exposure_longterm). Coal PM2.5 source impacts and facility information are available for download from Open Science Framework (https://osf.io/8gdau), and code for importing gridded coal PM2.5 into R is provided on Github (https://github.com/lhenneman/coal_unit_PM25).
The disperseR package enables parallelization of many HYSPLIT runs and provides tools for manipulating and summarizing the output data relevant for HyADS. All results presented here were run using the development version of the disperseR package maintained on Github (https://github.com/lhenneman/disperseR).
==== Refs
References
1. Squizzato S, Masiol M, Rich DQ, Hopke PK. 2018. PM2.5 and gaseous pollutants in New York state during 2005–2016: spatial variability, temporal trends, and economic influences. Atmos Environ 183 :209–224, 10.1016/j.atmosenv.2018.03.045.
2. Ivey CE, Holmes HA, Hu YT, Mulholland JA, Russell AG. 2015. Development of PM2.5 source impact spatial fields using a hybrid source apportionment air quality model. Geosci Model Dev 8 (7 ):2153–2165, 10.5194/gmd-8-2153-2015.
3. Blanchard CL, Tanenbaum S, Hidy GM. 2012. Source contributions to atmospheric gases and particulate matter in the southeastern United States. Environ Sci Technol 46 (10 ):5479–5488, PMID: , 10.1021/es203568t.22475316
4. National Research Council. 2004. Air Quality Management in the United States. Washington, DC: National Academies Press. 10.17226/10728.
5. U.S. EPA (U.S. Environmental Protection Agency). 2005. Regulatory Impact Analysis for the Final Clean Air Interstate Rule. EPA-452/R-05-002. Washington, DC: Office of Air and Radiation.
6. United States Environmental Protection Agency Office of Air and Radiation Office of Air Quality Planning and Standards Office of Atmospheric Programs (6204J) 1200 Pennsylvania Ave, NW Washington, DC 20460. https://www.epa.gov/sites/default/files/2015-08/documents/ozonenbp-2004.pdf.
7. U.S. EPA. 2015. Cross-State Air Pollution Rule Update for the 2008 Ozone NAAQS, Vol. 80, 74504–74650. https://www.epa.gov/sites/default/files/2017-09/documents/2014_full_report.pdf [accessed 27 February 2023].
8. Greenbaum DS, Shaikh R. 2018. Air Quality and Human Health: The Role of Health Science in Setting National Ambient Air Quality Standards. https://airandwmapa.sharepoint.com/sites/AWMA_Website/Shared%20Documents/Forms/AllItems.aspx?id=%2Fsites%2FAWMA%5FWebsite%2FShared%20Documents%2Fem%2Ddo%20not%20delete%2F2018%2F12%2Fgreenbaum%2Epdf&parent=%2Fsites%2FAWMA%5FWebsite%2FShared%20Documents%2Fem%2Ddo%20not%20delete%2F2018%2F12&p=true&ga=1 [accessed 10 December 2018].
9. U.S. Energy Information Administration. Form EIA-860 Detailed Data with Previous Form Data (EIA-860A/860B). https://www.eia.gov/electricity/data/eia860/ [accessed 17 June 2021].
10. U.S. EPA. Clean Air Markets Program Data. https://campd.epa.gov/ [accessed 22 July 2021].
11. de Gouw JA, Parrish DD, Frost GJ, Trainer M. 2014. Reduced emissions of CO2, NOx and SO2 from U.S. power plants due to the switch from coal to natural gas with combined cycle technology. Earths Future 2 :75–82, 10.1002/2014EF000196.
12. Harrington W, Morgenstern R, Shih J-S, Bell ML. 2012. Did the clean air act amendments of 1990 really improve air quality? Air Qual Atmosphere Health 5 (4 ):353–367, 10.1007/s11869-012-0176-5.
13. Lin S, Jones R, Pantea C, Özkaynak H, Rao STT, Hwang S-A, et al. 2013. Impact of NOx emissions reduction policy on hospitalizations for respiratory disease in New York state. J Expo Sci Environ Epidemiol 23 (1 ):73–80, PMID: , 10.1038/jes.2012.69.22829050
14. Godowitch JM, Gilliland AB, Draxler RR, Rao ST. 2008. Modeling assessment of point source NOx emission reductions on ozone air quality in the Eastern United States. Atmos Environ 42 (1 ):87–100, 10.1016/j.atmosenv.2007.09.032.
15. Russell MC, Belle JH, Liu Y. 2015. The impact of three recent coal-fired power plant closings on Pittsburgh air quality: a natural experiment. J Air Waste Manag Assoc 67 (1 ):3–16, 10.1080/10962247.2016.1170738.
16. Casey JA, Su JG, Henneman LRF, Zigler C, Neophytou AM, Catalano R, et al. 2020. Improved asthma outcomes observed in the vicinity of coal power plant retirement, retrofit and conversion to natural gas. Nat Energy 5 :398–408, 10.1038/s41560-020-0600-2.32483491
17. Abrams JY, Klein M, Henneman LR, Sarnat SE, Chang HH, Strickland MJ, et al. 2019. Impact of air pollution control policies on cardiorespiratory emergency department visits, Atlanta, GA, 1999–2013. Environ Int 126 :627–634, PMID: , 10.1016/j.envint.2019.01.052.30856450
18. Henneman LR, Choirat C, Zigler CM. 2019. Accountability assessment of health improvements in the United States associated with reduced coal emissions between 2005 and 2012. Epidemiology 30 (4 ):477–485, PMID: , 10.1097/EDE.0000000000001024.31162280
19. Levy JI, Wilson AM, Zwack LM. 2007. Quantifying the efficiency and equity implications of power plant air pollution control strategies in the United States. Environ Health Perspect 115 (5 ):743–750, PMID: , 10.1289/ehp.9712.17520062
20. Thind MPS, Tessum CW, Azevedo IL, Marshall JD. 2019. Fine particulate air pollution from electricity generation in the US: health impacts by race, income, and geography. Environ Sci Technol 53 (23 ):14010–14019, PMID: , 10.1021/acs.est.9b02527.31746196
21. The President of the United States. 1994. Executive Order 12898. Federal Actions to Address Environmental Justice in Minority Populations and Low-Income Populations. Fed Reg 59(32):7629–7633 (11 February 1994). https://www.govinfo.gov/content/pkg/FR-1994-02-16/html/94-3685.htm [accessed 4 January 2022].
22. The President of the United States. 2021. Executive Order 14008. Tackling the Climate Crisis at Home and Abroad. Fed Reg 26:7619–7633 (27 January 2021). https://www.federalregister.gov/documents/2021/02/01/2021-02177/tackling-the-climate-crisis-at-home-and-abroad [accessed 4 January 2022].
23. Harper S, Ruder E, Roman HA, Geggel A, Nweke O, Payne-Sturges D, et al. 2013. Using inequality measures to incorporate environmental justice into regulatory analyses. Int J Environ Res Public Health 10 (9 ):4039–4059, PMID: , 10.3390/ijerph10094039.23999551
24. Buonocore JJ, Dong X, Spengler JD, Fu JS, Levy JI. 2014. Using the Community Multiscale Air Quality (CMAQ) model to estimate public health impacts of PM2.5 from individual power plants. Environ Int 68 :200–208, PMID: , 10.1016/j.envint.2014.03.031.24769126
25. American Lung Association and American Public Health Association, Petitioners v. Environmental Protection Agency and Andrew Wheeler, Administrator, Respondents. 2021. U. S. Court Appeals Dist Columbia Circuit No. 19-1140 (D.C. Cir. 19 January 2021). https://statepowerproject.files.wordpress.com/2021/01/american-lung-assn-v.-epa-dc-cir.-no.-19-1140-per-curiam-decision.pdf [accessed 25 March 2022].
26. U.S. Environmental Protection Agency Environmental Justice Screening Report Carbon Pollution Emission Guidelines for Existing Stationary Sources: Electric Utility Generating Units Docket: EPA–HQ–OAR–2013–0602 Office of Air and Radiation, Office of Air Quality Planning and Standards Research Triangle Park, NC 27711. https://19january2017snapshot.epa.gov/sites/production/files/2016-04/documents/ejscreencpp.pdf.
27. U.S. EPA. Power Plants and Neighboring Communities. 2021. https://www.epa.gov/airmarkets/power-plants-and-neighboring-communities [accessed 29 March 2022].
28. Tessum CW, Hill JD, Marshall JD. 2017. InMAP: a model for air pollution interventions. PLoS One 12 (4 ):1–26, PMID: , 10.1371/journal.pone.0176131.28423049
29. Tessum CW, Apte JS, Goodkind AL, Muller NZ, Mullins KA, Paolella DA, et al. 2019. Inequity in consumption of goods and services adds to racial–ethnic disparities in air pollution exposure. Proc Natl Acad Sci 116 (13 ):201818859, PMID: , 10.1073/pnas.1818859116.30858319
30. Goodkind AL, Tessum CW, Coggins JS, Hill JD, Marshall JD. Fine-scale damage estimates of particulate matter air pollution reveal opportunities for location-specific mitigation of emissions. Proc Natl Acad Sci USA 116 (18 ):8775–8780. 10.1073/pnas.1816102116.
31. Muller NZ, Mendelsohn R. 2007. Measuring the damages of air pollution in the United States. J Environ Econ Manag 54 (1 ):1–14, 10.1016/j.jeem.2006.12.002.
32. Heo J, Adams PJ, Gao HO. 2016. Reduced-form modeling of public health impacts of inorganic PM2.5 and precursor emissions. Atmos Environ 137 :80–89, 10.1016/j.atmosenv.2016.04.026.
33. Heo J, Adams PJ, Gao HO. 2017. Public health costs accounting of inorganic PM2.5 pollution in metropolitan areas of the United States using a risk-based source-receptor model. Environ Int 106 :119–126, PMID: , 10.1016/j.envint.2017.06.006.28633084
34. Baker KR, Amend M, Penn S, Bankert J, Simon H, Chan E, et al. 2020. A database for evaluating the InMAP, APEEP, and EASIUR reduced complexity air-quality modeling tools. Data Brief 28 :104886, PMID: , 10.1016/j.dib.2019.104886.31872009
35. Dedoussi IC, Barrett SRH. 2014. Air pollution and early deaths in the United States. Part II: attribution of PM2.5 exposure to emissions species, time, location and sector. Atmos Environ 99 :610–617, 10.1016/j.atmosenv.2014.10.033.
36. Henneman LR, Choirat C, Ivey CE, Cummiskey K, Zigler CM. 2019. Characterizing population exposure to coal emissions sources in the United States using the HyADS model. Atmos Environ 203 (203 ):271–280, 10.1016/j.atmosenv.2019.01.043.
37. Henneman LR, Dedoussi IC, Casey JA, Choirat C, Barrett SRH, Zigler CM. 2020. Comparisons of simple and complex methods for quantifying exposure to individual point source air pollution emissions. J Expo Sci Environ Epidemiol 31 (4 ):654–663, PMID: , 10.1038/s41370-020-0219-1.32203059
38. Henneman LR, Mickley LJ, Zigler CM. 2019. Air pollution accountability of energy transitions: the relative importance of point source emissions and wind fields in exposure changes. Environ Res Lett 14 (11 ):115003, PMID: , 10.1088/1748-9326/ab4861.33408754
39. Stein AF, Draxler RR, Rolph GD, Stunder BJB, Cohen MD, Ngan F. 2015. NOAA’s HYSPLIT atmospheric transport and dispersion modeling system. Bull Am Meteorol Soc 96 (12 ):2059–2077, 10.1175/BAMS-D-14-00110.1.
40. Draxler RR, Hess GD. 1998. An overview of the HYSPLIT_4 modelling system for trajectories, dispersion, and deposition. Aust Meteorol Mag 47 :295–308.
41. Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, et al. 1996. The NCEP/NCAR 40-year reanalysis project. Bull Am Meteorol Soc 77 (3 ):437–471, 10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.
42. Ivey C, Holmes H, Shi G, Balachandran S, Hu Y, Russell AG. 2017. Development of PM2.5 source profiles using a hybrid chemical transport-receptor modeling approach. Environ Sci Technol 51 (23 ):13788–13796, PMID: , 10.1021/acs.est.7b03781.29110467
43. Dedoussi IC, Eastham SD, Monier E, Barrett SRH. 2020. Premature mortality related to United States cross-state air pollution. Nature 578 (7794 ):261–265, PMID: , 10.1038/s41586-020-1983-8.32051602
44. Gilmore EA, Heo J, Muller NZ, Tessum CW, Hill JD, Marshall JD, et al. 2019. An inter-comparison of the social costs of air quality from reduced-complexity models. Environ Res Lett 14 (7 ):074016, 10.1088/1748-9326/ab1ab5.
45. U.S. Census Bereau. County Population by Characteristics: 2010–2019. https://www.census.gov/data/datasets/time-series/demo/popest/2010s-counties-detail.html [accessed 10 June 2021].
46. Emami F, Masiol M, Hopke PK. 2018. Air pollution at Rochester, NY: long-term trends and multivariate analysis of upwind SO2 source impacts. Sci Total Environ 612 :1506–1515, PMID: , 10.1016/j.scitotenv.2017.09.026.28915545
47. Hand JL, Schichtel BA, Malm WC, Pitchford ML. 2012. Particulate sulfate ion concentration and SO2 emission trends in the United States from the early 1990s through 2010. Atmospheric Chem Phys 12 (21 ):10353–10365, 10.5194/acp-12-10353-2012.
48. Levy JI, Spengler JD, Hlinka D, Sullivan D, Moon D. 2002. Using CALPUFF to evaluate the impacts of power plant emissions in Illinois: model sensitivity and implications. Atmos Environ 36 (6 ):1063–1075, 10.1016/S1352-2310(01)00493-9.
49. Mullen LA, Bratt J. 2018. USAboundaries: historical and contemporary boundaries of the United States of America. J Open Source Softw 3 (23 ):314, 10.21105/joss.00314.
50. Daouda M, Henneman LR, Kioumourtzoglou M-A, Gemmill A, Zigler C, Casey JA. 2021. Association between county-level coal-fired power plant pollution and racial disparities in preterm births from 2000 to 2018. Environ Res Lett 16 (3 ):034055, PMID: , 10.1088/1748-9326/abe4f7.34531925
51. Vann A. 2010. The Federal Government’s Role in Electric Transmission Facility Siting. Washington, DC: Library of Congress, Congressional Research Service, 17.
52. U.S. EPA. 2015. Regulatory Impact Analysis for the Clean Power Plan Final Rule. Research Triangle Park, NC: U.S. Environmental Protection Agency Office of Air and Radiation, Office of Air Quality Planning and Standards, 343.
53. Wang S, Fisher EB, Feng L, Zhong X, Ellis JH, Hobbs BF. 2021. Linking energy sector and air quality models through downscaling: long-run siting of electricity generators to account for spatial variability and technological innovation. Sci Total Environ 772 :145504, PMID: , 10.1016/j.scitotenv.2021.145504.33581514
54. U.S. EPA. Air Pollutant Emissions Trends Data. https://www.epa.gov/air-emissions-inventories/air-pollutant-emissions-trends-data [accessed 23 August 2022].
55. Solomon PA, Crumpler D, Flanagan JB, Jayanty RKM, Rickman EE, McDade CE. 2014. U.S. National PM2.5 chemical speciation monitoring networks—CSN and IMPROVE: description of networks. J Air Waste Manag Assoc 64 (12 ):1410–1438, PMID: , 10.1080/10962247.2014.956904.25562937
56. Vasilakos P, Russell A, Weber R, Nenes A. 2018. Understanding nitrate formation in a world with less sulfate. Atmos Chem Phys Discuss 2018 :1–27, 10.5194/acp-2018-406.
57. Holt J, Selin NE, Solomon S. 2015. Changes in inorganic fine particulate matter sensitivities to precursors due to large-scale US emissions reductions. Environ Sci Technol 49 (8 ):4834–4841, PMID: , 10.1021/acs.est.5b00008.25816113
58. Spiller E, Proville J, Roy A, Muller NZ. 2021. Mortality risk from PM2.5: a comparison of modeling approaches to identify disparities across racial/ethnic groups in policy outcomes. Environ Health Perspect 129 (12 ):127004, PMID: , 10.1289/EHP9001.34878311
59. Mikati I, Benson AF, Luben TJ, Sacks JD, Richmond-Bryant J. 2018. Disparities in distribution of particulate matter emission sources by race and poverty status. Am J Public Health 108 (4 ):480–485, PMID: , 10.2105/AJPH.2017.304297.29470121
|
PMC010xxxxxx/PMC10010382.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36913236
EHP12167
10.1289/EHP12167
Invited Perspective
Invited Perspective: Air Pollutants, Genetics, and the Mucosal Paradigm for Rheumatoid Arthritis Risk
McDermott Gregory C. 1 2
https://orcid.org/0000-0002-5556-4618
Sparks Jeffrey A. 1 2
1 Division of Rheumatology, Inflammation, and Immunity, Brigham and Women’s Hospital, Boston, Massachusetts, USA
2 Harvard Medical School, Boston, Massachusetts, USA
Address correspondence to Jeffrey A. Sparks, Division of Rheumatology, Inflammation, and Immunity, Brigham and Women’s Hospital, 60 Fenwood Rd., #6016U, Boston, MA 02115 USA. Telephone: (617) 525-1040. Email: [email protected]
13 3 2023
3 2023
131 3 03130316 9 2022
08 11 2022
06 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
G.C.M. is supported by the Value and Evidence in Rheumatology using bioinformaTics and advanced analYtics (VERITY) Pilot and Feasibility Award and grant T32 AR007530, both through the National Institute of Arthritis and Musculoskeletal and Skin Diseases. J.A.S. is supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (grants R01 AR077607, P30 AR070253, and P30 AR072577), the R. Bruce and Joan M. Mickey Research Scholar Fund, and the Llura Gund Award for Rheumatoid Arthritis Research and Care. J.A.S. has also received research support from Bristol Myers Squibb and performed consultancy for AbbVie, Amgen, Boehringer Ingelheim, Bristol Myers Squibb, Gilead, Inova Diagnostics, Janssen, Optum, and Pfizer unrelated to this work. The funders had no role in the decision to publish or preparation of this manuscript, and no specific funding was received from any funding bodies in the public, commercial or not-for-profit sectors to carry out the work described in this manuscript. The content is solely the responsibility of the authors and does not necessarily represent the official views of Harvard University, its affiliated academic health care centers, or the National Institutes of Health.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org/10.1289/EHP10710
==== Body
pmcRheumatoid arthritis (RA) is a chronic, debilitating inflammatory condition that is triggered by a combination of environmental and genetic factors.1–3 Several decades of RA research have contributed to the development of the “mucosal paradigm” of RA disease pathogenesis, which posits that local inflammation of the mucosa in airways and other mucosal sites leads to the loss of immune tolerance and the production of autoantibodies.4,5 This process often occurs years before patients develop the joint-related symptoms that are the clinical hallmarks of RA.
Support for the mucosal origins hypothesis derives from several sources. Respiratory exposures, such as cigarette smoking and silica dust, have been associated with RA risk, demonstrating that airway damage and inflammation likely serve as key drivers of RA autoantibody formation and disease pathogenesis.6–8 Cigarette smoking in particular has been shown to strongly interact with known genetic risk factors for RA, including the major histocompatibility complex, class II, DR beta 1 (HLA-DRB1) “shared epitope,” suggesting that people with specific genetic variants may be more likely to generate autoantibodies that place them on the path toward RA development.9–11 Recent evidence showing an association between childhood passive smoking and future development of adult-onset seropositive RA further emphasizes the potential for a long latency period for environmental exposures and the development of clinical RA symptoms.12 However, because many people who develop RA do not smoke cigarettes or have significant exposure to secondhand smoke, there has been intense interest in identifying other inhalants that may also affect RA risk.
In this issue of Environmental Health Perspectives, Zhang et al. report on their comprehensive study investigating ambient air pollution, genetic factors, and RA risk.13 They found that higher levels of specific pollutants and a higher overall air pollution score were each associated with increased RA risk. They noted a strong dose effect for the air pollution score on RA risk among subjects with low and intermediate RA genetic risk.
These findings contribute to the growing literature investigating the impact of inhalants other than cigarette smoke on RA risk. An investigation in the Nurses’ Health Study of 90,297 women found that living within 50 m of a road conferred increased risk of RA compared with living 200m or farther away.14 A study of 640,041 people in the British Columbia Border Air Quality Study found a similar association between traffic proximity and RA risk.15 However, the data linking specific pollutants to RA risk have been conflicting. In both the Nurses’ Health Study and the British Columbia study, specific pollutants, including particulate matter with aerodynamic diameters of ≤2.5μm or ≥10μm, nitrogen dioxide, and sulfur dioxide, were not associated with RA risk.15,16 Similarly, there was no significant association between pollutants and RA risk after adjustment for smoking and education in a large case–control study of 1,497 incident RA cases and 2,536 age- and sex-matched controls in the Swedish Epidemiological Investigations of Rheumatoid Arthritis study.17 In contrast, large population-based studies of 322,301 people in Taiwan18 and 230,034 people in South Korea19 did find associations between specific pollutants and RA. However, none of these previous studies incorporated genetic factors to investigate possible gene–pollutant interactions as in the study by Zhang et al.13
Although Zhang et al. included several single-nucleotide polymorphisms of the human HLA genes in the RA genetic risk score,13 they did not specifically determine shared epitope status or other classical HLA alleles. Because these HLA proteins play a central role in the presentation of neoantigens to T cells,20 it is possible that the lack of strong gene–pollutant interaction could be due to omission of this important RA genetic risk factor. Important future directions include the incorporation of HLA into RA genetic risk scores, as well as investigating occupational inhalants and seropositive RA, while integrating RA genetic risk.
In conclusion, it is increasingly clear that air pollutants and genetic factors each likely contribute to RA risk. The important findings reported by Zhang et al.13 offer more rationale to encourage policies to improve air quality to lower risk of RA and other autoimmune conditions that have been associated with respiratory exposures. These findings also provide additional evidence of the mucosal paradigm of RA risk related to pulmonary mucosal injury from air pollutants, which will continue to be a key area of mechanistic RA research.
==== Refs
References
1. Sparks JA. 2019. Rheumatoid arthritis. Ann Intern Med 170 (1 ):ITC1–ITC16, PMID: , 10.7326/AITC201901010.30596879
2. Kowalski EN, Qian G, Vanni KMM, Sparks JA. 2022. A roadmap for investigating preclinical autoimmunity using patient-oriented and epidemiologic study designs: example of rheumatoid arthritis. Front Immunol 13 :890996, PMID: , 10.3389/fimmu.2022.890996.35693829
3. Sparks JA, Costenbader KH. 2014. Genetics, environment, and gene–environment interactions in the development of systemic rheumatic diseases. Rheum Dis Clin North Am 40 (4 ):637–657, PMID: , 10.1016/j.rdc.2014.07.005.25437282
4. Holers VM, Demoruelle MK, Kuhn KA, Buckner JH, Robinson WH, Okamoto Y, et al. 2018. Rheumatoid arthritis and the mucosal origins hypothesis: protection turns to destruction. Nat Rev Rheumatol 14 (9 ):542–557, PMID: , 10.1038/s41584-018-0070-0.30111803
5. Klareskog L, Stolt P, Lundberg K, Källberg H, Bengtsson C, Grunewald J, et al. 2006. A new model for an etiology of rheumatoid arthritis: smoking may trigger HLA-DR (shared epitope)–restricted immune reactions to autoantigens modified by citrullination. Arthritis Rheum 54 (1 ):38–46, PMID: , 10.1002/art.21575.16385494
6. Hutchinson D, Shepstone L, Moots R, Lear JT, Lynch MP. 2001. Heavy cigarette smoking is strongly associated with rheumatoid arthritis (RA), particularly in patients without a family history of RA. Ann Rheum Dis 60 (3 ):223–227, PMID: , 10.1136/ard.60.3.223.11171682
7. Stolt P, Källberg H, Lundberg I, Sjögren B, Klareskog L, Alfredsson L, et al. 2005. Silica exposure is associated with increased risk of developing rheumatoid arthritis: results from the Swedish EIRA study. Ann Rheum Dis 64 (4 ):582–586, PMID: , 10.1136/ard.2004.022053.15319232
8. Prisco LC, Martin LW, Sparks JA. 2020. Inhalants other than personal cigarette smoking and risk for developing rheumatoid arthritis. Curr Opin Rheumatol 32 (3 ):279–288, PMID: , 10.1097/BOR.0000000000000705.32141952
9. Kim K, Jiang X, Cui J, Lu B, Costenbader KH, Sparks JA, et al. 2015. Interactions between amino acid–defined major histocompatibility complex class II variants and smoking in seropositive rheumatoid arthritis. Arthritis Rheumatol 67 (10 ):2611–2623, PMID: , 10.1002/art.39228.26098791
10. Padyukov L, Silva C, Stolt P, Alfredsson L, Klareskog L. 2004. A gene–environment interaction between smoking and shared epitope genes in HLA-DR provides a high risk of seropositive rheumatoid arthritis. Arthritis Rheum 50 (10 ):3085–3092, PMID: , 10.1002/art.20553.15476204
11. Okada Y, Wu D, Trynka G, Raj T, Terao C, Ikari K, et al. 2014. Genetics of rheumatoid arthritis contributes to biology and drug discovery. Nature 506 (7488 ):376–381, PMID: , 10.1038/nature12873.24390342
12. Yoshida K, Wang J, Malspeis S, Marchand N, Lu B, Prisco LC, et al. 2021. Passive smoking throughout the life course and the risk of incident rheumatoid arthritis in adulthood among women. Arthritis Rheumatol 73 (12 ):2219–2228, PMID: , 10.1002/art.41939.34406709
13. Zhang J, Fang XY, Wu J, Fan YG, Leng RX, Liu B, et al. 2023. Association of combined exposure to ambient air pollutants, genetic risk and incident rheumatoid arthritis: a prospective cohort study in UK biobank. Environ Health Perspect 131 (3 ):037008, 10.1289/EHP10710.36913237
14. Hart JE, Laden F, Puett RC, Costenbader KH, Karlson EW. 2009. Exposure to traffic pollution and increased risk of rheumatoid arthritis. Environ Health Perspect 117 (7 ):1065–1069, PMID: , 10.1289/ehp.0800503.19654914
15. De Roos AJ, Koehoorn M, Tamburic L, Davies HW, Brauer M. 2014. Proximity to traffic, ambient air pollution, and community noise in relation to incident rheumatoid arthritis. Environ Health Perspect 122 (10 ):1075–1080, PMID: , 10.1289/ehp.1307413.24905961
16. Hart JE, Källberg H, Laden F, Costenbader KH, Yanosky JD, Klareskog L, et al. 2013. Ambient air pollution exposures and risk of rheumatoid arthritis. Arthritis Care Res (Hoboken) 65 (7 ):1190–1196, PMID: , 10.1002/acr.21975.23401426
17. Hart JE, Källberg H, Laden F, Bellander T, Costenbader KH, Holmqvist M, et al. 2013. Ambient air pollution exposures and risk of rheumatoid arthritis: results from the Swedish EIRA case–control study. Ann Rheum Dis 72 (6 ):888–894, PMID: , 10.1136/annrheumdis-2012-201587.22833374
18. Jung CR, Hsieh HY, Hwang BF. 2017. Air pollution as a potential determinant of rheumatoid arthritis: a population-based cohort study in Taiwan. Epidemiology 28 (suppl 1 ):S54–S59, PMID: , 10.1097/EDE.0000000000000732.29028676
19. Park JS, Choi S, Kim K, Chang J, Kim SM, Kim SR, et al. 2021. Association of particulate matter with autoimmune rheumatic diseases among adults in South Korea. Rheumatology (Oxford) 60 (11 ):5117–5126, PMID: , 10.1093/rheumatology/keab127.33560298
20. van Drongelen V, Holoshitz J. 2017. Human leukocyte antigen–disease associations in rheumatoid arthritis. Rheum Dis Clin North Am 43 (3 ):363–376, PMID: , 10.1016/j.rdc.2017.04.003.28711139
|
PMC010xxxxxx/PMC10010385.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36913235
EHP12630
10.1289/EHP12630
Invited Perspective
Invited Perspective: Examining Chemicals in Food as a Priority for Toxicity Testing
https://orcid.org/0000-0002-5500-0222
Carignan Courtney C. 1
1 Department of Food Science and Human Nutrition, Department of Pharmacology and Toxicology, Michigan State University, East Lansing, Michigan, USA
Address correspondence to Courtney C. Carignan, 469 Wilson Rd., Room 208, East Lansing, MI 48864 USA. Telephone: (517) 884.2039. Email: [email protected]
13 3 2023
3 2023
131 3 03130419 12 2022
10 2 2023
14 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
C.C.C. has served as a plaintiff’s expert witness for a case involving exposure to per- and polyfluoroalkyl substances.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11305
==== Body
pmcIn their new paper,1 Zhao et al. present an elegant computational approach to estimate concentrations of exogenous chemicals in human blood and calculated corresponding toxic equivalencies to prioritize toxicity testing. Application of their new model identified a surprisingly high proportion of food additives, indirect additives, and food-contact substances predicted to have the highest total toxic equivalencies [referred to by Zhao et al. as bioanalytical equivalencies (BEQ%)] for the general population. Other predominant categories included industrial chemicals, pesticides, and household, fragrance, and personal care products.
Zhao et al. used biomonitoring2 and ExpoCast3 data for 216 compounds to train and test a machine learning algorithm. They employed the model to predict blood concentrations for 7,858 chemicals from ToxCast,4 which they used to calculate %BEQ for 12 endocrine-disruption assays. The authors listed the top 25 chemicals with the highest BEQ% for each of the assays, a total of 145 unique chemicals, in their Excel Table S8. I visualized these results in Figure 1 by summing BEQ% by application category and within each category by assay end point.
Figure 1. Proportional summed toxic equivalencies by application category and within each category by ToxCast end point (n=145). Application categories derived by summing bioanalytic equivalencies (BEQ%1 + BEQ%2) for chemicals within each category in Zhao et al., Excel Table S8.1
Figure 1 is a pie chart displays the following information: Pesticides are 9 percent, Food additives, indirect additives and contact substances are 38 percent, household fragrance and personal care are 21 percent, and industrial uses are 31 percent. The following information is given: Pesticides: Metabolic are 32 percent, thyroid hormone are 31 percent, estrogen are 25 percent, androgen are 12 percent. Food additives, indirect additives and contact substances: Metabolic are 51 percent, Androgen are 31 percent, Thyroid hormone are 13 percent, and Estrogen are 5 percent. Household fragrance and personal care: Estrogen are 33 percent, Androgen are 25 percent, Thyroid hormone are 28 percent, and Metabolic are 14 percent. Industrial uses: Androgen are 45 percent, Metabolic are 36 percent, Estrogen are 16 percent, and Thyroid hormone are 3 percent.
The unexpected predominance of the 50 food additives, indirect additives, and food-contact substances identified by this research highlights the need for comprehensive and agnostic approaches to toxicity testing prioritization. Resulting BEQ% were especially notable for two flavoring agents (2,3-butanedione and methyl formate), a colorant (FD&C Yellow 5), and three plasticizers used in food-contact substances (dimethyl isophthalate, diisobutyl phthalate, and diethyl phthalate), indicating the need to prioritize these chemicals for closer evaluation. Rigorous evaluation is also a priority among consumers, who are increasingly choosing organic foods and those with fewer ingredients and less packaging in an effort to avoid potentially harmful chemical exposures.5
Chemical production is on the rise,6 and only a small fraction of the estimated 350,000-plus chemicals in commerce have undergone careful screening or testing.7 Identification of harmful chemicals in commerce with widespread exposure has been ongoing for decades.8 Notorious examples include dichlorodiphenyltrichloroethane (DDT), lead, radium, dioxins, polybrominated diphenyl ethers, phthalates, bisphenols, and per- and polyfluoroalkyl substances.9,10 These discoveries have occurred alongside soaring rates of chronic diseases and conditions—including infertility, metabolic syndrome, thyroid disease, cancer, and neurodevelopmental and neurodegenerative conditions—that are not fully accounted for by genetic, lifestyle, or nutritional factors11 and that have been linked with exposure to many of these contaminants.12
The new model can also be used to screen alternatives that may have been introduced without adequate toxicity testing, potentially avoiding the continued use of regrettable substitutions. For example, one of the plasticizers identified by Zhao et al. for prioritization, diisobutyl phthalate, has been used as a replacement for di(2-ethylhexyl) phthalate.13 The authors also identified tri-isobutyl phosphate for prioritization; this compound is in the class of organophosphate flame retardants that have been widely used as replacements for polybrominated diphenyl ethers.14,15
Zhao et al. trained the model using biomonitoring data from the nationally representative National Health and Nutrition Examination Survey (NHANES),2 so it should be reasonably inclusive of communities with disparate burdens of chemical exposures. Improved chemical prioritization for toxicity testing is vital to better protect low-income communities and communities of color. A future evaluation of model performance could test the accuracy of predictions among vulnerable and sensitive populations. The authors noted several other ways to improve the model, such as use of more robust toxicity data and periodic future updates to incorporate new estimates of exposure and measured blood concentrations of chemicals.1
The task of predicting population blood concentrations for thousands of chemicals is seemingly impossible, yet it is a necessary step in overhauling current methods for prioritization of chemicals for toxicity testing. I applaud Zhao et al. for their innovative approach to tackling it.
Acknowledgments
C.C.C. is supported in part by the National Institute of Environmental Health Sciences, National Institutes of Health (R01ES028311), the U.S. Environmental Protection Agency National Priorities Program (No. 84025201 and Assistance Agreement No. R839482), and the U.S. Department of Agriculture National Institute of Food and Agriculture (Hatch Project MICL02565). This perspective has not been formally reviewed by the funding agencies. The views expressed in this document are solely from the author and do not necessarily reflect those of the funding agencies. The agencies do not endorse any products or commercial services mentioned in this paper.
==== Refs
References
1. Zhao F, Li L, Lin P, Chen Y, Xing S, Du H, et al. 2023. HExpPredict: in vivo exposure prediction of human blood exposome using a random forest model and its application in chemical risk prioritization. Environ Health Perspect 131 (3 ):037009, 10.1289/EHP11305.36913238
2. Calafat AM. 2012. The U.S. National Health and Nutrition Examination Survey and human exposure to environmental chemicals. Int J Hyg Environ Health 215 (2 ):99–101, PMID: , 10.1016/j.ijheh.2011.08.014.21937270
3. Wambaugh JF, Rager JE. 2022. Exposure forecasting–ExpoCast–for data-poor chemicals in commerce and the environment. J Expo Sci Environ Epidemiol 32 (6 ):783–793, PMID: , 10.1038/s41370-022-00492-z.36347934
4. Richard AM, Huang R, Waidyanatha S, Shinn P, Collins BJ, Thillainadarajah I, et al. 2021. The Tox21 10K compound library: collaborative chemistry advancing toxicology. Chem Res Toxicol 34 (2 ):189–216, PMID: , 10.1021/acs.chemrestox.0c00264.33140634
5. Crawford E. 2018. Eight trends that are reshaping the natural industry and opening doors in the conventional channel. Food Navigator USA. https://www.foodnavigator-usa.com/Article/2018/03/16/Eight-trends-that-are-reshaping-the-natural-industry-and-opening-doors-in-the-conventional-channel [accessed 19 December 2022].
6. Bernhardt E, Rosi, E, Gessner, M. 2017. Synthetic chemicals as agents of global change. Front Ecol Environ 15 (2 ):84–90, 10.1002/fee.1450.
7. European Environment Agency. 2001. Late Lessons from Early Warnings: The Precautionary Principle 1896–2000. Luxembourg: Office for Official Publications of the European Communities. https://www.eea.europa.eu/publications/environmental_issue_report_2001_22#:∼:text=Late%20lessons%20from%20early%20warnings%20is%20about%20the%20gathering%20of,then%20living%20with%20the%20consequences [accessed 19 December 2022].
8. Wang Z, Walker GW, Muir DCG, Nagatani-Yoshida K. 2020. Toward a global understanding of chemical pollution: a first comprehensive analysis of national and regional chemical inventories. Environ Sci Technol 54 (5 ):2575–2584, PMID: , 10.1021/acs.est.9b06379.31968937
9. Naidu R, Biswas B, Willett IR, Cribb J, Kumar Singh B, Paul Nathanail C, et al. 2021. Chemical pollution: a growing peril and potential catastrophic risk to humanity. Environ Int 156 :106616, PMID: , 10.1016/j.envint.2021.106616.33989840
10. Karlsson O, Rocklöv J, Lehoux AP, Bergquist J, Rutgersson A, Blunt MJ, et al. 2021. The human exposome and health in the Anthropocene. Int J Epidemiol 50 (2 ):378–389, PMID: , 10.1093/ije/dyaa231.33349868
11. Sears ME, Genuis SJ. 2012. Environmental determinants of chronic disease and medical approaches: recognition, avoidance, supportive therapy, and detoxification. J Environ Public Health 2012 :356798, PMID: , 10.1155/2012/356798.22315626
12. Sly PD, Carpenter DO, Van den Berg M, Stein RT, Landrigan PJ, Brune-Drisse M-N, et al. 2016. Health consequences of environmental exposures: causal thinking in global environmental epidemiology. Ann Glob Health 82 (1 ):3–9, PMID: , 10.1016/j.aogh.2016.01.004.27325063
13. Chiang C, Flaws JA. 2019. Subchronic exposure to di(2-ethylhexyl) phthalate and diisononyl phthalate during adulthood has immediate and long-term reproductive consequences in female mice. Toxicol Sci 168 (2 ):620–631, PMID: , 10.1093/toxsci/kfz013.30649530
14. Blum A, Behl M, Birnbaum LS, Diamond ML, Phillips A, Singla V, et al. 2019. Organophosphate ester flame retardants: are they a regrettable substitution for polybrominated diphenyl ethers? Environ Sci Technol Lett 6 (11 ):638–649, PMID: , 10.1021/acs.estlett.9b00582.32494578
15. Zhang X, Sühring R, Serodio D, Bonnell M, Sundin N, Diamond ML. 2016. Novel flame retardants: estimating the physical–chemical properties and environmental fate of 94 halogenated and organophosphate PBDE replacements. Chemosphere 144 :2401–2407, PMID: , 10.1016/j.chemosphere.2015.11.017.26613357
|
PMC010xxxxxx/PMC10010393.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36913238
EHP11305
10.1289/EHP11305
Research
HExpPredict: In Vivo Exposure Prediction of Human Blood Exposome Using a Random Forest Model and Its Application in Chemical Risk Prioritization
Zhao Fanrong 1 2 3
Li Li 4
Lin Penghui 3
Chen Yue 5
Xing Shipei 6
Du Huili 3 7
Wang Zheng 5
Yang Junjie 3
Huan Tao 6
Long Cheng 5
Zhang Limao 3
Wang Bin 8 9
https://orcid.org/0000-0002-2204-9783
Fang Mingliang 1 3 10
1 Department of Environmental Science and Engineering, Fudan University, Shanghai, P.R. China
2 Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore
3 School of Civil and Environmental Engineering, Nanyang Technological University, Singapore
4 School of Community Health Sciences, University of Nevada, Reno, Reno, Nevada, USA
5 School of Computer Science and Engineering, Nanyang Technological University, Singapore
6 Department of Chemistry, University of British Columbia, Vancouver, British Columbia, Canada
7 College of Resources and Environment, University of Chinese Academy of Sciences, Beijing, P.R. China
8 Institute of Reproductive and Child Health, Peking University/Key Laboratory of Reproductive Health, National Health Commission of the People’s Republic of China, Beijing, P.R. China
9 Department of Epidemiology and Biostatistics, School of Public Health, Peking University, Beijing, P.R. China
10 Institute of Eco-Chongming, Shanghai, P.R. China
Address correspondence to Bin Wang, Institute of Reproductive and Child Health, Peking University/National Health Commission’s Key Laboratory of Reproductive Health, Beijing 100191, China. Email: [email protected]. And, Mingliang Fang, Department of Environmental Science and Engineering, Fudan University, Shanghai, China, 200433. Email: [email protected]
13 3 2023
3 2023
131 3 03700925 3 2022
15 12 2022
14 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Due to many substances in the human exposome, there is a dearth of exposure and toxicity information available to assess potential health risks. Quantification of all trace organics in the biological fluids seems impossible and costly, regardless of the high individual exposure variability. We hypothesized that the blood concentration (CB) of organic pollutants could be predicted via their exposure and chemical properties. Developing a prediction model on the annotation of chemicals in human blood can provide new insight into the distribution and extent of exposures to a wide range of chemicals in humans.
Objectives:
Our objective was to develop a machine learning (ML) model to predict blood concentrations (CBs) of chemicals and prioritize chemicals of health concern.
Methods:
We curated the CBs of compounds mostly measured at population levels and developed an ML model for chemical CB predictions by considering chemical daily exposure (DE) and exposure pathway indicators (δij), half-lives (t1/2), and volume of distribution (Vd). Three ML models, including random forest (RF), artificial neural network (ANN) and support vector regression (SVR) were compared. The toxicity potential or prioritization of each chemical was represented as a bioanalytical equivalency (BEQ) and its percentage (BEQ%) estimated based on the predicted CB and ToxCast bioactivity data. We also retrieved the top 25 most active chemicals in each assay to further observe changes in the BEQ% after the exclusion of the drugs and endogenous substances.
Results:
We curated the CBs of 216 compounds primarily measured at population levels. RF outperformed the ANN and SVF models with the root mean square error (RMSE) of 1.66 and 2.07μM, the mean absolute error (MAE) values of 1.28 and 1.56μM, the mean absolute percentage error (MAPE) of 0.29 and 0.23, and R2 of 0.80 and 0.72 across test and testing sets. Subsequently, the human CBs of 7,858 ToxCast chemicals were successfully predicted, ranging from 1.29×10−6 to 1.79×10−2 μM. The predicted CBs were then combined with ToxCast in vitro bioassays to prioritize the ToxCast chemicals across 12 in vitro assays with important toxicological end points. It is interesting that we found the most active compounds to be food additives and pesticides rather than widely monitored environmental pollutants.
Discussion:
We have shown that the accurate prediction of “internal exposure” from “external exposure” is possible, and this result can be quite useful in the risk prioritization. https://doi.org/10.1289/EHP11305
Supplemental Material is available online (https://doi.org/10.1289/EHP11305).
The authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Because many chemical substances have been developed and used in commerce over numerous recent decades, there is a dearth of exposure and toxicity information available to assess potential health risks of most of these chemicals to humans.1,2 To address concerns over the potential health effects of untested chemicals, high-throughput screening (HTS) assessments that incorporate both exposure and toxicity data are needed for risk-based screening and prioritization.1–4 The U.S. Environmental Protection Agency (U.S. EPA) has developed the ToxCast program to provide in vitro bioactivity data that may inform chemical toxicity.5,6 However, to use the in vitro bioactivity data of ToxCast to evaluate the potential risk to human health, chemical blood concentration (CB) is essential to link the internal exposure to external human exposure.7
One challenge to chemical exposure and risk assessments has been the demand for a large number of chemical CB measurements.8 Clearly, experimental quantification is cumbersome and time-consuming. The standards used for analysis are also costly or difficult to obtain. In addition, the concentrations of most compounds are too low to be detectable.9,10 Moreover, there is high variability in chemical levels between biospecimens from different people, sometimes even for samples collected from the same donors on different days in cases of exposure to rapidly metabolized chemicals.11,12 The National Health and Nutrition Examination Survey (NHANES) has spent years monitoring several hundred chemicals, which is still insufficient for the evaluation of chemical exposure risk in the era of the exposome. Therefore, without extensive direct measurements of chemicals at the population level, there is an urgent need to explore whether we can develop in silico methods to predict the CBs of chemicals. Although the U.S. EPA has also developed the ExpoCast program to predict human exposure to the large number of chemicals with the balanced accuracies of the source-based exposure pathway models ranging from 73% to 81% and with a coefficient of determination (R2) between predictions and biomonitoring-based inferences of 0.8,3 the ExpoCast can only predict the intake rates, which is an indicator of external exposure. Because different chemicals have different bioavailability and clearance, to assess health risks using ToxCast activity test data, it is necessary to convert the external exposure data into internal concentration in bodily fluids.7 Previous efforts built quantitative approaches to translate in vitro toxicity potencies to equivalent in vivo doses using in vitro−in vivo extrapolation (IVIVE) techniques.13 These approaches used pharmacokinetic equations to estimate steady-state plasma concentrations (CSS) using the High-Throughput Toxicokinetic (HTTK) the open-source R package (version 4.2.1; R Development Core Team).13 However, the CB values predicted by the HTTK model were derived by assuming steady-state and 100% oral bioavailability under a dose rate of 1mg/kg/d, which did not consider the exposure and the corresponding uncertainty; and chemicals such as perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS), which were thought to be actively resorbed by the kidney, were not captured by the current HTTK model.13 In addition, for the recent studies, the high-throughput PROduction-To-EXposure (PROTEX-HT) model developed by Li et al. could already predict the Css without assuming 100% oral absorption,2 and the Physiologically based Toxicokinetic (PBTK) model developed by Armitage et al. could already capture the renal clearance and reabsorption of ions such as polyfluoroalkyl substances (PFAS).14 However, most of those theoretical methods used to predict the chemical Css resulting from repeated daily exposure were limited to oral route of exposure.4,15,16
We hypothesized that the CB of organic pollutants could be predicted via their exposure and chemical properties, especially for those with similar exposure routes and physicochemical parameters. We seek to increase the prediction accuracy of CB using machine learning (ML) methods. In this study, we curated the CBs of pollutants in the general population from available databases and literature and applied ML algorithms for CB predictions by optimizing the key parameters that mediate the CB. We compared three ML algorithms, including random forest (RF), artificial neural network (ANN) and support vector regression (SVR), based on the publicly available experimental data. The best-performing RF model was then used to predict the CBs of >7,500 ToxCast chemicals. The predicted CB values were further combined with ToxCast in vitro bioassays to prioritize those ToxCast chemicals in terms of CB/AC50 ratios, using different assay end points. This advanced human internal exposure prediction (HExpPredict) approach provides the ability to evaluate and prioritize chemicals for potential risk to human health.
Methods
A detailed data processing workflow is depicted in Figure 1. Key parameters and models regarding the models developed for this study are described in the following sections. According to the pharmacokinetics and toxicokinetics models, factors that are known or expected to influence the relationship between external exposure and the chemical CB are the elimination half-life, bioavailability, volume of distribution (Vd), dosage, and dosing interval.17 When defining dosing interval equal to 1 day, the maintenance dose refers to the daily exposure (DE, milligrams per kilogram body weight per day). Because of the lack of data, the parameters such as renal clearance half-life and bioavailability were treated as an unknown parameter and trained by ML model. As one of the major pathways of elimination, the predictable biotransformation half-life (t1/2) was included in our prediction model.
Figure 1. Overview of framework for human CB prediction (HExpPredict) modeling and risk prioritization in this study. Note: CB, blood concentration.
Figure 1 is schematic illustration with two parts. On the left, the illustration is titled Blood exposure database has four steps. Step 1: Biomonitoring California C D C, exposome explorer, and literature lead to 216 chemicals and parameter characterization. Step 2: Q S A R ready structures with 7858 chemicals leads to H L B prediction, experimental C begin subscript uppercase b end subscript, V begin subscript lowercase d end subscript, ExpoCast S E E M 3, including D E and lowercase delta begin subscript lowercase italic i j end subscript. Step 3: ToxCast trademark leads to Q S A R ready structures with 7858 chemicals. Step 4: ExpoCast trademark leads to ExpoCast S E E M 3, including D E and lowercase delta begin subscript lowercase italic i j end subscript. On the right, the illustration titled machine learning prediction model has four steps. Step 1: 172 training set and 44 testing set with an icon of a flowchart. Step 2: A set of three dot graphs titled AN N, random forest S V R plots experimental inc begin subscript uppercase b end subscript (y-axis) across predicted inc begin subscript uppercase b end subscript (x-axis) leads to 7858 prediction set. Step 3: It is set of one line graph and pie chart. The line graph titled predicted uppercase c begin subscript uppercase b end subscript plots predicted uppercase c begin subscript uppercase b end subscript (y-axis) across Rank (x-axis). T E Q begin subscript lowercase i end subscript percentage equals uppercase c begin subscript uppercase b end subscript per Activity concentration, 50 percent begin subscript lowercase i end subscript over uppercase sigma uppercase c begin subscript uppercase b end subscript per Activity concentration, 50 percent begin subscript lowercase i end subscript uppercase c begin subscript uppercase b end subscript per Activity concentration, 50 percent begin subscript lowercase i end subscript. Under risk prioritization, a pie chart depict ratios of Top 1, Top 2, Top 3.
Chemical Selection
The chemicals were selected based on a subset of the ToxCast Database (version 3.0, publicly released October 2018) in this study, for which the exposure data and in vitro bioactivity assay data were readily available.18 The U.S. EPA’s ToxCast chemical list includes more than 9,000 compounds, including industrial chemicals, pesticides, consumer product ingredients, and pharmaceuticals. The full list of chemicals considered is available in Excel Table S1. All chemical descriptors including CAS registry number, chemical name, Simplified Molecular Input Line Entry Specification (SMILES), molecular formula, average mass, and monoisotopic mass are available through the U.S. EPA’s CompTox Chemicals Dashboard (version 2.1.1; https://comptox.epa.gov/dashboard/batch-search).19
Exposure Estimates
The median of estimated DE level (milligrams per kilogram body weight per day) with uncertainty [95% confidence interval (CI)] for the ToxCast chemical as shown in Excel Table S1 was acquired from the U.S. EPA’s ExpoCast exposure estimates, which were developed using the General Population Consensus Model (SEEM3).3,20 The exposure pathway indicators (δij) for four source-based pathways (far-field pesticide use, nonpesticide dietary exposure, far-field industrial exposure, and consumer) in the SEEM3 model were also included in our prediction model.3 The δij is an estimated probability of whether a given pathway j is relevant to a given chemical i.
Chemical Biotransformation Half-Life Prediction
The predicted half-life values (t1/2) for the ToxCast chemicals were taken from the Human Exposome and Metabolite Database (HExpMetDB).21 The prediction was based on the quantitative structure−activity relationship (QSAR) approach called Iterative Fragment Selection (IFS).22
The Distribution Volume (Vd) Prediction
The Vd values were predicted by a comprehensive exposure model named Risk Assessment, IDentification And Ranking-Indoor and Consumer Exposure (RAIDAR-ICE) according to previous study.23
Molecular Descriptors and QSAR Parameter Calculation
The QSAR parameters such as Log KOW and Log KOA were calculated using solute descriptors provided by the online UFZ-LSER Database.24 Water solubility (WS) and substructure molecular descriptors were calculated by the Toxicity Estimation Software Tool (TEST, version 5.1.1).25
Chemical CB Search
To investigate the occurrence and levels of xenobiotics in human blood, we conducted a database and literature search on chemicals in human blood. The measured CBs of xenobiotics in this study were first retrieved from the NHANES 2003–2017,26 the California Environmental Contaminant Biomonitoring Program (also known as Biomonitoring California),27 or the Exposome-Explorer.28 We excluded drugs and endogenous compounds by filtering the U.S. EPA’s CompTox Chemicals Dashboard Drugbank list (https://comptox.epa.gov/dashboard/chemical-lists/DRUGBANK) and manually searching the chemical category through PubChem (https://pubchem.ncbi.nlm.nih.gov/). When a given chemical was present in both of these databases, we used the NHANES concentrations. To further obtain concentration data for more compounds, we performed a literature search on typical pollutants that were not in the databases, based on the chemicals of concerns previously summarized in our research.29 The National Center for Biotechnology Information (NCBI) PubMed database (https://pubmed.ncbi.nlm.nih.gov/) was searched from the year 2005 to 2022. The keywords used to search the PubMed database included those describing sample types “blood,” “plasma,” or “serum” and terms for the typical pollutant classes summarized in our previous study,29 including “perfluorinated compounds,” “volatile organic compound,” “pesticide,” “organophosphorus flame retardant,” or “polycyclic aromatic hydrocarbons,” together with keywords including “exposome,” “exposure,” “detection,” “level” or “concentration.” We included only the studies from healthy human populations using a mass spectrometry–based analytical method during our manual screening of the possible literature hits. We also excluded the studies from polluted areas or special environment areas. The CB of each compound was calculated based on the sample size weighted geometric mean (GM, if provided) or median concentrations measured in serum, plasma, or blood. To develop models for different age and sex groups, we also collected the GMs of CBs for different age and sex groups from the NHANES Database (n=48).
ML Models
Methods of random search and 5-fold cross-validation were used for parameter optimization to train three ML models (i.e., RF, ANN, and SVR) with various prediction features of DE, δij, Vd, t1/2, and other chemical properties, of which the optimal parameters was evaluated by and root mean square error (RMSE). The publicly available data sets Exposome-Explorer database,28 the Fourth National Report on Human Exposure to Environmental Chemicals,26 and the California Environmental Contaminant Biomonitoring Program27 were searched for experimentally measured human in vivo CB values. Literature mining was performed by manually searching reviews or articles as mentioned above. The measured CBs were employed to train ML models for in silico CB prediction. We excluded the drug and endogenous compounds by filtering the U.S. EPA’s CompTox Chemicals Dashboard Drugbank list and manually searching the chemical category through PubChem (https://pubchem.ncbi.nlm.nih.gov/), because our model only considers the CBs produced by external exposures. For the collected experimental CBs and predicted t1/2, Vd, and DE values, we unified their units into micromolar, day, L, and micromole per day, respectively, and normalized the right-skewed data by natural logarithmic transformation before feeding them to a ML model. The training and testing splits were 80:20 to train and test RF, ANN, and SVR models. Training and testing set chemicals were randomly selected. In this work, the RMSE, mean absolute error (MAE), mean absolute percentage error (MAPE), and fitness degree R2 of the three models were compared. Finally, the trained model was used to predict CB for the ToxCast compounds. All analyses were performed in R (version 4.2.1; R Development Core Team). All chemical predictors are provided in Excel Table S1. To improve the applicability of our model, the R script and tutorial for users are also available in the Supplemental File HExpPredict_scripts.rar and Supplemental Material, “Text S1,” as well as at https://github.com/FangLabNTU/HExpPredict.
Monte Carlo (MC) Simulation and Parameter Distributions
MC simulation was implemented to simulate the impact of DE and t1/2 uncertainty on calculating the CB 10,000 times, using a similar model as in our previous studies.21,30,31 Three separate MC simulations were performed referring to previous studies: DE prediction uncertainty only, t1/2 prediction uncertainty only, and both DE and t1/2 prediction uncertainty.20,21 For each chemical, the CB was calculated 10,000 times for three separate MC simulations respectively, allowing estimation of the 5th, median, and 95th percentiles.
In Vitro Bioactivity Data
All ToxCast in vitro HTS data (version 3.0, publicly released October 2018)18 were downloaded from the U.S. EPA’s CompTox Chemicals Dashboard (version 2.1) Assay Endpoints List (https://comptox.epa.gov/dashboard/assay-endpoints?filtered) to estimate the endocrine-related activity. The 12 targeted assays covering the estrogen receptor alpha (ERα) (TOX21_Era_BLA_Agonist_ratio and TOX21_Era_BLA_Antagonist_ratio), androgen receptor (AR) (TOX21_AR_BLA_Agonist_ratio, Tox21_AR_LUC_MDAKB2_Agonist, TOX21_AR_BLA_Antagonist_ratio and TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881), peroxisome proliferator–activated receptor gamma (PPARγ) (Tox21_PPARg_BLA_Agonist_ratio, TOX21_PPARg_BLA_Agonist_ch2, TOX21_PPARg_BLA_Antagonist_ch1 and TOX21_PPARg_BLA_antagonist_viability), and thyroid hormone receptor (TR) (TOX21_TR_LUC_GH3_Agonist and TOX21_TR_LUC_GH3_Antagonist) were chosen for further study. The bioactivity potential or prioritization of each chemical was represented as CB-to-AC50 ratio (CB/AC50). We used the concentration at 50% of maximum activity (AC50) estimates from the U.S. EPA’s CompTox Chemicals Dashboard (version 2.1) ToxCast Assay Endpoints List32 provided by the ToxCast program18 as well as the predicted CB to calculate the CB/AC50 ratios of ToxCast chemicals.
The relative ranking of CB/AC50 can be used for priority setting; that is, higher CB/AC50 can be considered to be a higher priority. The toxicity potential or prioritization of each chemical was represented as a bioanalytical equivalency (BEQ). The BEQ values of each chemical and its percentage in the total BEQ (BEQ%) were estimated based on the below equations29: (1) BEQi=CB i/AC50 i×AC50 ref
(2) BEQi%=BEQi÷∑BEQi×100%,
where CB i is the predicted blood concentration of compound i; AC50 i is the concentration of compound i that causes 50% response; and AC50 ref is the concentration of the reference compound (the compound with the minimum AC50 for each assay) that causes 50% response. We further retrieved the applications of the top 25 most active chemicals of each assay from the NCBI PubMed databases (https://pubmed.ncbi.nlm.nih.gov).
Results
A total of 7,858 chemicals were selected in this study from 9,403-chemical U.S. EPA’s ToxCast Database.18 The chemicals that were not selected (1,545) comprised those that did not have available DE data through ExpoCast SEEM3 and those that were categorized as ionogenic chemicals, organic mixtures, or chemicals with molecular weights over 1,000 Da and therefore unable to be used by the iterated function system (IFS) algorithm.
Chemical CB Search
To investigate the occurrence and levels of the selected chemicals in human blood, we conducted a database26–28 and literature search,10,33–37 extracting CB from identified data and studies. In total, the measured CBs of 216 chemicals were documented for the further ML modeling. In general, the CBs of the documented chemicals ranged from 1.65×10−8 to 1.59μM staggering 8 orders of magnitude. The final list is presented in Excel Table S2, including CAS registry number, chemical name, formula, average mass, monoisotopic mass, weighted CB, and data sources for our RF model. Overall, the NHANES, the Exposome-Explorer Database, and the literature search were the dominant contributors to the training set and contributed to 23%, 39%, and 36% of the data set, respectively. Due to limited measured CB data of the population, we collected data from only 48 chemicals for which the age- and sex-specific geometric means of measured CB was available from NHANES Database. The GM CB ranges for individuals age 12–19 y and those older than 20 y were 4.65×10−6–5.32×10−3 μM and 1.11×10−5–9.00×10−3 μM, respectively. The GM CB ranges were 8.31×10−6–1.07×10−2 μM for males and 8.31×10−6–6.84×10−3 μM for females (Excel Table S3).
Human Exposure Evaluation
The predicted exposure values of 7,858 chemicals were obtained from ExpoCast (Excel Table S1). The estimated human chemical DE ranged from 3.17×10−15 (95% CI: 3.82×10−17, 4.19×10−13) to 4.92 (95% CI: 1.65×10−7, 2.21×105) mg/kg body weight/d, spanning across 15 orders of magnitude. The δij values of 7,858 chemicals ranged from 0 to 1 for the four pathways (Excel Table S1; i.e., far-field pesticide use, nonpesticide dietary exposure, far-field industrial exposure, and consumer), with values near zero indicating low probability and values near one indicating high probability exposure to the chemical.
Chemical t1/2 Evaluation
The t1/2 of 7,858 chemicals listed in Excel Table S1 were successfully predicted using the IFS approach. Of these 7,858 chemicals, the median t1/2 was predicted to be 4.64 h (h). Rolitetracycline was predicted to have the shortest t1/2 of 0.05 h, and mirex was predicted to have the longest t1/2 of 2,020,000 h with a wide range of 8 orders of magnitude.
Chemical Vd Prediction
We used the RAIDAR-ICE model to predict the Vd values of 7,858 chemicals (Excel Table S1). The median Vd was predicted to be 14.4L/kg whole blood. The Vds span over 3 orders of magnitude, from 7.36×10−1 to 20.3L/kg whole blood.
CB Prediction ML Modeling
We developed a workflow to use experimental CB data to train and test ML models (Figure 1). Such models were then applied to the 7,858 chemicals from U.S. EPA ToxCast Program for which in vitro bioactivity data were available. We collected available experimentally measured human CB values through publicly available databases and literature to train ML models for in silico CB prediction. After excluding the drug and endogenous compounds, a total of 216 experimental CB data points were included in the ML model (Figure 2A). We randomly divided the 216 data points into 172 compounds for training and 44 compounds for further testing (i.e., 80%:20%). We downloaded the chemical QSAR-ready SMILES from the U.S. EPA’s CompTox Chemicals Dashboard Batch Search (version 2.1.1),38 which we used to predict the Vd, and t1/2. Chemical-specific inputs to ML models included DE, Vd, t1/2, and δij for parameter tuning.
Figure 2. (A) Overlapping analysis of major sources for measured CB used in machine learning training. (B) Prediction performance of RF ML model for training (n=172) and testing (n=44) sets (referring to the data in Excel Table S4); Black line is the y=x line, and blue dotted lines are 10-fold boundaries; (C) Prediction performance of RF ML model for different groups of chemicals (referring to the data in Excel Table S2); Black line is the y=x line. (D) Violin plots for training and testing set prediction errors by calculating the ratio between measured and predicted concentration from RF ML model (referring to the data in Excel Table S4). Blue dashed lines are the median line, and red dotted lines are quartiles. Note: BC, Biomonitoring California; CB, blood concentration; EE, Exposome-Explorer; ML, machine learning; NHANES, National Health and Nutrition Examination Survey; OPFRs, organophosphorus flame retardants; OP, organochlorine pesticide; PAE, phthalate ester; PBDE; polybrominated diphenyl ether; PCB, polychlorinated biphenyl; PFC, perfluorinated compounds; PPCP, personal care and consumer product; RF, random forest; VOC, volatile organic compound.
Figure 2A is a Venn diagram. On the left, there are N H A N E S, including 50, 0, 0, 0 and B C, including 4, 12, 0, 0. The intersection area includes the following data: 0, 0, 0, 0. On the right, there are E E, including 73, 12, 0, 0 and Literature, including 77, 0, 0, 0. The intersection area includes the following data: 0, 0, 0, 0. Figure 2B is a dot graph, plotting Lanthanum (Predicted uppercase c begin subscript uppercase b end subscript per micrometer), ranging from negative 18 to 0 in increments of negative 2 (y-axis) across Lanthanum (observed uppercase c begin subscript uppercase b end subscript per micrometer), ranging from negative 18 to 0 in increments of negative 2 (x-axis) for training set and testing set. Figure 2C is a line graph, plotting Lanthanum (Predicted uppercase c begin subscript uppercase b end subscript per micrometer), ranging from negative 18 to 0 in increments of negative 2 (y-axis) across Lanthanum (observed uppercase c begin subscript uppercase b end subscript per micrometer), ranging from negative 18 to 0 in increments of negative 2 (x-axis) for Dioxins, O C Ps, P A Es, PO P E Rs, P B D Es, P C Bs, P P C Ps, P A Hs, V O Cs, and P F Cs. Figure 2D is a graph, plotting predicted uppercase c begin subscript uppercase b end subscript per experimental uppercase c begin subscript uppercase b end subscript ratios, ranging from 0.001 to 0.01 in increments of 0.009; 0.01 to 0.1 in increments of 0.09, 0.1 to 1 in increments of 0.9, 1 to 10 in increments of 9, 10 to 100 in increments of 90, and 100 to 1000 in increments of 900 (y-axis) across training set and testing set.
Model Validation
To optimize the CB prediction model performance by training set, tuning parameters including maximum depth (5–100), mtry ratio (0.2–0.8), number of trees (10–500), maximum tuning times (20), and tuning method (“random_search”) were executed using the learner “ranger” of “mlr3” learning platform (https://github.com/mlr-org/mlr3). The RMSE, MAE, MAPE, and R2 were calculated to compare the predicted and experimental CB in the test data set. We investigated three widely used ML models (RF, ANN, and SVR) for CB predictions with seven basic variables, including DE, Vd, t1/2, and four δijs. RF outperformed the other two models with RMSE values of 1.66 and 2.07μM, MAE of 1.28 and 1.56μM, MAPE of 0.29 and 0.23, and R2 of 0.80 and 0.72 across training and testing predictions of CB, respectively (Table 1). In comparison, ANN and SVR showed less robustness, with RMSE values of 2.83 and 3.07μM, MAE of 2.13 and 2.56μM, MAPE of 0.39 and 0.35, and R2 of 0.41 and 0.39 for ANN, and with RMSE values of 3.52 and 3.79μM, MAE of 2.81 and 3.22μM, MAPE of 0.69 and 0.47, and R2 of 0.08 and 0.06 for SVR across training and testing sets (Table 1), respectively. Approximately 90% (165 of 174) and 84% (37 of 44) predicted CB values of training and testing sets showed to be within the 10-fold boundary when compared with measured CB values (Figure 2B), showing much better regression than ANN and SVR models (Figure S1, referring to the data in Excel Table S4). To further optimize the RF model, we considered adding sex, age, and variables of varying complexity including Log KOW, Log KOA, WS, and additional molecular descriptors to our RF model. However, the prediction performance was not dramatically improved when more parameters were included into the RF model. Detailed results were provided in the Supplemental Material, “Text S2.”
Table 1 The prediction performance of three CB prediction machine learning models.
Model Training set (n=172) Testing set (n=44)
RMSE MAE MAPE R2 RMSE MAE MAPE R2
Random forest 1.66 1.28 0.29 0.80 2.07 1.56 0.23 0.72
Artificial neural network 2.83 2.13 0.39 0.41 3.07 2.56 0.35 0.39
Support vector regression 3.52 2.81 0.69 0.08 3.79 3.22 0.47 0.06
Note: CB, blood concentration; MAE, mean absolute error; MAPE, mean absolute percentage error; RMSE, root mean square error.
Good prediction performance of the RF model were observed for some typical environmental pollutants, such as polychlorinated biphenyls (PCBs), dioxins, phthalate esters (PAEs), dioxins, polycyclic aromatic hydrocarbons (PAHs), perfluorinated compounds (PFCs), organophosphorus flame retardants (OPFRs), and volatile organic compounds (VOCs) (Figure 2C), with the RMSE of 0.64, 0.70, 0.71, 0.73, 0.83, 0.85, and 0.86, respectively (Table S1). In contrast, some substances, like personal care and consumer products (PPCPs) and organochlorine pesticides (OPs), showed relatively poor prediction performance, with the RMSE of 1.18 and 1.68, respectively. The RF model covered 50% compounds within 0.32 to 2.6 and 0.24 to 3.4 times of predicted CB/experimental CB ratios for training set and testing set, respectively (Figure 2D).
Using the final RF model, CBs were determined for each of the 7,858 ToxCast chemicals. In general, the predicted human blood CB of 7,858 ToxCast chemicals ranged from 1.02×10−6 to 3.25×10−2 μM (Excel Table S1), ranging four orders of magnitude (Figure 3).
Figure 3. The cumulative distribution of chemical predicted CB using RF model (n=7,858). The bar indicates the median predicted CB for each chemical; the pink area represents the predicted CB range (5%−95%) derived from the Monte Carlo simulations. Some typical environmental pollutants are labeled. Note: CB, blood concentration; RF, random forest.
Figure 3 is a line graph, plotting predicted uppercase c begin subscript uppercase b end subscript (micromolar), ranging as 10 begin superscript negative 6 end superscript, 10 begin superscript negative 5 end superscript, 10 begin superscript negative 4 end superscript, 10 begin superscript negative 3 end superscript, 10 begin superscript negative 2 end superscript, 10 begin superscript negative 1 end superscript (y-axis) across chemical rank by uppercase c begin subscript uppercase b end subscript, ranging from 0 to 8000 in increments of 1000 (x-axis) for uppercase c begin subscript uppercase b end subscript and 95 percent confidence interval.
Uncertainty Analysis
Three MC simulations (DE prediction uncertainty alone, t1/2 prediction uncertainty alone, and both) were performed to determine the predicted CB upper 95th percentile. The ratio of the CB for the 95th percentile to the median indicates the relative contribution uncertainty, with larger ratios indicating greater uncertainty. We observed that the ratio value of median t1/2 prediction uncertainty (1.17) was close to DE prediction uncertainty (1.28). The ratio value of both uncertainty (2.17) was close to the sum of t1/2 and DE, which indicated that the prediction of t1/2 and DE contributed approximately the same degree of uncertainty to the prediction model.
Chemical Prioritization Using the U.S. EPA’s ToxCast Database
We evaluated bioactivity potential for each chemical across 12 in vitro assays from ToxCast using AC50/CB ratios calculated as ToxCast AC50/CB ratios. The 12 ToxCast in vitro HT screening assays,18 including the targets of ERα, AR, PPAR-γ and TR, were chosen as case studies. The total 12 assays covered two AR agonists, two AR antagonists, one ERα agonist, one ERα antagonist, two PPARγ agonists, two PPARγ antagonists, one TR agonist, and two TR antagonist assays (Excel Table S5). The CB/AC50 ratios across all the 12 assays are listed in Excel Table S6, and the distribution (BEQ%) of each target assay result is shown in Excel Table S7. We found that each end point had obviously different chemical toxicity prioritization and had its own dominant contributor(s). For different assays of the same receptor toxicity end point, the results varied widely due to the distinct compounds tested by the different assays. It was interesting to find that drugs or endogenous chemicals were dominant contributors with the top CB/AC50 ratios for most assays. For example, salidroside and N-vinyl-2-pyrrolidone were the most dominant contributors for Tox21_AR_LUC_MDAKB2_Agonist and TOX21_AR_LUC_MDAKB2_Antagonist_0.5 nM_R1881 assays, respectively, with the high CB/AC50 ratios of 2,288, and BEQ% of 24.0% and 37.0%, respectively (Excel Table S7; Figure S2). Salidroside is a major component of Rhodiola rosea, which has been used in traditional Chinese medicine39 and N-vinyl-2-pyrrolidone is used for treatment of infectious conjunctivitis.40 For the TOX21_PPARg_BLA_antagonist_viability assay, the top contributor was ribavirin (CB/AC50: 327, BEQ: 79.1%), followed by ramipril (36.6, 8.84%) and diphenoxylate hydrochloride (31.0, 7.49%). Drugs like acipimox, 5-methyl-1-phenyl-2(1H)-pyridone, piconol, triacetin, and hexylcaine hydrochloride were the most abundant chemicals, accounting for 9.17%, 19.9%, 11.5%, 10.7%, and 4.05% in TOX21_AR_BLA_Agonist_ratio, TOX21_ERa_BLA_Agonist_ratio, TOX21_ERa_BLA_Antagonist_ratio, TOX21_PPARg_BLA_Agonist_ch2, and TOX21_TR_LUC_GH3_Agonist assay, respectively.
Because the predicted CB in this study was based only on the internal CB generated by the external exposure, we further excluded endogenous chemicals and drugs, and performed the analysis on the remaining 4,893 chemicals. After excluding endogenous chemicals and drugs, methyl formate, di(2-methoxyethyl) phthalate, propylammonium nitrate, 2,3-butanedione, and (3,5-dimethyl-1H-pyrazol-1-yl)methanol became the most dominant chemicals in TOX21_AR_BLA_Antagonist_ratio, TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881, TOX21_PPARg_BLA_Antagonist_ch1, TOX21_PPARg_BLA_antagonist_viability, and TOX21_TR_LUC_GH3_Antagonist assay, with the BEQ% of 22.1%, 23.8%, 51.7%, 61.4%, and 46.3%, respectively (Excel Table S8). 2-Acetylpyrrole, thiamine thiozole, and aminopyridine a showed the highest CB/AC50 ratios of 549, 494, and 401, respectively (Excel Table S6), which was the dominant contributor with the BEQ% of 10.3%, 9.31%, and 7.56%, for the TOX21_AR_BLA_Agonist_ratio assay (Excel Table S8), suggesting that they had a relatively high potential risk of androgen disruption. In the Tox21_AR_LUC_MDAKB2_Agonist, the largest contributions were 3,3′-(ethylenedioxy)dipropiononitrile (CB/AC50 ratio: 275, BEQ%: 10.3%), 1,3-dichloropropanone (226, 8.42%), and MCPB (214, 7.97%). The dominant contributor of the TOX21_AR_BLA_Antagonist_ratio assay was methyl formate (1218, 22.1%), followed by 1-bromoheptadecane (644, 11.7%), and 1,2-dimethylhydrazine dihydrochloride (488, 8.83%). In the TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881 assay, di(2-methoxyethyl) phthalate (7.71, 23.8%), FD&C yellow 5 (7.50, 23.1%), and acetone (7.43, 22.9%) contributed the most.
In the TOX21_ERa_BLA_Agonist_ratio assay, the major contributions came from 1,1':4',1''-terphenyl (538, 22.8%), sodium nicotinate (379, 16.0%), and sodium 2,5-dimethylbenzenesulfonate (361, 15.3%). In the TOX21_ERa_BLA_Antagonist_ratio assay, 2-bromo-1-ethanol (550, 13.6%), benzyl nicotinate (379, 9.32%), and ethyl bromoacetate (224, 5.51%) were the dominant contributors. The dominant contributors were 1-bromopentadecane (328, 24.1%), beta-nitrostyrene (221, 16.2%), and (6Z)-non6-en-1-ol (216, 15.9%) for the Tox21_PPARg_BLA_Agonist_ratio assay; triacetin (1132, 18.2%), succinic anhydride (841, 17.2%), 2-(2-aminoethoxy)ethanol (680, 13.9%), and 2-pyrrolidinone (510, 10.5%) for the TOX21_PPARg_BLA_Agonist_ch2 assay; propylammonium nitrate (599, 51.7%), citronellol (246, 21.2%), geranyl formate (188, 16.3%), and isopentyl benzoate (39.5, 3.41%) for the TOX21_PPARg_BLA_Antagonist_ch1assay; and 2,3-butanedione (7.47, 61.4%), 3-acetyldihydro-2(3H)-furanone (2.61, 21.4%) and 3-mercaptopropyltrimethoxysilane (0.75, 6.15%) for the TOX21_PPARg_BLA_antagonist_viability assay.
The top contributions of the TOX21_TR_LUC_GH3_Agonist and assay were (4-methoxyphenyl)methanol (549, 6.20%), 2-butene-1,4-diol (528, 5.96%), 2,3-butanedione (523, 5.90%), and ethyl phthalyl ethyl glycolate (487, 5.50%). In the TOX21_TR_LUC_GH3_Antagonist assay, the dominant contributors were (3,5-dimethyl-1H-pyrazol-1-yl)methanol (638, 46.5%), phenethyl anthranilate (183, 13.4%), dimethyl isophthalate (89.5, 10.3%), and sodium 2-mercaptobenzothiolate (73.2, 6.91%).
We further retrieved the applications of the top 25 chemicals of each assay from the NCBI PubMed databases (https://pubmed.ncbi.nlm.nih.gov) (Excel Table S8), and we recalculated their BEQ% values after excluding drugs and endogenous substances: Food additives such as 2,3-butanedione, methyl formate, and FD&C Yellow 5 are used as flavoring agents or colorants, with the BEQ% values of 61.4%, 22.1%, and 23.1% in TOX21_PPARg_BLA_antagonist_viability, TOX21_AR_BLA_Antagonist_ratio, and TOX21_AR_LUC_MDAKB2_Antagonist_0.5nM_R1881 assay, respectively. Plasticizers such as dimethyl isophthalate (6.50%), diisobutyl phthalate (4.51%), and diethyl phthalate (4.16%), which are defined as U.S. Food and Drug Administration indirect additives used in food-contact substances, also showed significant activity after excluding drugs and endogenous substances in TOX21_TR_LUC_GH3_Antagonist, TOX21_PPARg_BLA_Agonist_ch2, and TOX21_AR_BLA_Antagonist_ratio assays, respectively. Chemicals such as propylammonium nitrate (51.7%) and (3,5-dimethyl-1H-pyrazol-1-yl)methanol (46.3%), used for solvents and cosmetic products, were the top contributors in TOX21_PPARg_BLA_Antagonist_ch1 and TOX21_TR_LUC_GH3_Antagonist assay, respectively.
Discussion
The framework described in this study provides several implications for HT chemical screening and prioritization. We used an HT machine learning algorithm for CB predictions with key parameters, including DE, δij, Vd, and t1/2. This HT HExpPredict approach can rapidly relate environmental chemical exposures to in vitro bioactivity, helping drive priorities based on risk potential.
Based on direct comparison of RMSE, MAE, MAPE, and R2 between models, we concluded that the RF model showed better performance than the other models. The ML model developed in this study was based on the physical and chemical properties and exposure of the chemicals. The input data of the ML models only included the key parameters DE, δij, Vd, and t1/2, and we used the ML models to combine these variables to make predictions without the other parameters, such as bioavailability and plasma protein binding data. We noted that only 10.3% and 15.9% of our evaluation chemicals were predicted to be over the 10-fold boundary for the RF training and testing sets, respectively, showing good predictive ability. To build this ML model, some well-performed predictive models including the IFS approach and SEEM3 were applied. Although these models were evaluated and tested, it is important to note that these prediction models can continue to be improved with the generation of more data, which could also improve our present ML model in the future.
Uncertainty in predicting CB can be accounted for in risk prioritization if the degree of uncertainty can be predicted for each chemical. According to the results of the three MC simulations, the prediction uncertainties of t1/2 and DE contributed approximate uncertainty to the ML prediction model. However, the uncertainty of the ML model was underestimated because of the lack of the Vd uncertainty. Although t1/2 and DE contributed approximately the same degree of uncertainty, some chemicals out of the model’s applicability domain, such as chemicals that contain silicon, were observed to have large standard errors in the prediction, which leads to high uncertainties for the t1/2.
The CB of phthalates such as di(2-methoxyethyl) phthalate, dipentyl phthalate, dipropyl phthalate, dihexyl phthalate, and bis(2-butoxyethyl) phthalate were predicted to be 7.86×10−3 (2.75×10−3–1.99×10−2), 3.08×10−3 (7.31×10−4–5.55×10−3), 1.93×10−3 (4.71×10−4–3.01×10−3), 1.50×10−3 (2.95×10−4–3.01×10−3) and 9.07×10−5 (2.97×10−5–3.18×10−4) μM, respectively. A phthalate metabolite such as monobutyl phthalate was predicted to be with the CB of 2.12×10−3 (6.64×10−4–3.82×10−3) μM [i.e., 0.47 (0.15–0.85) ng/mL], which was consistent with the concentration of 0.5 ng/mL observed in the previous study.41 Because the exposure of phthalates is usually characterized by monitoring the concentrations of their metabolites in the urine, our model can HT predict the CB of these easily metabolized substances, which is convenient for subsequent HT prioritization of their toxicity and risk. The CBs of bisphenol A (BPA) alternatives, such as bisphenol AF (BPAF), were predicted to be 0.020 (0.019–0.021) ng/mL, which was similar to the GM concentration of 0.01 ng/mL determined in the previous study.42 Perfluorinated compounds such as perfluorononanoic acid (PFNA) and perfluoroundecanoic acid (PFUnA) were predicted to have the CBs of 0.21 (0.19–0.52) and 0.17 (0.13–0.27) ng/mL, respectively, which were within the GM concentration ranges of 0.11–1.88 and 0.07–1.38 ng/mL, respectively, as observed in the general populations in 13 Chinese cities.43 However, for perfluorohexanoic acid (PFHxA), the predicted CB value (0.24; 95% CI: 0.21–0.26 ng/mL) was a little bit higher than the GM concentration range of 0.02–0.21 ng/mL of the 13 Chinese cities’ general populations.43 The estimated concentration can be very useful in the exposure or toxicity prioritization or even the mixture effect of blood exposome.31,44,45 In this study, the potential health effects and the causal compounds of ToxCast were summarized, revealing several key biomarker assays. A total of 12 ToxCast assays were used to assess the health effects of 4,893 chemicals, which showed different risk-based prioritization patterns. In addition to the top risk substances listed in the “Results” section, we found it interesting that some typical AR agonists, such as 2,3,7,8-Tetrachlorodibenzo-p-dioxin with the CB/AC50 ratio of 0.12 for Tox21_AR_LUC_MDAKB2_Agonist assay, also showed relative higher (97th of 4,893 chemicals) AR agonist activity owing to its extremely low AC50 (6.45×10−5 μM). In contrast, due to its low CB (7.91×10−6 μM), the BEQ% was only 0.0045%. Nonylparaben showed relatively strong AR antagonist activity, with the CB/AC50 ratio of 2.49 and BEQ% of 0.05% in the TOX21_AR_BLA_Agonist_ratio assay, and diethyl phthalate showed very strong AR antagonist activity, with the CB/AC50 ratio of 230 and BEQ% of 3.59% in the TOX21_AR_BLA_Antagonist_ratio assay.
Due to the very low AC50 values, some pesticides such as siduron and tribufos were observed to have relatively strong ER agonist activity in the TOX21_ERa_BLA_Agonist_ratio assay, with the CB/AC50 ratios of 15.1 and 5.92, and BEQ% of 30.49% and 0.19%, respectively, and benzyl nicotinate (379, 9.26%) and diallate (19.5, 0.48%) were found to have strong antagonist activity in the TOX21_ERa_BLA_Antagonist_ratio assay, with the CB/AC50 ratios of 379 and 19.5 and BEQ% of 9.26% and 0.48%, respectively. In the ER agonist and antagonist assays, phthalates, BPA, and BPA alternatives were negligible due to their relatively higher AC50. For example, the CB/AC50 ratios of BPA in the TOX21_ERa_BLA_Agonist_ratio assay and di(2-ethylhexyl) phthalate (DEHP) in the TOX21_ERa_BLA_Antagonist_ratio assay were only 4.22×10−3 and 5.07×10−4, respectively, due to their higher AC50 of 0.96 and 6.46μM, respectively, although they had relative high CB values of 4.06×10−3 and 3.27×10−3 μM, respectively. For the organophosphate compounds, the CB/AC50 ratios of dibenzyl phosphate and triisobutyl phosphate were 202 and 13.6, respectively, in the TOX21_TR_LUC_GH3_Agonist assay, showing strong TR agonist activity. Triisobutyl phosphate also showed strong PPAR agonist activity, with the CB/AC50 ratio of 41.2 in the Tox21_PPARg_BLA_Agonist_ratio assay.
An interesting finding was that when drugs and endogenous substances were excluded, food additives were the major contributors of BEQ% to the majority of assays (Figure 4; Figure S3) due to high predicted exposure by SEEM3. Food additives such as 2,3-butanedione, methyl formate, FD&C Yellow 5, and succinic anhydride showed a high potential receptor activity in AR or PPARγ (Figure 4). However, these substances are not typical pollutants, and there are little data on their biomonitoring in humans, raising concerns about their potential health risks. U.S. FDA indirect additives used in food-packing materials, such as dimethyl isophthalate, diisobutyl phthalate, and diethyl phthalate, also showed modest potential receptor activity in TR, PPARγ, and AR. The health risk of food additives and indirect food additives should be studied further. It should be noted that, besides nuclear receptors, adverse outcome pathways (AOP) (https://aopkb.oecd.org) with more toxicological end points could be further considered in future risk prioritization.
Figure 4. Toxicity contributions (percentage) of ToxCast chemicals (excluding the drugs and endogenous compounds) in assays of AR and PPARγ as examples (referring to the data in Excel Table S8). Note: AR, androgen receptor; FA, food additive; FIA, food indirect additive; PPARγ, peroxisome proliferator–activated receptor.
Figure 4 is a set of four pie charts. On the top-left, the pie chart is titled T O X 21 underscore A R underscore B L A underscore antagonist underscore ratio displays the following information: 3.3 percent diethyl maleate, food additive; 4.2 percent diethyl phthalate, food indirect additive; 4.3 percent 2, 3-dimethylimidazolium hexafluorophosphate, 4.5 percent ethyl hexadecanoate, food additive; 4.8 percent 3-acetyl-2,5-dimethylfuran, food addictive; 6.1 percent Bis(1-methylethyl) methylphosphonate; 22.1 percent methyl formate, food additive; 11.7 percent -bromoheptadecane; 8.8 percent 1,2-dimethylhydrazine dihydrochloride; 8 percent n,n-diisopropyl ethanolamine. At the bottom-left, the pie chart is titled T O X 21 underscore P P A Rg underscore B L A underscore antagonist underscore viability displays the following information: 3.4 percent 2-penty-2-cyclopenten-1-one; 6.2 percent 3-mercaptopropyltrimethoxysilane; 21.5 percent 3-acetyldihydro-2(3 H)-furanone; 61.5 percent 2,3-butanedione, food additive. On the top-right, the pie chart is titled T O X 21 underscore A R underscore L U C underscore M D A K B 2 underscore antagonist displays the following information: 6.4 percent diallyl disulfide, food additive; 19.1 percent acetic acid, mercapto-,ethylester, 22.9 percent acetone, 23.8 percent di(2-methoxyethyl) phthalate, 23.2 percent F D and C yellow 5, food additive. At the bottom-right, the pie chart is titled T O X 21 underscore underscore P P A Rg underscore B L A underscore agonist underscore ch2 displays the following information: 3.2 percent 2-butenoic acid, food additive; 4 percent R-(plus)-pulegone, food additive; 4.1 percent 2-phenylmercaptoethanol; 4.1 percent (6 Z)-Non-6-en-l-ol, food additive; 4.5 percent diisobutyl phthalate, food indirect additive; 17.2 percent succinic anhydride, food additive; 13.9 percent 2-(2-aminoethoxy)ethanol; 10.5 percent 2-pyrolidinone; 5 percent (Z)-5-octen-l-ol, food additive, and 4.9 percent 1,2-benzendicarbonitrile, food additive.
This study has several limitations. First, we could not predict the t1/2 of chemicals with a metal atom or molecular weight over 1,000 using the IFS approach. In addition, some chemicals had extreme properties that were out of the model’s applicability domain, such as silicon-containing chemicals. These chemicals were observed with large standard errors higher than the predicted mean. As far as we are concerned, no computational model can handle silicon-containing molecules at this point. Second, the ExpoCast database was unable to cover all the ToxCast compounds, and ExpoCast merely represents the exposure of typical Americans for their historical exposure. Because the amount of chemicals used varies with the year, the variation of chemical exposure and the year of blood collection has a certain impact on the predicted results. Our prediction should be periodically updated to incorporate new estimated exposure and measured CBs of chemicals in the future. Third, models based on subsets of measured data for chemical groups were not considered in the prediction model due to limited measured data. In the future, more accurate prediction models based on different chemical subsets could be built if we can collect sufficient data as a training set. In addition, we regarded the concentrations of blood, plasma, and serum as CBs and did not consider parameters such as plasma protein binding. Nonetheless, the predicted CBs in this study can still contribute to the concentrations’ ranking of substances in human blood and the prioritization of potential biological effects. An accurate PBPK model could be combined with the CB prediction model of this study in the future to predict concentrations in other organs, and animal experiments for validation of the model should be made in the future as well. Fourth, although the AC50 value has become a standard way to compare potencies of chemicals in in vitro pharmacology and toxicology studies, it may not be the best metric for prioritization or estimating toxicological risk based on well-designed in vitro tests. Fifth, the mode of toxic action (MOA), which was not considered in our prediction model, is related to the CB and metabolism of the chemical. The MOA could be considered in future work to refine the model. Last, the present prioritization results based on ToxCast data have limitations in predicting the toxicities of the chemicals due to the limited assays adopted by the ToxCast exercise, and different chemicals were tested in different assays. Therefore, it is still impossible to systematically evaluate the contribution of one chemical in different toxicological end points.
In conclusion, we curated the CBs of 216 compounds and developed ML algorithms for CB prediction, and our work improved HT risk prioritization for large numbers of environmental chemicals. Many of the high-risk chemicals in some assays were also unexpected. This study has implications for current efforts to overhaul existing chemical testing methods to address the disparity in the number of tested and untested chemicals. By using the HT method, chemicals could be screened in a cost-effective and efficient manner, which provides a better basis for informed decisions on chemical testing priorities and regulatory attention.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was funded by the National Key R&D Program (No. 2022YFC3702600 and 2022YFC3702601), the Singapore Ministry of Education Academic Research Fund Tier 1 (04MNP000567C120), and the Startup Grant of Fudan University (No. JIH 1829010Y).
In addition, to improve the applicability of our model, the R scripts are also provided at https://github.com/FangLabNTU/HExpPredict.
==== Refs
References
1. Shin H-M, Ernstoff A, Arnot JA, Wetmore BA, Csiszar SA, Fantke P, et al. 2015. Risk-based high-throughput chemical screening and prioritization using exposure models and in vitro bioactivity assays. Environ Sci Technol 49 (11 ):6760–6771, PMID: , 10.1021/acs.est.5b00498.25932772
2. Li L, Sangion A, Wania F, Armitage JM, Toose L, Hughes L, et al. 2021. Development and evaluation of a holistic and mechanistic modeling framework for chemical emissions, fate, exposure, and risk. Environ Health Perspect 129 (12 ):127006, PMID: , 10.1289/EHP9372.34882502
3. Ring CL, Arnot JA, Bennett DH, Egeghy PP, Fantke P, Huang L, et al. 2019. Consensus modeling of median chemical intake for the U.S. population based on predictions of exposure pathways. Environ Sci Technol 53 (2 ):719–732, PMID: , 10.1021/acs.est.8b04056.30516957
4. Wambaugh JF, Bare JC, Carignan CC, Dionisio KL, Dodson RE, Jolliet O, et al. 2019. New approach methodologies for exposure science. Curr Opin Toxicol 15 :76–92, 10.1016/j.cotox.2019.07.001.
5. Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ. 2007. The ToxCast program for prioritizing toxicity testing of environmental chemicals. Toxicol Sci 95 (1 ):5–12, PMID: , 10.1093/toxsci/kfl103.16963515
6. Honda GS, Pearce RG, Pham LL, Setzer RW, Wetmore BA, Sipes NS, et al. 2019. Using the concordance of in vitro and in vivo data to evaluate extrapolation assumptions. PLoS One 14 (5 ):e0217564, PMID: , 10.1371/journal.pone.0217564.31136631
7. Blaauboer BJ. 2010. Biokinetic modeling and in vitro-in vivo extrapolations. J Toxicol Environ Health B Crit Rev 13 (2–4 ):242–252, PMID: , 10.1080/10937404.2010.483940.20574900
8. Wambaugh JF, Wetmore BA, Pearce R, Strope C, Goldsmith R, Sluka JP, et al. 2015. Toxicokinetic triage for environmental chemicals. Toxicol Sci 147 (1 ):55–67, PMID: , 10.1093/toxsci/kfv118.26085347
9. David A, Chaker J, Price EJ, Bessonneau V, Chetwynd AJ, Vitale CM, et al. 2021. Towards a comprehensive characterisation of the human internal chemical exposome: challenges and perspectives. Environ Int 156 :106630, PMID: , 10.1016/j.envint.2021.106630.34004450
10. Rappaport SM, Barupal DK, Wishart D, Vineis P, Scalbert A. 2014. The blood exposome and its role in discovering causes of disease. Environ Health Perspect 122 (8 ):769–774, PMID: , 10.1289/ehp.1308015.24659601
11. Jia S, Xu T, Huan T, Chong M, Liu M, Fang W, et al. 2019. Chemical isotope labeling exposome (CIL-EXPOSOME): one high-throughput platform for human urinary global exposome characterization. Environ Sci Technol 53 (9 ):5445–5453, PMID: , 10.1021/acs.est.9b00285.30943026
12. Zhao F, Kang Q, Zhang X, Liu J, Hu J. 2019. Urinary biomarkers for assessment of human exposure to monomeric aryl phosphate flame retardants. Environ Int 124 :259–264, PMID: , 10.1016/j.envint.2019.01.022.30660026
13. Sipes NS, Wambaugh JF, Pearce R, Auerbach SS, Wetmore BA, Hsieh J-H, et al. 2017. An intuitive approach for predicting potential human health risk with the Tox21 10k library. Environ Sci Technol 51 (18 ):10786–10796, PMID: , 10.1021/acs.est.7b00650.28809115
14. Armitage JM, Hughes L, Sangion A, Arnot JA. 2021. Development and intercomparison of single and multicompartment physiologically-based toxicokinetic models: implications for model selection and tiered modeling frameworks. Environ Int 154 :106557, PMID: , 10.1016/j.envint.2021.106557.33892222
15. Wetmore BA, Wambaugh JF, Allen B, Ferguson SS, Sochaski MA, Setzer RW, et al. 2015. Incorporating high-throughput exposure predictions with dosimetry-adjusted in vitro bioactivity to inform chemical toxicity testing. Toxicol Sci 148 (1 ):121–136, PMID: , 10.1093/toxsci/kfv171.26251325
16. Wetmore BA, Wambaugh JF, Ferguson SS, Sochaski MA, Rotroff DM, Freeman K, et al. 2012. Integration of dosimetry, exposure, and high-throughput screening data in chemical toxicity assessment. Toxicol Sci 125 (1 ):157–174, PMID: , 10.1093/toxsci/kfr254.21948869
17. Wadhwa R, Cascella M. Steady State Concentration. Treasure Island, FL: StatPearls Publishing. https://www.ncbi.nlm.nih.gov/books/NBK553132/ [accessed 12 December 2022].
18. U.S. EPA (U.S. Environmental Protection Agency). Previously Published ToxCast Data. Updated data released October 2018. https://epa.figshare.com/articles/dataset/Previously_Published_ToxCast_Data/6062551/3 [accessed 4 March 2023].
19. Williams AJ, Grulke CM, Edwards J, McEachran AD, Mansouri K, Baker NC, et al. 2017. The CompTox Chemistry Dashboard: a community data resource for environmental chemistry. J Cheminform 9 (1 ):61, PMID: , 10.1186/s13321-017-0247-6.29185060
20. Wambaugh JF, Wetmore BA, Ring CL, Nicolas CI, Pearce RG, Honda GS, et al. 2019. Assessing toxicokinetic uncertainty and variability in risk prioritization. Toxicol Sci 172 (2 ):235–251, PMID: , 10.1093/toxsci/kfz205.31532498
21. Zhao F, Li L, Chen Y, Huang Y, Keerthisinghe TP, Chow A, et al. 2021. Risk-Based chemical ranking and generating a prioritized human exposome database. Environ Health Perspect 129 (4 ):47014, PMID: , 10.1289/EHP7722.33929905
22. Arnot JA, Brown TN, Wania F. 2014. Estimating screening-level organic chemical half-lives in humans. Environ Sci Technol 48 (1 ):723–730, PMID: , 10.1021/es4029414.24298879
23. Li L, Westgate JN, Hughes L, Zhang X, Givehchi B, Toose L, et al. 2018. A model for risk-based screening and prioritization of human exposure to chemicals from near-field sources. Environ Sci Technol 52 (24 ):14235–14244, PMID: , 10.1021/acs.est.8b04059.30407800
24. Ulrich N, Endo S, Brown TN, Watanabe N, Bronner G, Abraham MH, et al. 2017. UFZ-LSER database v 3.2.1. Leipzig, Germany: Helmholtz Centre for Environmental Research-UFZ.
25. U.S. EPA. 2022 . Toxicity Estimation Software Tool (TEST). https://www.epa.gov/chemical-research/toxicity-estimation-software-tool-test [accessed 4 March 2023].
26. U.S. CDC (U.S. Centers for Disease Control and Prevention). 2022. National Report on Human Exposure to Environmental Chemicals. https://www.cdc.gov/exposurereport/ [accessed 8 December 2022].
27. Office of Environmental Health Hazard Assessment, California Department of Public Health. 2020. Explore Results: Biomonitoring California’s Results Database. https://biomonitoring.ca.gov/results/explore [accessed 5 December 2021].
28. Neveu V, Moussy A, Rouaix H, Wedekind R, Pon A, Knox C, et al. 2017. Exposome-Explorer: a manually-curated database on biomarkers of exposure to dietary and environmental factors. Nucleic Acids Res 45 (D1 ):D979–D984, PMID: , 10.1093/nar/gkw980.27924041
29. Dong T, Zhang Y, Jia S, Shang H, Fang W, Chen D, et al. 2019. Human indoor exposome of chemicals in dust and risk prioritization using EPA’s ToxCast database. Environ Sci Technol 53 (12 ):7045–7054, PMID: , 10.1021/acs.est.9b00280.31081622
30. Jia S, Sankaran G, Wang B, Shang H, Tan ST, Yap HM, et al. 2019. Exposure and risk assessment of volatile organic compounds and airborne phthalates in Singapore’s child care centers. Chemosphere 224 :85–92, PMID: , 10.1016/j.chemosphere.2019.02.120.30818198
31. Zhang Y, Liu M, Peng B, Jia S, Koh D, Wang Y, et al. 2020. Impact of mixture effects between emerging organic contaminants on cytotoxicity: a systems biological understanding of synergism between tris(1,3-dichloro-2-propyl)phosphate and triphenyl phosphate. Environ Sci Technol 54 (17 ):10722–10734, PMID: , 10.1021/acs.est.0c02188.32786581
32. U.S. EPA. 2022. CompTox Chemicals Dashboard (Version 2.1) Assay Endpoints List. https://comptox.epa.gov/dashboard/assay-endpoints?filtered= [accessed 8 December 2022].
33. Glynn A, Aune M, Darnerud PO, Cnattingius S, Bjerselius R, Becker W, et al. 2007. Determinants of serum concentrations of organochlorine compounds in Swedish pregnant women: a cross-sectional study. Environ Health 6 :2, PMID: , 10.1186/1476-069X-6-2.17266775
34. Koukoulakis KG, Kanellopoulos PG, Chrysochou E, Koukoulas V, Minaidis M, Maropoulos G, et al. 2020. Leukemia and PAHs levels in human blood serum: preliminary results from an adult cohort in Greece. Atmos Pollut Res 11 (9 ):1552–1565, 10.1016/j.apr.2020.06.018.
35. Mathur H, Agarwal H, Johnson S, Saikia N. 2005. Analysis of Pesticide Residues in Blood Samples from Villages of Punjab. CSE/PML/PR-21, India, 1–15.
36. Mochalski P, King J, Klieber M, Unterkofler K, Hinterhuber H, Baumann M, et al. 2013. Blood and breath levels of selected volatile organic compounds in healthy volunteers. Analyst 138 (7 ):2134–2145, PMID: , 10.1039/c3an36756h.23435188
37. Zhao F, Wan Y, Zhao H, Hu W, Mu D, Webster TF, et al. 2016. Levels of blood organophosphorus flame retardants and association with changes in human sphingolipid homeostasis. Environ Sci Technol 50 (16 ):8896–8903, PMID: , 10.1021/acs.est.6b02474.27434659
38. U.S. EPA. 2022. CompTox Chemicals Dashboard (Version 2.1.1) Batch Search. https://comptox.epa.gov/dashboard/batch-search [accessed 4 March 2023].
39. Zhang X, Xie L, Long J, Xie Q, Zheng Y, Liu K, et al. 2021. Salidroside: a review of its recent advances in synthetic pathways and pharmacological properties. Chem Biol Interact 339 :109268, PMID: , 10.1016/j.cbi.2020.109268.33617801
40. National Library of Medicine, PubChem. PubChem Compound Summary for CID 6917, N-Vinyl-2-pyrrolidone. https://pubchem.ncbi.nlm.nih.gov/compound/N-Vinyl-2-pyrrolidone [accessed 22 January 2022].
41. Wang Y, Zhu H, Kannan K. 2019. A review of biomonitoring of phthalate exposures. Toxics 7 (2 ):21, PMID: , 10.3390/toxics7020021.30959800
42. Zhang B, He Y, Zhu H, Huang X, Bai X, Kannan K, et al. 2020. Concentrations of bisphenol A and its alternatives in paired maternal-fetal urine, serum and amniotic fluid from an e-waste dismantling area in China. Environ Int 136 :105407, PMID: , 10.1016/j.envint.2019.105407.31955035
43. Zhang S, Kang Q, Peng H, Ding M, Zhao F, Zhou Y, et al. 2019. Relationship between perfluorooctanoate and perfluorooctane sulfonate blood concentrations in the general population and routine drinking water exposure. Environ Int 126 :54–60, PMID: , 10.1016/j.envint.2019.02.009.30776750
44. Liu M, Jia S, Dong T, Zhao F, Xu T, Yang Q, et al. 2020. Metabolomic and transcriptomic analysis of MCF-7 cells exposed to 23 chemicals at human-relevant levels: estimation of individual chemical contribution to effects. Environ Health Perspect 128 (12 ):127008, PMID: , 10.1289/EHP6641.33325755
45. Fang M, Hu L, Chen D, Guo Y, Liu J, Lan C, et al. 2021. Exposome in human health: utopia or wonderland? Innovation (Camb) 2 (4 ):100172, PMID: , 10.1016/j.xinn.2021.100172.34746906
|
PMC010xxxxxx/PMC10010395.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36913237
EHP10710
10.1289/EHP10710
Research
Association of Combined Exposure to Ambient Air Pollutants, Genetic Risk, and Incident Rheumatoid Arthritis: A Prospective Cohort Study in the UK Biobank
Zhang Jie 1 2 *
Fang Xin-Yu 1 2 *
Wu Jun 3
Fan Yin-Guang 1 2
Leng Rui-Xue 1 2
Liu Bo 1 2
Lv Xiao-Jie 1 2
Yan Yu-Lu 1 2
Mao Chen 4
https://orcid.org/0000-0001-6604-9614
Ye Dong-Qing 1 2
1 Department of Epidemiology and Biostatistics, School of Public Health, Anhui Medical University, Hefei, Anhui, China
2 Inflammation and Immune Mediated Diseases Laboratory of Anhui Province, Hefei, Anhui, China
3 Department of Epidemiology and Biostatistics, School of Public Health, Guangdong Medical University, Dongguan, Guangdong, China
4 Department of Epidemiology, School of Public Health, Southern Medical University, Guangzhou, Guangdong, China
Address correspondence to Dong-Qing Ye, Department of Epidemiology and Biostatistics, School of Public Health, Anhui Medical University, Hefei, China. Email: [email protected]. And, Chen Mao, Department of Epidemiology, School of Public Health, Southern Medical University, Guangzhou, China. Email: [email protected]
13 3 2023
3 2023
131 3 03700801 12 2021
03 2 2023
06 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Evidence for a potential link between air pollution and rheumatoid arthritis (RA) is inconsistent, and the modified effect of genetic susceptibility on the relationship between air pollution and RA has not been well studied.
Objective:
Using a general population cohort from the UK Biobank, this study aimed to investigate the associations between various air pollutants and the risk of incident RA and to further estimate the impact of combined exposure to ambient air pollutants on the risk of developing RA under the modification effect of genetic predisposition.
Methods:
A total of 342,973 participants with completed genotyping data and who were free of RA at baseline were included in the study. An air pollution score was constructed by summing the concentrations of each pollutant weighted by the regression coefficients with RA from single-pollutant models to assess the combined effect of air pollutants, including particulate matter (PM) with diameters ≤2.5μm (PM2.5), between 2.5 and 10μm (PM2.5–10), and ≤10μm (PM10), as well as nitrogen dioxide (NO2) and nitrogen oxides (NOx). In addition, the polygenic risk score (PRS) of RA was calculated to characterize individual genetic risk. The Cox proportional hazard model was used to estimate hazard ratios (HRs) and 95% confidence intervals (95% CIs) of associations of single air pollutant, air pollution score, or PRS with incident RA.
Results:
During a median follow-up time of 8.1 y, 2,034 incident events of RA were recorded. The HRs (95% CIs) of incident RA per interquartile range increment in PM2.5, PM2.5–10, PM10, NO2, and NOx were 1.07 (1.01, 1.13), 1.00 (0.96, 1.04), 1.01 (0.96, 1.07), 1.03 (0.98, 1.09), and 1.07 (1.02, 1.12), respectively. We also found a positive exposure–response relationship between air pollution score and RA risk (pTrend=0.000053). The HR (95% CI) of incident RA was 1.14 (1.00, 1.29) in the highest quartile group compared with the lowest quartile group of the air pollution score. Furthermore, the results of the combined effect of air pollution score and PRS on the RA risk showed that the risk of RA incidence in the highest genetic risk and air pollution score group was almost twice that of the lowest genetic risk and air pollution score group [incidence rate (IR) per 100,000 person-years: 98.46 vs. 51.19, and HR= 1.73 (95% CI: 1.39, 2.17) vs. 1 (reference)], although no statistically significant interaction between the air pollution and genetic risk for incident RA was found (pInteraction>0.05).
Discussion:
The results revealed that long-term combined exposure to ambient air pollutants might increase the risk of RA, particularly in those with high genetic risk. https://doi.org/10.1289/EHP10710
Supplemental Material is available online (https://doi.org/10.1289/EHP10710).
* These authors contributed equally to this work.
All authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Rheumatoid arthritis (RA) is a chronic systematic autoimmune disorder characterized by progressive joint erosion that leads to severe disability.1 As one of the most prevalent chronic inflammatory diseases, it affects ∼1% of the world’s adult population.2,3 Further, RA disease burden experienced an unexpected steep rise from 2012 to 2017, reaching an age-standardized disability-adjusted life years (DALY) rate of 43.3 (95% Uncertainty Interval: 33.0 to 54.5) per 100,000 population.4 Despite extensive studies on the exact etiology of RA, it remains unknown but is assumed to be multifactorial, involving both genetic and environmental factors.5,6
Currently, air pollution is nominated by the World Health Organization as one of the most significant health threats. It is well established that exposure to smoking7 or silica exposure,8 which causes an inflammatory and oxidative stress response, can increase the risk of RA. In addition, a previous study9 also revealed that the lung may be the site of early related autoimmune injury in RA. Exposure to air pollution has been demonstrated to disrupt oxidation–reduction homeostasis in respiratory mucosal and triggers pro-inflammatory immune responses across multiple immune cells,10,11 indicating that air pollution may be a potential risk factor for RA.12 Several studies have focused on the relationship between air pollution and RA12–17; however, results were conflicting. The Nurses’ Health Study (NHS) examined the association between distance to the nearest major road, a proxy marker of traffic pollution exposure, and the incidence of RA in 90,297 females, and the findings indicated that higher exposure to traffic pollution may be associated with RA risk.13 Similarly, a retrospective cohort study in Taiwan, China,14 found that newly diagnosed RA was significantly associated with NO2 exposure and a study in South Korea15 also reported that the incidence rate (IR) of RA was positively correlated with the concentration of particulate matter (PM) with an aerodynamic diameter of ≤2.5μm (PM2.5). In contrast, a case–control study based on the Swedish Epidemiological Investigation of Rheumatoid Arthritis (EIRA) has shown that after adjusting for the confounding factors of education and smoking status, the associations between particulate pollutants [PM with an aerodynamic diameter of ≤10μm (PM10)], gaseous pollutants [nitrogen dioxide (NO2) and sulfur dioxide (SO2)], and incident RA were not statistically significant.16 In addition, air pollutants, including PM2.5, PM10, NO2, and SO2, were associated with RA were also not observed, neither in the NHS study17 nor in the British Colombian study.18 The inconsistency among these findings implies that air pollution needs to be further investigated to evaluate whether it is a potential determinant of RA.
Genetic determinants provide initial insights into the presence of systemic autoimmunity and the identification of potentially at-risk individuals in the pre-RA stage.19 In recent years, genome-wide association studies (GWAS) have identified several nonhuman leukocyte antigen (non-HLA) risk loci, and >100 genetic loci have been confirmed to be associated with the risk of RA.20 Although each variant accounts for only a small-to-moderate proportion of the genetic risk of RA, using polygenic risk scores (PRSs) has proven to be an effective method for measuring the cumulative effects of multiple risk-related variants.21,22 Moreover, quantifying the joint effects of genetic risk and epidemiological factors can greatly improve risk stratification or explore potential environment–gene interactions, thereby providing new insight for precise prediction and targeted interventions in RA. For instance, a previous study used a combination of 39 independent RA risk alleles to establish a PRS, which, combined with family history and epidemiological risk factors, was used to develop a well-performed risk prediction model for seropositive and seronegative RA.23
Previous studies have explored only the relationship between ambient air pollutants and RA risk and have largely ignored the modification effects of genetic susceptibility. Therefore, based on a general population cohort from the UK Biobank, the present study aimed to assess whether combined exposure to air pollution and genetic factors contributes to the incidence of RA.
Methods
Study Population
The UK Biobank resource includes ∼0.5 million UK participants 39–73 years of age during the period of recruitment between 2006 and 2010. Participants attended one of the 22 assessment centers across England, Scotland, and Wales, where they completed touchscreen and nurse-led questionnaires, underwent physical measurements, and provided biological samples. More details of the study design have been described elsewhere.24 The UK Biobank study was conducted under the approval of the North West Multi-center Research Ethical Committee (11/NW/0382). All participants provided written informed consent.
In the present study, the baseline time was defined as the time when participants first attended the assessment center, between 2006 and 2010. As genetic quality control (QC) measures, we excluded participants with sex mismatch, heterozygosity rate outliers, missing genotypes, excess relatives, and non-White race/ethnicity (n=109,499). The White participants included those whose self-reported ancestry was White British (based on UK Biobank Data-Field 1657 “self-reported ethnic group”) and was also further confirmed as of Caucasian ancestry (based on UK Biobank Data-Field 22006 “genetic ethnic group”). Non-White participants were excluded under the consideration that the RA GWAS is mainly of European ancestry,20 and the proportion of non-White participants in the UK Biobank is <5%25; thus, there may not be enough incident RA cases for the statistical analyses for the non-White population. Further, participants with prevalent RA (n=2,936) and those with incomplete information on residential air pollution (n=32,001) at baseline were also excluded. A total of 342,973 participants who had complete data for the concentration of five air pollutants at baseline and genotyping were included in the final analysis. A flowchart of the study participants selection is shown in Figure S1.
Ascertainment of Outcomes
At the baseline (2006–2010), we combined the self-reported and related therapeutic drugs, including steroids, synthetic disease-modifying anti-rheumatic drugs (DMARDs), and biologic DMARDS, the use of which represents outpatient visits, as well as the hospital inpatient records [using the International Classification of Diseases, Manual of the International Statistical Classification of Diseases, Injuries and Causes of Death (ICD-9,26 codes 71400, 71401, 71403, 71404, 71405, 71406, and 71409) or the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10,27 codes M05 and M06) to identify prevalent cases of RA.] The self-reported RA and records of those who had used related therapeutic drugs for RA were collected through verbal interviews with well-trained nurses, the specific information about RA in verbal interviews was entered as free text and subsequently used as unique coded data. Hospital inpatient records are directly linked to the Health Episode Statistics in England and Wales and the Scottish Morbidity Records in Scotland, allowing for accurate identification of the first recorded date of each diagnosis. After removing cases of RA at baseline, incident RA events were identified during the subsequent following-up from the admission data using ICD-9 or ICD-10 codes. Detailed information on codes used to identify RA cases in this study can be found in Table S1.
Estimation of Air Pollutants and Air Pollution Score
Estimates of ambient air pollutants, including PM2.5, PM2.5–10, PM10, NO2, and NOx, were collected by the UK Biobank between 2005–2007 and 2010. Data on ambient air pollution for the 2005–2007 period were derived from EU-wide air pollution maps (resolution 100×100m), and the UK Biobank overlaid the coordinates of each subject’s residential address onto these maps to obtain the corresponding air pollution concentrations of 100×100-m-grid cells.28 Meanwhile, the 2010 annual average air pollution concentration was calculated by using a land use regression (LUR) model that was combined with the participants’ residential addresses given at the baseline visit and the monitoring data from the European Study of Cohorts for Air Pollution Effects (ESCAPE) from 26 January 2010 to 18 January 2011. The LUR model was developed as part of the ESCAPE project, and the validation of the models has been described elsewhere (http://www.escapeproject.eu/).29
Annual concentration data for PM2.5, PM2.5–10, and NOx were only available for 2010, whereas the concentration data for NO2 was available for 2005–2007 and 2010. Finally, data for PM10 were available for 2007 and 2010. Pollutants for which only the annual concentration data in 2010 was available were directly defined as their respective exposure variable at baseline. Meanwhile, for pollutants for which several years of annual concentration data within the baseline time range were available, we took the average value of multiple annual concentrations as their exposure variable. For example, for PM10, we considered the average value of the annual concentrations in 2007 and 2010 as its exposure level. Furthermore, the air pollution score was constructed by weighted summing concentrations of the five air pollutants, and the weighted method was based on the multivariable-adjusted risk estimates (β coefficients) of RA. The use of the air pollution score has been well accepted in previous UK Biobank studies to assess the association of combined exposure to multiple air pollutants with the risk of chronic diseases, such as type 2 diabetes30 and heart failure.31 The air pollution score was obtained using the following formula: Air pollution score=[β(PM2.5)×PM2.5+β(PM2.5−10)×PM2.5–10+β(PM10)×PM10+β(NO2)×NO2+β(NOx)×NOx]×(5/∑β)
Genotype Data, QC, and PRS
Genotype calling, QC, phasing, and imputation were performed centrally and have been previously described in detail.25 In brief, ∼50,000 and 450,000 participants were separately genotyped on two closely related purpose-designed arrays (UK BiLEVE Axiom and UK Biobank Axiom), and 805,426 markers were identified in the finally released genotype data. In addition, the data set was staged using computationally efficient methods and combined with the Haplotype Reference Consortium and UK10K Haplotype Resources to impute a total of 96 million genotypes. Based on the above genotyping imputed data, we carried out downstream QC measures; specifically, we removed single nucleotide polymorphisms (SNPs) with imputation information (INFO) score of <0.3, with minor allele frequency of <0.5%, or those that failed Hardy-Weinberg tests with a p<1×10−6 using QCTOOL (version 2; https://www.well.ox.ac.uk/∼gav/qctool/index.html).
A PRS, which captures an individual’s load of common genetic variants associated with RA risk, was constructed. The score was based on RA summary statistics from a meta-analysis of GWAS data from individuals of European ancestry (http://plaza.umin.ac.jp/∼yokada/datasource/software.htm).32 We applied the clumping and threshold (C+T) method for calculating the PRS of RA, which involved computing PRSs based on a subset of partially independent (clumped) SNPs exceeding a specific GWAS association p-value threshold.21 To select the SNPs that would be included in the calculation of the genetic risk score, we first filtered the GWAS statistics results to exclude variants with a p≥5×10−8 (genome-wide significant). Linkage-disequilibrium clumping was performed to identify independently associated variants (r2<0.01). If the selected variants were not present in the UK Biobank genotyping data, proxy variants were sought (r2>0.8). Moreover, SNPs were also filtered based on the INFO score (INFO score >0.95) to ensure imputed genotyping quality.
In this study, the additive genetic model,33 including 154 SNPs (see Excel Table S1 for details), was used for PRS calculation, and the final PRS was standardized. The calculation formula is as follows: PRSj =[∑ijSi×Gij−mean(PRS)]/SD(PRS),
where S is the summary statistic for the effective allele; G is the number of the effective allele (0, 1, 2) observed; i is the ith SNP; j is the jth individual; and SD is the standard deviation. The above procedure was performed in PRSice-234 with the RA GWAS summary statistics and the genotyping imputed data from the UK Biobank. According to the PRS distribution, individuals were categorized into low (tertile 1)-, intermediate (tertile 2)-, and high (tertile 3)-grade RA genetic risks.
Covariate Measurements
Covariates in this study included age (years, continuous), sex (female, male), assessment center (22 centers), average total household income before tax (<£18,000, £18,000–£29,999, £30,000–£51,999, £52,000–£100,000, >£100,000), educational level [college or university degree, others (including advanced levels (A levels)/(advanced subsidiary levels (AS levels) or equivalent, ordinary levels (O levels)/general certificate of secondary education (GCSEs) or equivalent, CSEs or equivalent, national vocational qualifications (NVQ) or higher national diplomas (HND) or higher national certificates (HNC) or equivalent, and other professional qualifications, such as nursing and teaching)], smoking status (current, previous, never), alcohol consumption (standard-drinks per day, continuous), sedentary activity time (hours per day), physical activity duration (minutes per day), body mass index (BMI, in kilograms per meter squared), and healthy diet score (0–5 points). Some general characteristics (including date of birth, sex, and the name of the recruitment center) of participants were known before arrival at the assessment center, and all were obtained from local NHS Primary Care Trust registries. Sociodemographics (average total household income before tax, educational level) and lifestyle and behaviors data (smoking status, alcohol consumption, sedentary activity time, and physical activity duration) were collected from touchscreen questionnaires at the baseline assessment center visit. The participants’ height and weight were measured by trained nurses, and BMI was determined by dividing weight (in kilograms) by height (in meters) squared. Sedentary activity time was obtained from the sum of hours per day spent driving, watching TV, and using a computer. Physical activity duration was defined as the sum of minutes per day spent walking and engaging in moderate and vigorous activity. The healthy diet score was based on the American Heart Association Guidelines35 and included five dietary components: vegetables, fruit, fish, processed meat, and unprocessed red meat. The total diet score ranged from 0 to 5 points; each time a dietary component intake goal was achieved, it was given 1 point, with a higher score representing a healthier diet. Alcohol consumption was calculated by conversion of alcohol intake collected from a touchscreen questionnaire to standard-drink (8g pure alcohol) of alcohol per day.36 More information is shown in Table S2.
Statistical Analysis
The follow-up period was defined as from the date of initial recruitment to the onset of RA or competitive events (death), the date of loss to follow-up, or the date of the current end of follow-up (31 March 2017 for England, 31 October 2016 for Scotland, and 31 January 2017 for Wales). Participants who were lost to follow-up or died before RA occurred were censored at the time of the respective event.
Hazard ratios (HRs) and 95% confidence intervals (CIs) for incident RA associated with the single air pollutant, the air pollution score, or PRS were estimated with Cox proportional hazard models. We evaluated the assumption of the proportionality of hazards by examining the association between standardized Schoenfeld residuals and time.37 Missing values of covariates were imputed by a multiple chain equation,38 and the missing value patterns were assumed to be randomly missing. The missing data in the continuous variable were imputed using predictive mean matching, whereas data in the categorical variable were imputed using logistic regression factor (2 levels) and multinomial or ordered logit models (>2 levels).
All the models for the association between a single pollutant and incident RA were adjusted for the covariates mentioned above. Then, the three nested Cox models, which included a basic model adjusted for age and sex, a multivariable adjustment model further adjusted for all listed confounding factors, and a fully adjusted model further adjusted for PRS, were used to estimate the impact of air pollution score on the risk of RA. For analysis models including PRS, we further adjusted for the genotyping batches (11 batches in the UK BiLEVE Axiom array, 95 batches in the Biobank Axiom array) and the first 10 genetic principal components (PC1–PC10). To evaluate the joint associations of PRS and air pollution score with RA risk, we classified participants into 12 groups according to genetic risk and quartiles of the air pollution score. The HRs of incident RA in different groups were estimated compared with those with low genetic risk and the lowest quartile of air pollution score. We performed the Cochran–Armitage test for trends in binomial RA status across the levels of the variable of interest. To quantify the interactions on additive and multiplicative scales, we added a product term that combined high genetic risk and fourth quartile air pollution score in the model. The HR for the product term and the relative excess risk due to interaction (RERI) were used as the measures of interaction on the multiplicative and additive scales, respectively. The exposure–response relationship of the air pollution score with RA risk was assessed using restricted cubic spline analysis with 3 knots. Spearman’s correlation coefficients were also calculated to assess correlations among air pollutants.
A 10-fold cross-validation analysis was performed to further validate the results.39 The overall data were randomly divided into 10 equal parts, with 9 of them taken as the training data set and the remaining 1 as the testing data set. In the training data set, the single air pollutant was refitted in the Cox model to obtain a new air pollution score–weighting coefficient. Accordingly, a new air pollutant score was constructed and its association strength with incident RA was evaluated in the testing data set. All process steps were repeated 10 times until each of the 10 parts was used once as the testing data set. A fixed-effects meta-analysis was performed to calculate the pooled HR. To further confirm the robustness of the weighted air pollution score, we also performed common mixture pollutants exposure estimation methods, quantile-based g-computation,40 and Bayesian kernel machine regression (BKMR),41 to recalculate the combined effect of the mixture air pollution.
Moreover, we conducted subgroup analyses in various dichotomous subgroups according to age (≥65 and <65 y), sex (female or male), education levels (with and without university degrees), and smoking status (previous/current and never). Apart from adjusting for the variables aforementioned in the main analysis, menopausal status (yes or no) and hormone replacement therapy use (yes or no) were additionally adjusted among females in the subgroup analyses. Furthermore, the RA cases were stratified according to the rheumatoid factor (RF) level and divided into positive and negative groups in line with a cutoff value of 20 IU/mL,42,43 and the association of air pollution exposure with different RF-status RA risks was further checked. The statistical methods in subgroup analysis used were consistent with the main analysis; however, when conducting interaction analysis in each subgroup, the product term included in the model were ≥65 years of age, female, no university degree, previous/current smoking, RF-positive, and fourth quartile air pollution score, respectively.
A series of sensitivity analyses were also conducted to demonstrate the robustness and reliability of the results. First, we established a new air pollution score that included only PM2.5, NO2, and NOx to further explore the association of air pollution score with the risk of incident RA. Second, PM2.5 absorbance, as a measurement of the blackness of PM2.5 filters and a proxy for elemental carbon, was assessed by the LUR model, as previously described.29,44 We included the PM2.5 absorbance (2010 available) in the air pollution score to serve as a further supplement to the PM exposure information. Third, to evaluate concerns for potential reverse causation, we restricted incident RA cases to >2y from the baseline time. Fourth, we additionally adjusted the latitude of the participants’ residence because latitude is closely related to exposure to ultraviolet radiation, which in turn affects immune regulation or vitamin D synthesis and thus, potentially, the occurrence of RA.45,46 Fifth, furthermore, to avoid inaccurate assignments of air pollution estimates caused by changes in residence, we included only participants who had lived at their current address for at least 5 y in the analysis. Finally, the diagnosis of RA may be delayed owing to the use of nonsteroidal anti-inflammatory drugs (NSAIDs) to alleviate symptoms, such as arthritis, so we excluded individuals using any dose of NSAIDs (Table S1) within 3 months before the baseline to avoid latent prevalent RA cases masked by NSAID use being included in our analysis.
All analyses were performed using R software (version 4.0.3; R Development Core Team). All p-values for the tests were two sided, and p<0.05 were considered statistically significant. Cox models were constructed using the “survival” package. Exposure–response relationship analyses were performed using the “rms” package. Interaction analyses were realized by using the “interactionR” package. “qgcomp” and “bkmr” packages were used to conduct quantile-based g-computation and BKMR. The location coordinates of participants’ residential addresses measured by the Ordnance Survey in Great Britain (OSGB) grid reference were transformed into latitude/longitude through a JavaScript library47 with the function of coordinate system conversions. The administrative boundary data of the UK were sourced from UK Government Open Data portal.48 All maps were drawn in R with the “rgdal” and “ggplot2” packages.
Results
The baseline characteristics of the study participants are presented in Table 1. Among 342,973 participants, 2,034 incident cases of RA were recorded during 2,760,119 person-years of follow-up (median follow-up time=8.1 y). Compared with participants without RA, the individuals with RA had lower income levels (<£18,000: 36.6% vs. 22.3%), and they were more likely to be previous or current smokers (54.7% vs. 45.3%), have a higher BMI (28.9 kg/m2 vs. 27.4 kg/m2), and have a more sedentary lifestyle (5.2 h/d vs. 4.9 h/d sedentary time). A comparison of baseline characteristics between the current study population and the full UK Biobank cohort was also reported in Table S3. The median [interquartile range (IQR)] estimates of PM2.5, PM2.5–10, PM10, NO2, and NOx were respectively 9.97 (1.32), 6.10 (0.79), 38.09 (4.38), 27.80 (9.87), and 42.30 (16.15)μg/m3 among participants with incident RA at baseline. The corresponding median (IQR) estimates were 9.88 (1.26), 6.08 (0.76), 38.01 (4.44), 27.30 (10.22), and 41.27 (16.10) μg/m3 for those without incident RA. The Spearman correlation coefficients among the five air pollutants are shown in Table S4. The dispersed distribution of air pollution levels and the incident RA cases in the areas where participants lived in 2010 are shown in Figure 1, and the distribution patterns were in line with patterns available in the public UK Air Information Resources.49
Table 1 Baseline characteristics of 342,973 participants in the UK Biobank study of the association of air pollution and genetic risk with rheumatoid arthritis (RA) incidence from 2006 to 2017.
Characteristicsa Incident RA Total population (N=342,973)
Yes (n=2,034) No (n=340,939)
Age [y (mean±SD)] 59.94±7.02 56.97±7.93 56.99±7.93
Follow-up time [median (IQR)] 5.2 (3.6) 8.1 (1.2) 8.1 (1.2)
Sex [n (%)]
Female 1,368 (67.26) 181,850 (53.33) 183,218 (53.42)
Male 666 (32.74) 159,089 (46.66) 159,755 (46.58)
Household income [n (%)]
<£18,000 745 (36.63) 76,176 (22.34) 76,921 (22.43)
£18,000–29,999 573 (28.17) 88,752 (26.03) 88,325 (26.04)
£30,000–51,999 430 (21.14) 89,897 (26.37) 90,327 (26.34)
£52,000–100,000 240 (11.80) 68,633 (20.13) 68,873 (20.08)
>£100,000 46 (2.26) 17,481 (5.13) 17,527 (5.11)
Education level [n (%)]
College or university degree 542 (26.65) 121,615 (35.67) 122,157 (35.62)
Other 1,492 (73.35) 219,324 (64.33) 220,816 (64.38)
Smoking status [n (%)]
Never smoking 921 (45.28) 186,337 (54.65) 187,258 (54.60)
Previous smoking 832 (40.90) 120,838 (35.44) 121,670 (35.48)
Current smoking 281 (13.81) 33,764 (9.91) 34,045 (9.93)
Alcohol consumption [standard-drink/d (mean±SD)b] 1.63±2.27 2.03±2.45 2.03±2.44
Healthy diet score [n (%)]
0–1 461 (22.66) 77,519 (22.74) 77,980 (22.74)
2–3 1,064 (52.31) 184,495 (54.12) 185,559 (54.10)
4–5 509 (25.02) 78,925 (23.14) 79,434 (23.16)
Sedentary time [h/d (mean±SD)] 5.18±2.45 4.87±2.38 4.87±2.38
Physical activity [min/d (mean±SD)] 126.93±107.24 128.81±102.79 128.80±102.82
Body mass index [kg/m2 (mean±SD)] 28.89±5.56 27.41±4.74 27.42±4.75
Rheumatoid factor (RF) status [n (%)]
RF-positive 417 (20.50) 11,803 (3.46) 12,220 (3.56)
RF-negative 1,543 (75.86) 312,557 (91.68) 314,100 (91.58)
Latitude of residence [degree (mean±SD)] 53.06±1.14 52.87±1.19 52.86±1.19
PM2.5 [μg/m3, median (IQR)] 9.97 (1.32) 9.88 (1.26) 9.88 (1.26)
PM2.5–10 [μg/m3, median (IQR)] 6.10 (0.79) 6.08 (0.76) 6.08 (0.76)
PM10 [μg/m3, median (IQR)] 38.09 (4.38) 38.01 (4.44) 38.01 (4.44)
NO2 [μg/m3, median (IQR)] 27.80 (9.87) 27.30 (10.22) 27.30 (10.22)
NOx [μg/m3, median (IQR)] 42.30 (16.15) 41.27 (16.10) 41.27 (16.10)
Air pollution score (mean±SD) 71.68±11.35 70.63±11.13 70.63±11.12
Note: IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM2.5–10, particulate matter with an aerodynamic diameter between 2.5 and 10μm; PM10, particulate matter with an aerodynamic diameter ≤10μm.
a Missing values for each characteristic: household income (n=48,144), education (n=62,218), smoking status (n=1,182), alcohol consumption (n=2,848), healthy diet score (n=56,104), sedentary time (n=8,334), physical activity (n=65,263), body mass index (n=1,079), and rheumatoid factor (n=16,653).
b A standard-drink of alcohol consumption=8g of pure alcohol intake.
Figure 1. Map of air pollution (NO2, NOx, PM2.5, PM2.5–10, and PM10) of areas where participants lived in 2010 and map of incident RA scatter distribution. The administrative boundary data of the UK are sourced from UK Government Open Data portal (https://data.gov.uk/dataset/3fd8d2d2-b591-42ff-b333-c53a6a513e96/countries-december-2017-full-clipped-boundaries-in-great-britain). These data are UK government–released open data. (A) NO2, (B) NOx, (C) PM2.5, (D) PM2.5–10, (E) PM10, and (F) incident RA. Note: NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM2.5–10, particulate matter with aerodynamic diameter 2.5–10μm; PM10, particulate matter with aerodynamic diameter ≤10μm; RA, rheumatoid arthritis.
Figure 1A to 1F are graphs, plotting Latitude, ranging from 50.0 to 60.0 in increments of 2.5 (y-axis) across Longitude, ranging from negative 9 to 0 in increments of negative 3 (x-axis). Figures 1A to 1E have a color scale depicting nitrogen dioxide (micrograms per meter cubed), ranging from 15 to 30 in increments of 5; a color scale depicting nitrogen oxides (micrograms per meter cubed) ranging from 20 to 60 in increments of 10; a color scale depicting particulate matter begin subscript 2.5 end subscript dioxide (micrograms per meter cubed) ranging from 9 to 13 in unit increments; a color scale depicting particulate matter begin subscript 2.5 to 10 end subscript dioxide (micrograms per meter cubed) ranging from 6 to 10 in unit increments; and a color scale depicting particulate matter begin subscript 10 end subscript dioxide (micrograms per meter cubed) ranging from 12 to 20 in increments of 2.
The associations between individual air pollutants and RA are shown in Table 2. We observed that PM2.5, NO2, and NOx were positively associated with the risk of RA (pTrend for PM2.5=0.000043, pTrend for NO2 =0.00042, and pTrend for NOx =0.000011, respectively) after adjusting for age, sex, UK Biobank assessment centers, household income, education, smoking status, BMI, alcohol consumption, sedentary time, physical activity duration, and healthy diet score. The HRs (95% CI) of RA per IQR increase in PM2.5 (IQR: 1.26 μg/m3), PM10 (IQR: 4.44 μg/m3), PM2.5–10 (IQR: 1.26 μg/m3), NO2 (IQR: 10.22 μg/m3), and NOx (IQR: 16.10 μg/m3) were 1.07 (1.01, 1.13), 1.01 (0.96, 1.07), 1.00 (0.96, 1.04), 1.03 (0.98, 1.09), and 1.07 (1.02, 1.12), respectively. The exposure–response relationship of the association of each air pollutant with incident RA was also checked, and PM2.5 showed a relatively strong effect (Figure S2 and Excel Table S2).
Table 2 Association between single air pollutant and incident rheumatoid arthritis (RA) among UK Biobank participants (N=342,973; n=2,034 incident RA cases).
Air pollutantsa Case/n HR (95% CI) of incident RAb pTrend
PM2.5
Per IQR increment — 1.07 (1.01, 1.13)
Q1 467/86,448 Ref 0.000043
Q2 480/86,172 0.97 (0.86, 1.11)
Q3 504/84,731 1.00 (0.88, 1.14)
Q4 583/85,622 1.12 (1.00, 1.26)
PM10
Per IQR increment — 1.01 (0.96, 1.07)
Q1 486/86,006 Ref 0.13
Q2 511/85,457 0.99 (0.88, 1.13)
Q3 516/85,870 0.98 (0.87, 1.11)
Q4 521/85,640 1.05 (0.92, 1.19)
PM2.5–10
Per IQR increment — 1.00 (0.96, 1.04)
Q1 498/86,570 Ref 0.12
Q2 504/85,471 1.00 (0.88, 1.13)
Q3 499/85,500 1.00 (0.88, 1.13)
Q4 533/85,432 1.05 (0.93, 1.19)
NO2
Per IQR increment — 1.03 (0.98, 1.09)
Q1 442/85,746 Ref 0.00042
Q2 512/85,758 1.06 (0.93, 1.20)
Q3 536/85,722 1.08 (0.95, 1.23)
Q4 544/85,747 1.14 (1.00, 1.30)
NOx
Per IQR increment — 1.07 (1.02, 1.12)
Q1 442/85,750 Ref 0.000011
Q2 498/85,758 1.05 (0.92, 1.19)
Q3 518/85,740 1.05 (0.93, 1.20)
Q4 576/85,725 1.17 (1.03, 1.33)
Note: —, not applicable; CI, confidence interval; HR, hazard ratio; IQR, interquartile range; NO2, nitrogen dioxide; NOx, nitrogen oxides; PM2.5, particulate matter with aerodynamic diameter ≤2.5μm; PM2.5–10, particulate matter with an aerodynamic diameter between 2.5 and 10mm; PM10, particulate matter with an aerodynamic diameter ≤10μm; Q, quartile.
a PM2.5 ranges: quartile 1: (8.17–9.23) μg/m3, quartile 2: (9.24–9.88) μg/m3, quartile 3: (9.89–10.49) μg/m3, and quartile 4: (10.50–21.31) μg/m3 and IQR is 1.26 μg/m3; PM10 ranges: quartile 1: (25.73–35.89) μg/m3, quartile 2: (35.90–38.01) μg/m3, quartile 3: (38.02–40.33) μg/m3, and quartile 4: (40.34–60.16) μg/m3 and IQR is 4.44 μg/m3; PM2.5–10 ranges: quartile 1: (5.57–5.83) μg/m3, quartile 2: (5.84–6.08) μg/m3, quartile 3: (6.09–6.59) μg/m3, and quartile 4: (6.60–12.82) μg/m3 and IQR is 1.26 μg/m3; NO2 ranges: quartile 1: (8.86–22.40) μg/m3, quartile 2: (22.41–27.30) μg/m3, quartile 3: (27.31–32.62) μg/m3, and quartile 4: (32.63–125.12) μg/m3 and IQR is 10.22 μg/m3; NOx ranges: quartile 1: (19.74–33.38) μg/m3, quartile 2: (33.39–41.27) μg/m3, quartile 3: (41.28–49.48) μg/m3, and quartile 4: (49.49–265.94) μg/m3 and IQR is 16.10 μg/m3.
b Adjusted for age, sex, UK Biobank assessment center, household income, education level, smoking status, body mass index, alcohol consumption, sedentary time, physical activity duration, and healthy diet score.
The mean air pollution score±standard deviation (SD) was 70.63±11.12, ranging from 48.54 to 202.00, with a higher score indicating a higher combined exposure to air pollutants. The weights of each air pollutant included in the calculation of the air pollution score are shown in Table S5, and the distribution of air pollution scores among participants is shown in Figure S3 and Excel Table S3. As shown in Table 3, we found a positive exposure–response relationship between the air pollution score and RA risk (pTrend=0.000053). After adjusting for age and sex, the risk of incident RA in the highest quartile of the air pollution score was 34% (95% CI: 18%, 51%) higher than in the lowest quartile group. Moreover, the association of air pollution score with incident RA remained significant after adjusting for the UK Biobank assessment center, household income, education, smoking status, BMI, alcohol consumption, sedentary time, physical activity duration, and healthy diet score, with the HR (95% CI) for the fourth quartile group being 1.14 (1.00, 1.29). Furthermore, the above results remained stable when we further included the PRS of RA, the first 10 genetic principal components, and genotyping batches into the models. The results from cross-validation analysis further confirmed the robustness of the findings, with the value of HRFixed-effect (95% CI) being 1.05 (1.01, 1.10) per SD increment in air pollution score (Figure S4). We compared the impacts of the air pollutants mixture using three methods, including BKMR, the quantile g-computation model, and our air pollution score. In the air pollution score we constructed, PM2.5 and NOx were the highest contributors (βPM2.5=0.05 and βNOx=0.004), which was relatively consistent with the BKMR [conditional posterior inclusion probabilities (PIPPM2.5=0.77 and PIPNOx=0.94); Table S6]. In quantile g-computation models, the magnitude of the overall mixture effects (Ψ); that is, the HR (95% CI) of incident RA was 1.08 (1.02, 1.15) per quartile increase in the concentration of all air pollutants, showing an association pattern similar to our air pollution score (Table S7).
Table 3 Association between air pollution score and incident rheumatoid arthritis (RA) among UK Biobank participants (N=342,973; n=2,034 incident RA cases).
Air pollution scorea Case/n HR (95% CI) of incident RA pTrend
Model 1b Model 2c Model 3d
Per standard deviation increment — 1.12 (1.07, 1.16) 1.06 (1.01, 1.10) 1.06 (1.01, 1.10)
Q1 454/85,743 Ref Ref Ref 0.000053
Q2 490/85,743 1.09 (0.96, 1.24) 1.00 (0.89, 1.14) 1.00 (0.88, 1.14)
Q3 514/85,743 1.17 (1.03, 1.33) 1.01 (0.89, 1.15) 1.01 (0.89, 1.15)
Q4 576/85,744 1.34 (1.18, 1.51) 1.14 (1.00, 1.29) 1.14 (1.00, 1.29)
Note: —, not applicable; CI, confidence interval; HR, hazard ratio; Q, quartile; Ref, reference; SD, standard deviation.
a Air pollution score ranges: quartile 1: (48.54–63.35); quartile 2: (63.36–69.94); quartile 3: (69.95–76.43); and quartile 4: (76.44–202.00). Mean±SD of the air pollution score is 70.63±11.12.
b Adjusted for age and sex.
c Adjusted for age, sex, UK Biobank assessment center, household income, education level, smoking status, body mass index, alcohol consumption, sedentary time, physical activity duration, and healthy diet score.
d Adjusted for age, sex, UK Biobank assessment center, household income, education level, smoking status, body mass index, alcohol consumption, sedentary time, physical activity duration, healthy diet score, polygenic risk score, first 10 genetic principal components, and genotyping batch.
The exposure–response relationships of the air pollution score with incident RA according to stratification of age, sex, smoking status, education level, and RF status are shown in Figures S5–S9. Age is an important modifier in the impact of air pollution on the risk of RA (pMultiplicative interaction=0.044), and the HR (95% CI) per SD increased air pollution score of those ≥65 years of age was 1.09 (1.01, 1.18); Table S8, Figure S5, and Excel Table S4]. A strong interaction between the air pollution and sex on the RA risk was also observed (pMultiplicative interaction=0.005 and pAdditive interaction=0.0009); the effect of air pollution on the RA risk in females was more apparent than in males, and the HR (95% CI) per SD increased air pollution score in females was 1.10 (1.05, 1.16); Table S8, Figure S6, and Excel Table S5]. Compared with nonsmokers, the exposure–response curve of smokers’ air pollution scores and RA incidence risk seemed to rise more rapidly (Figure S7 and Excel Table S6); however, there was no significant interaction between the air pollution and smoking status on the RA risk (pMultiplicative interaction=0.99 and pAdditive interaction=0.98; Table S8). Although there were significant differences in the strength of association between air pollution and RA risk in different education levels (Figure S8 and Excel Table S7), only marginal significant additive interaction was found [RERI=0.23 (95% CI: 0.002, 0.46); Table S8]. In addition, the risk of seronegative-RF RA increased significantly with an increase in air pollution score, suggesting there may be a negative interaction between air pollution and RF-positive RA (pMultiplicative interaction=0.0071; Table S8, Figure S9, and Excel Table S8).
As shown in Table 4, a significant positive association was observed between the PRS of RA and the risk of incident RA. After adjusting for sex, age, assessment center, first 10 genetic principal components, and genotyping batches, the HR (95% CI) of incident RA per increment in SD in the PRS of RA was 1.22 (1.17, 1.27). Moreover, the HR (95% CI) of incident RA in the high genetic risk group was 1.48 (1.33, 1.65) when compared with the low genetic risk group. All the findings remained stable in the multivariate-adjusted models.
Table 4 Association between genetic risk and incident rheumatoid arthritis (RA) among UK Biobank participants (N=342,973; n=2,034 incident RA cases).
Polygenic risk scorea Case/n HR (95% CI) of incident RA pTrend
Model 1b Model 2c
Per standard deviation increment — 1.22 (1.17, 1.27) 1.22 (1.17, 1.27)
Low genetic risk 554/113,181 Ref Ref 2.63×10−13
Intermediate genetic risk 640/113,181 1.16 (1.03, 1.29) 1.16 (1.04, 1.30)
High genetic risk 840/116,611 1.48 (1.33, 1.65) 1.48 (1.33, 1.65)
Note: —, not applicable; CI, confidence interval; HR, hazard risk; PRS, polygenic risk score; Ref, reference.
a PRS ranges: low genetic risk (tertile 1): (−3.524 to −0.497); intermediate genetic risk (tertile 2): (−0.498 to 0.297); high genetic risk (tertile 3): (0.298 to 5.066). Mean±standard deviation of PRS is 0±1.
b Adjusted for age, sex, first 10 genetic principal components, and genotyping batch.
c Adjusted for age, sex, UK Biobank assessment center, household income, education level, smoking status, body mass index, alcohol consumption, sedentary time, physical activity duration, healthy diet score, PRS, first 10 genetic principal components, and genotyping batch.
The joint association of the air pollution score and the PRS with the risk of RA incidence was further assessed, and a significant exposure–response relationship between air pollutants and incident RA was found in the low (pTrend=0.00064) and intermediate genetic risk groups (pTrend=0.0096) (Figure 2). Furthermore, the results of the combined effect of air pollution score and PRS on the RA risk showed that the risk of RA incidence in the highest genetic risk and air pollution score group was almost twice that of the lowest genetic risk and air pollution score group [incidence rate (IR) per 100,000 person-years=98.46 vs. 51.19, and HR=1.73 (95% CI: 1.39, 2.17) vs. 1 (reference)], although no statistically significant interaction between the air pollution and genetic risk on incident RA was found (pMultiplicative interaction=0.057 and pAdditive interaction=0.54; Figure 2).
Figure 2. Association of combined air pollution and genetic risk with incident RA among UK Biobank [N=342,973 participants (n=2,034 incident RA)]. Adjusted for age, sex, UK Biobank assessment center, household income, education level, smoking status, body mass index, alcohol consumption, sedentary time, physical activity duration, healthy diet score, PRS, first 10 genetic principal components and genotyping batch. Genetic risk was categorized into three levels by tertiles of PRS: low (tertile 1): (−3.524 to −0.497); intermediate (tertile 2): (−0.498 to 0.297); and high (tertile 3): (0.298 to 5.066). Air pollution score ranges: quartile 1: (48.54–63.35); quartile 2: (63.36–69.94); quartile 3: (69.95–76.43); and quartile 4: (76.44–202.00). Note: CI, confidence interval; HR, hazard risk; PRS, polygenic risk score; Q, quartile; RA, rheumatoid arthritis; RERI, relative excess risk due to interaction.
Figure 2 is a forest plot, plotting risk with cases and incidence per 10,000 person-years (bottom to top) High genetic risk, including air pollution score Quartile 1 with 199 per 29,114 cases and 84.84 incidence, air pollution score Quartile 2 with 210 per 29,275 cases and 89.39 incidence, air pollution score Quartile 3 with 199 per 29,054 cases and 85.63 incidence, air pollution score Quartile 4 with 232 per 29,168 cases and 98.46 incidence; Intermediate genetic risk, including air pollution score Quartile 1 with 138 per 28,277 cases and 60.54 incidence, air pollution score Quartile 2 with 153 per 28,042 cases and 67.79 incidence, air pollution score Quartile 3 with 171 per 28,341 cases and 75.43 incidence, air pollution score Quartile 4 with 178 per 28,521 cases and 77.13 incidence; Low genetic risk, including air pollution score Quartile 1 with 117 per 28,352 cases and 51.19 incidence, air pollution score Quartile 2 with 127 per 28,426 cases and 55.54 incidence, air pollution score Quartile 3 with 144 per 28,348 cases and 63.41 incidence, air pollution score Quartile 4 with 166 per 28,055 cases and 73.08 incidence (y-axis) across hazard ratios (95 percent confidence interval), ranging from 0.5 to 2.5 in increments of 0.5 (x-axis) for hazard ratios (95 percent confidence interval), lowercase italic p values, and uppercase italic p begin subscript trend end subscript.
In addition, several sensitivity analyses were performed to confirm our findings. We first recalculated air pollution scores that excluded PM2.5–10 and PM10 (the remaining weights of PM2.5, NO2, and NOx were the same as those in the main analyses) and examined the relationship between RA incidence and the recalculated air pollution scores, and the results remained unchanged (Table S9). In addition, we found no significant impact of PM2.5 absorbance on RA risk in the single-pollutant model (pTrend=0.10), and the magnitude of the association was slightly attenuated in the multivariable-adjusted model analysis after incorporating it into the air pollution score (Table S10). Participants with a follow-up time of <2y were removed from the analysis, and this did not appreciably change the results (Table S11). Furthermore, the latitude of participants’ residence was further adjusted (Table S12) and participants in the analysis were limited to those who had lived at their current address for at least 5 y (Table S13), and the findings were found to be robust. Finally, after excluding participants who had used NSAIDs, the results were still comparable (Table S14).
Discussion
To the best of our knowledge, this is the first prospective cohort study to investigate the association of combined exposure to multiple air pollutants with the risk of incident RA while considering the modification effect of genetic risk. By weighting the regression coefficient of each air pollutant (PM2.5, PM2.5–10, PM10, NO2, and NOx), we constructed an air pollution score to represent comprehensive air pollution and assessed the association with the risk of RA. The results showed a positive association with RA incidence. Furthermore, we found that the IR of RA almost increased monotonically with increasing air pollution scores across different genetic risk strata, particularly in the low and intermediate genetic risk groups. The nonstatistically significant exposure–response relationship in the high genetic risk group reflected that the health effect of air pollution may play a minor role compared with a high genetic predisposition. However, the highest IR and HR of RA risk in the population that had the highest air pollution and genetic risk was still more concerning.
To date, the effects of single air pollutants, such as NO2 and PM2.5, on health effects have been widely demonstrated.50 However, given that humans are exposed to a mixture of air pollutants, seeking to use a mixture of pollutant exposure estimation methods is important.51,52 In this study, we attempted to construct an air pollution score using the weighted regression coefficient method to characterize the mixed exposure to multiple air pollutants and observed a modest positive association between the air pollution score and the risk of RA. Previous UK Biobank studies30,31 used the same algorithm to deal with the additive linear effects of different air pollutants and developed an air pollution score that proved that joint exposure to air pollutants is significantly associated with the risk of type 2 diabetes and heart failure. Similar integrated scores have been applied not only in the field of environmental health53,54 but also in other epidemiological studies.55,56 Moreover, we additionally checked whether the magnitude of our current air pollution score health effect for RA was comparable to that using other common mixture pollutants exposure estimate methods, such as quantile-based g-computation40 and BKMR,41 in our further validation (Tables S6 and S7). In the quantile g-computation model, the overall mixture effects (Ψ) were close to our air pollution score. The results from the BKMR models also demonstrated a similar positive association between air pollution and RA risk. In general, our air pollution scores were relatively accurate and reliable, and cross-validation methods avoided the problem of overfitting. Moreover, as a continuous variable, the air pollution score can contribute to the determination of the possible risk threshold for disease prevention and also facilitate complex interaction analysis with other risk factors of interest. In the multivariable-adjusted model, only the air pollution score in the highest quartile was significantly associated with incident RA. However, it cannot be ignored that the air pollution scores in this study did not cover all air pollutants and may have resulted in an underestimated relationship between air pollution and the risk of RA.
Environmental agents are thought to interact with genetic factors and jointly trigger the immunologic processes before clinical RA.57 A classic example supporting the environment–gene interaction is that between human leucocyte antigen-shared epitope gene (HLA-SE) alleles and smoking, which strongly increases the risk of seropositive RA.7,58 Previous studies13–18 have often ignored the modification effects of genetic susceptibility and have reported conflicting evidence on the relationship between air pollution and RA. In this study, to explore the potential interaction between air pollution and genetic risk, we constructed a PRS that represents the overall genetic risk of RA. We found that the intensity of the association between RA risk and air pollution level was higher with the increase of PRS even though the interaction between the high air pollution exposure and high genetic risk was not significant. However, the RA risk–associated genomic loci identified from population-based genetic association studies collectively account for only ∼15% of the phenotypic variance of RA59,60 and may partially explain the nonstatistically significant interaction.
The pathogenic mechanism of air pollution in RA development has been comprehensively explored, with several hypotheses being proposed about potentially involved factors, including T cell imbalance, production of pro-inflammatory cytokines, local pulmonary inflammation, oxidative stress, and methylation changes.61–65 The association of air pollution with RA was not limited to the risk of incidence, the two case-crossover studies66,67 further revealed that exposure to high levels of air pollutants will also affect disease activity and drug response for RA. In addition, age, sex, socioeconomic status, and lifestyle are all potential modifiers of the magnitude of the effects of air pollution on RA.68–73 Therefore, to confirm the robustness of the results, we further performed stratified analyses of these factors. The results revealed that larger effect estimates of air pollution exposure on the risk of RA were among participants who were ≥65 years of age, female, and lacking a university degree. These results could be related to several potential mechanisms. For example, the immune system and hormonal functions are different in people of different ages and genders68,69; for instance, the immune function and lung compensation ability of the elderly are often weakened or even impaired, making them particularly susceptible to air pollution.70 A larger health impact of air pollution among females be may be partly explained by lifestyle factors such that more time spent at home results in better accuracy of residential air pollution exposure assessment, as well as biological factors, such as greater airway reactivity and hormonal action.71 The increasing trend of the risk of RF-negative RA with higher air pollution scores seemed to be more apparent than that of RF-positive RA, which is consistent with the evidence from a study in the Studies of the Etiology of Rheumatoid Arthritis (SERA) that reported that ambient annual PM levels are not associated with the early development of RA-related autoimmunity prior to the development of articular RA.74 However, the study was limited to PM and did not include other air pollutants, so the linkage of overall air pollution and early RA-related autoimmunity may need further confirmation in the future.
The results of the present study should be interpreted in the context of its strengths and limitations. The main strength of this study is that it is based on the UK Biobank prospective cohort design, with a large sample size. Our study strictly controlled confounding factors, including socioeconomic status and lifestyle, and used cross-validation to ensure the stability of the results. This study has some limitations. First, this observational study could not fully control for all unknown or unmeasured confounding factors and was unable to determine a causal relationship between air pollution and RA. Then, the measurement of air pollution was only within the baseline time range, making it impossible to further explore the important window period of the impact of air pollution exposure on the risk of RA. Further studies with repeated air pollution measurements are required to confirm our findings. Moreover, the coverage of air pollutants was not comprehensive and the exposure to air pollutants, such as SO2, carbon monoxide, and ozone, was unavailable. In the future, it may be necessary to integrate more air pollutants to establish a more powerful and explainable air pollution score. Furthermore, although the number of incident RA cases was sufficient for the main analysis, in the specific stratified analysis the statistical power may have been limited by the decreased number of cases. In addition, we lacked information on exposure to these pollutants in locations other than participants’ residential addresses, such as outdoor or work sites. This prevented us from exploring the impact of the total air pollution exposure of each participant on the incidence of RA. Moreover, UK Biobank participants are not representative of the UK general population owing to evidence of a healthy volunteer selection bias,75 and the coverage of only the 40- to 70-y-old population may have limited us from finding more incident RA events in younger individuals. Furthermore, we determined incident RA cases based on hospital inpatient records, and latent incident RA cases with less severe clinical symptoms may not have been admitted and recorded in the hospital inpatient records. Finally, our study participants were all White, so it may be necessary to validate the generalizability of the conclusions to other ethnic populations.
Conclusions
Our findings showed that long-term combined exposure to ambient air pollutants was associated with an elevated risk of RA, and this association was more pronounced in populations with high genetic risk. We highlight the importance of comprehensive assessment for air pollution and genetic predisposition in the prevention of RA.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
J.Z., X.-Y.F. and D.-Q.Y. conceived the idea for the paper. J.Z. conducted the analysis. J.Z. and X.-Y.F. are joint first authors, and had primary responsibility for drafting the manuscript. J.W., Y.-G.F., R.-X.L., B.L., X.-J.L., Y.-L.Y. contributed to the data cleaning. D.-Q.Y., C.M., J.W., Y.-G.F., R.-X.L. contributed to the analysis or interpretation of the data. All authors critically reviewed the manuscript for important intellectual content. D.-Q.Y. directed the study. D.-Q.Y. is the study guarantor and has full access to data.
The authors thank the UK Biobank participants. This research has been conducted using the UK Biobank resource under application no. 62663.
Funding was provided by the Chinese national high level personnel special support plan.
The funders played no role in the study design or implementation; data collection, management, analysis or interpretation; manuscript preparation, review or approval; or the decision to submit the manuscript for publication.
==== Refs
References
1. Smolen JS, Aletaha D, McInnes IB. 2016. Rheumatoid arthritis. Lancet 388 (10055 ):2023–2038, PMID: , 10.1016/S0140-6736(16)30173-8.27156434
2. van der Woude D, van der Helm-van Mil AHM. 2018. Update on the epidemiology, risk factors, and disease outcomes of rheumatoid arthritis. Best Pract Res Clin Rheumatol 32 (2 ):174–187, PMID: , 10.1016/j.berh.2018.10.005.30527425
3. Alamanos Y, Drosos AA. 2005. Epidemiology of adult rheumatoid arthritis. Autoimmun Rev 4 (3 ):130–136, PMID: , 10.1016/j.autrev.2004.09.002.15823498
4. Safiri S, Kolahi AA, Hoy D, Smith E, Bettampadi D, Mansournia MA, et al. 2019. Global, regional and national burden of rheumatoid arthritis 1990–2017: a systematic analysis of the Global Burden of Disease study 2017. Ann Rheum Dis 78 (11 ):1463–1471, PMID: , 10.1136/annrheumdis-2019-215920.31511227
5. Arleevskaya M, Takha E, Petrov S, Kazarian G, Novikov A, Larionova R, et al. 2021. Causal risk and protective factors in rheumatoid arthritis: a genetic update. J Transl Autoimmun 4 :100119, PMID: , 10.1016/j.jtauto.2021.100119.34522877
6. Deane KD, Demoruelle MK, Kelmenson LB, Kuhn KA, Norris JM, Holers VM. 2017. Genetic and environmental risk factors for rheumatoid arthritis. Best Pract Res Clin Rheumatol 31 (1 ):3–18, PMID: , 10.1016/j.berh.2017.08.003.29221595
7. Karlson EW, Chang SC, Cui J, Chibnik LB, Fraser PA, De Vivo I, et al. 2010. Gene–environment interaction between HLA-DRB1 shared epitope and heavy cigarette smoking in predicting incident rheumatoid arthritis. Ann Rheum Dis 69 (1 ):54–60, PMID: , 10.1136/ard.2008.102962.19151010
8. Stolt P, Yahya A, Bengtsson C, Källberg H, Rönnelid J, Lundberg I, et al. 2010. Silica exposure among male current smokers is associated with a high risk of developing ACPA-positive rheumatoid arthritis. Ann Rheum Dis 69 (6 ):1072–1076, PMID: , 10.1136/ard.2009.114694.19966090
9. Demoruelle MK, Weisman MH, Simonian PL, Lynch DA, Sachs PB, Pedraza IF, et al. 2012. Brief report: airways abnormalities and rheumatoid arthritis-related autoantibodies in subjects without arthritis: early injury or initiating site of autoimmunity? Arthritis Rheum 64 (6 ):1756–1761, PMID: , 10.1002/art.34344.22183986
10. Glencross DA, Ho TR, Camiña N, Hawrylowicz CM, Pfeffer PE. 2020. Air pollution and its effects on the immune system. Free Radic Biol Med 151 :56–68, PMID: , 10.1016/j.freeradbiomed.2020.01.179.32007522
11. Huff RD, Carlsten C, Hirota JA. 2019. An update on immunologic mechanisms in the respiratory mucosa in response to air pollutants. J Allergy Clin Immunol 143 (6 ):1989–2001, PMID: , 10.1016/j.jaci.2019.04.012.31176381
12. Sigaux J, Biton J, André E, Semerano L, Boissier MC. 2019. Air pollution as a determinant of rheumatoid arthritis. Joint Bone Spine 86 (1 ):37–42, PMID: , 10.1016/j.jbspin.2018.03.001.29524589
13. Hart JE, Laden F, Puett RC, Costenbader KH, Karlson EW. 2009. Exposure to traffic pollution and increased risk of rheumatoid arthritis. Environ Health Perspect 117 (7 ):1065–1069, PMID: , 10.1289/ehp.0800503.19654914
14. Jung CR, Hsieh HY, Hwang BF. 2017. Air pollution as a potential determinant of rheumatoid arthritis: a population-based cohort study in Taiwan. Epidemiology 28 (suppl 1 ):S54–S59, PMID: , 10.1097/EDE.0000000000000732.29028676
15. Park JS, Choi S, Kim K, Chang J, Kim SM, Kim SR, et al. 2021. Association of particulate matter with autoimmune rheumatic diseases among adults in South Korea. Rheumatology (Oxford) 60 (11 ):5117–5126, PMID: , 10.1093/rheumatology/keab127.33560298
16. Hart JE, Källberg H, Laden F, Bellander T, Costenbader KH, Holmqvist M, et al. 2013. Ambient air pollution exposures and risk of rheumatoid arthritis: results from the Swedish EIRA case–control study. Ann Rheum Dis 72 (6 ):888–894, PMID: , 10.1136/annrheumdis-2012-201587.22833374
17. Hart JE, Källberg H, Laden F, Costenbader KH, Yanosky JD, Klareskog L, et al. 2013. Ambient air pollution exposures and risk of rheumatoid arthritis. Arthritis Care Res (Hoboken) 65 (7 ):1190–1196, PMID: , 10.1002/acr.21975.23401426
18. De Roos AJ, Koehoorn M, Tamburic L, Davies HW, Brauer M. 2014. Proximity to traffic, ambient air pollution, and community noise in relation to incident rheumatoid arthritis. Environ Health Perspect 122 (10 ):1075–1080, PMID: , 10.1289/ehp.1307413.24905961
19. Hunt L, Emery P. 2014. Defining populations at risk of rheumatoid arthritis: the first steps to prevention. Nat Rev Rheumatol 10 (9 ):521–530, PMID: , 10.1038/nrrheum.2014.82.24934191
20. Okada Y, Eyre S, Suzuki A, Kochi Y, Yamamoto K. 2019. Genetics of rheumatoid arthritis: 2018 status. Ann Rheum Dis 78 (4 ):446–453, PMID: , 10.1136/annrheumdis-2018-213678.30530827
21. Choi SW, Mak TS, O’Reilly PF. 2020. Tutorial: a guide to performing polygenic risk score analyses. Nat Protoc 15 (9 ):2759–2772, PMID: , 10.1038/s41596-020-0353-1.32709988
22. Khera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, et al. 2018. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat Genet 50 (9 ):1219–1224, PMID: , 10.1038/s41588-018-0183-z.30104762
23. Sparks JA, Chen CY, Jiang X, Askling J, Hiraki LT, Malspeis S, et al. 2015. Improved performance of epidemiologic and genetic risk models for rheumatoid arthritis serologic phenotypes using family history. Ann Rheum Dis 74 (8 ):1522–1529, PMID: , 10.1136/annrheumdis-2013-205009.24685909
24. Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. 2015. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med 12 (3 ):e1001779, PMID: , 10.1371/journal.pmed.1001779.25826379
25. Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. 2018. The UK Biobank resource with deep phenotyping and genomic data. Nature 562 (7726 ):203–209, PMID: , 10.1038/s41586-018-0579-z.30305743
26. CDC (Centers for Disease Control and Prevention). 2013. International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM). www.cdc.gov/nchs/icd/icd9cm.htm [accessed 2 April 2021].
27. WHO. 2016. International Statistical Classification of Diseases and Related Health Problems, 10th Revision. http://apps.who.int/classifications/icd10/browse/2016/en [accessed 2 April 2021].
28. Vienneau D, de Hoogh K, Bechle MJ, Beelen R, van Donkelaar A, Martin RV, et al. 2013. Western European land use regression incorporating satellite- and ground-based measurements of NO2 and PM10. Environ Sci Technol 47 (23 ):13555–13564, PMID: , 10.1021/es403089q.24156783
29. Eeftens M, Beelen R, de Hoogh K, Bellander T, Cesaroni G, Cirach M, et al. 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PMcoarse in 20 European study areas; results of the ESCAPE project. Environ Sci Technol 46 (20 ):11195–11205, PMID: , 10.1021/es301948k.22963366
30. Li X, Wang M, Song Y, Ma H, Zhou T, Liang Z, et al. 2021. Obesity and the relation between joint exposure to ambient air pollutants and incident type 2 diabetes: a cohort study in UK Biobank. PLoS Med 18 (8 ):e1003767, PMID: , 10.1371/journal.pmed.1003767.34460827
31. Wang M, Zhou T, Song Y, Li X, Ma H, Hu Y, et al. 2021. Joint exposure to various ambient air pollutants and incident heart failure: a prospective analysis in UK Biobank. Eur Heart J 42 (16 ):1582–1591, PMID: , 10.1093/eurheartj/ehaa1031.33527989
32. Okada Y, Wu D, Trynka G, Raj T, Terao C, Ikari K, et al. 2014. Genetics of rheumatoid arthritis contributes to biology and drug discovery. Nature 506 (7488 ):376–381, PMID: , 10.1038/nature12873.24390342
33. Babb de Villiers C, Kroese M, Moorthie S. 2020. Understanding polygenic models, their development and the potential application of polygenic scores in healthcare. J Med Genet 57 (11 ):725–732, PMID: , 10.1136/jmedgenet-2019-106763.32376789
34. Choi SW, O’Reilly PF. 2019. PRSice-2: polygenic risk score software for biobank-scale data. Gigascience 8 (7 ):giz082, PMID: , 10.1093/gigascience/giz082.31307061
35. Lloyd-Jones DM, Hong Y, Labarthe D, Mozaffarian D, Appel LJ, Van Horn L, et al. 2010. Defining and setting national goals for cardiovascular health promotion and disease reduction: the American Heart Association’s Strategic Impact Goal through 2020 and beyond. Circulation 121 (4 ):586–613, PMID: , 10.1161/CIRCULATIONAHA.109.192703.20089546
36. Kalinowski A, Humphreys K. 2016. Governmental standard drink definitions and low-risk alcohol consumption guidelines in 37 countries. Addiction 111 (7 ):1293–1298, PMID: , 10.1111/add.13341.27073140
37. Schoenfeld D. 1982. Partial residuals for the proportional hazards regression model. Biometrika 69 (1 ):239–241, 10.1093/biomet/69.1.239.
38. White IR, Royston P, Wood AM. 2011. Multiple imputation using chained equations: issues and guidance for practice. Stat Med 30 (4 ):377–399, PMID: , 10.1002/sim.4067.21225900
39. Simon RM, Subramanian J, Li MC, Menezes S. 2011. Using cross-validation to evaluate predictive accuracy of survival risk classifiers based on high-dimensional data. Brief Bioinform 12 (3 ):203–214, PMID: , 10.1093/bib/bbr001.21324971
40. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 128 (4 ):47004, PMID: , 10.1289/EHP5838.32255670
41. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16 (3 ):493–508, PMID: , 10.1093/biostatistics/kxu058.25532525
42. McQueenie R, Nicholl BI, Jani BD, Canning J, Macdonald S, McCowan C, et al. 2020. Patterns of multimorbidity and their effects on adverse outcomes in rheumatoid arthritis: a study of 5658 UK Biobank participants. BMJ Open 10 (11 ):e038829, PMID: , 10.1136/bmjopen-2020-038829.33234629
43. Beckman Coulter, Inc. 2020. Chemistry Information Sheet on Rheumatoid Factor (REF44707). https://www.beckmancoulter.com/download/file/phx988646AP-EN_US/988646AP?type=pdf [accessed 10 August 2022].
44. Zhang Z, Chen L, Wang X, Wang C, Yang Y, Li H, et al. 2023. Associations of air pollution and genetic risk with incident dementia: a prospective cohort study. Am J Epidemiol 192 (2 ):182–194, PMID: , 10.1093/aje/kwac188.36269005
45. Vieira VM, Hart JE, Webster TF, Weinberg J, Puett R, Laden F, et al. 2010. Association between residences in U.S. northern latitudes and rheumatoid arthritis: a spatial analysis of the Nurses’ Health Study. Environ Health Perspect 118 (7 ):957–961, PMID: , 10.1289/ehp.0901861.20338859
46. Arkema EV, Hart JE, Bertrand KA, Laden F, Grodstein F, Rosner BA, et al. 2013. Exposure to ultraviolet-B and risk of developing rheumatoid arthritis among women in the Nurses’ Health Study. Ann Rheum Dis 72 (4 ):506–511, PMID: , 10.1136/annrheumdis-2012-202302.23380431
47. Veness C. 2022. Geodesy Library: Movable Type scripts. https://www.movable-type.co.uk/scripts/geodesy-library.html [accessed 10 August 2022].
48. UK Office for National Statistics. 2018. Countries (December 2017) Full Clipped Boundaries in Great Britain. Last updated December 2019. https://www.data.gov.uk/dataset/3fd8d2d2-b591-42ff-b333-c53a6a513e96/countries-december-2017-full-clipped-boundaries-in-great-britain [accessed 23 November 2022].
49. UK Department for Environment Food & Rural Affairs. 2019. Modelled background pollution data. https://uk-air.defra.gov.uk/data/pcm-data [accessed 6 April 2022].
50. Costa S, Ferreira J, Silveira C, Costa C, Lopes D, Relvas H, et al. 2014. Integrating health on air quality assessment—review report on health risks of two major European outdoor air pollutants: PM and NO2. J Toxicol Environ Health B Crit Rev 17 (6 ):307–340, PMID: , 10.1080/10937404.2014.946164.25333993
51. Dominici F, Peng RD, Barr CD, Bell ML. 2010. Protecting human health from air pollution: shifting from a single-pollutant to a multipollutant approach. Epidemiology 21 (2 ):187–194, PMID: , 10.1097/EDE.0b013e3181cc86e8.20160561
52. Johns DO, Stanek LW, Walker K, Benromdhane S, Hubbell B, Ross M, et al. 2012. Practical advancement of multipollutant scientific and risk assessment approaches for ambient air pollution. Environ Health Perspect 120 (9 ):1238–1242, PMID: , 10.1289/ehp.1204939.22645280
53. Hong YC, Leem JH, Ha EH, Christiani DC. 1999. PM10 exposure, gaseous pollutants, and daily mortality in Inchon, South Korea. Environ Health Perspect 107 (11 ):873–878, PMID: , 10.1289/ehp.99107873.10544154
54. Park SK, Tao Y, Meeker JD, Harlow SD, Mukherjee B. 2014. Environmental risk score as a new tool to examine multi-pollutants in epidemiologic research: an example from the NHANES study using serum lipid levels. PLoS One 9 (6 ):e98632, PMID: , 10.1371/journal.pone.0098632.24901996
55. Lourida I, Hannon E, Littlejohns TJ, Langa KM, Hyppönen E, Kuzma E, et al. 2019. Association of lifestyle and genetic risk with incidence of dementia. JAMA 322 (5 ):430–437, PMID: , 10.1001/jama.2019.9879.31302669
56. Ahola-Olli AV, Mustelin L, Kalimeri M, Kettunen J, Jokelainen J, Auvinen J, et al. 2019. Circulating metabolites and the risk of type 2 diabetes: a prospective study of 11,896 young adults from four Finnish cohorts. Diabetologia 62 (12 ):2298–2309, PMID: , 10.1007/s00125-019-05001-w.31584131
57. Klareskog L, Padyukov L, Lorentzen J, Alfredsson L. 2006. Mechanisms of disease: genetic susceptibility and environmental triggers in the development of rheumatoid arthritis. Nat Clin Pract Rheumatol 2 (8 ):425–433, PMID: , 10.1038/ncprheum0249.16932734
58. Linn-Rasker SP, van der Helm-van Mil AHM, van Gaalen FA, Kloppenburg M, de Vries RRP, le Cessie S, et al. 2006. Smoking is a risk factor for anti-CCP antibodies only in rheumatoid arthritis patients who carry HLA-DRB1 shared epitope alleles. Ann Rheum Dis 65 (3 ):366–371, PMID: , 10.1136/ard.2005.041079.16014670
59. Kim K, Bang SY, Lee HS, Bae SC. 2017. Update on the genetic architecture of rheumatoid arthritis. Nat Rev Rheumatol 13 (1 ):13–24, PMID: , 10.1038/nrrheum.2016.176.27811914
60. Lenz TL, Deutsch AJ, Han B, Hu X, Okada Y, Eyre S, et al. 2015. Widespread non-additive and interaction effects within HLA loci modulate the risk of autoimmune diseases. Nat Genet 47 (9 ):1085–1090, PMID: , 10.1038/ng.3379.26258845
61. Zhao CN, Xu Z, Wu GC, Mao YM, Liu LN, Dan YL, et al. 2019. Emerging role of air pollution in autoimmune diseases. Autoimmun Rev 18 (6 ):607–614, PMID: , 10.1016/j.autrev.2018.12.010.30959217
62. Farhat SCL, Silva CA, Orione MAM, Campos LMA, Sallum AME, Braga ALF. 2011. Air pollution in autoimmune rheumatic diseases: a review. Autoimmun Rev 11 (1 ):14–21, PMID: , 10.1016/j.autrev.2011.06.008.21763467
63. Essouma M, Noubiap JJN. 2015. Is air pollution a risk factor for rheumatoid arthritis? J Inflamm (Lond) 12 :48, PMID: , 10.1186/s12950-015-0092-1.26225124
64. Chen R, Li H, Cai J, Wang C, Lin Z, Liu C, et al. 2018. Fine particulate air pollution and the expression of microRNAs and circulating cytokines relevant to inflammation, coagulation, and vasoconstriction. Environ Health Perspect 126 (1 ):017007, PMID: , 10.1289/EHP1447.29342453
65. Fiorito G, Vlaanderen J, Polidoro S, Gulliver J, Galassi C, Ranzi A, et al. 2018. Oxidative stress and inflammation mediate the effect of air pollution on cardio- and cerebrovascular disease: a prospective study in nonsmokers. Environ Mol Mutagen 59 (3 ):234–246, PMID: , 10.1002/em.22153.29114965
66. Adami G, Viapiana O, Rossini M, Orsolini G, Bertoldo E, Giollo A, et al. 2021. Association between environmental air pollution and rheumatoid arthritis flares. Rheumatology (Oxford) 60 (10 ):4591–4597, PMID: , 10.1093/rheumatology/keab049.33470401
67. Adami G, Rossini M, Viapiana O, Orsolini G, Bertoldo E, Pontalti M, et al. 2021. Environmental air pollution is a predictor of poor response to biological drugs in chronic inflammatory arthritides. ACR Open Rheumatol 3 (7 ):451–456, PMID: , 10.1002/acr2.11270.34060251
68. van Onna M, Boonen A. 2016. The challenging interplay between rheumatoid arthritis, ageing and comorbidities. BMC Musculoskelet Disord 17 (1 ):184, PMID: , 10.1186/s12891-016-1038-3.27118031
69. Favalli EG, Biggioggero M, Crotti C, Becciolini A, Raimondo MG, Meroni PL. 2019. Sex and management of rheumatoid arthritis. Clinic Rev Allerg Immunol 56 (3 ):333–345, PMID: , 10.1007/s12016-018-8672-5.29372537
70. Kelly FJ, Dunster C, Mudway I. 2003. Air pollution and the elderly: oxidant/antioxidant issues worth consideration. Eur Respir J Suppl 40 :70s–75s, PMID: , 10.1183/09031936.03.00402903.12762578
71. Doiron D, de Hoogh K, Probst-Hensch N, Fortier I, Cai Y, De Matteis S, et al. 2019. Air pollution, lung function and COPD: results from the population-based UK Biobank study. Eur Respir J 54 (1 ):1802140, PMID: , 10.1183/13993003.02140-2018.31285306
72. Kan H, Heiss G, Rose KM, Whitsel E, Lurmann F, London SJ. 2007. Traffic exposure and lung function in adults: the Atherosclerosis Risk in Communities study. Thorax 62 (10 ):873–879, PMID: , 10.1136/thx.2006.073015.17442705
73. O’Neill MS, Jerrett M, Kawachi I, Levy JI, Cohen AJ, Gouveia N, et al. 2003. Health, wealth, and air pollution: advancing theory and methods. Environ Health Perspect 111 (16 ):1861–1870, PMID: , 10.1289/ehp.6334.14644658
74. Gan RW, Deane KD, Zerbe GO, Demoruelle MK, Weisman MH, Buckner JH, et al. 2013. Relationship between air pollution and positivity of RA-related autoantibodies in individuals without established RA: a report on SERA. Ann Rheum Dis 72 (12 ):2002–2005, PMID: , 10.1136/annrheumdis-2012-202949.23572338
75. Fry A, Littlejohns TJ, Sudlow C, Doherty N, Adamska L, Sprosen T, et al. 2017. Comparison of sociodemographic and health-related characteristics of UK Biobank participants with those of the general population. Am J Epidemiol 186 (9 ):1026–1034, PMID: , 10.1093/aje/kwx246.28641372
|
PMC010xxxxxx/PMC10010399.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36913239
EHP10812
10.1289/EHP10812
Research
Cognitive Development and Prenatal Air Pollution Exposure in the CHAMACOS Cohort
https://orcid.org/0000-0001-5200-1716
Holm Stephanie M. 1 2 3
Balmes John R. 2 3 4
Gunier Robert B. 5
Kogut Katherine 5
Harley Kim G. 5
Eskenazi Brenda 5
1 Division of Epidemiology, School of Public Health, University of California Berkeley, Berkeley, California, USA
2 Western States Pediatric Environmental Health Specialty Unit, University of California San Francisco, San Francisco, California, USA
3 Division of Occupational and Environmental Medicine, University of California San Francisco San Francisco, California, USA
4 Division of Environmental Health Sciences, School of Public Health, University of California Berkeley, Berkeley, California, USA
5 Center for Environmental Research and Children’s Health, School of Public Health, University of California Berkeley, Berkeley, California, USA
Address correspondence to Stephanie M. Holm, UCSF Division of Occupational and Environmental Medicine, 2330 Post St. #460, San Francisco, CA 94115 USA. Telephone: (415) 514-0878. Email: [email protected]
13 3 2023
3 2023
131 3 03700717 12 2021
19 1 2023
26 1 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Because fine particulate matter [PM, with aerodynamic diameter ≤2.5μm (PM2.5)] is a ubiquitous environmental exposure, small changes in cognition associated with PM2.5 exposure could have great societal costs. Prior studies have demonstrated a relationship between in utero PM2.5 exposure and cognitive development in urban populations, but it is not known whether these effects are similar in rural populations and whether they persist into late childhood.
Objectives:
In this study, we tested for associations between prenatal PM2.5 exposure and both full-scale and subscale measures of IQ among a longitudinal cohort at age 10.5 y.
Methods:
This analysis used data from 568 children enrolled in the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS), a birth cohort study in California’s agricultural Salinas Valley. Exposures were estimated at residential addresses during pregnancy using state of the art, modeled PM2.5 surfaces. IQ testing was performed by bilingual psychometricians in the dominant language of the child.
Results:
A 3-μg/m3 higher average PM2.5 over pregnancy was associated with −1.79 full-scale IQ points [95% confidence interval (CI): −2.98, −0.58], with decrements specifically in Working Memory IQ (WMIQ) and Processing Speed IQ (PSIQ) subscales [WMIQ −1.72 (95% CI: −2.98, −0.45) and PSIQ −1.19 (95% CI: −2.54, 0.16)]. Flexible modeling over the course of pregnancy illustrated mid-to-late pregnancy (months 5–7) as particularly susceptible times, with sex differences in the timing of susceptible windows and in which subscales were most affected [Verbal Comprehension IQ (VCIQ) and WMIQ in males; and PSIQ in females].
Discussion:
We found that small increases in outdoor PM2.5 exposure in utero were associated with slightly lower IQ in late childhood, robust to many sensitivity analyses. In this cohort there was a larger effect of PM2.5 on childhood IQ than has previously been observed, perhaps due to differences in PM composition or because developmental disruption could alter the cognitive trajectory and thus appear more pronounced as children get older. https://doi.org/10.1289/EHP10812
Supplemental Material is available online (https://doi.org/10.1289/EHP10812).
The authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Particulate matter (PM) exposure is ubiquitous in the United States, and newly available estimates of PM that combine ground-based measurements with remotely sensed data1 allow for characterization of exposure of populations, especially in rural areas, for which such estimates were previously unavailable. This characterization of the PM in rural areas is important because the composition of PM can vary between urban and rural areas,2 with the potential for differences in biologic effects.3 Urban areas tend to have more combustion-related particles with a higher content of metals in comparison with rural areas, which have more PM from natural sources and less ultrafine PM.2
There are many known health effects of PM,4 including increasing evidence for neurocognitive effects across the life course. Because childhood is a particularly critical period of rapid brain growth and neurodevelopment,5 recent evidence linking fine PM [PM with aerodynamic diameter ≤2.5μm PM2.5] exposure to decrements in childhood cognitive function is particularly concerning.6,7 Yet, much of what we know about these effects of PM2.5 exposure comes from studies in major metropolitan areas. For example, in a Los Angeles cohort, monthly PM2.5 levels averaged over the 1–3 y prior to the assessments were associated with increased risk of delinquent behavior in adolescents.8 Using a combination of two German cohorts, a large population of adolescents in major cities was also found to have increased risk of hyperactivity and inattention at age 15 y associated with PM2.5 exposure estimated at their childhood (age 10 y) or current home addresses.9 A large reanalysis of multiple European birth cohorts, mostly based in large population centers, did not find any relationship between in utero PM2.5 exposure and neurodevelopmental outcomes; however, their exposure data were back-extrapolated to dates many years prior, raising the possibility that exposure misclassification may have obscured a relationship.10 Multiple studies have also shown that when comparing schools with higher roadway pollution exposure to those with lower, students in lower pollution–exposure schools perform better on cognitive testing even when controlling for socioeconomic status (SES).11,12
The prenatal period could be a particularly critical time for neurodevelopmental insults, given the rapid growth of brain structures during that period. A 2016 systematic review concluded that in utero exposure to urban air pollution was associated with decreases in measured intelligence in preschool-age children,6 and noted that a few studies of air pollution’s effect on neurodevelopment through the life course suggest a larger effect of air pollution exposure on boys in comparison with girls. Higher prenatal PM2.5 exposure levels have been associated with lower cognitive functioning in early childhood (ages 1–6 y),13–15 as has prenatal exposure to larger PM (ages 2–6 y),16,17 to polycyclic aromatic hydrocarbons (age 5 y),18,19 to NO2 (age 7 y),20 and to roadway proximity (age 7–8 y).20,21 To our knowledge, no studies have assessed effects of prenatal exposures on cognitive function later in childhood.
The Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS) Study is a birth cohort study conducted in an agricultural community in California with extensive exposure and health outcome characterization, including prenatal residential address data and detailed neurodevelopmental follow-up throughout childhood. Prenatal exposures to organochlorine pesticides22 and organophosphate pesticides as measured by urinary metabolites,23 ambient exposure to organophosphate pesticides and carbamates as well as other pesticides estimated near the prenatal residence,24 organophosphate flame retardant exposure,25 and polybrominated diphenyl ether (PBDE) flame retardant exposure26,27 have been previously adversely associated with neurocognitive development in the cohort. In this study, our aim was to use PM2.5 estimates—newly available for areas like Salinas with rural and small urban areas—to assign prenatal PM2.5 exposures to CHAMACOS cohort members and assess the relationship with IQ at age 10.5 y, using the rich data available in the cohort on other exposures.
Methods
Study Population
This analysis uses data from the CHAMACOS study, a decades-long birth cohort study in California’s agricultural Salinas Valley, which is part of the Center for Environmental Research and Community Health (CERCH) at the University of California Berkeley (UC Berkeley). Details of the cohort recruitment and follow-up have been previously reported in detail.22,28,29 The CHAMACOS study is approved by the UC Berkeley Institutional Review Board, and informed consent was given by the participating parents on behalf of themselves and their children.
Enrollment into the cohort occurred in two phases. In the first phase (“CHAM1”), pregnant women were recruited from community clinics between October 1999 and October 2000. Inclusion criteria included: being over 18 y of age, being <20 wk pregnant, speaking Spanish or English, qualifying for low-income health insurance, and planning to deliver at the (single) county hospital. The initial CHAM1 cohort included 601 women enrolled in 1999–2000 while the CHAMACOS participant(s) were in utero, of whom 537 live-born infants were followed to delivery. Enrollment into the second phase (“CHAM2”) occurred in 2009–2011, when the original cohort was 9–10 y of age, with an additional 305 9-y-old children recruited, using inclusion criteria to closely match the original cohort. Children recruited into CHAM2 had mothers who: were 18 y old or older at the time of the child’s birth, were Spanish- or English-speaking, were eligible for low-income health insurance at the time of delivery, and were residents of the Salinas Valley at birth (they did not necessarily deliver at the county hospital, though approximately 70% of these mothers did). By the time of the assessment for this study (roughly age 10.5 y), the participants from the two subcohorts have very similar demographics with the participants from the CHAM2 cohort being slightly younger, more male, and of lower SES (Table 1). CHAM1 participants who remained in the cohort for analysis of these outcomes at age 10.5 y were very similar to those lost to follow-up; one area of difference is children whose mothers smoked during pregnancy (or were exposed to smoke during pregnancy) were slightly more likely to be lost to follow up (see Table S1, “Characteristics of the CHAM1 Cohort remaining at 10.5 y”).
Table 1 Characteristics of the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2010–2013, n=568).
Full cohort (n=568) Missing CHAM1 (n=286) CHAM2 (n=282)
Demographics
Age (y) 10.52 (0.18) 0 10.62 (0.18) 10.42 (0.11)
Female 291 (51.2%) 0 152 (53.1%) 139 (49.3%)
Primarily Spanish speaking 181 (31.9%) 0 92 (32.2%) 89 (31.6%)
Household above the poverty level 156 (27.5%) 0 83 (29.0%) 73 (25.9%)
Mother born in Mexico 489 (86.4%) 2 247 (86.4%) 242 (86.4%)
Mother has seventh grade education or higher 327 (57.6%) 0 164 (57.3%) 163 (57.8%)
Maternal Peabody Picture Vocabulary test score 91.6 (19.4) 13 92.8 (18.8) 90.4 (20.0)
Exposures
HOME score at age 10.5 (z-scores) 0.01 (1.01) 0 0.00 (1.05) 0.01 (0.96)
Mother smoked during pregnancy 12 (2.1%) 2 9 (3.1%) 3 (1.1%)
Mother exposed to smoke (secondhand) during pregnancy 54 (9.5%) 2 21 (7.3%) 33 (11.8%)
Outcomes
Full-scale IQ score 89.56 (11.09) 2 90.58 (11.04) 88.53 (11.06)
Verbal Comprehension subscale 86.96 (12.37) 1 87.37 (12.73) 86.53 (12.00)
Perceptual Reasoning subscale 92.65 (13.93) 0 94.10 (13.90) 91.19 (13.84)
Working Memory subscale 91.04 (12.00) 0 92.01 (11.96) 90.05 (11.98)
Processing Speed subscale 98.48 (11.97) 1 98.61 (12.26) 98.35 (11.69)
Note: All cells contain either mean (SD) or number (%).
Study visits were conducted repeatedly every year or two throughout childhood and adolescence. For this study, we use data from the visits at age 10.5 y and included 611 participants, who had no history of neurodevelopmental disease (e.g., down syndrome, hydrocephalus), and underwent cognitive testing at age 10.5 y (320 CHAM1 participants and 291 CHAM2 participants). Of these, we excluded 24 with no prenatal residential history data, 17 born at <36-wk gestation and two full-term infants who were from twin births in which the twin with the lower assigned ID number was selected for this analysis. Gestational age was based on last menstrual period (when that had been reported by the mother AND the resulting gestational age was withing 2 wk of that listed in the medical record) or directly from the medical record. The final sample size for the analysis at 10.5 y was 568 children.
We conducted sensitivity analyses using cognitive outcomes at age 7 y on 310 CHAM1 participants, who had no history of neurodevelopmental disease, sat for neurodevelopmental testing at age 7 y, and had additional prenatal and early-life exposure data that were not available on the CHAM2 cohort. We excluded 11 children who were born at <36-wk gestation and and two full-term infants who were from twin births in which the twin with the lower assigned ID number was selected for this analysis. Thus, the sample size for the 7-y-old sensitivity analysis was 297 children.
Exposure Assessment
The date of conception was estimated from the child’s birth date and estimated gestational age at the time of birth. Information on prenatal residential history was collected prospectively for CHAM1 and at 9-y and 16-y visits for CHAM2. Though every effort was made to ensure accurate and complete residential histories, there is a somewhat higher likelihood of exposure misclassification among CHAM2 participants because parents may be less likely to remember exactly the month in which they moved in or out of a particular address during their child’s gestation. Average PM2.5 exposure was calculated as a continuous variable for each month of gestation at each residential location, rounding to the nearest calendar month because of the availability of pollutant data in calendar months (e.g., if the pregnancy started 5 January, the first month exposure was estimated using the January PM2.5 spatial surface; however, if the pregnancy started 25 January, the first month exposure was estimated using the February PM2.5 surface). Exposures were estimated through the ninth month of gestation (i.e., the 36th wk), because even among full-term infants there may not be a full 10th month of gestation.
We estimated annual ground-level PM2.5 at each residential address from the publicly available data sets provided by the Atmospheric Composition Analysis Group at Washington University in St. Louis, Missouri (the “ACAG data sets”),1 which combine remotely sensed data from satellites, chemical transport models, and ground-based measurements in a geographically weighted regression model. The ACAG data sets include monthly average estimates for PM2.5 across North America beginning in January 2000, at roughly 1-km resolution. Because some infants in the cohort were in utero for a portion of 1999, we needed to calculate monthly surfaces for 1999. Using explained variability from regression models and prior knowledge, we chose to classify areas by whether they are goods movement corridors, have major nongoods movement corridor roadways, or are not characterized by either type of roadway because these differences are known to impact California PM2.5 concentrations.30 After we calculated exposure surfaces for 1999, we applied our method to the year 2000 and compared our calculated 2000 data set to the known ACAG data sets for that year. When comparing our 2000 data to the ACAG 2000 data, our model was highly correlated with the ACAG data (R2=0.8). Testing the method over an additional 10-y period (2000–2010) the R2 was 0.7. We are thus confident that the modeled PM2.5 surfaces for 1999 capture most of the variability in the true values (more detail is available in the Statistical Analysis Plan in the Supplemental Material, Part A, section 6a). A total of 87 children had some prenatal exposure time in the year 1999; out of 5,112 exposure-months in the cohort, 284 (6%) occurred in 1999.
Outcome Assessment
Children’s cognitive abilities were assessed at ages 7 and 10.5 y, using the Wechsler Intelligence Scale for Children, Fourth Edition (WISC-IV) English or Spanish versions as appropriate.31 All assessments at age 7 y were conducted by a single bilingual psychometrician; two bilingual psychometricians conducted the assessments at age 10.5 y. All psychometricians were trained and supervised by a pediatric neuropsychologist. Scores for four domains were calculated based on the following subtests31: Verbal Comprehension (VCIQ, composed of Vocabulary and Similarities subtests), Perceptual Reasoning (PRIQ, Block Design and Matrix Reasoning subtests), Working Memory (WMIQ, Digit Span and Letter-Number Sequencing subtests), and Processing Speed (PSIQ, Coding and Symbol Search subtests). All subtests were administered in the dominant language of the child, which was determined through administration of the Oral Vocabulary subtest of the Woodcock–Johnson/Woodcock–Muñoz Tests of Cognitive Ability in both English and Spanish at the beginning of the assessment.32 The psychometrician was blinded to PM2.5 exposure status. Full-Scale IQ (FSIQ) as well as subscale scores were used for each testing time point, standardized against U.S. population–based norms for English- and Spanish-speaking children. Note that during the first 3 months of assessments at age 7 y, two subtests were not administered, meaning that 27 children are missing working memory and processing speed subscales, and thus FSIQ as well.25
Missing Data
Only children who had at least one prenatal residential address recorded and underwent neurodevelopmental assessments were included in analyses. However, because of some missing residential addresses due to moves during pregnancy, we used multiple imputation by chained equations for missing PM2.5 values for 531 of 5,112 exposure-months (roughly 10% of the months).33 We imputed these using predictive mean matching with 5 donors, for each month of pregnancy. The whole-pregnancy mean was then calculated as an average of the full set of months, once each missing month had been imputed. We also used multiple imputation for missing covariate data.
Statistical Analyses
All analyses were done in accordance with a prespecified analysis plan that was posted to Open Science Framework prior to the start of any exposure–outcome analyses (prespecified analysis plan with versioning and comments is available at https://osf.io/zwbgs/?view_only=a01c321901024b4ab7bbc9e82920d025; the most recent version is also Part A in the Supplemental Material).
The tidyverse implemented in R (version 3.6.3, codenamed “Holding the Windsock”; The R Foundation for Statistical Computing)34 was used for all exposure modeling and the analysis of the exposure–outcome relationship (see Supplemental Material, Part A, section 6).
Confounder selection was prespecified based on subject matter knowledge and encoded in a directed acyclic graph (Figure S1, “Directed Acyclic Graph”); all recoding of variables was also prespecified. The analyses included the following covariates to block backdoor confounding pathways: maternal verbal intelligence (assessed when the children were 9 y of age), maternal education (dichotomized into ≥seventh grade education, or sixth grade education and lower), household poverty (dichotomized to less than or equal to the poverty level vs. above the poverty level, at the time of the neurocognitive testing, but as a proxy for lifetime poverty) and Home Observation for Measurement of the Environment (HOME) inventory35,36 (assessed at the time of the neurocognitive testing but as a proxy for lifetime HOME). In addition, we included age, sex assigned at birth (binarized into female and male), language of assessment, and smoke exposure in utero (binary variables for whether the mother smoked and whether she was exposed to smoking during pregnancy) because these are expected to be strongly associated with the outcome variable.
Rothman and Greenland recently suggested that study design decisions should be based on precision rather than power,37 because precision estimates do not change based on the alternative hypothesis considered. Power and precision analyses were calculated for the 10.5-y assessment (Supplemental Material, Statistical Analysis Plan Part A, section 3b, and Figure S2, “Power for analyses using Distributed Lag Models”). Because of variable degrees of correlation between the exposure at the various months of gestation, the power to detect a 3-point difference in IQ score with a 3-μg/m3 difference in PM2.5 ranged between ∼20%–99%, with most months of gestation having >80% power. Thus, we also calculated the precision expected at each month of gestation (e.g., the full width of the confidence interval); these ranged from 2.2 to 6.0 IQ points for a 3-μg/m3 increase in PM2.5 (Supplemental Material, Part A, section 3b). A 3-μg/m3 increase in the PM2.5 exposure was chosen because this is approximately the interquartile range (IQR) of the whole pregnancy average in this sample.
The main analyses were done using distributed lag models (DLM), which allowed us to include all the monthly exposure lags in a single analysis, thereby controlling for exposure during all other months.38–40 We assumed that the relationship between PM2.5 and IQ was linear across our range of PM2.5 values, but we allowed the shape of the time-lag dimension to vary flexibly. We examined multiple options for smoothing functions (including allowing nonlinearity in the exposure dimension) and examined the Akaike Information Criteria (AIC).41 The model specification with the best model fit (lowest AIC) in both the 7-y and 10.5-y analyses used b-splines for the smoothing functions (Table S2, “Akaike Information Criteria for Various Smoothing Functions”). Linear models were also used to associate whole-pregnancy average PM2.5 exposure with IQ and subscales, and assumptions were checked.
Even though the power and precision analyses were conducted for the larger, 10.5-y cohort, we made the decision a priori to also analyze the 7-y data, which consisted of only CHAM1 participants (n=297), because there are more early-life variables available for sensitivity analysis. To assess for changes in our results based on misclassification of covariates used at the time of assessment, HOME score and poverty category assessed at age 6 months instead of at the time of assessment were used together to perform a sensitivity analysis. Presence of a gas stove while the child was in utero (with or without a working range hood) could be associated with IQ (but not ambient prenatal PM2.5 exposure) and thus could affect the precision of our estimate. A separate sensitivity analysis was performed to assess for meaningful changes in the precision of our estimates after adding a variable for gas stove presence. Finally, exposure to pesticides and other environmental contaminants can vary seasonally, and multiple chemicals have previously been associated with neurodevelopmental outcomes in the CHAMACOS cohort. Thus, to assess for potential confounding by these environmental exposures, these variables were added (one at a time) to further sensitivity analyses: mean dialkyl phosphate metabolites (DAPs) measured in parental urine during pregnancy,23 mean organophosphate flame retardants measured in maternal urine during pregnancy,25 and prenatal polybrominated diethyl ether (PBDE) concentrations in nanograms per gram lipid measured during pregnancy26 or estimated from mothers’ levels when children were 9 y of age.27,42 The intent of this set of sensitivity analyses at the 7-y assessment was to look for major discrepancies in the pattern of associations between PM2.5 exposure and IQ between the 7-y and 10.5-y analyses and to identify other exposures that may bias the results in the analysis of the children at age 10.5 y. To achieve this goal, we also performed a post hoc sensitivity analyses on the data collected at age 10.5 y, restricted to the CHAM1 participants only (n=286).
Prespecified sensitivity analyses with the 10.5-y IQ measurements from both CHAM1 and CHAM2 included the addition (one at a time) of prenatal organophosphate and carbamate pesticide exposure estimated as described previously24,43 from the California Department of Pesticide Regulation pesticide use reporting (PUR) data available with both spatial (1 square mile) and temporal (daily) resolution. We also added each of prenatal concentrations of PBDEs and DDE and DDT in nanograms per gram lipid measured during pregnancy or estimated from maternal levels when their children were 9 y of age. These analyses were performed to assess for potential confounding of the relationships by pesticides that have previously been shown to related to neurodevelopmental outcomes in the CHAMACOS cohort.
Additional post hoc sensitivity analyses compared a) results between the entire 10.5-y cohort and only those children whose entire in utero period was in 2000; b) results excluding the younger of 14 pairs of siblings; and c) results accounting for either duration of schooling or season of birth as covariate.
The only planned subgroup analysis was in the 10.5-y data, where we analyzed the PM2.5–IQ relationship stratified by fetal sex.
Results
This cohort has many children of Mexican mothers (86% of mothers were born in Mexico) (Table 1). The cohort is low-income (72.5% of households at or below the poverty line), and nearly 40% of mothers had less than a seventh grade education. Though a large majority of recruited mothers (91% in CHAM1) primarily spoke Spanish as their dominant home language, by age 10.5 y roughly two-thirds of children had transitioned to English as their dominant academic language, meaning that they tested higher in English than Spanish on the screener and thus completed IQ testing in English. Mothers lived at 1.3 different addresses on average during pregnancy, and prenatal PM2.5 exposure averaged 10.6 μg/m3 (Table 2), less than the current annual U.S. Environmental Protection Agency standard for fine particles (12 μg/m3).4
Table 2 Exposure summary for the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2011–2012, n=568).
Exposure window Ambient PM2.5 (μg/m3) Missing
Mean (SD) 25th, 50th, 75th percentiles
Overall in utero 10.63 (2.25) 8.94, 10.62, 12.53 246
First month of pregnancy 8.85 (4.89) 5.64, 6.47, 9.47 230
Second month of pregnancy 9.91 (5.45) 6.04, 7.69, 11.83 78
Third month of pregnancy 10.11 (5.61) 6.08, 7.95, 13.02 64
Fourth month of pregnancy 10.58 (5.70) 6.14, 8.32, 13.72 55
Fifth month of pregnancy 10.83 (5.75) 6.26, 8.40, 14.59 51
Sixth month of pregnancy 10.92 (5.92) 6.36, 8.36, 14.89 45
Seventh month of pregnancy 10.93 (5.67) 6.35, 8.92, 13.69 44
Eighth month of pregnancy 11.09 (5.74) 6.84, 8.99, 13.66 42
Ninth month of pregnancy 11.48 (6.36) 6.91, 8.90, 14.10 94
Addresses where the mothers lived while pregnant with the cohort children are mapped (Figure 1). Though prenatal addresses are located throughout the Salinas Valley, many are clustered in the city of Salinas. As the map indicates, the presence of disparate ambient exposure levels at very similar residential locations (without a clear spatial pattern) makes clear that there is a substantial temporal component to the PM2.5 exposure received while in utero (i.e., seasonal variability), even though there is minimal seasonal variation in temperature in this area.
Figure 1. Map of prenatal addresses for the CHAMACOS cohort (n=568). The larger map shows the entire Salinas Valley, and in the inset is the city of Salinas. The locations of CHAMACOS households are indicated with dots; these are shaded to represent the mean ambient PM2.5 exposure that an in utero CHAMACOS participant received at this address (with all in utero periods occurring between 1999–2002). All residential locations have had a small amount of random noise added to their location to protect participant privacy (i.e., have been jittered). This addition is the reason for a few implausible residential locations, such as in agricultural fields and on the airport runway. One address is not shown; that address was in the Los Angeles Valley and associated with a prenatal PM2.5 exposure between 25 and 30 μg/m3. Base map and data from OpenStreetMap and OpenStreetMap Foundation (https://www.openstreetmap.org/). Note: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas.
Figure 1 is a set of two maps of Salinas Valley, California. On the left, the map depicts a broad view of the Salinas Valley. A scale depicts kilometers ranging from 0 to 50 in increments of 10. On the right, the map is the inset city of Salinas depicting the households included in the study. The dots indicating household locations are colored to depict mean ambient particulate matter begin subscript 2.5 end subscript exposure that an in-utero participant experienced during 1999 to 2002). A scale depicts kilometers ranging from 0.0 to 2.5 in increments of 0.5. The Prenatal mean particulate matter begin subscript 2.5 end subscript concentration (microgram per meter cubed) is broken into six categories, namely, 0 to 5, 5 to 10, 10 to 15, 15 to 20, 20 to 25, and 25 to 30.
Linear models associating whole-pregnancy PM2.5 exposure with IQ at age 10.5 y demonstrated lower FSIQ [−1.79 IQ points (95% CI: −2.98, −0.58)] and WMIQ [−1.72 (95% CI: −2.98, −0.45)] associated with 3-μg/m3 higher PM2.5 (3 μg/m3 is roughly the IQR difference in this sample). Inverse associations were also observed with VCIQ [−1.25 (95% CI: −2.61, 0.11)], PRIQ [−1.14 (95% CI: −2.74, 0.46)], and PSIQ [−1.19 (95% CI: −2.54, 0.16)], though the CI crossed the null for these indices (Figure 2).
Figure 2. Estimated difference in FSIQ and subscale IQ at age 10.5 y associated with 3-μg/m3 higher PM2.5 exposure averaged over all of pregnancy for the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2010–2013, n=568). Note: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; FSIQ, full-scale IQ.
Figure 2 is an error bar graph titled “estimated change in I Q at age 10.5 years (full-scale and subscales)”, plotting estimated difference in I Q points associated with a 3 micrograms per meter cubed interval of particulate matter begin subscript 2.5 end subscript, ranging from negative 3 to 0 in unit increments (y-axis) across full scale, verbal comprehension, perceptual reasoning, working memory, and processing speed (x-axis).
DLMs allowed the relationship between PM2.5 and IQ to vary over gestation, and the period from the fifth to seventh months of gestation emerged as the most susceptible window of exposure for potential effects of PM2.5 exposure on FSIQ [Figure 3; Table S3, “Main Distributed Lag Model Results for Full Scale IQ (age 10.5)”]. The figure presents the point estimate for IQ points per 3-μg/m3 higher average ambient PM2.5 at each month of gestation (the solid line), with the 95% CI for those estimates in the shaded gray area. For example, at the sixth month following conception, a 3-μg/m3 higher average ambient PM2.5 at the residential address is associated with a 2.4-point lower FSIQ at age 10.5 y (95% CI: −4.0, −0.8).
Figure 3. Estimated difference in FSIQ points at age 10.5 y associated with 3-μg/m3 higher PM2.5 exposure each month of gestation, controlling for exposure at other months and flexibly modeling in the time dimension. These are calculated from the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2010–2013, n=568). The solid line is the point estimate for the difference in IQ points at each month from distributed lag models, with the shaded area representing the 95% confidence interval of the estimate. Note: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; FSIQ, full-scale IQ.
Figure 3 is a ribbon plus line graph titled “full-scale I Q difference associated with 3 micrograms per meter cubed interval in particulate matter begin subscript 2.5 end subscript”, plotting estimated difference in full-scale I Q points at age 10.5 years, ranging from negative 4 to 2 in increments of 2 (y-axis) across months post-conception, ranging from 1 to 9 in unit increments (x-axis).
The pattern between PM2.5 and the subscales show different patterns over the course of gestation [Figure 4; Table S4, “Main Distributed Lag Model Results for IQ Subscales (age 10.5)”]. The Verbal Comprehension and Working Memory indices show a pattern similar to that of FSIQ; Verbal Comprehension and Working Memory scores are also lower with higher PM2.5, with the largest effects seen at mid- to late gestation (5–7 months for VCIQ and 4–8 months for WMIQ). The patterns are less clear in Perceptual Reasoning and Processing Speed (with CIs that cross the null for all of gestation). Results were robust to the exclusion of the participants for whom a portion of the in utero period occurred in the year 1999 and also to the exclusion of children who had a sibling in the cohort (Table S5; Figure S3, “Sensitivity Analyses using LM and DLM models excluding children who: (a) had prenatal exposure time in 1999” and “(b) had a nontwin sibling in the cohort”).
Figure 4. Estimated difference in IQ subscales at age 10.5 y associated with 3-μg/m3 higher PM2.5 exposure each month of gestation, controlling for exposure at other months, and flexibly modeling in the time dimension. These are calculated from the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2010–2013, n=568). The solid line is the point estimate for the difference in IQ points at each month from distributed lag models, with the shaded area representing the 95% confidence interval of the estimate. (A) VCIQ (B) PRIQ (C) WMIQ and (D) PSIQ. Note: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; PRIQ, perceptual reasoning IQ; PSIQ, processing speed IQ; VCIQ, verbal comprehension IQ; WMIQ, working memory, IQ.
Figures 4A to 4D are ribbon plus line graphs titled verbal comprehension, perceptual reasoning, working memory, and processing speed under I Q difference associated with a 3 micrograms per meter cubed interval in particulate matter begin subscript 2.5 end subscript, plotting estimated difference in I Q subscales at age 10.5 years, ranging from negative 4 to 2 in increments of 2 (y-axis) across months post-conception, ranging from 1 to 9 in unit increments (x-axis), respectively.
When considering the exposure–outcome relationship based on the binary sex of the fetus (female vs. male), there are different patterns in the effects of PM2.5 on IQ subscale scores in childhood (Figure 5). A 3-μg/m3 higher whole-pregnancy PM2.5 is associated with lower scores in Verbal Comprehension and Working Memory for males [in males, VCIQ −2.16 (95% CI: −4.28, −0.05); WMIQ −2.54 (95% CI: −4.50, −0.59)] in comparison with females [in females, VCIQ −0.32 (95% CI: −2.07, 1.44); WMIQ −1.24 (95% CI: −2.89, 0.42)], whereas females show larger decrements in PSIQ in comparison with males [female −2.42 (95% CI: −4.14, −0.69); male 0.20 (95% CI: −1.96, 2.35)]. CIs for sex by PM2.5 interaction terms for these three subscales (VCIQ, WMIQ, PSIQ) crossed the null, though the majority of the confidence band is below zero for VCIQ and WMIQ and above zero for PSIQ [per 3 μg/m3 for males in comparison with females, there was an additional difference of −1.95 (95% CI: −4.70, 0.78) for VCIQ; −1.45 (95% CI: −4.00, 1.09) for WMIQ; 2.50 (95% CI: −0.21, 5.22) for PSIQ]. Lower IQ associated with PM2.5 exposure was similar in males and females for both FSIQ and PRIQ [male FSIQ −1.95 (95% CI: −3.85, −0.04); female FSIQ −1.56 (95% CI: −3.10, −0.01); male PRIQ −0.90 (95% CI: −3.40, 1.60); female PRIQ −1.09 (95% CI: −3.18, 1.01)]. Differences between males and females are also present in the time patterning of exposure (Figure 6; Table S6, “Distributed Lag Model Results for Full Scale IQ (age 10.5), stratified by sex”), with females showing a pattern of increasing difference in IQ associated with PM2.5 exposure throughout gestation, whereas males do not show a clear pattern and have a suggestion of a peak difference earlier in gestation (third–fourth months). A similar difference is present across all four subscales, with female fetuses having a steady decrease in IQ associated with PM2.5 exposure across gestation and male fetuses having a nonsignificant nadir around the fourth month of gestation (Figure S4, “Sex Differences in the time patterning of PM2.5 effects on IQ across the different subscales,” and Table S7, “Distributed Lag Model Results for IQ subscales, stratified by sex”).
Figure 5. Estimated difference in FSIQ and subscale IQ at age 10.5 y associated with 3-μg/m3 higher PM2.5 exposure averaged over all of pregnancy, stratified by assigned sex of the fetus, dichotomized as female vs. male. These are calculated from the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2010–2013, n=568). Note: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; FSIQ, full-scale IQ.
Figure 5 is an error bar graph titled “estimated difference in I Q at age 10.5 years, grouped by fetal sex (full-scale and subscales)”, plotting estimated difference in I Q points associated with a 3 micrograms per meter cubed interval of particulate matter begin subscript 2.5 end subscript, ranging from negative 2.5 to 2.5 in increments of 2.5 (y-axis) across Full-Scale Intelligence Quotient, Verbal Comprehension Intelligence Quotient, Perceptual Reasoning Intelligence Quotient, Working Memory Intelligence Quotient, and Processing Speed Intelligence Quotient (x-axis) for sex, including female and male.
Figure 6. Estimated difference in FSIQ at age 10.5 y associated with 3-μg/m3 higher PM2.5 exposure each month of gestation, controlling for exposure at other months and flexibly modeling in the time dimension. These are calculated from the CHAMACOS cohort in the Salinas Valley, California, at the 10.5-y analysis (2010–2013, n=568). The solid line is the point estimate for the difference in IQ points at each month from distributed lag models, with the shaded area representing the 95% confidence interval of the estimate. (A) results for those assigned female at birth and (B) results for those assigned male at birth. Note: CHAMACOS, Center for the Health Assessment of Mothers and Children of Salinas; FSIQ, full-scale IQ.
Figure 6a and 6b are ribbon plus line graphs titled Full-Scale Intelligence Quotient female and Full-Scale Intelligence Quotient male under full-scale I Q difference associated with a 3 micrograms per meter cubed interval in particulate matter begin subscript 2.5 end subscript, plotting estimated difference in full-scale I Q subscales at age 10.5 years, ranging from negative 5 to 5 in increments of 5 (y-axis) across months post-conception, ranging from 1 to 9 in unit increments (x-axis), respectively.
Analyses of the smaller, CHAM1-only cohort at age 7 y showed a similar pattern of results in both linear and nonlinear models, though CIs were wider, as expected (Figure S5 and Table S8, “Linear model results for IQ (full and subscales) testing at age 7”; Figure S6 and Table S9, “DLM model results for IQ (full and subscales) testing at age 7”). The subcohort present at age 7 y was similar to those tested at 10.5 y, except that more children were using Spanish as their primary language at age 7 y than at age 10.5 y. (see Table S10, “Characteristics of the Cohort at the 7 y analysis”).
Sensitivity analyses using the CHAM1 data and IQ at age 7 y generally showed minimal differences in the PM2.5–IQ association after adjusting for other exposures (Figure S7; Table S11, “Sensitivity Analyses, DLM, among CHAM1 participants only, age 7”). PM2.5 was associated with a larger IQ difference in the later months of pregnancy after adjusting for urinary DAPs or for the income and HOME score assessed in infancy rather than at the time of neurodevelopmental testing. In addition, the association between PM2.5 and Processing Speed was moved toward the null once organophosphate flame retardants were added to the model, but the point estimate for the PM2.5-Processing Speed association was near null regardless.
Based on findings in the sensitivity analyses from the CHAM1 cohort, the PUR estimate of nearby agricultural pesticide use was added as a sensitivity analysis at the 10.5-y assessment for CHAM1 and CHAM2. The sensitivity analysis using organophosphate and carbamate pesticide exposure as estimated with PUR data was unremarkable, and there were also minimal changes in the PM2.5–IQ relationships seen after adjusting for maternal serum PBDEs, DDE, or DDT (Figure S8; Table S12, “Sensitivity Analyses, DLM, among full cohort at age 10.5”).
Sensitivity analyses using the CHAM1 data and IQ at age 10.5 y also generally showed minimal differences in the PM2.5–IQ association after adjusting for other exposures (Figure S9; Table S13, “Sensitivity Analyses, DLM, CHAM1, age 10.5”). PM2.5 in late in pregnancy was associated with even lower FSIQ after adjusting for gas stoves in the home and when using the income and HOME score assessed in infancy rather than at the time of neurodevelopmental testing. In addition, the association between PM2.5 and processing speed was moved toward the positive once organophosphate flame retardants were added to the models (it moved the estimate from negative to the null for most months, but from null into the positive for the last month of pregnancy), but the association between PM2.5 and processing speed were near null regardless.
Finally, sensitivity analyses exploring the effect of season of birth on the PM2.5–IQ association suggest that the pattern of the relationships appear to also be robust to the inclusion multiple different markers for seasonality, though in some cases the results are attenuated. These analyses included explicitly controlling for season of birth and adjusting for schooling using the best estimate of schooling available to us [parent-reported grade at testing (Figure S10; Table S14, “Sensitivity Analyses, LM and DLM, accounting for schooling and season of birth”)].
Discussion
In this large, well-characterized cohort of preteens living in a rural to semirural agricultural community, we found lower WISC FSIQ and some subscales associated with average in utero exposure to fine PM. Using models that varied flexibly in the time dimension, we demonstrated that childhood IQ was particularly associated with PM2.5 exposure in mid- to late pregnancy (months 5–7). These results were robust to multiple sensitivity analyses.
An interesting finding was that the PM2.5−IQ relationship may be modified by the sex of the fetus, with males having lower verbal comprehension and working memory associated with average prenatal exposure to PM2.5, and females having lower processing speed, though all had lower FSIQ. The prenatal period with highest susceptibility also seemed to vary between male and female fetuses: The lowest childhood IQ–PM2.5 association occurred earlier in those assigned male at birth than in those assigned female. However, these differences between the association for male and female fetuses may also reflect some random variability in the exposures and outcomes within the sample.
Several other cohort studies have found decrements in cognition associated with prenatal PM2.5 exposure, though these have been in younger children and in more urban environments. A birth cohort in Mexico City found that prenatal PM2.5 exposure was negatively associated with cognitive and language development at multiple testing points through age 2 y, with the largest difference seen for exposure in the third trimester.14 A birth cohort in multiple Spanish cities demonstrated a negative association between prenatal PM2.5 exposure and cognition at age 15 months15 as well as with measures of memory at age 4–6 y, but the latter only among children identified as male.44 A recent study in New York assessed the relationship between prenatal PM2.5 exposure and IQ at age 6.5 y in a birth cohort that, like CHAMACOS, was low-income and largely Hispanic, but which was smaller and located in an urban area.13 Using a flexible modeling strategy that was similar to ours, those investigators also showed a pattern of lower IQ score associated with higher PM2.5 exposure late in gestation (after approximately 30 wk). We were intrigued to see that their pattern for changes in FSIQ with late-pregnancy PM2.5 exposure suggested a larger difference in children identified as boys (with a 1-2 point lower FSIQ associated with a 10-μg/m3 increase in PM2.5), whereas we found a larger effect in those identified as female at birth with late gestation exposure. In a birth cohort in Italy, no significant relationships were found between prenatal PM2.5 exposure and IQ at age 7 y, though there was a trend toward lower Perceptual Reasoning and Processing Speed subscales [a 10-μg/m3 higher average pregnancy PM2.5 was associated with 3.1-point lower perceptual organization (95% CI: −9.5, 3.4) and a 4.0 lower processing speed (95% CI: −10, 2.4)].20 A prospective birth cohort in urban and suburban Massachusetts found that prenatal traffic-related pollution exposure was associated with lower cognition at age 8 y but did not find a specific relationship with PM2.5 exposure.45
Sensitivity analyses suggest that our results are robust to consideration of a variety of other chemicals to which this population has been exposed, as well as to excluding portions of the cohort. As suggested by the sensitivity analysis using the HOME score and poverty data from infancy (among those cohort members for whom it was available), the use of these variables from the time of testing may have introduced some misclassification that biased our results to the null. Thus, the use of these variables from late childhood may mean that we are underestimating the effects of prenatal PM2.5 on IQ. Though explicitly accounting for season in the model moves the estimates toward the null, there is still a pattern of decreases in IQ scores associated with PM2.5 exposures. Moreover, because the exposure is seasonal, including seasonality in the model may be an overadjustment, adjusting away some of the true relationship between PM2.5 and IQ. The fact that the pattern of decreases in IQ remains, even with potential overadjustment, suggests that there may be a true relationship between PM2.5 and IQ, acknowledging that it may be only one factor among a seasonal milieu of factors that affect IQ.
Though we are not aware of any other studies associating prenatal PM2.5 exposure with cognitive function in preteen years/late childhood, a recent neuroimaging study demonstrated white matter changes in 9–12 y old children was associated with PM2.5 exposure in utero,46 providing further support for neurocognitive effects later in childhood, as we have found. Thus, to our knowledge, our study is the first in a cohort of preteens outside a major metropolitan area, and we found a somewhat larger effect of PM2.5 on childhood IQ than has previously been seen. Because less-urbanized settings have more PM2.5 from sources not related to traffic or power generation, such as biomass burning and windblown dust, and fewer combustion-related particles,2 these differences in effect could be the result of differences in PM composition, or that developmental disruption causes effects that alter the cognitive trajectory and thus appear more pronounced as children get older.
Recently demonstrated effects of PM2.5 on neurodevelopment using laboratory and animal data provide a strong scientific premise for an association between PM2.5 exposure and neurodevelopment in children.47 PM is directly toxic to human neurons,48 can change neuron gene expression,49,50 and may have a role in central nervous system myelination as well.51 These direct effects are problematic, given that PM (especially the ultrafine particulate fraction, those <0.1 microns in diameter) can enter the systemic circulation, can cross the blood–brain barrier, and has been found in the brain parenchyma.47 In addition to potential direct effects of PM, increases in inflammation and oxidative stress secondary to air pollution exposures have long been associated with cardiovascular and lung diseases but are increasingly recognized as contributors to central nervous system pathology as well.52 Rodents have demonstrated increases in neuroinflammation following exposure to fine particles, including prenatal exposure.53–55 In particular, prenatal diesel exhaust exposure, which has a large PM component,47 induced long-term changes in the microglia, the resident immune cells in the brain, into adulthood.53 Rodents exposed to PM have also shown deficits on tests of memory.56,57 Furthermore, young people from urban centers in Mexico who died accidentally had more neuroinflammation on autopsy than those from smaller cities, a finding the authors ascribe to long-term air pollution exposure, mirroring the animal studies.58 It is known that the functional connectivity of the human fetal brain increases substantially in the second half of pregnancy,59,60 meaning this is a period that could be particularly susceptible to environmental insults.
How brain growth and development relate to biologic sex is incompletely understood, with potential roles for hormonal modifications, direct effects of genes on sex chromosomes, and epigenetic differences.61,62 Studies in transgender people, which often show structural brain patterns more congruent with their experienced gender than with their assigned sex, highlight how much there is to learn in this area.63 Yet, studies using binary categorizations of sex into male and female suggest that fetal brain development has some differences associated with fetal sex, including differences in neuronal connectivity,64 which could differentially affect susceptibility to environmental insults. Animal studies have also demonstrated sex differences in critical windows of PM exposure; for example in a group of rats exposed to ultrafine particles, the male rats had increased impulsivity when exposed during the period of neurodevelopment, whereas female rats had these effects when exposed in adulthood.56 As mentioned above, microglial cells are thought to play a key role in priming of the neurological system by air pollution,65 and sex differences in the activation of microglial cells have been noted, with larger effects seen in male rodents. Thus, sex differences in the effects of PM2.5 exposure on neurodevelopment and cognition are entirely plausible. Our results suggest that the effects are somewhat different depending on the subscale considered, and that differences in the timing of exposure could be relevant.
Small changes in cognition associated with air pollution would be particularly important to understand because of the ubiquity of the exposure. Studies of other environmental exposures, such as lead, have indicated a large social cost to the loss of IQ points.66 For example, a Belgian study estimated that in a population of adolescents from 2003 to 2004, among those with elevated lead there was an average IQ loss of 1.67 points per individual, and that this had a social cost of 1.8 billion Euros per 100,000 people.66
Strengths of this study include the use of a well-characterized cohort in which we were able to conduct many sensitivity analyses to look for changes in the PM2.5–IQ relationship associated with coexposures. We also have robust neuropsychological testing in the primary language of the child, whether Spanish or English. The use of newly available pollution surfaces that provide good spatial variability in rural areas allowed us to assess a cohort that previously had no such exposure assessment available. In addition, the flexible modeling strategy allowed us to evaluate exposures at multiple time points in a conservative manner, such that each month’s exposure–response relationship controlled for exposure at the other months and allowed for identification of a susceptible window (5–7 months) that might not have been apparent if using the common trimester periods. If the body of literature were able to clearly establish susceptible windows, these could be used to counsel pregnant people to especially minimize air pollutant exposure during those periods.
Although the unique features of this cohort (including rural and semirural location and older children with prenatal exposure data) add a new dimension to the extant literature, these features also limit the generalizability of findings from the cohort, meaning that the findings may not generalize to the entire population of children in the United States, especially those living in more urban environments. It would be valuable to have further study in a more representative cohort, especially one large enough to analyze births occurring within a single season. Because residential history was incomplete for some of the cohort, we had to impute data for those exposures; uncertainty due to the multiple imputation has been included in the results.
In summary, in this large, well-characterized cohort of rural preteens growing up in an agricultural area, we have found that slightly higher outdoor PM2.5 exposure in utero was associated with small decrements in IQ in late childhood. Though a 1- or 2-point change in IQ is unlikely to be meaningful for an individual, shifting the entire distribution of IQ down by a point or two could greatly change the number of individuals that qualify for intervention services. Because air pollution exposures are inequitably distributed, with low-income communities and communities of color exposed to higher levels of air pollution,67 shifting the entire distribution of IQ down for these populations could have important implications, further contributing to systemic injustices. Our findings suggest that, at levels allowable within current U.S. EPA air quality standards, small fluctuations in exposure to fine PM might have enduring changes on childhood cognition.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank K. Long and S. Rauch, the CHAMACOS data managers, for their prompt replies to queries and unending patience as we worked through this analysis. Eternal gratitude to the CHAMACOS families for allowing us snapshots of their lives and, in doing so, helping to teach us all.
The collection of data used in this analysis was supported by the following grants: National Institute of Environmental Health Sciences (NIEHS) P01-ES009605, NIEHS R01-ES015572, NIEHS R01-ES017054, U.S. EPA RD 83171001, U.S. EPA RD 82670901, U.S. EPA RD 83451301, National Institute for Occupational Safety and Health R01 OH007400. This analysis did not receive any specific support from funding agencies in the public, commercial, or not-for-profit sectors.
The contents of this publication are solely the authors’ responsibility and do not necessarily represent the official views of the NIEHS, National Institutes of Health (NIH), U.S. EPA, or the Centers for Disease Control and Prevention.
This manuscript is based on a chapter of a dissertation published by the University of California Berkeley in partial fulfillment of S.M.H.’s doctoral degree, but that dissertation is currently embargoed and not publicly available.
These data (de-identified) and associated documentation/metadata will be made available to others in the scientific community by the principal investigator of the study, under a data-sharing agreement consistent with both the University of California and NIH data-sharing policies.
==== Refs
References
1. van Donkelaar A, Martin RV, Li C, Burnett RT. 2019. Regional estimates of chemical composition of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors. Environ Sci Technol 53 (5 ):2595–2611, PMID: , 10.1021/acs.est.8b06392.30698001
2. Castanheiro A, Wuyts K, Hofman J, Nuyts G, De Wael K, Samson R. 2021. Morphological and elemental characterization of leaf-deposited particulate matter from different source types: a microscopic investigation. Environ Sci Pollut Res 28 (20 ):25716–25732, PMID: , 10.1007/s11356-021-12369-z.33471309
3. Hime NJ, Marks GB, Cowie CT. 2018. A comparison of the health effects of ambient particulate matter air pollution from five emission sources. Int J Environ Res Public Health 15 (6 ):1206, PMID: , 10.3390/ijerph15061206.29890638
4. U.S. Environmental Protection Agency, National Center for Environmental Assessment RTP Division. 2019 . Integrated Science Assessment (ISA) for Particulate Matter (Final Report, Dec 2019). Washington, DC: U.S. Environmental Protection Agency. https://cfpub.epa.gov/ncea/isa/recordisplay.cfm?deid=347534 [accessed 7 May 2020].
5. Suades-González E, Gascon M, Guxens M, Sunyer J. 2015. Air pollution and neuropsychological development: a review of the latest evidence. Endocrinology 156 (10 ):3473–3482, PMID: , 10.1210/en.2015-1403.26241071
6. Clifford A, Lang L, Chen R, Anstey KJ, Seaton A. 2016. Exposure to air pollution and cognitive functioning across the life course-a systematic literature review. Environ Res 147 :383–398, PMID: , 10.1016/j.envres.2016.01.018.26945620
7. WHO (World Health Organization). 2018. Air pollution and child health: prescribing clean air. https://www.who.int/publications-detail-redirect/air-pollution-and-child-health [accessed 27 March 2021].
8. Younan D, Tuvblad C, Franklin M, Lurmann F, Li L, Wu J, et al. 2018. Longitudinal analysis of particulate air pollutants and adolescent delinquent behavior in Southern California. J Abnorm Child Psychol 46 :1283–1293, 10.1007/s10802-017-0367-5.29234991
9. Fuertes E, Standl M, Forns J, Berdel D, Garcia-Aymerich J, Markevych I, et al. 2016. Traffic-related air pollution and hyperactivity/inattention, dyslexia and dyscalculia in adolescents of the German GINIplus and LISAplus birth cohorts. Environ Int 97 :85–92, PMID: , 10.1016/j.envint.2016.10.017.27835751
10. Guxens M, Garcia-Esteban R, Giorgis-Allemand L, Forns J, Badaloni C, Ballester F, et al. 2014. Air pollution during pregnancy and childhood cognitive and psychomotor development: six European birth cohorts. Epidemiology 25 (5 ):636–647, PMID: , 10.1097/EDE.0000000000000133.25036432
11. Sunyer J, Esnaola M, Alvarez-Pedrerol M, Forns J, Rivas I, López-Vicente M, et al. 2015. Association between traffic-related air pollution in schools and cognitive development in primary school children: a prospective cohort study. PLoS Med 12 (3 ):e1001792, PMID: , 10.1371/journal.pmed.1001792.25734425
12. Wang S, Zhang J, Zeng X, Zeng Y, Wang S, Chen S. 2009. Association of traffic-related air pollution with children’s neurobehavioral functions in Quanzhou, China. Environ Health Perspect 117 (10 ):1612–1618, PMID: , 10.1289/ehp.0800023.20019914
13. Chiu Y-HM, Hsu H-HL, Coull BA, Bellinger DC, Kloog I, Schwartz J, et al. 2016. Prenatal particulate air pollution and neurodevelopment in urban children: examining sensitive windows and sex-specific associations. Environ Int 87 :56–65, PMID: , 10.1016/j.envint.2015.11.010.26641520
14. Hurtado-Díaz M, Riojas-Rodríguez H, Rothenberg SJ, Schnaas-Arrieta L, Kloog I, Just A, et al. 2021. Prenatal PM2.5 exposure and neurodevelopment at 2 years of age in a birth cohort from Mexico City. Int J Hyg Environ Health 233 :113695, PMID: , 10.1016/j.ijheh.2021.113695.33582606
15. Lertxundi A, Baccini M, Lertxundi N, Fano E, Aranbarri A, Martínez MD, et al. 2015. Exposure to fine particle matter, nitrogen dioxide and benzene during pregnancy and cognitive and psychomotor developments in children at 15 months of age. Environ Int 80 :33–40, PMID: , 10.1016/j.envint.2015.03.007.25881275
16. Kim E, Park H, Hong Y-C, Ha M, Kim Y, Kim B-N, et al. 2014. Prenatal exposure to PM10 and NO2 and children’s neurodevelopment from birth to 24 months of age: Mothers and Children’s Environmental Health (MOCEH) study. Sci Total Environ 481 :439–445, PMID: , 10.1016/j.scitotenv.2014.01.107.24631606
17. Yorifuji T, Kashima S, Higa Diez M, Kado Y, Sanada S, Doi H. 2016. Prenatal exposure to traffic-related air pollution and child behavioral development milestone delays in Japan. Epidemiology 27 (1 ):57–65, PMID: , 10.1097/EDE.0000000000000361.26247490
18. Edwards SC, Jedrychowski W, Butscher M, Camann D, Kieltyka A, Mroz E, et al. 2010. Prenatal exposure to airborne polycyclic aromatic hydrocarbons and children’s intelligence at 5 years of age in a prospective cohort study in Poland. Environ Health Perspect 118 (9 ):1326–1331, PMID: , 10.1289/ehp.0901070.20406721
19. Perera FP, Li Z, Whyatt R, Hoepner L, Wang S, Camann D, et al. 2009. Prenatal airborne polycyclic aromatic hydrocarbon exposure and child IQ at age 5 years. Pediatrics 124 (2 ):e195–e202, PMID: , 10.1542/peds.2008-3506.19620194
20. Porta D, Narduzzi S, Badaloni C, Bucci S, Cesaroni G, Colelli V, et al. 2016. Air pollution and cognitive development at age 7 in a prospective Italian birth cohort. Epidemiology 27 (2 ):228–236, PMID: , 10.1097/EDE.0000000000000405.26426942
21. Harris MH, Gold DR, Rifas-Shiman SL, Melly SJ, Zanobetti A, Coull BA, et al. 2016. Prenatal and childhood traffic-related air pollution exposure and childhood executive function and behavior. Neurotoxicol Teratol 57 :60–70, PMID: , 10.1016/j.ntt.2016.06.008.27350569
22. Eskenazi B, Marks AR, Bradman A, Fenster L, Johnson C, Barr DB, et al. 2006. In utero exposure to dichlorodiphenyltrichloroethane (DDT) and dichlorodiphenyldichloroethylene (DDE) and neurodevelopment among young Mexican American children. Pediatrics 118 (1 ):233–241, PMID: , 10.1542/peds.2005-3117.16818570
23. Bouchard MF, Chevrier J, Harley KG, Kogut K, Vedar M, Calderon N, et al. 2011. Prenatal exposure to organophosphate pesticides and IQ in 7-year-old children. Environ Health Perspect 119 (8 ):1189–1195, PMID: , 10.1289/ehp.1003185.21507776
24. Gunier RB, Bradman A, Harley KG, Kogut K, Eskenazi B. 2017. Prenatal residential proximity to agricultural pesticide use and IQ in 7-year-old children. Environ Health Perspect 125 (5 ):057002, PMID: , 10.1289/EHP504.28557711
25. Castorina R, Bradman A, Stapleton HM, Butt C, Avery D, Harley KG, et al. 2017. Current-use flame retardants: maternal exposure and neurodevelopment in children of the CHAMACOS cohort. Chemosphere 189 :574–580, PMID: , 10.1016/j.chemosphere.2017.09.037.28963974
26. Eskenazi B, Chevrier J, Rauch SA, Kogut K, Harley KG, Johnson C, et al. 2013. In utero and childhood polybrominated diphenyl ether (PBDE) exposures and neurodevelopment in the CHAMACOS study. Environ Health Perspect 121 (2 ):257–262, PMID: , 10.1289/ehp.1205597.23154064
27. Sagiv SK, Kogut K, Harley K, Bradman A, Morga N, Eskenazi B. 2021. Gestational exposure to organophosphate pesticides and longitudinally assessed behaviors related to attention-deficit/hyperactivity disorder and executive function. Am J Epidemiol 190 (11 ):2420–2431, PMID: , 10.1093/aje/kwab173.34100072
28. Eskenazi B, Harley K, Bradman A, Weltzien E, Jewell NP, Barr DB, et al. 2004. Association of in utero organophosphate pesticide exposure and fetal growth and length of gestation in an agricultural population. Environ Health Perspect 112 (10 ):1116–1124, PMID: , 10.1289/ehp.6789.15238287
29. Sagiv SK, Kogut K, Gaspar FW, Gunier RB, Harley KG, Parra K, et al. 2015. Prenatal and childhood polybrominated diphenyl ether (PBDE) exposure and attention and executive function at 9–12 years of age. Neurotoxicol Teratol 52 (Pt B ):151–161, PMID: , 10.1016/j.ntt.2015.08.001.26271888
30. Su JG, Meng YY, Chen X, Molitor J, Yue D, Jerrett M. 2020. Predicting differential improvements in annual pollutant concentrations and exposures for regulatory policy assessment. Environ Int 143 :105942, PMID: , 10.1016/j.envint.2020.105942.32659530
31. Wechsler D. 2003. Wechsler Intelligence Scale for Children, Fourth Edition. San Antonio, TX: Harcourt Assessment Inc.
32. Woodcock RW. 1990. Woodcock-Johnson Psycho-Educational Battery– Revised: Woodcock-Johnson Tests of Achievement. Boston, MA: Teaching Resources Corp.
33. Harel O, Mitchell EM, Perkins NJ, Cole SR, Tchetgen Tchetgen EJ, Sun BLuo, et al. 2018. Multiple imputation for incomplete data in epidemiologic studies. Am J Epidemiol 187 (3 ):576–584, PMID: , 10.1093/aje/kwx349.29165547
34. Wickham H, Averick M, Bryan J, Chang W, McGowan LD, François R, et al. 2019. Welcome to the Tidyverse. J Open Source Softw 4 (43 ):1686, 10.21105/joss.01686.
35. Bradley RH. 2015. Constructing and adapting causal and formative measures of family settings: the HOME inventory as illustration. J Fam Theory Rev 7 (4 ):381–414, PMID: , 10.1111/jftr.12108.26997978
36. Totsika V, Sylva K. 2004. The home observation for measurement of the environment revisited. Child Adolesc Ment Health 9 (1 ):25–35, PMID: , 10.1046/j.1475-357X.2003.00073.x.32797621
37. Rothman KJ, Greenland S. 2018. Planning study size based on precision rather than power. Epidemiology 29 (5 ):599–603, PMID: , 10.1097/EDE.0000000000000876.29912015
38. Buckley JP, Hamra GB, Braun JM. 2019. Statistical approaches for investigating periods of susceptibility in children’s environmental health research. Curr Environ Health Rep 6 (1 ):1–7, PMID: , 10.1007/s40572-019-0224-5.30684243
39. Gasparrini A. 2011. Distributed lag linear and non-linear models in R: the package dlnm. J Stat Softw 43 (8 ):1–20, PMID: .22003319
40. Gasparrini A, Armstrong B, Kenward MG. 2010. Distributed lag non-linear models. Stat Med 29 (21 ):2224–2234, PMID: , 10.1002/sim.3940.20812303
41. Gasparrini A. 2014. Modeling exposure–lag–response associations with distributed lag non-linear models. Stat Med 33 (5 ):881–899, PMID: , 10.1002/sim.5963.24027094
42. Verner M-A, Gaspar FW, Chevrier J, Gunier RB, Sjödin A, Bradman A, et al. 2015. Increasing sample size in prospective birth cohorts: back-extrapolating prenatal levels of persistent organic pollutants in newly enrolled children. Environ Sci Technol 49 (6 ):3940–3948, PMID: , 10.1021/acs.est.5b00322.25698216
43. Coker E, Gunier R, Bradman A, Harley K, Kogut K, Molitor John, et al. 2017. Association between pesticide profiles used on agricultural fields near maternal residences during pregnancy and IQ at age 7 years. Int J Environ Res Public Health 14 (5 ):506, PMID: , 10.3390/ijerph14050506.28486423
44. Lertxundi A, Andiarena A, Martínez MD, Ayerdi M, Murcia M, Estarlich M, et al. 2019. Prenatal exposure to PM2.5 and NO2 and sex-dependent infant cognitive and motor development. Environ Res 174 :114–121, PMID: , 10.1016/j.envres.2019.04.001.31055169
45. Harris MH, Gold DR, Rifas-Shiman SL, Melly SJ, Zanobetti A, Coull BA, et al. 2015. Prenatal and childhood traffic-related pollution exposure and childhood cognition in the Project Viva cohort (Massachusetts, USA). Environ Health Perspect 123 (10 ):1072–1078, PMID: , 10.1289/ehp.1408803.25839914
46. Lubczyńska MJ, Muetzel RL, El Marroun H, Basagaña X, Strak M, Denault W, et al. 2020. Exposure to air pollution during pregnancy and childhood, and white matter microstructure in preadolescents. Environ Health Perspect 128 (2 ):027005, PMID: , 10.1289/EHP4709.32074458
47. Costa LG, Cole TB, Coburn J, Chang YC, Dao K, Roqué PJ. 2017. Neurotoxicity of traffic-related air pollution. Neurotoxicology 59 :133–139, PMID: , 10.1016/j.neuro.2015.11.008.26610921
48. Fagundes LS, Fleck A da S, Zanchi AC, Saldiva PHN, Rhoden CR. 2015. Direct contact with particulate matter increases oxidative stress in different brain structures. Inhal Toxicol 27 (10 ):462–467, PMID: , 10.3109/08958378.2015.1060278.26327340
49. Solaimani P, Saffari A, Sioutas C, Bondy SC, Campbell A. 2017. Exposure to ambient ultrafine particulate matter alters the expression of genes in primary human neurons. NeuroToxicology 58 :50–57, PMID: , 10.1016/j.neuro.2016.11.001.27851901
50. Wei H, Liang F, Meng G, Nie Z, Zhou R, Cheng W, et al. 2016. Redox/methylation mediated abnormal DNA methylation as regulators of ambient fine particulate matter-induced neurodevelopment related impairment in human neuronal cells. Sci Rep 6 (1 ):33402, PMID: , 10.1038/srep33402.27624276
51. Klocke C, Allen JL, Sobolewski M, Blum JL, Zelikoff JT, Cory-Slechta DA. 2018. Exposure to fine and ultrafine particulate matter during gestation alters postnatal oligodendrocyte maturation, proliferation capacity, and myelination. NeuroToxicology 65 :196–206, PMID: , 10.1016/j.neuro.2017.10.004.29079486
52. Block ML, Calderón-Garcidueñas L. 2009. Air pollution: mechanisms of neuroinflammation and CNS disease. Trends Neurosci 32 (9 ):506–516, PMID: , 10.1016/j.tins.2009.05.009.19716187
53. Bolton JL, Smith SH, Huff NC, Gilmour MI, Foster WM, Auten RL, et al. 2012. Prenatal air pollution exposure induces neuroinflammation and predisposes offspring to weight gain in adulthood in a sex-specific manner. FASEB J 26 (11 ):4743–4754, PMID: , 10.1096/fj.12-210989.22815382
54. Bos I, De Boever P, Emmerechts J, Buekers J, Vanoirbeek J, Meeusen R, et al. 2012. Changed gene expression in brains of mice exposed to traffic in a highway tunnel. Inhal Toxicol 24 (10 ):676–686, PMID: , 10.3109/08958378.2012.714004.22906174
55. Levesque S, Surace MJ, McDonald J, Block ML. 2011. Air pollution & the brain: subchronic diesel exhaust exposure causes neuroinflammation and elevates early markers of neurodegenerative disease. J Neuroinflammation 8 (1 ):105, PMID: , 10.1186/1742-2094-8-105.21864400
56. Allen JL, Liu X, Weston D, Prince L, Oberdörster G, Finkelstein JN, et al. 2014. Developmental exposure to concentrated ambient ultrafine particulate matter air pollution in mice results in persistent and sex-dependent behavioral neurotoxicity and glial activation. Toxicol Sci 140 (1 ):160–178, PMID: , 10.1093/toxsci/kfu059.24690596
57. Ku T, Ji X, Zhang Y, Li G, Sang N. 2016. PM2.5, SO2 and NO2 co-exposure impairs neurobehavior and induces mitochondrial injuries in the mouse brain. Chemosphere 163 :27–34, PMID: , 10.1016/j.chemosphere.2016.08.009.27521637
58. Calderón-Garcidueñas L, Solt AC, Henríquez-Roldán C, Torres-Jardón R, Nuse B, Herritt L, et al. 2008. Long-term air pollution exposure is associated with neuroinflammation, an altered innate immune response, disruption of the blood-brain barrier, ultrafine particulate deposition, and accumulation of amyloid β-42 and α-synuclein in children and young adults. Toxicol Pathol 36 (2 ):289–310, PMID: , 10.1177/0192623307313011.18349428
59. Jakab A, Schwartz E, Kasprian G, Gruber GM, Prayer D, Schöpf V, et al. 2014. Fetal functional imaging portrays heterogeneous development of emerging human brain networks. Front Hum Neurosci 8 :852, PMID: , 10.3389/fnhum.2014.00852.25374531
60. Thomason ME, Dassanayake MT, Shen S, Katkuri Y, Alexis M, Anderson AL, et al. 2013. Cross-hemispheric functional connectivity in the human fetal brain. Sci Transl Med 5 (173 ):173ra24, PMID: , 10.1126/scitranslmed.3004978.23427244
61. Matsuda KI, Mori H, Kawata M. 2012. Epigenetic mechanisms are involved in sexual differentiation of the brain. Rev Endocr Metab Disord 13 (3 ):163–171, PMID: , 10.1007/s11154-012-9202-z.22327342
62. Reinius B, Jazin E. 2009. Prenatal sex differences in the human brain. Mol Psychiatry 14 (11 ):988–989, PMID: , 10.1038/mp.2009.79.19851278
63. Nguyen HB, Loughead J, Lipner E, Hantsoo L, Kornfield SL, Epperson CN. 2019. What has sex got to do with it? The role of hormones in the transgender brain. Neuropsychopharmacology 44 (1 ):22–37, PMID: , 10.1038/s41386-018-0140-7.30082887
64. Wheelock MD, Hect JL, Hernandez-Andrade E, Hassan SS, Romero R, Eggebrecht AT, et al. 2019. Sex differences in functional connectivity during fetal brain development. Dev Cogn Neurosci 36 :100632, PMID: , 10.1016/j.dcn.2019.100632.30901622
65. Hanamsagar R, Bilbo SD. 2017. Environment matters: microglia function and dysfunction in a changing world. Curr Opin Neurobiol 47 :146–155, PMID: , 10.1016/j.conb.2017.10.007.29096243
66. Remy S, Hambach R, Van Sprundel M, Teughels C, Nawrot TS, Buekers J, et al. 2019. Intelligence gain and social cost savings attributable to environmental lead exposure reduction strategies since the year 2000 in Flanders, Belgium. Environ Health 18 (1 ):113, PMID: , 10.1186/s12940-019-0548-5.31881883
67. Jbaily A, Zhou X, Liu J, Lee T-H, Kamareddine L, Verguet S, et al. 2022. Air pollution exposure disparities across US population and income groups. Nature 601 (7892 ):228–233, PMID: , 10.1038/s41586-021-04190-y.35022594
|
PMC010xxxxxx/PMC10013688.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36917477
EHP12762
10.1289/EHP12762
Science Selection
The Echoes of Noise: Residential Exposure to Traffic and Risk of Tinnitus
https://orcid.org/0000-0002-9793-8024
Adegboye Oyelola
14 3 2023
3 2023
131 3 03400118 1 2023
15 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Traffic at night on a street lined by small apartment buildings and houses
==== Body
pmcNoisy occupational settings place workers at risk for adverse auditory outcomes, including tinnitus.1 But exposure to loud noise off the job—for example, when using earbuds, shooting firearms, visiting nightclubs, or cranking up the volume on computer games—also puts people at risk for tinnitus.2–4 Tinnitus is characterized by the perception of sound in the absence of an external source—sometimes experienced as a ringing in the ears—and it is a common problem, affecting up to 38% of adults.5–7 In a cohort study recently published in Environmental Health Perspectives, Manuella Lech Cantuaria and colleagues quantified the risk of tinnitus associated with residential road and railway noise exposure.8
Loud noises are thought to cause tinnitus by damaging the inner ear hair cells, which transmit sound signals to the brain.9 However, it is still unclear how, or even whether, chronic noise exposures lead to the condition. The authors of the new study point to papers suggesting that “stressful situations and sleep disturbances precede tinnitus occurrence and contribute to the transition from mild to severe symptoms.”10–12
The researchers considered road traffic noise levels on the quiet side of buildings as a proxy for nighttime exposure or noise during sleep. Roadway noise exposures on both the quietest and noisiest sides were associated with higher risk of developing tinnitus. Railway noise was not associated with tinnitus risk. Image: © iStock.com/mathess.
Traffic at night on a street lined by small apartment buildings and houses
The researchers used data from health, administrative, and housing registers for more than 3.5 million Danish adults from 2000 to 2017. By the end of the follow-up period, 40,692 new cases of tinnitus had been diagnosed.8 The authors used precisely geocoded data for the location and floors of homes and apartment buildings. They modeled noise exposures using detailed information on road traffic (average daily traffic, vehicle distribution, road type, and travel speed) and railway traffic (average train length, train type, and travel speed). Then they quantified associations between road and railway noise at the most- and least-exposed façades of the dwelling and development of tinnitus, based on estimated average exposure over 1-, 5-, and 10-year periods.
No association was found for railway noise. But for road traffic, the estimated risk of tinnitus rose as the average 10-year noise level increased. The risk increased by 6% for every 10-decibel (dB) increase in average exposure at the quietest façade. At the loudest façade, the risk increased by 2% for every 10-dB increase in noise level.
“By using high-quality Danish registers, we had access to the address history of all people in Denmark, and we could then estimate the exact amount of road traffic and railway noise each person was exposed to over a long period of time,” says first author Cantuaria, an assistant professor of epidemiology and data science at the University of Southern Denmark’s Maersk McKinney Møller Institute. “We found that road traffic noise was associated with a higher risk of tinnitus, with a linear exposure–response relationship when noise was estimated at the least-exposed façade.”
Cantuaria suggests that noise levels at the least-exposed façade—or the quieter side of the building—tend to reflect exposures during sleep, because people may choose to sleep in the quieter rooms.13 “Our results suggest that noise exposure during sleep can have an even greater effect in increasing tinnitus risk than daytime exposure,” she says.
According to the paper’s senior author Mette Sørensen, the authors believe this to be the first epidemiological study investigating associations between residential exposure to transportation noise and tinnitus. “Environmental health studies to date have focused on nonauditory health effects of noise,” says Sørensen, a senior researcher at the Danish Cancer Society Research Center in Copenhagen. These include cardiovascular diseases,14,15 stroke,16,17 and diabetes.18 Her point is underscored by the World Health Organization (WHO), which in 2018 reported finding no studies investigating road transportation noise and hearing-related outcomes, even though these are among the critical outcomes identified for development of policies regarding noise pollution.19
Martin Röösli, an associate professor of environmental epidemiology and head of the Environmental Exposures and Health Unit at the Swiss Tropical and Public Health Institute, says previous studies on tinnitus focused on noise levels above 85 dB (such as occupational noise), at which there is solid evidence for a link.1,20 Röösli, who was not involved in the Danish study, noted, “The pattern of higher risks of tinnitus among women, people without hearing loss, people with high education and income, and people who had never been in a noisy job is plausible as these groups may have less competing risk from occupational exposure; thus, the road traffic noise effect is not diluted.” Moreover, the authors suggested that women and those with higher education and income might be more likely to seek—and receive—medical help for the problem than other groups.
Noise exposure is just one risk factor for tinnitus; others include anxiety, depression, hearing loss, and other health conditions.9,21,22 The WHO estimated that, in Europe, environmental noise—which was defined as including noise from transportation (road traffic, railway, and aircraft), wind turbines, and leisure activities—caused a loss of 22,000 disability-adjusted life-years (years of living without disability) due to tinnitus.23
“Since the number of studies linking traffic noise with adverse health effects continues to grow, there are reasons to believe that the health consequences of traffic noise are likely much greater than what we assume today,” says Cantuaria. “It is thus essential to know more about the harmful effects of noise so that effective public health policies can be implemented.”
Oyelola Adegboye, PhD, is a senior biostatistics lecturer in public health and tropical medicine in the James Cook University College of Public Health, Medical and Veterinary Sciences in Townsville, Australia.
==== Refs
References
1. Lie A, Skogstad M, Johannessen HA, Tynes T, Mehlum IS, Nordby KC, et al. 2016. Occupational noise exposure and hearing: a systematic review. Int Arch Occup Environ Health 89 (3 ):351–372, PMID: , 10.1007/s00420-015-1083-5.26249711
2. Rhee J, Lee D, Suh MW, Lee JH, Hong Y-C, Oh SH, et al. 2020. Prevalence, associated factors, and comorbidities of tinnitus in adolescents. PLoS One 15 (7 ):e0236723, PMID: , 10.1371/journal.pone.0236723.32735626
3. Weilnhammer V, Gerstner D, Huß J, Schreiber F, Alvarez C, Steffens T, et al. 2022. Exposure to leisure noise and intermittent tinnitus among young adults in Bavaria: longitudinal data from a prospective cohort study. Int J Audiol 61 (2 ):89–96, PMID: , 10.1080/14992027.2021.1899312.33787447
4. Kim H-J, Lee H-J, An S-Y, Sim S, Park B, Kim SW, et al. 2015. Analysis of the prevalence and associated risk factors of tinnitus in adults. PLoS One 10 (5 ):e0127578, PMID: , 10.1371/journal.pone.0127578.26020239
5. Jarach CM, Lugo A, Scala M, van den Brandt PA, Cederroth CR, Odone A, et al. 2022. Global prevalence and incidence of tinnitus: a systematic review and meta-analysis. JAMA Neurol 79 (9 ):888–900, PMID: , 10.1001/jamaneurol.2022.2189.35939312
6. Biswas R, Lugo A, Akeroyd MA, Schlee W, Gallus S, Hall DA, et al. 2022. Tinnitus prevalence in Europe: a multi-country cross-sectional population study. Lancet Reg Health Eur 12 :100250, PMID: , 10.1016/j.lanepe.2021.100250.34950918
7. Schubert NM, Rosmalen JG, van Dijk P, Pyott SJ. 2021. A retrospective cross-sectional study on tinnitus prevalence and disease associations in the Dutch population-based cohort Lifelines. Hear Res 411 :108355, PMID: , 10.1016/j.heares.2021.108355.34607212
8. Cantuaria ML, Pedersen ER, Poulsen AH, Raaschou-Nielsen O, Hvidtfeldt UA, Levin G, et al. 2023. Transportation noise and risk of tinnitus: a nationwide cohort study from Denmark. Environ Health Perspect 131 (2 ):027001, 10.1289/EHP11248.36722980
9. Le TN, Straatman LV, Lea J, Westerberg B. 2017. Current insights in noise-induced hearing loss: a literature review of the underlying mechanism, pathophysiology, asymmetry, and management options. J Otolaryngol Head Neck Surg 46(1) :41, PMID: , 10.1186/s40463-017-0219-x.28535812
10. Betz LT, Mühlberger A, Langguth B, Schecklmann M. 2017. Stress reactivity in chronic tinnitus. Sci Rep 7 :1–9, 10.1038/srep41521.28127051
11. Rauschecker JP, Leaver AM, Mühlau M. 2010. Tuning out the noise: limbic-auditory interactions in tinnitus. Neuron 66 (6 ):1–16, PMID: , 10.1016/j.neuron.2010.04.032.20399721
12. Mazurek B, Boecking B, Brueggemann P. 2019. Association between stress and tinnitus – new aspects. Otol Neurotol 40 :467–473, PMID: , 10.1097/MAO.0000000000002180.30870382
13. Bartels S, Ögren M, Kim J-L, Fredriksson S, Persson Waye K. 2021. The impact of nocturnal road traffic noise, bedroom window orientation, and work-related stress on subjective sleep quality: results of a cross-sectional study among working women. Int Arch Occup Environ Health 94 (7 ):1523–1536, PMID: , 10.1007/s00420-021-01696-w.34043056
14. Münzel T, Sørensen M, Daiber A. 2021. Transportation noise pollution and cardiovascular disease. Nat Rev Cardiol 18 (9 ):619–636, PMID: , 10.1038/s41569-021-00532-5.33790462
15. Hao G, Zuo L, Weng X, Fei Q, Zhang Z, Chen L, et al. 2022. Associations of road traffic noise with cardiovascular diseases and mortality: longitudinal results from UK Biobank and meta-analysis. Environ Res 212 (pt A ):113129, PMID: , 10.1016/j.envres.2022.113129.35358546
16. Hegewald J, Schubert M, Lochmann M, Seidler A. 2021. The burden of disease due to road traffic noise in Hesse, Germany. Int J Environ Res Public Health 18 (17 ):9337, PMID: , 10.3390/ijerph18179337.34501923
17. Roswall N, Pyko A, Ögren M, Oudin A, Rosengren A, Lager A, et al. 2021. Long-term exposure to transportation noise and risk of incident stroke: a pooled study of nine Scandinavian cohorts. Environ Health Perspect 129 (10 ):107002, PMID: , 10.1289/EHP8949.34605674
18. Thacher JD, Poulsen AH, Hvidtfeldt UA, Raaschou-Nielsen O, Brandt J, Geels C, et al. 2021. Long-term exposure to transportation noise and risk for type 2 diabetes in a nationwide cohort study from Denmark. Environ Health Perspect 129 (12 ):127003, PMID: , 10.1289/EHP9146.34855467
19. WHO (World Health Organization) Regional Office for Europe. 2018. Environmental Noise Guidelines for the European Region. https://www.who.int/europe/publications/i/item/9789289053563 [accessed 28 February 2023].
20. Somma G, Pietroiusti A, Magrini A, Coppeta L, Ancona C, Gardi S, et al. 2008. Extended high‐frequency audiometry and noise induced hearing loss in cement workers. Am J Ind Med 51 (6 ):452–462, PMID: , 10.1002/ajim.20580.18393354
21. Bhatt JM, Bhattacharyya N, Lin HW. 2017. Relationships between tinnitus and the prevalence of anxiety and depression. Laryngoscope 127 (2 ):466–469, PMID: , 10.1002/lary.26107.27301552
22. Wang W, Zhang LS, Zinsmaier AK, Patterson G, Leptich EJ, Shoemaker SL, et al. 2019. Neuroinflammation mediates noise-induced synaptic imbalance and tinnitus in rodent models. PLoS Biology 17 :e3000307, PMID: , 10.1371/journal.pbio.3000307.31211773
23. WHO Regional Office for Europe. 2011. Burden of Disease from Environmental Noise: Quantification of Healthy Life Years Lost in Europe. https://apps.who.int/iris/handle/10665/326424 [accessed 28 February 2023].
|
PMC010xxxxxx/PMC10015888.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36920051
EHP10723
10.1289/EHP10723
Research
Birth Outcomes in Relation to Prenatal Exposure to Per- and Polyfluoroalkyl Substances and Stress in the Environmental Influences on Child Health Outcomes (ECHO) Program
https://orcid.org/0000-0003-1435-4814
Padula Amy M. 1
https://orcid.org/0000-0002-8523-0483
Ning Xuejuan 2
Bakre Shivani 2
Barrett Emily S. 3
Bastain Tracy 4
Bennett Deborah H. 5
Bloom Michael S. 6
Breton Carrie V. 4
Dunlop Anne L. 7
Eick Stephanie M. 1
Ferrara Assiamira 8
Fleisch Abby 9 10
Geiger Sarah 11
Goin Dana E. 1
Kannan Kurunthachalam 12
Karagas Margaret R. 13
Korrick Susan 10 14
Meeker John D. 15
Morello-Frosch Rachel 16
O’Connor Thomas G. 17
Oken Emily 18
Robinson Morgan 12
Romano Megan E. 13
Schantz Susan L. 19
Schmidt Rebecca J. 5
Starling Anne P. 20 21
Zhu Yeyi 8
Hamra Ghassan B. 2 *
Woodruff Tracey J. 1 *
the program collaborators for Environmental influences on Child Health Outcomes †
1 Program for Reproductive Health and the Environment, Department of Obstetrics, Gynecology and Reproductive Sciences, University of California, San Francisco, San Francisco, California, USA
2 Department of Epidemiology, Johns Hopkins University, Baltimore, Maryland, USA
3 Department of Biostatistics and Epidemiology, Rutgers School of Public Health, Environmental and Occupational Health Sciences Institute, Piscataway, New Jersey, USA
4 Department of Population and Public Health Sciences, Keck School of Medicine, University of Southern California, Los Angeles, California, USA
5 Department of Public Health Sciences, University of California, Davis, Davis, California, USA
6 Department of Global and Community Health, George Mason University, Fairfax, Virginia, USA
7 Department of Gynecology and Obstetrics, Emory University School of Medicine, Atlanta, Georgia, USA
8 Division of Research, Kaiser Permanente Northern California, Oakland, California, USA
9 Center for Outcomes Research and Evaluation, Maine Medical Center Research Institute, Portland, Maine, USA
10 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
11 Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Champaign, Illinois, USA
12 Department of Pediatrics and Department of Environmental Medicine, New York University Grossman School of Medicine, New York, New York, USA
13 Department of Epidemiology, Dartmouth Geisel School of Medicine, Lebanon, New Hampshire, USA
14 Channing Division of Network Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts, USA
15 Department of Environmental Health Sciences, University of Michigan, Ann Arbor, Michigan, USA
16 School of Public Health and Department of Environmental Science, Policy and Management, University of California, Berkeley, Berkeley, California, USA
17 Department of Psychiatry, University of Rochester School of Medicine and Dentistry, Rochester, New York, USA
18 Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, Massachusetts, USA
19 Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Champaign, Illinois, USA
20 Center for Lifecourse Epidemiology of Adiposity and Diabetes, University of Colorado Anschutz Medical Campus, Aurora, Colorado, USA
21 Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA
Address correspondence to Amy M. Padula, Program for Reproductive Health and the Environment, Department of OBGYN & RS, Box 0132, University of California, San Francisco, 490 Illinois St., San Francisco, CA 94158 USA. E-mail: [email protected]
15 3 2023
3 2023
131 3 03700603 12 2021
01 12 2022
06 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Per- and polyfluoroalkyl substances (PFAS) are persistent and ubiquitous chemicals associated with risk of adverse birth outcomes. Results of previous studies have been inconsistent. Associations between PFAS and birth outcomes may be affected by psychosocial stress.
Objectives:
We estimated risk of adverse birth outcomes in relation to prenatal PFAS concentrations and evaluate whether maternal stress modifies those relationships.
Methods:
We included 3,339 participants from 11 prospective prenatal cohorts in the Environmental influences on the Child Health Outcomes (ECHO) program to estimate the associations of five PFAS and birth outcomes. We stratified by perceived stress scale scores to examine effect modification and used Bayesian Weighted Sums to estimate mixtures of PFAS.
Results:
We observed reduced birth size with increased concentrations of all PFAS. For a 1-unit higher log-normalized exposure to perfluorooctanoic acid (PFOA), perfluorooctanesulfonic acid (PFOS), perfluorononanoic acid (PFNA), and perfluorohexane sulfonic acid (PFHxS), we observed lower birthweight-for-gestational-age z-scores of β=−0.15 [95% confidence interval (CI): −0.27, −0.03], β=−0.14 (95% CI: −0.28, −0.002), β=−0.22 (95% CI: −0.23, −0.10), β=−0.06 (95% CI: −0.18, 0.06), and β=−0.25 (95% CI: −0.37, −0.14), respectively. We observed a lower odds ratio (OR) for large-for-gestational-age: ORPFNA=0.56 (95% CI: 0.38, 0.83), ORPFDA=0.52 (95% CI: 0.35, 0.77). For a 1-unit increase in log-normalized concentration of summed PFAS, we observed a lower birthweight-for-gestational-age z-score [−0.28; 95% highest posterior density (HPD): −0.44, −0.14] and decreased odds of large-for-gestational-age (OR=0.49; 95% HPD: 0.29, 0.82). Perfluorodecanoic acid (PFDA) explained the highest percentage (40%) of the summed effect in both models. Associations were not modified by maternal perceived stress.
Discussion:
Our large, multi-cohort study of PFAS and adverse birth outcomes found a negative association between prenatal PFAS and birthweight-for-gestational-age, and the associations were not different in groups with high vs. low perceived stress. This study can help inform policy to reduce exposures in the environment and humans. https://doi.org/10.1289/EHP10723
Supplemental Material is available online (https://doi.org/10.1289/EHP10723).
* These authors contributed equally to this work.
† See the “Acknowledgments” section for full listing of collaborators.
The authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Per- and polyfluoroalkyl substances (PFAS) are a group of synthetic chemicals used in nonstick and stain- and water-resistant consumer products, as well as in industrial processes. PFAS are persistent in the environment and in the human body.1 Pathways of human exposure include ingestion of contaminated drinking water and food, and inhalation.2,3 As a result, PFAS are widely detectable in human biomonitoring studies, including studies showing that nearly 100% of pregnant women studied have measurable levels of PFAS in their bodies.4 Reported human health associations include carcinogenicity (kidney and testicular cancers),5 cardiovascular effects (dyslipidemia6), pregnancy-induced hypertension,7 impaired renal function,8,9 endocrine disruption (thyroid disease and altered age at menarche),10,11 obesity,12 and immune effects (immunotoxicity and decreased antibody production).13–15
PFAS have been associated with adverse effects on fetal development in both animal and human studies.16,17 Reductions in birthweight have been reported with higher exposure to perfluorooctanesulfonic acid (PFOS), perfluorooctanoic acid (PFOA), and perfluorononanoic acid (PFNA).18–31 A systematic review and meta-analysis of animal and human research found sufficient evidence for an inverse association between PFOA and birthweight.16 Fewer studies have examined PFAS in relation to preterm birth; however, a recent review and meta-analysis found maternal PFOS was associated with increased risk of preterm birth.32 Only one study examined PFAS in relation to large-for-gestational-age, and it reported no association.33
Psychosocial stressors and responses to stress during pregnancy are associated with perinatal outcomes and may also contribute to the persistence of disparities in adverse birth outcomes by socioeconomic status and racial and ethnic groups.31,34 Experiences of psychosocial stress during pregnancy may be more prevalent among women of lower socioeconomic status as indicated by lower education or income level.35 Perceived stress may also be higher among women of color because of racial and gender-based discrimination.36–38 Environmental chemical exposures can co-occur with chronic psychosocial risk factors during pregnancy.39,40 This combination may have a greater impact than each individual factor alone and result in amplified risk of adverse pregnancy outcomes.40–42 Furthermore, these environmental and psychosocial stressors may operate via similar biological systems and mechanisms (i.e., endocrine or metabolic disruption, inflammation, and epigenetic changes).41
The Environmental influences on Child Health Outcomes (ECHO) program is a National Institutes of Health initiative to address pediatric outcomes with high public health impact.43 ECHO comprises 69 cohorts from across the United States and includes over 57,000 mother–child dyads.44 The program is well powered to analyze environmental exposures in a demographically and geographically diverse study population including 56 cohorts with chemical biomonitoring data for mothers and children.45 The present study estimates associations using ECHO data from 11 pregnancy cohorts to examine the extent to which prenatal exposure to PFAS is associated with increased risk of adverse birth outcomes and whether these associations are modified by stress.
Methods
Overview
ECHO cohorts were invited to participate based on consent for data sharing with ECHO of the mother–child pairs46 and were harmonized and pooled for analysis. Mothers were required to have either extant prenatal PFAS data or at least one serum or plasma biospecimen collected during pregnancy that was available for assessment of PFAS concentration. Data on child birthweight or gestational age at birth were required for participation, and the study population was restricted to singleton births and included 3,339 mother–child pairs from 11 cohorts between 1999 and 2019 (Figure S1). Cohorts submitted data to the ECHO Data Analysis Center for analysis. Cohort was not considered when determining inclusion for this analytic data set. All cohorts had institutional review board approvals from their local institutions. Written consent to participate in the ECHO study was obtained for all participants. Participants received various stipends for their time according to the individual cohort.
PFAS
Laboratory methods varied by cohort (Table S1). PFAS were measured (in nanograms per milliliter) in plasma or serum at three laboratories: the California Department of Toxic Substances Control,34 the Centers for Disease Control and Prevention (CDC),30,47,48 and the Wadsworth Human Health Exposure Analysis Resource Laboratory.49 All laboratories participated in the CDC’s quality assurance program to test interlaboratory comparisons. The number of PFAS measured in each cohort varied from 8 to 14 (Table S2). PFAS were included in the present analysis if more than 60% of values were above the method limit of detection (LOD) and no cohort had <40% below the LOD (Table S2). Five PFAS met these criteria: PFOA, PFOS, PFNA, perfluorohexane sulfonic acid (PFHxS), and perfluorodecanoic acid (PFDA). If a cohort had separate sums of branched and linear chain isomers for PFOA or PFOS, the two were summed as total PFOA or PFOS.50 Distributions of PFAS were examined by cohort, year, and perceived stress scale (PSS; Table S3). LOD varied between labs and within cohorts owing to batches performed years apart (Table S3). For those observations that were below the LOD, we imputed exposure values as the LOD divided by the square root of 2. PFAS measures were nonnormally distributed, and, thus, were natural log transformed (Figure S2). Most cohorts collected prenatal biospecimens during the second trimester (9 cohorts, n=2,531, Table S1). For three cohorts (n=565) with PFAS measured at multiple time points, concentrations above the LOD were averaged. We tested the correlations between the different PFAS and each PFAS across different trimesters of exposure. Spearman correlations of PFAS concentrations measured multiple times during pregnancy were strong (ρ>0.8), with one exception, which was moderately correlated [PFDA in the first and third trimesters (ρ=0.53)] (Table S4). We compared PFAS concentrations to those measured by the National Health and Nutrition Examination Surveys (NHANES) during the study period (Table S5).
Prenatal Stress
We examined maternal stress as an effect modifier of the relationship between PFAS and birth outcomes. For a subset of cohorts (8 of 11, N=2,032), maternal stress was assessed using the PSS administered in the prenatal period; the PSS measures perceptions of life as uncontrollable, unpredictable, and overwhelming.51 The PSS is a widely used self-report instrument for measuring stress perception and is available in three versions, with 4, 10, or 14 items [PSS-4 (1 cohort, n=402), PSS-10 (5 cohorts, n=1,148), and PSS-14 (2 cohorts, n=459), respectively], each containing items rated on a five-point Likert scale. Psychometric data support reliability and validity of the PSS-10 in comparison with the PSS-14 and perceived helplessness (r=0.85) and perceived self-efficacy (r=0.82) scales, respectively.52 In addition, the PSS-4 has been validated in pregnant women and correlated strongly (ρ=0.71) with the Assessment of Stress portion of the Prenatal Psychosocial Profile and was valid in predicting maternal depression (Edinburgh Postnatal Depression Scale, r=0.67), and quality of life (mental health component of the Short-Form-12, r=−0.62).53 Cohorts were administered one version of the PSS (Table S1), and item response theory was used to harmonize PSS to a t-score metric by the ECHO Patient-Reported Outcomes Core [ECHO PRO Core Data Harmonization Group, ECHO-wide Cohort Protocol (version 2.0), Harmonization Technical Report (version 5.2, 24 March 2021)].54 PSS scores were unavailable for participants in three cohorts (the Project Viva cohort, the Kaiser Permanente Research Bank Pregnancy Cohort, and the New Hampshire Birth Cohort) and partially missing in other cohorts except for Illinois Kids Development Studies, which had complete data on PSS [N=2,009 (60%) of 3,339].
Birth Outcomes
Outcomes included gestational age at birth (completed weeks), preterm birth (birth <37 vs. ≥37 wk gestation), term low birthweight (birthweight <2,500 vs. ≥2,500g among births at ≥37 wk gestation), birthweight-for-gestational-age and sex-specific z-scores, and both small- and large-for-gestational-age (<10th percentile and >90th percentile, respectively) using a 2017 referent population in the United States.55 Birth outcomes and covariates were obtained according to the protocol for each cohort (from medical records or self-report).
Statistical Analysis
We analyzed two continuous and four dichotomous birth outcomes using linear and logistic regression, respectively, in relation to single PFAS exposures. Covariates selected as potential confounders a priori based on a directed acyclic graph (Figure S3) included cohort (base model), maternal age at delivery (<25, 25–29, 30–34, ≥35 y), parity (0, ≥1), maternal educational attainment [<high school; high school degree, General Educational Development (GED), or equivalent; some college, no degree; bachelor’s degree and above], and maternal race/ethnicity (non-Hispanic White, non-Hispanic Black, non-Hispanic Asian/Pacific Islander, non-Hispanic other, and Hispanic). Race/ethnicity was included as a social construct and proxy for racism and discrimination. The non-Hispanic other category included Native Hawaiian or other Pacific Islander, American Indian or Alaska Native, multiple race, or any other race group not included in a more specific category. We examined race/ethnicity in relation to PSS scores because we hypothesized racism and discrimination might be associated with perceived stress. Our study was restricted to participants with nonmissing data on these covariates. Factors related to the outcome and to stress but not PFAS (e.g., maternal tobacco use, prenatal secondhand smoke exposure) were considered in sensitivity analyses. Additional covariates were considered potential mediators [e.g., maternal body mass index (BMI), gestational diabetes, gestational hypertension, and preeclampsia] and were not considered confounders and not included in analytic models. We performed stratified analyses by PSS scores, which were dichotomized at the median of the t-scores and examined the p-value of the interaction term to determine potential effect modification (results with p<0.1 were noted).
We estimated the effect of summed concentrations of five PFAS (PFOA, PFOS, PFNA, PFHxS, and PFDA) using Bayesian Weighted Sums, a recent Bayesian approach that provides the effect of the mixture of PFAS, as well as the percentage contribution of each of the PFAS. This approach allows the data and model to estimate the weights56 and uses a Dirichlet prior that restricts values of the weights to sum to 1 and restricts individual values to a 0–1 range.57 These analyses were similarly adjusted for the covariates and stratified by PSS. We provide 95% highest posterior density (HPD) intervals as opposed to 95% credible intervals.
We performed several sensitivity analyses to assess the robustness of our results and explore additional effect modifiers and confounders. We performed a stratified analysis by infant sex to identify potential sex-specific associations of PFAS and birth outcomes and examined the p-value of the interaction term to determine potential effect modification (results with p<0.1 were noted). We conducted a trimester-stratified analysis to compare results by timing of PFAS measurements during pregnancy. Because results may be sensitive to inclusion of specific cohorts, we conducted leave-one-out analyses, excluding each cohort from calculation of the main effects of PFAS. We examined quartiles of exposure in relation to the outcomes to assess the linearity of the exposure–response relationship. We performed the birthweight-for-gestational-age z-score analysis with cohort as a random effect in mixed effects models to determine if our main findings were impacted by cohort heterogeneity. We adjusted for prenatal tobacco smoke exposure (indicators of either any maternal smoking or secondhand smoke during pregnancy) as an additional potential confounder for birthweight-for-gestational-age z-score and large-for-gestational-age. Last, we provided estimates of the association between non–log-transformed PFAS and continuous birth weight (adjusted for gestational age) given the difficulty of interpreting log-transformed values of PFAS in relation to z-scores of birthweight-for-gestational-age and the potential that log transformation may bias the results. We chose not to correct for multiple comparisons given the few a priori tests and our preference to present actual observations.58 Primary statistical analyses were conducted using Stata (version 17.0; StataCorp), and correlation maps and Bayesian mixtures analyses were conducted in R (version 4.1.0; R Development Core Team) using the JAGS software program (version 4.3.1). Software code to recreate results of this work is maintained by the ECHO Data Analysis Center (https://dcricollab.dcri.duke.edu/sites/echomaterials/SitePages/Home.aspx).
Results
This study included 3,339 mother–child pairs from 11 cohorts in ECHO. Mothers were demographically and racially/ethnically diverse, with about half non-Hispanic White (53.8%) and having a bachelor’s degree or higher educational attainment level (53.0%) (Table 1). The mean age of mothers at delivery was 30.9±5.8 y. The years of birth for all cohorts ranged from 1999 through 2019 (Table S1).
Table 1 Characteristics of the study population among selected ECHO cohorts (N=3,339).
Characteristic N (%) or mean±SD
Maternal race/ethnicity
Hispanic/Latina 653 (20.8)
Non-Hispanic White 1,687 (53.8)
Non-Hispanic Black 509 (16.2)
Non-Hispanic Asian 193 (6.2)
Non-Hispanic other 96 (3.1)
Unknown 201
Maternal educational attainment
<High school 312 (9.5)
High school degree, GED, or equivalent 530 (16.1)
Some college, no degree 702 (21.4)
Bachelor’s degree and above 1,742 (53.0)
Unknown 53
Maternal age at delivery (y)
<25 497 (15.7)
25–29 672 (21.3)
30–34 1,124 (35.6)
≥35 867 (27.4)
Unknown 179
PSS scale
PSS t-score category 49.8±9.9 a
<Median (50.6) 1,003 (49.9)
≥Median (50.6) 1,006 (50.1)
Unknown 1,330
Gestational age (wk) 38.9±1.9
Preterm birth (<37 wk)
Yes 252 (7.5)
No 3,087 (92.5)
Birthweight (g) 3,337.4±563.3
Low birthweight (<2,500g)
Yes 182 (5.5)
No 3,157 (94.5)
Size for gestational ageb
Small-for-gestational-age 357 (10.7)
Appropriate-for-gestational-age 2,623 (78.6)
Large-for-gestational-age 359 (10.8)
Child sex
Male 1,643 (49.2)
Female 1,696 (50.8)
Parity prior to indexed birth
0 1,783 (53.4)
≥1 1,556 (46.6)
Prenatal tobacco use
Yes 200 (7.2)
No 2,573 (92.8)
Unknown 165
Prenatal secondhand smoke
Yes 1,386 (64.0)
No 781 (36.0)
Unknown 1,172
Prepregnancy BMI (kg/m2) 26.1±6.3 c
Gestational diabetes
Yes 321 (10.5)
No 2,722 (89.5)
Unknown 296
Gestational hypertension
Yes 154 (8.5)
No 1,667 (91.5)
Unknown 1,518
Preeclampsia
Yes 104 (5.6)
No 1,751 (94.4)
Unknown 1,484
Year of birth
1999 28 (0.8)
2000 330 (9.9)
2001 310 (9.3)
2002 170 (5.1)
2003 4 (0.1)
2009 9 (0.3)
2010 122 (3.7)
2011 225 (6.7)
2012 283 (8.5)
2013 243 (7.3)
2014 204 (6.1)
2015 278 (8.3)
2016 287 (8.6)
2017 283 (8.5)
2018 286 (8.6)
2019 98 (2.9)
Cohort
Chemicals in Our Bodies (CiOB) 402 (12.0)
Illinois Kids Development Studies (IKIDS) 184 (5.5)
Project Viva 842 (25.2)
Healthy Start 652 (19.5)
New Hampshire Birth Cohort Study (NHBCS) 324 (9.7)
Markers of Autism Risk in Babies Learning Early Signs (MARBLES) 39 (1.2)
Emory (Atlanta) 424 (12.7)
Maternal And Developmental Risks from Environmental and Social Stressors (MADRES) 347 (10.4)
Pregnancy and EnvironmenT And Lifestyle Study (PETALS) 124 (3.7)
Rochester 35 (1.0)
Kaiser Permanente Research Bank Pregnancy Cohort (KPRB-PC) 13 (0.4)
Note: BMI, body mass index; ECHO, Environmental influences on Child Health Outcomes; GED, General Educational Development; PSS, perceived stress scale; SD, standard deviation.
a n=2,009.
b Small-, appropriate-, and large-for-gestational-age were defined, respectively, as singleton infants with weight <10th percentile, 10th–90th percentile, and >90th percentile of birthweight-for-gestational-age and sex using a 2017 U.S. reference population.
c n=3,219.
Four PFAS were detected in 96%–100% of participants (PFOS, PFOA, PFNA, and PFHxS) and concentrations were lower than those measured in NHANES (Table S5). Most PFAS were moderately positively correlated with Spearman correlations between ρ=0.14 (PFDA and PFHxS) and ρ=0.83 (PFOA and PFOS) (Figure S4). PFAS concentrations were highest among participants from older cohorts, although not monotonically, and PFAS decreased across years except for PFHxS, which increased between 2015 and 2019, although levels were not as high as earlier (1999–2003) (Table S3).
As compared with participants who were white, a higher proportion of participants who were Asian and other race/ethnicity had above-median levels of PSS. A lower proportion of participants who were Hispanic or unknown race/ethnicity had above-median levels of PSS, and levels of PSS were similar among participants who were Black (Table S6).
We estimated the associations between each PFAS and birth outcome with adjusted linear and logistic regression models (Table 2). We observed lower birthweight-for-gestational-age z-scores with increasing concentrations of all PFAS. For a 1-unit higher log-normalized exposure to PFOA, PFOS, PFNA, PFHxS, and PFDA, we observed a lower birthweight-for-gestational-age z-score of β=−0.15 [95% confidence interval (CI): −0.27, −0.03], β=−0.14 (95% CI: −0.28, −0.002), β=−0.22 (95% CI: −0.33, −0.10), β=−0.06 (95% CI: −0.18, 0.06), and β=−0.25 (95% CI: −0.37, −0.14), respectively. Positive point estimates for PFAS and risk of small-for-gestational-age were consistent for all PFAS, with ORs ranging from 1.06 to 1.29, although 95% CIs for all estimates included the null. We observed lower odds ratios (ORs) of large-for-gestational-age, with estimates for PFNA and PFDA excluding the null: ORPFNA=0.56 (95% CI: 0.38, 0.83), and ORPFDA=0.52 (95% CI: 0.35, 0.77). Point estimates for all PFAS showed increased risk of term low birth weight, with ORs ranging from 1.13 to 2.24, although 95% CIs included the null. All PFAS showed increased risk of preterm birth and decreased gestational age at birth, although all but one estimate included the null in fully adjusted models (βPFOA=−0.22; 95% CI: −0.43, −0.01) with the exception of PFHxS (Table 2).
Table 2 Associations of continuous measures of prenatal natural log-transformed PFAS (ng/mL) concentrations and risk of adverse birth outcomes in selected ECHO cohorts.
PFAS Birthweight-for-gestational-age z-scores Small-for-gestational-agea Large-for-gestational-agea Term low birth weight Preterm birth Gestational age at birth (wk)
N β (95% CI) N OR (95% CI) N OR (95% CI) N OR (95% CI) N OR (95% CI) N β (95% CI)
PFOA
Model 1 3,099 −0.27 (−0.38, −0.15) 2,752 1.20 (0.83, 1.74) 2,791 0.55 (0.38, 0.79) 2,815 1.29 (0.53, 3.16) 3,063 1.27 (0.86, 1.87) 3,102 0.03 (−0.16, 0.23)
Model 2 3,099 −0.15 (−0.27, −0.03) 2,752 1.07 (0.72, 1.59) 2,791 0.75 (0.51, 1.12) 2,815 1.43 (0.54, 3.80) 3,063 1.41 (0.93, 2.14) 3,102 −0.22 (−0.43, −0.01)
PFOS
Model 1 3,099 −0.27 (−0.41, −0.14) 2,752 1.23 (0.80, 1.89) 2,791 0.61 (0.40, 0.94) 2,815 1.40 (0.50, 3.92) 3,063 1.18 (0.71, 1.96) 3,102 0.06 (−0.18, 0.29)
Model 2 3,099 −0.14 (−0.28, −0.00) 2,752 1.06 (0.68, 1.65) 2,791 0.87 (0.55, 1.39) 2,815 1.21 (0.43, 3.39) 3,063 1.29 (0.76, 2.18) 3,102 −0.16 (−0.40, 0.09)
PFNA
Model 1 3,099 −0.29 (−0.41, −0.18) 2,752 1.18 (0.81, 1.71) 2,791 0.46 (0.32, 0.67) 2,815 1.42 (0.58, 3.49) 3,063 1.31 (0.87, 1.97) 3,102 0.01 (−0.20, 0.21)
Model 2 3,099 −0.22 (−0.33, −0.10) 2,752 1.09 (0.74, 1.60) 2,791 0.56 (0.38, 0.83) 2,815 1.67 (0.64, 4.35) 3,063 1.43 (0.93, 2.19) 3,102 −0.17 (−0.38, 0.04)
PFHxS
Model 1 3,099 −0.12 (−0.23, 0.00) 2,752 1.29 (0.89, 1.89) 2,791 0.73 (0.51, 1.04) 2,815 1.13 (0.47, 2.71) 3,063 0.86 (0.55, 1.34) 3,102 0.24 (0.04, 0.44)
Model 2 3,099 −0.06 (−0.18, 0.06) 2,752 1.25 (0.84, 1.87) 2,791 0.86 (0.59, 1.25) 2,815 1.14 (0.46, 2.84) 3,063 0.97 (0.61, 1.55) 3,102 0.02 (−0.19, 0.23)
PFDA
Model 1 3,047 −0.30 (−0.41, −0.18) 2,701 1.22 (0.85, 1.76) 2,744 0.47 (0.32, 0.69) 2,770 1.93 (0.86, 4.34) 3,011 1.16 (0.77, 1.75) 3,050 −0.01 (−0.22, 0.19)
Model 2 3,047 −0.25 (−0.37, −0.14) 2,701 1.18 (0.81, 1.73) 2,744 0.52 (0.35, 0.77) 2,770 2.24 (0.96, 5.24) 3,011 1.22 (0.80, 1.86) 3,050 −0.11 (−0.32, 0.09)
Note: Beta coefficients (βs) and ORs represent 1 log-unit increase in PFAS concentration (ng/mL) and are presented with 95% CI. Model 1 was adjusted for cohort (dummy variables). Model 2 was additionally adjusted for maternal race/ethnicity (Hispanic, non-Hispanic White, Black, Asian, other), for maternal educational attainment (<high school, high school degree/GED, some college, bachelor’s degree or higher), maternal age at delivery (<25, 25–29, 30–34, ≥35 y), parity (0, ≥1). CI, confidence interval; ECHO, Environmental influences on Child Health Outcomes; GED, General Educational Development; OR, odds ratio; PFAS, per- and polyfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid.
a Appropriate-for-gestational-age is the referent from both small- and large-for-gestational age estimates.
When stratified by PSS, associations between some PFAS and birthweight-for-gestatonal-age z-scores were stronger (i.e., larger decreases) among those who reported below-median levels of perceived stress, although tests did not show evidence of statistical interaction (Table 3). Similar results were observed for large-for-gestational-age with stronger decreased risk among those with lower perceived stress (Table 3). Three estimates had interaction terms with p<0.1, although not in a consistent direction; PFOS was associated with increased risk of small-for-gestational-age among those with lower perceived stress (OR=1.57; 95% CI: 0.73, 3.38), PFHxS with increased risk of large-for-gestational-age among those with higher perceived stress (OR=1.13; 95% CI: 0.51, 2.49), and PFDA with increased risk of term low birth weight among those with higher perceived stress (OR=5.25; 95% CI: 1.08, 25.64) (Table 3). Some associations were stronger in the subsample with PSS scores, including increased PFHxS and lower birthweight-for-gestational-age, increased PFOA and lower risk of large-for-gestational-age, and increased PFOA and PFNA and increased risk of preterm birth (Table 3).
Table 3 Associations of prenatal natural log-transformed PFAS (ng/mL) and birth outcomes stratified by perceived stress study population among selected ECHO cohorts.
PFAS Birthweight-for-gestational-age z-scores Small-for-gestational-agea Large-for-gestational-agea Term low birth weight Preterm birth Gestational age at birth (wk)
N β (95% CI) N OR (95% CI) N OR (95% CI) N OR (95% CI) N OR (95% CI) N β (95% CI)
PFOA
Pooled 1,830 −0.18 (−0.32, −0.05) 1,658 1.04 (0.66, 1.64) 1,620 0.54 (0.33, 0.88) 1,628 1.30 (0.40, 4.19) 1,766 1.72 (1.05, 2.82) 1,831 −0.21 (−0.44, 0.02)
Low PSS 900 −0.19 (−0.37, −0.00) 804 0.97 (0.53, 1.79) 797 0.53 (0.28, 1.01) 673 0.98 (0.20, 4.71) 839 1.56 (0.82, 2.98) 901 −0.14 (−0.44, 0.15)
High PSS 930 −0.18 (−0.38, 0.03) 842 1.16 (0.57, 2.38) 813 0.66 (0.31, 1.41) 599 3.56 (0.41, 30.66) 904 1.94 (0.88, 4.24) 930 −0.32 (−0.67, 0.04)
p-Valueb 0.95 0.41 0.80 0.13 0.18 0.22
PFOS
Pooled 1,830 −0.08 (−0.25, 0.09) 1,658 0.92 (0.55, 1.54) 1,620 0.88 (0.48, 1.61) 1,628 0.96 (0.29, 3.14) 1,766 1.50 (0.79, 2.85) 1,831 −0.11 (−0.39, 0.16)
Low PSS 900 −0.14 (−0.39, 0.11) 804 1.57 (0.73, 3.38) 797 0.83 (0.35, 1.97) 673 0.62 (0.13, 2.98) 839 1.98 (0.74, 5.29) 901 −0.14 (−0.54, 0.26)
High PSS 930 −0.01 (−0.24, 0.22) 842 0.52 (0.26, 1.04) 813 1.12 (0.46, 2.73) 599 1.70 (0.28, 10.20) 904 1.24 (0.53, 2.93) 930 −0.12 (−0.51, 0.27)
p-Valueb 0.37 0.06 0.70 0.19 0.94 0.58
PFNA
Pooled 1,830 −0.22 (−0.35, −0.08) 1,658 1.11 (0.71, 1.73) 1,620 0.54 (0.33, 0.88) 1,628 1.83 (0.57, 5.90) 1,766 1.71 (1.03, 2.85) 1,831 −0.16 (−0.39, 0.07)
Low PSS 900 −0.33 (−0.53, −0.14) 804 1.76 (0.94, 3.30) 797 0.49 (0.25, 0.98) 673 1.71 (0.38, 7.60) 839 1.92 (0.95, 3.88) 901 −0.23 (−0.54, 0.08)
High PSS 930 −0.08 (−0.28, 0.12) 842 0.67 (0.35, 1.30) 813 0.69 (0.33, 1.46) 599 2.44 (0.39, 15.17) 904 1.59 (0.74, 3.42) 930 −0.11 (−0.45, 0.23)
p-Valueb 0.23 0.23 0.70 0.23 0.63 0.85
PFHxS
Pooled 1,830 −0.17 (−0.33, −0.01) 1,658 1.54 (0.91, 2.59) 1,620 0.66 (0.38, 1.14) 1,628 0.78 (0.24, 2.50) 1,766 0.98 (0.52, 1.81) 1,831 0.06 (−0.21, 0.32)
Low PSS 900 −0.16 (−0.40, 0.07) 804 1.27 (0.59, 2.75) 797 0.47 (0.22, 1.03) 673 0.28 (0.05, 1.44) 839 1.87 (0.73, 4.81) 901 −0.03 (−0.41, 0.35)
High PSS 930 −0.15 (−0.37, 0.07) 842 1.87 (0.89, 3.91) 813 1.13 (0.51, 2.49) 599 1.94 (0.33, 11.54) 904 0.53 (0.23, 1.23) 930 0.16 (−0.22, 0.54)
p-Valueb 0.41 0.86 0.09 0.15 0.11 0.87
PFDA
Pooled 1,781 −0.23 (−0.38, −0.08) 1,610 1.08 (0.68, 1.74) 1,576 0.57 (0.32, 1.00) 1,586 2.23 (0.74, 6.68) 1,717 1.38 (0.81, 2.34) 1,782 −0.09 (−0.34, 0.16)
Low PSS 874 −0.32 (−0.53, −0.11) 778 1.28 (0.67, 2.46) 775 0.44 (0.20, 0.96) 648 1.00 (0.19, 5.34) 813 1.27 (0.61, 2.63) 875 −0.28 (−0.61, 0.06)
High PSS 907 −0.12 (−0.34, 0.10) 820 0.86 (0.43, 1.73) 791 0.83 (0.36, 1.93) 582 5.25 (1.08, 25.64) 881 1.42 (0.65, 3.13) 907 0.15 (−0.22, 0.53)
p-Valueb 0.42 0.56 0.38 0.07 0.79 0.06
Note: Pooled rows represent the combined high and low PSS groups for comparison. Beta coefficients (βs) and ORs represent 1 log-unit increase in PFAS concentration (ng/mL) and are presented with 95% CIs. Low PSS, below-median PSS t-scores. High PSS, above-median PSS t-scores. Models were adjusted for cohort, maternal race/ethnicity (Hispanic, non-Hispanic White, Black, Asian, other), for maternal educational attainment (<high school, high school degree/GED, some college, bachelor’s degree or higher), maternal age at delivery (<25, 25–29, 30–34, ≥35 y), parity (0, ≥1). CI, confidence interval; ECHO, Environmental influences on Child Health Outcomes; GED, General Educational Development; OR, odds ratio; PFAS, per- and polyfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid; PSS, perceived stress scale.
a Appropriate-for-gestational-age is the referent from both small- and large-for-gestational age estimates.
b p-Value of interaction term (PSS×PFAS).
Bayesian Weighted Sums results were largely consistent with the main findings (Table 4). The change in birthweight-for-gestational-age z-scores for a 1-unit increase in the sum of logged PFAS was −0.28 (95% HPD: −0.44, −0.14). The odds of small- and large-for-gestational-age associated with summed PFAS were 1.13 (95% HPD: 0.68, 1.83) and 0.49 (95% HPD: 0.29, 0.82), respectively. The percentages of the summed effect for birthweight-for-gestational-age z-scores and large-for-gestational-age explained by PFDA were 40%. The percentages of the summed effect explained by each PFAS for small-for-gestational-age were approximately equal to one another (Table 5). Odds of preterm birth and term low birth weight were both elevated for the summed effect of PFAS: 1.45 (95% HPD: 0.82, 2.55) and 1.04 (95% HPD: 1.00, 1.07), respectively. Among those with low PSS, associations between PFAS and each birth outcome were consistent and stronger than for those with high PSS, although all 95% HPD included the null (Table 4).
Table 4 Estimates of the Risk Difference (RD) from Bayesian Weighted Sums analysis of selected prenatal natural log-transformed PFAS (ng/mL) and risk of adverse birth outcomes in 11 selected ECHO cohorts.
Birthweight-for-gestational-age z-scores Small-for-gestational-agea Large-for-gestational-agea Preterm birth Term low birth weight Gestational age at birth (weeks)
Categories RD 95% HPD N OR 95% HPD N OR 95% HPD N OR 95% HPD N OR 95% HPD N RD 95% HPD N
Summed effect −0.28 (−0.44, −0.14) 3,083 1.13 (0.68, 1.83) 2,734 0.49 (0.29, 0.82) 2,776 1.45 (0.82, 2.55) 3,086 1.04 (1.00, 1.07) 3,086 −0.23 (−0.51, 0.05) 3,086
Low PSS −0.41 (−0.68, −0.14) 878 1.39 (0.52, 3.48) 782 0.31 (0.11, 0.79) 788 2.03 (0.73, 5.54) 879 1.03 (0.97, 1.09) 879 −0.24 (−0.69, 0.20) 879
High PSS −0.15 (−0.43, 0.12) 911 0.80 (0.31, 2.17) 836 0.67 (0.24, 1.81) 805 1.56 (0.53, 4.93) 911 1.06 (1.00, 1.12) 911 −0.08 (−0.58, 0.42) 911
Note: Model was adjusted for cohort, maternal race/ethnicity (Hispanic, non-Hispanic White, Black, Asian, other), for maternal educational attainment (<high school, high school degree/GED, some college, Bachelor’s degree or higher), maternal age at delivery (<25, 25–29, 30–34, ≥35 years), parity (0, 1+). RDs and ORs represent 1 log unit increase in PFAS concentrations (ng/mL) and are presented with 95% HDP. ECHO, Environmental influences on Child Health Outcomes; GED, General Educational Development, HDP, highest posterior density; OR, odds ratio; PFAS, per-and polyfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid; RD, risk difference.
a Appropriate-for-gestational-age is the referent from both small- and large-for-gestational age estimates.
Table 5 Weights of each PFAS in Bayesian Weighted Sums analysis of selected prenatal natural log-transformed PFAS (ng/mL) and risk of adverse birth outcomes in 11 selected ECHO cohorts.
Birthweight-for-gestational-age z-scores Small-for-gestational-agea Large-for-gestational-agea Preterm birth Term low birth weight Gestational age at birth (weeks)
Categories Weights 95% HPD Weights 95% HPD Weights 95% HPD Weights 95% HPD Weights 95% HPD Weights 95% HPD
Summed effect
PFOA 0.15 (0.00, 0.38) 0.20 (0.00, 0.51) 0.13 (0.00, 0.35) 0.22 (0.00, 0.57) 0.24 (0.00, 0.58) 0.26 (0.00, 0.61)
PFOS 0.12 (0.00, 0.34) 0.21 (0.00, 0.55) 0.11 (0.00, 0.30) 0.19 (0.00, 0.49) 0.16 (0.00, 0.42) 0.20 (0.00, 0.51)
PFNA 0.22 (0.00, 0.52) 0.20 (0.00, 0.53) 0.23 (0.00, 0.55) 0.22 (0.00, 0.55) 0.18 (0.00, 0.46) 0.20 (0.00, 0.53)
PFHxS 0.10 (0.00, 0.27) 0.20 (0.00, 0.53) 0.14 (0.00, 0.37) 0.16 (0.00, 0.45) 0.13 (0.00, 0.37) 0.14 (0.00, 0.40)
PFDA 0.40 (0.02, 0.71) 0.20 (0.00, 0.52) 0.40 (0.03, 0.74) 0.20 (0.00, 0.50) 0.29 (0.00, 0.62) 0.20 (0.00, 0.51)
Low PSS
PFOA 0.14 (0.00, 0.39) 0.17 (0.00, 0.47) 0.18 (0.00, 0.46) 0.19 (0.00, 0.50) 0.20 (0.00, 0.52) 0.19 (0.00, 0.49)
PFOS 0.13 (0.00, 0.36) 0.21 (0.00, 0.56) 0.13 (0.00, 0.37) 0.19 (0.00, 0.51) 0.18 (0.00, 0.47) 0.18 (0.00, 0.49)
PFNA 0.30 (0.00, 0.63) 0.22 (0.00, 0.57) 0.19 (0.00, 0.47) 0.24 (0.00, 0.57) 0.22 (0.00, 0.54) 0.22 (0.00, 0.55)
PFHxS 0.18 (0.00, 0.44) 0.20 (0.00, 0.52) 0.27 (0.00, 0.58) 0.20 (0.00, 0.51) 0.19 (0.00, 0.49) 0.17 (0.00, 0.47)
PFDA 0.24 (0.00, 0.54) 0.19 (0.00, 0.50) 0.22 (0.00, 0.53) 0.18 (0.00, 0.47) 0.21 (0.00, 0.52) 0.24 (0.00, 0.58)
High PSS
PFOA 0.22 (0.00, 0.56) 0.17 (0.00, 0.49) 0.21 (0.00, 0.53) 0.24 (0.00, 0.57) 0.22 (0.00, 0.54) 0.23 (0.00, 0.60)
PFOS 0.17 (0.00, 0.47) 0.24 (0.00, 0.60) 0.18 (0.00, 0.49) 0.19 (0.00, 0.50) 0.16 (0.00, 0.44) 0.20 (0.00, 0.53)
PFNA 0.19 (0.00, 0.50) 0.20 (0.00, 0.52) 0.22 (0.00, 0.56) 0.21 (0.00, 0.53) 0.17 (0.00, 0.46) 0.19 (0.00, 0.51)
PFHxS 0.22 (0.00, 0.54) 0.19 (0.00, 0.54) 0.19 (0.00, 0.50) 0.17 (0.00, 0.48) 0.14 (0.00, 0.38) 0.19 (0.00, 0.52)
PFDA 0.21 (0.00, 0.53) 0.20 (0.00, 0.53) 0.20 (0.00, 0.51) 0.20 (0.00, 0.51) 0.31 (0.00, 0.67) 0.19 (0.00, 0.51)
Note: Model was adjusted for cohort, maternal race/ethnicity (Hispanic, non-Hispanic White, Black, Asian, other), for maternal educational attainment (<high school, high school degree/GED, some college, Bachelor’s degree or higher), maternal age at delivery (<25, 25–29, 30–34, ≥35 years), parity (0, 1+). RDs and ORs represent 1 log unit increase in PFAS concentrations (ng/mL) and are presented with 95% HDP. ECHO, Environmental influences on Child Health Outcomes; GED, General Educational Development; HDP, highest posterior density; OR, odds ratio; PFAS, per-and polyfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid; RD, risk difference.
a Appropriate-for-gestational-age is the referent from both small- and large-for-gestational age estimates.
Associations between most PFAS and birth outcomes were stronger among female compared with male infants. Eight of 30 interaction terms had p<0.1, and three of those with p<0.05 are noted here. Among female infants, PFOA, PFOS, and PFNA were associated with decreased birthweight-for-gestational-age (βPFNA=−0.29; 95% CI: −0.46, −0.12). Decreased odds of large-for-gestational-age were also stronger in females for several PFAS (ORPFOA=0.54; 95% CI: 0.31, 0.93; ORPFNA=0.38; 95% CI: 0.22, 0.66; ORPFDA=0.39; 95% CI: 0.22, 0.69) (Table 6).
Table 6 Associations of prenatal natural log-transformed PFAS concentrations (ng/mL) and risk of adverse birth outcomes stratified by infant sex in selected ECHO cohorts.
PFAS Birthweight-for-gestational-age z-scores Small-for-gestational-agea Large-for-gestational-agea Term low birth weight Preterm birth Gestational age at birth (wk)
N β (95% CI) N OR (95% CI) N OR (95% CI) N OR (95% CI) N OR (95% CI) N β (95% CI)
PFOA
Male 1,525 −0.12 (−0.29, 0.05) 1,344 1.23 (0.67, 2.24) 1,378 1.11 (0.62, 1.98) 1,028 1.78 (0.29, 11.05) 1,487 1.69 (0.95, 3.03) 1,526 −0.19 (−0.49, 0.11)
Female 1,574 −0.18 (−0.35, −0.02) 1,407 0.93 (0.55, 1.58) 1,398 0.54 (0.31, 0.93) 1,441 1.25 (0.37, 4.23) 1,556 1.04 (0.56, 1.91) 1,576 −0.22 (−0.52, 0.08)
p-Valueb 0.06 0.66 0.03 0.26 0.54 0.24
PFOS
Male 1,525 −0.06 (−0.27, 0.14) 1,344 0.92 (0.47, 1.78) 1,378 1.30 (0.67, 2.53) 1,028 0.56 (0.08, 3.80) 1,487 1.36 (0.64, 2.88) 1,526 −0.20 (−0.56, 0.15)
Female 1,574 −0.21 (−0.39, −0.02) 1,407 1.14 (0.62, 2.10) 1,398 0.60 (0.31, 1.16) 1,441 1.41 (0.41, 4.92) 1,556 1.05 (0.49, 2.25) 1,576 −0.08 (−0.42, 0.26)
p-Valueb 0.04 0.07 0.16 0.28 0.16 0.21
PFNA
Male 1,525 −0.14 (−0.31, 0.03) 1,344 1.09 (0.61, 1.95) 1,378 0.84 (0.47, 1.51) 1,028 1.16 (0.22, 6.24) 1,487 1.74 (0.96, 3.13) 1,526 −0.27 (−0.57, 0.03)
Female 1,574 −0.29 (−0.46, −0.12) 1,407 1.05 (0.62, 1.79) 1,398 0.38 (0.22, 0.66) 1,441 1.81 (0.53, 6.24) 1,556 1.02 (0.54, 1.90) 1,576 −0.04 (−0.34, 0.26)
p-Valueb 0.04 0.92 <0.01 0.12 0.92 0.89
PFHxS
Male 1,525 −0.05 (−0.23, 0.13) 1,344 1.17 (0.62, 2.19) 1,378 0.91 (0.53, 1.56) 1,028 0.19 (0.03, 1.18) 1,487 0.96 (0.49, 1.90) 1,526 0.11 (−0.19, 0.42)
Female 1,574 −0.05 (−0.22, 0.11) 1,407 1.30 (0.77, 2.21) 1,398 0.85 (0.50, 1.46) 1,441 1.98 (0.70, 5.64) 1,556 0.92 (0.47, 1.80) 1,576 −0.04 (−0.33, 0.25)
p-Valueb 0.35 0.13 0.87 0.04 0.61 0.21
PFDA
Male 1,496 −0.20 (−0.37, −0.03) 1,316 1.19 (0.67, 2.13) 1,353 0.69 (0.40, 1.21) 1,005 1.68 (0.32, 8.93) 1,458 1.32 (0.73, 2.38) 1,497 −0.20 (−0.50, 0.09)
Female 1,551 −0.31 (−0.47, −0.14) 1,384 1.12 (0.67, 1.87) 1,376 0.39 (0.22, 0.69) 1,419 2.23 (0.80, 6.26) 1,533 1.09 (0.59, 2.00) 1,553 −0.04 (−0.33, 0.26)
p-Valueb 0.34 0.87 0.08 0.36 0.86 0.64
Note: Models were adjusted for cohort (dummy variables), maternal race/ethnicity (Hispanic, non-Hispanic White, Black, Asian, other), for maternal educational attainment (<high school, high school degree/GED, some college, bachelor’s degree or higher), maternal age at delivery (<25, 25–29, 30–34, ≥35 y), parity (0, ≥1). CI, confidence interval; ECHO, Environmental influences on Child Health Outcomes; GED, General Educational Development; OR, odds ratio; PFAS, per- and polyfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctanesulfonic acid.
a Appropriate-for-gestational-age is the referent from both small- and large-for-gestational age estimates.
b p-Value of interaction term (infant sex×PFAS).
When stratified by trimester of exposure, some results were stronger in the first trimester for PFNA and PFDA and birthweight-for-gestational-age (Table S7). The estimates were less precise, and study populations differed between trimesters, with fewer participants in the third trimester.
When each cohort was removed from the pooled analysis at a time, most of the results were similar (Figure S5). In some cases, excluding the Project Viva, Atlanta, or Maternal And Developmental Risks from Environmental and Social Stressors cohorts influenced the results in various directions, but the results were overall consistent.
In general, associations of PFAS quartiles were consistent with the continuous main results for birthweight-for-gestational-age z-scores and risk of large-for-gestational-age. Quartile analyses showed associations with increased odds of preterm birth when exposed to the highest quartile of PFOA (OR=2.87; 95% CI: 1.28, 6.44) and PFNA (OR=1.74; 95% CI: 1.05, 2.89) (Figure S6, Table S8), where continuous associations were in the same direction with 95% CIs that included the null (Table 3). When using a mixed effects model, allowing for random effects by cohort, we not see notable changes in either point estimates or CIs (Table S9). Results did not differ when adjusted for prenatal exposure to tobacco smoke, which included maternal smoking and secondhand smoke during pregnancy (Table S10). Estimates of changes in birthweight (in grams) associated with an interquartile increase in PFAS (not log transformed) showed consistent results in terms of directionality of the association (Table S11). The largest decrements in birthweight were associated with increases in PFNA (β=−15.99; 95% CI: −29.77, −2.22) and PFDA (β=−15.76; 95% CI: −26.81, −4.71) (Table S11).
Discussion
This is the largest study, to the best of our knowledge, in the United States of pregnancy exposures to PFAS and adverse birth outcomes. We found that higher levels of several PFAS were associated with lower birthweight-for-gestational-age z-scores and lower risk of being large-for-gestational-age. Associations between PFAS and preterm birth and term low birth weight were also observed, although results were less robust.
Associations between PFAS and birth outcomes were not modified by perceived stress. These findings were unexpected because of our hypothesis that exposure to chemical and social stressors would result in stronger associations; however, given the known associations between stress and birthweight, the additional effect of PFAS may be minimal.59
The ECHO study population for this analysis included pregnancies from 11 cohorts in seven states across the United States. This unique and demographically diverse study population enabled us to examine five PFAS measured prenatally and their association with continuous and categorical birth outcomes related to gestational age and birthweight. Statistical power allowed for stratified analyses to explore potential effect modifiers and sensitivity analyses to explore potential bias stemming from timing during pregnancy, assumptions of linearity or threshold effects, additional confounders, and influence by different cohorts spanning time and place.
In our study, birth years of the children spanned 21 years (1999–2019), during which time there was an overall decrease in exposures to PFAS owing to the phase-out of some PFAS. Correlations were generally high across trimesters, providing evidence that PFAS levels remain relatively consistent across pregnancy. Our analysis removing one cohort at a time showed that a few cohorts deviated from the pattern, but overall they were notably consistent. Their results were published previously,30 as were results for several other individual cohorts, including Chemicals in Our Bodies, Illinois Kids Development Studies, and Healthy Start.31,60–62
Our findings of lower birthweight-for-gestational-age z-scores confirm previous studies wherein PFAS were associated with lower birthweight-for-gestational-age, intrauterine growth restriction, and small-for-gestational-age, and reduced fetal growth.16,27,32,60 Our findings overall support a shift in the distribution of birthweight toward decreased birth size measured continuously (i.e., birthweight-for-gestational-age z-scores) and categorically (i.e., large-for-gestational-age) and are suggestive of increased risk of preterm birth.
Despite some inconsistencies in previous studies and meta-analyses, our findings confirm the recent report from the National Academy of Science (NAS) stating there is sufficient evidence of an association between PFAS and decreased infant and fetal growth, which weighted evidence based on low risk of bias.63 For example, two meta-analyses of birthweight in relation to PFOA64 and PFOS65 found decreases in birthweight, which is consistent with our results; notably, when restricted to studies earlier in pregnancy, associations in these meta-analyses were null. In contrast, results of our study show stronger associations between increased PFOA in the first trimester and lower birthweight-for-gestational-age z-scores and increased risk of term low birth weight and small-for-gestational-age. There are several possibilities as to why these meta-analyses may differ, such as the inclusion of studies that did not adjust for gestational age and/or parity, were cross-sectional in design, were conducted in study populations outside of the United States, or were driven by a single study.64,65 Among samples with PFAS exposures at multiple times in pregnancy in our study, concentrations were strongly correlated across trimesters (Table S4). Glomerular filtration rate (GFR) has been suggested as a confounder of trimester-specific associations of PFAS with birth outcomes; however, a recent systematic analysis of the PFAS literature by the NAS found that the available evidence of PFAS on GFR were insufficient to determine a relationship.63 Further, the Project Viva cohort included in this study previously found that GFR did not confound the relationship between PFAS and birth outcomes.30 GFR levels were not available for other cohorts; however, if GFR were to confound the PFAS-birthweight relationship, it would be expected to do so later in pregnancy. Our trimester-specific analysis did not support this potential confounding or reverse causality. Finally, effects on birthweight have been found in multiple animal species including mouse, rat, zebrafish, and fruit flies.66
Our study found an association between PFOS and preterm birth consistent with prior work,30,67,68 including a recent review and meta-analysis showing a linear positive association between PFOS and risk of preterm birth32; however, our results were not as precisely estimated. Given that preterm birth is a multifactorial outcome and PFAS may contribute to a small risk increase, large studies (and/or highly exposed participants) are needed to find such effects.
Our findings are consistent with a previous study in which associations between PFAS and birthweight-for-gestational-age z-scores were stronger among females,27 but they contradict another study that found stronger associations among males.69 Biological mechanisms by which PFAS may affect birth outcomes are largely unknown, but research has investigated potential pathways including endocrine disruption,70 systemic inflammation,71 metabolic dysfunction,72 placental function,73 and epigenetic changes.74
Despite our large sample size, uncertainties in our estimates remain. Our study was limited to participants with nonmissing data on key variables. In addition, some PFAS were not able to be examined because levels were below the LOD. As legacy PFAS are phased out and replaced with alternative PFAS, our studies must be updated with changing levels to be examined in relation to multifactorial health outcomes. Methodologically, there is no agreed-upon approach to evaluate the effects of PFAS, or other chemicals, as a mixture. Our Bayesian Weighted Sums approach assumes linearity of the summed effect of PFAS, which appeared defensible based on the results exploring effects of PFAS by exposure quartiles (Table S8, Figure S6).
Future studies can address some of these limitations. A large study such as ECHO may be able to better investigate mediation effects of prepregnancy BMI and maternal conditions, such as gestational diabetes and hypertensive disorders in pregnancy, that may be on the causal pathway between PFAS and fetal growth once more of those data become available. Similarly, future studies can examine interaction with other environmental chemicals. Furthermore, birthweight is a single measurement in time, and further studies are needed to investigate the potential impact of PFAS on infant and child health outcomes.
In conclusion, we found that maternal PFAS concentrations during pregnancy are associated with lower birthweight-for-gestational-age z-scores and suggestive of an association with preterm birth. These associations are consistent with previous studies showing decreased birth weight/fetal growth. Associations were stronger among females, although fewer previous studies were able to confirm these findings. We did not find these associations to differ between mothers with high vs. low perceived stress. Given the persistence of PFAS in the environment and human bodies, ubiquitous exposure, and the transfer of maternal PFAS in utero and during breastfeeding, disruption of fetal growth remains a health threat in offspring and needs to be addressed as part of efforts evaluating interventions and prevention.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We acknowledge the contribution of the following Environmental influences on Child Health Outcomes (ECHO) program collaborators: Coordinating Center: Duke Clinical Research Institute, Durham, North Carolina: P.B. Smith, K.L. Newby, and D.K. Benjamin; Data Analysis Center: Johns Hopkins University Bloomberg School of Public Health, Baltimore, Maryland: L.P. Jacobson; Research Triangle Institute, Durham, North Carolina: C.B. Parker; Person-Reported Outcomes Core: Northwestern University, Evanston, Illinois: R. Gershon and D. Cella; Children’s Health Exposure Analysis Resource: Wadsworth Center, Albany, New York: P. Parsons and K. Kurunthacalam.
G.B.H, X.N., and S.B. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
We thank our ECHO colleagues; the medical, nursing, and program staff; as well as the children and families participating in the ECHO cohorts.
Research reported in this publication was supported by the ECHO program, Office of The Director, National Institutes of Health (NIH), under award nos. U2COD023375 (Coordinating Center), U24OD023382 (Data Analysis Center, G.B.H., X.N., and S.B.); U24OD023319 (PRO Core); P01ES022841; RD83543301, R01ES027051 (T.J.W.) UH3OD023272 (S.L.S., T.J.W., R.M.-F., S.K., A.M.P., S.G., S.M.E., and D.E.G.); UH3OD023349, R01HD083369, UH3OD023349, P30ES005022 (T.G.O. and E.S.B.); UH3OD023286, UH3OD023318, R01NR014800, R24ESO29490, P50ESO2607, EPA 83615301 (A.L.D.); P30ES007048, P50ES026086, 83615801, P50MD01570, UH3OD023287 (C.V.B. and T.B.), UH3OD02333, UG3OD023316 (M.S.B.), UH3OD023289 (A.F.), UH3OD023275, NIGMS P20GM104416 (M.R.K. and M.E.R.), UH30D023342 (D.H.B. and R.J.S.), U2CES026542 (K.K.), and UH3OD023248 (A.P.S.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
==== Refs
References
1. Li Y, Fletcher T, Mucs D, Scott K, Lindh CH, Tallving P, et al. 2018. Half-lives of PFOS, PFHxS and PFOA after end of exposure to contaminated drinking water. Occup Environ Med 75 (1 ):46–51, PMID: , 10.1136/oemed-2017-104651.29133598
2. U.S. EPA (U.S. Environmental Protection Agency). 2021. PFOA, PFOS and Other PFAS. https://www.epa.gov/pfas/basic-information-pfas [accessed 8 November 2021].
3. Wu Q, Kannan K. 2019. Perfluoroalkyl Substances (PFASs) in Foodstuffs and Human Dietary Exposure. In: Advances in the Determination of Xenobiotics in Foods. Gomara B, Marina ML, eds. Potomac, MD: Bentham Science Publishers, 259–313.
4. Woodruff TJ, Zota AR, Schwartz JM. 2011. Environmental chemicals in pregnant women in the United States: NHANES 2003–2004. Environ Health Perspect 119 (6 ):878–885, PMID: , 10.1289/ehp.1002727.21233055
5. Bartell SM, Vieira VM. 2021. Critical review on PFOA, kidney cancer, and testicular cancer. J Air Waste Manag Assoc 71 (6 ):663–679, PMID: , 10.1080/10962247.2021.1909668.33780327
6. Lin PID, Cardenas A, Hauser R, Gold DR, Kleinman KP, Hivert MF, et al. 2019. Per- and polyfluoroalkyl substances and blood lipid levels in pre-diabetic adults—longitudinal analysis of the Diabetes Prevention Program Outcomes Study. Environ Int 129 :343–353, PMID: , 10.1016/j.envint.2019.05.027.31150976
7. Bommarito PA, Ferguson KK, Meeker JD, McElrath TF, Cantonwine DE. 2021. Maternal levels of perfluoroalkyl substances (PFAS) during early pregnancy in relation to preeclampsia subtypes and biomarkers of preeclampsia risk. Environ Health Perspect 129 (10 ):107004, PMID: , 10.1289/EHP9091.34637358
8. Watkins DJ, Josson J, Elston B, Bartell SM, Shin HM, Vieira VM, et al. 2013. Exposure to perfluoroalkyl acids and markers of kidney function among children and adolescents living near a chemical plant. Environ Health Perspect 121 (5 ):625–630, PMID: , 10.1289/ehp.1205838.23482063
9. Xie LN, Wang XC, Su LQ, Ji SS, Dong XJ, Zhu HJ, et al. 2022. Serum concentrations of per-/polyfluoroalkyl substances and its association with renal function parameters among teenagers near a Chinese fluorochemical industrial plant: a cross-sectional study. Environ Pollut 302 :119020, PMID: , 10.1016/j.envpol.2022.119020.35183668
10. Rickard BP, Rizvi I, Fenton SE. 2022. Per- and poly-fluoroalkyl substances (PFAS) and female reproductive outcomes: PFAS elimination, endocrine-mediated effects, and disease. Toxicology 465 :153031, PMID: , 10.1016/j.tox.2021.153031.34774661
11. Carwile JL, Seshasayee SM, Aris IM, Rifas-Shiman SL, Claus Henn B, Calafat AM, et al. 2021. Prospective associations of mid-childhood plasma per- and polyfluoroalkyl substances and pubertal timing. Environ Int 156 :106729, PMID: , 10.1016/j.envint.2021.106729.34171588
12. Liu P, Yang F, Wang Y, Yuan Z. 2018. Perfluorooctanoic acid (PFOA) exposure in early life increases risk of childhood adiposity: a meta-analysis of prospective cohort studies. Int J Environ Res Public Health 15 (10 ):2070, PMID: , 10.3390/ijerph15102070.30241417
13. Grandjean P. 2018. Delayed discovery, dissemination, and decisions on intervention in environmental health: a case study on immunotoxicity of perfluorinated alkylate substances. Environ Health 17 (1 ):62, PMID: , 10.1186/s12940-018-0405-y.30060739
14. Steenland K, Fletcher T, Savitz DA. 2010. Epidemiologic evidence on the health effects of perfluorooctanoic acid (PFOA). Environ Health Perspect 118 (8 ):1100–1108, PMID: , 10.1289/ehp.0901827.20423814
15. Sunderland EM, Hu XC, Dassuncao C, Tokranov AK, Wagner CC, Allen JG. 2019. A review of the pathways of human exposure to poly- and perfluoroalkyl substances (PFASs) and present understanding of health effects. J Expo Sci Environ Epidemiol 29 (2 ):131–147, PMID: , 10.1038/s41370-018-0094-1.30470793
16. Lam J, Koustas E, Sutton P, Johnson PI, Atchley DS, Sen S, et al. 2014. The Navigation Guide—evidence-based medicine meets environmental health: integration of animal and human evidence for PFOA effects on fetal growth. Environ Health Perspect 122 (10 ):1040–1051, PMID: , 10.1289/ehp.1307923.24968389
17. ATSDR (Agency for Toxic Substances and Disease Registry). 2018. Toxicological Profile for Perfluoroalkyls. Atlanta, GA: ATSDR. https://www.atsdr.cdc.gov/ToxProfiles/tp200-p.pdf [accessed 24 February 2023].
18. Apelberg BJ, Witter FR, Herbstman JB, Calafat AM, Halden RU, Needham LL, et al. 2007. Cord serum concentrations of perfluorooctane sulfonate (PFOS) and perfluorooctanoate (PFOA) in relation to weight and size at birth. Environ Health Perspect 115 (11 ):1670–1676, PMID: , 10.1289/ehp.10334.18008002
19. Chen MH, Ha EH, Wen TW, Su YN, Lien GW, Chen CY, et al. 2012. Perfluorinated compounds in umbilical cord blood and adverse birth outcomes. PLoS One 7 (8 ):e42474, PMID: , 10.1371/journal.pone.0042474.22879996
20. Darrow LA, Stein CR, Steenland K. 2013. Serum perfluorooctanoic acid and perfluorooctane sulfonate concentrations in relation to birth outcomes in the Mid-Ohio Valley, 2005–2010. Environ Health Perspect 121 (10 ):1207–1213, PMID: , 10.1289/ehp.1206372.23838280
21. Fei C, McLaughlin JK, Tarone RE, Olsen J. 2007. Perfluorinated chemicals and fetal growth: a study within the Danish National Birth Cohort. Environ Health Perspect 115 (11 ):1677–1682, PMID: , 10.1289/ehp.10506.18008003
22. Bach CC, Bech BH, Nohr EA, Olsen J, Matthiesen NB, Bonefeld-Jørgensen EC, et al. 2016. Perfluoroalkyl acids in maternal serum and indices of fetal growth: the Aarhus Birth Cohort. Environ Health Perspect 124 (6 ):848–854, PMID: , 10.1289/ehp.1510046.26495857
23. Hamm MP, Cherry NM, Chan E, Martin JW, Burstyn I. 2010. Maternal exposure to perfluorinated acids and fetal growth. J Expo Sci Environ Epidemiol 20 (7 ):589–597, PMID: , 10.1038/jes.2009.57.19865074
24. Maisonet M, Terrell ML, McGeehin MA, Christensen KY, Holmes A, Calafat AM, et al. 2012. Maternal concentrations of polyfluoroalkyl compounds during pregnancy and fetal and postnatal growth in British girls. Environ Health Perspect 120 (10 ):1432–1437, PMID: , 10.1289/ehp.1003096.22935244
25. Manzano-Salgado CB, Casas M, Lopez-Espinosa MJ, Ballester F, Iñiguez C, Martinez D, et al. 2017. Prenatal exposure to perfluoroalkyl substances and birth outcomes in a Spanish birth cohort. Environ Int 108 :278–284, PMID: , 10.1016/j.envint.2017.09.006.28917208
26. Washino N, Saijo Y, Sasaki S, Kato S, Ban S, Konishi K, et al. 2009. Correlations between prenatal exposure to perfluorinated chemicals and reduced fetal growth. Environ Health Perspect 117 (4 ):660–667, PMID: , 10.1289/ehp.11681.19440508
27. Wikström S, Lin PI, Lindh CH, Shu H, Bornehag CG. 2020. Maternal serum levels of perfluoroalkyl substances in early pregnancy and offspring birth weight. Pediatr Res 87 (6 ):1093–1099, PMID: , 10.1038/s41390-019-0720-1.31835271
28. Wu K, Xu X, Peng L, Liu J, Guo Y, Huo X. 2012. Association between maternal exposure to perfluorooctanoic acid (PFOA) from electronic waste recycling and neonatal health outcomes. Environ Int 48 :1–8, PMID: , 10.1016/j.envint.2012.06.018.22820015
29. Callan AC, Rotander A, Thompson K, Heyworth J, Mueller JF, Odland JØ, et al. 2016. Maternal exposure to perfluoroalkyl acids measured in whole blood and birth outcomes in offspring. Sci Total Environ 569–570 :1107–1113, PMID: , 10.1016/j.scitotenv.2016.06.177.27387804
30. Sagiv SK, Rifas-Shiman SL, Fleisch AF, Webster TF, Calafat AM, Ye X, et al. 2018. Early-pregnancy plasma concentrations of perfluoroalkyl substances and birth outcomes in Project Viva: confounded by pregnancy hemodynamics? Am J Epidemiol 187 (4 ):793–802, PMID: , 10.1093/aje/kwx332.29155920
31. Eick SM, Hom Thepaksorn EK, Izano MA, Cushing LJ, Wang Y, Smith SC, et al. 2020. Associations between prenatal maternal exposure to per- and polyfluoroalkyl substances (PFAS) and polybrominated diphenyl ethers (PBDEs) and birth outcomes among pregnant women in San Francisco. Environ Health 19 (1 ):100, PMID: , 10.1186/s12940-020-00654-2.32938446
32. Gao X, Ni W, Zhu S, Wu Y, Cui Y, Ma J, et al. 2021. Per- and polyfluoroalkyl substances exposure during pregnancy and adverse pregnancy and birth outcomes: a systematic review and meta-analysis. Environ Res 201 :111632, PMID: , 10.1016/j.envres.2021.111632.34237336
33. Whitworth KW, Haug LS, Baird DD, Becher G, Hoppin JA, Skjaerven R, et al. 2012. Perfluorinated compounds in relation to birth weight in the Norwegian Mother and Child Cohort Study. Am J Epidemiol 175 (12 ):1209–1216, PMID: , 10.1093/aje/kwr459.22517810
34. Eick SM, Enright EA, Geiger SD, Dzwilewski KLC, DeMicco E, Smith S, et al. 2021. Associations of maternal stress, prenatal exposure to per- and polyfluoroalkyl substances (PFAS), and demographic risk factors with birth outcomes and offspring neurodevelopment: an overview of the ECHO.CA.IL prospective birth cohorts. Int J Environ Res Public Health 18 (2 ):742, PMID: , 10.3390/ijerph18020742.33467168
35. Okechukwu CA, El Ayadi AM, Tamers SL, Sabbath EL, Berkman L. 2012. Household food insufficiency, financial strain, work–family spillover, and depressive symptoms in the working class: the Work, Family, and Health Network study. Am J Public Health 102 (1 ):126–133, PMID: , 10.2105/AJPH.2011.300323.22095360
36. Braveman PA, Heck K, Egerter S, Marchi KS, Dominguez TP, Cubbin C, et al. 2015. The role of socioeconomic factors in Black–White disparities in preterm birth. Am J Public Health 105 (4 ):694–702, PMID: , 10.2105/AJPH.2014.302008.25211759
37. Bécares L, Atatoa-Carr P. 2016. The association between maternal and partner experienced racial discrimination and prenatal perceived stress, prenatal and postnatal depression: findings from the Growing Up in New Zealand cohort study. Int J Equity Health 15 (1 ):155, PMID: , 10.1186/s12939-016-0443-4.27658457
38. Rich-Edwards JW, Kleinman K, Abrams A, Harlow BL, McLaughlin TJ, Joffe H, et al. 2006. Sociodemographic predictors of antenatal and postpartum depressive symptoms among women in a medical group practice. J Epidemiol Community Health 60 (3 ):221–227, PMID: , 10.1136/jech.2005.039370.16476752
39. Marques AH, O’Connor TG, Roth C, Susser E, Bjørke-Monsen AL. 2013. The influence of maternal prenatal and early childhood nutrition and maternal prenatal stress on offspring immune system development and neurodevelopmental disorders. Front Neurosci 7 :120, PMID: , 10.3389/fnins.2013.00120.23914151
40. Barrett ES, Padula AM. 2019. Joint impact of synthetic chemical and non-chemical stressors on children’s health. Curr Environ Health Rep 6 (4 ):225–235, PMID: , 10.1007/s40572-019-00252-6.31637664
41. Padula AM, Monk C, Brennan PA, Borders A, Barrett ES, McEvoy CT, et al. 2020. A review of maternal prenatal exposures to environmental chemicals and psychosocial stressors-implications for research on perinatal outcomes in the ECHO program. J Perinatol 40 (1 ):10–24, PMID: , 10.1038/s41372-019-0510-y.31616048
42. Morello-Frosch R, Shenassa ED. 2006. The environmental “riskscape” and social inequality: implications for explaining maternal and child health disparities. Environ Health Perspect 114 (8 ):1150–1153, PMID: , 10.1289/ehp.8930.16882517
43. Blaisdell CJ, Park C, Hanspal M, Roary M, Arteaga SS, Laessig S, et al. 2022. The NIH ECHO program: investigating how early environmental influences affect child health. Pediatr Res 92 :1215–1216, PMID: , 10.1038/s41390-021-01574-8.34131291
44. LeWinn KZ, Caretta E, Davis A, Anderson AL, Oken E, program collaborators for Environmental influences on Child Health Outcomes. 2022. SPR perspectives: Environmental influences on Child Health Outcomes (ECHO) program: overcoming challenges to generate engaged, multidisciplinary science. Pediatr Res 92 :1262–1269, PMID: , 10.1038/s41390-021-01598-0.34131290
45. Buckley JP, Barrett ES, Beamer PI, Bennett DH, Bloom MS, Fennell TR, et al. 2020. Opportunities for evaluating chemical exposures and child health in the United States: the Environmental influences on Child Health Outcomes (ECHO) Program. J Expo Sci Environ Epidemiol 30 (3 ):397–419, PMID: , 10.1038/s41370-020-0211-9.32066883
46. Jacobson LP, Lau B, Catellier D, Parker CB. 2018. An environmental influences on Child Health Outcomes viewpoint of data analysis centers for collaborative study designs. Curr Opin Pediatr 30 (2 ):269–275, PMID: , 10.1097/MOP.0000000000000602.29474274
47. Kato K, Basden BJ, Needham LL, Calafat AM. 2011. Improved selectivity for the analysis of maternal serum and cord serum for polyfluoroalkyl chemicals. J Chromatogr A 1218 (15 ):2133–2137, PMID: , 10.1016/j.chroma.2010.10.051.21084089
48. Oh J, Bennett DH, Calafat AM, Tancredi D, Roa DL, Schmidt RJ, et al. 2021. Prenatal exposure to per- and polyfluoroalkyl substances in association with autism spectrum disorder in the MARBLES study. Environ Int 147 :106328, PMID: , 10.1016/j.envint.2020.106328.33387879
49. Honda M, Robinson M, Kannan K. 2018. A rapid method for the analysis of perfluorinated alkyl substances in serum by hybrid solid-phase extraction. Environ Chem 15 (2 ):92–99, 10.1071/EN17192.
50. CDC (Centers for Disease Control and Prevention). 2016. National Health and Nutrition Examination Survey: 2013–2014 Data Documentation, Codebook, and Frequencies. Dietary Interview Technical Support File—Food Codes (DRXFD_H). Atlanta, GA: CDC. https://wwwn.cdc.gov/Nchs/Nhanes/2013-2014/DRXFCD_H.htm [accessed 24 February 2023].
51. Cohen S, Kamarck T, Mermelstein R. 1983. A global measure of perceived stress. J Health Soc Behav 24 (4 ):385–396, PMID: , 10.2307/2136404.6668417
52. Roberti JW, Harrington LN, Storch EA. 2006. Further psychometric support for the 10-item version of the perceived stress scale. J Coll Couns 9 (2 ):135–147, 10.1002/j.2161-1882.2006.tb00100.x.
53. Solivan AE, Xiong X, Harville EW, Buekens P. 2015. Measurement of perceived stress among pregnant women: a comparison of two different instruments. Matern Child Health J 19 (9 ):1910–1915, PMID: , 10.1007/s10995-015-1710-5.25652063
54. McDonald RP. 1999. Test Theory: A Unified Treatment. Mahwah, NJ: Lawrence Earlbaum Associates Inc.
55. Aris IM, Kleinman KP, Belfort MB, Kaimal A, Oken E. 2019. A 2017 US reference for singleton birth weight percentiles using obstetric estimates of gestation. Pediatrics 144 (1 ):e20190076, PMID: , 10.1542/peds.2019-0076.31201230
56. Hamra GB, Buckley JP. 2018. Environmental exposure mixtures: questions and methods to address them. Curr Epidemiol Rep 5 (2 ):160–165, PMID: , 10.1007/s40471-018-0145-0.30643709
57. Hamra GB, Maclehose RF, Croen L, Kauffman EM, Newschaffer C. 2021. Bayesian Weighted Sums: a flexible approach to estimate summed mixture effects. Int J Environ Res Public Health 18 (4 ):1373, PMID: , 10.3390/ijerph18041373.33546139
58. Rothman KJ. 1990. No adjustments are needed for multiple comparisons. Epidemiology 1 (1 ):43–46, PMID: , 10.1097/00001648-199001000-00010.2081237
59. Vieira VM, Levy JI, Fabian MP, Korrick S. 2021. Assessing the relation of chemical and non-chemical stressors with risk-taking related behavior and adaptive individual attributes among adolescents living near the New Bedford Harbor Superfund site. Environ Int 146 :106199, PMID: , 10.1016/j.envint.2020.106199.33126063
60. Rokoff LB, Rifas-Shiman SL, Coull BA, Cardenas A, Calafat AM, Ye X, et al. 2018. Cumulative exposure to environmental pollutants during early pregnancy and reduced fetal growth: the Project Viva cohort. Environ Health 17 (1 ):19, PMID: , 10.1186/s12940-018-0363-4.29458383
61. Starling AP, Adgate JL, Hamman RF, Kechris K, Calafat AM, Ye X, et al. 2017. Perfluoroalkyl substances during pregnancy and offspring weight and adiposity at birth: examining mediation by maternal fasting glucose in the Healthy Start study. Environ Health Perspect 125 (6 ):067016, PMID: , 10.1289/EHP641.28669937
62. Eick SM, Enright EA, Padula AM, Aung M, Geiger SD, Cushing L, et al. 2022. Prenatal PFAS and psychosocial stress exposures in relation to fetal growth in two pregnancy cohorts: applying environmental mixture methods to chemical and non-chemical stressors. Environ Int 163 :107238, PMID: , 10.1016/j.envint.2022.107238.35436721
63. National Academies of Sciences, Engineering, and Medicine. 2022. Guidance on PFAS Exposure, Testing, and Clinical Follow-Up. Washington, DC: National Academies Press.
64. Steenland K, Barry V, Savitz D. 2018. Serum perfluorooctanoic acid and birthweight: an updated meta-analysis with bias analysis. Epidemiology 29 (6 ):765–776, PMID: , 10.1097/EDE.0000000000000903.30063543
65. Dzierlenga MW, Crawford L, Longnecker MP. 2020. Birth weight and perfluorooctane sulfonic acid: a random-effects meta-regression analysis. Environ Epidemiol 4 (3 ):e095, PMID: , 10.1097/EE9.0000000000000095.33778349
66. Koustas E, Lam J, Sutton P, Johnson PI, Atchley DS, Sen S, et al. 2014. The Navigation Guide—evidence-based medicine meets environmental health: systematic review of nonhuman evidence for PFOA effects on fetal growth. Environ Health Perspect 122 (10 ):1015–1027, PMID: , 10.1289/ehp.1307177.24968374
67. Meng Q, Inoue K, Ritz B, Olsen J, Liew Z. 2018. Prenatal exposure to perfluoroalkyl substances and birth outcomes; an updated analysis from the Danish National Birth Cohort. Int J Environ Res Public Health 15 (9 ):1832, PMID: , 10.3390/ijerph15091832.30149566
68. Gardener H, Sun Q, Grandjean P. 2021. PFAS concentration during pregnancy in relation to cardiometabolic health and birth outcomes. Environ Res 192 :110287, PMID: , 10.1016/j.envres.2020.110287.33038367
69. Negri E, Metruccio F, Guercio V, Tosti L, Benfenati E, Bonzi R, et al. 2017. Exposure to PFOA and PFOS and fetal growth: a critical merging of toxicological and epidemiological data. Crit Rev Toxicol 47 (6 ):482–508, PMID: , 10.1080/10408444.2016.1271972.28617200
70. Du G, Hu J, Huang H, Qin Y, Han X, Wu D, et al. 2013. Perfluorooctane sulfonate (PFOS) affects hormone receptor activity, steroidogenesis, and expression of endocrine-related genes in vitro and in vivo. Environ Toxicol Chem 32 (2 ):353–360, PMID: , 10.1002/etc.2034.23074026
71. Starling AP, Liu C, Shen G, Yang IV, Kechris K, Borengasser SJ, et al. 2020. Prenatal exposure to per- and polyfluoroalkyl substances, umbilical cord blood DNA methylation, and cardio-metabolic indicators in newborns: the Healthy Start study. Environ Health Perspect 128 (12 ):127014, PMID: , 10.1289/EHP6888.33356526
72. Chang CJ, Barr DB, Ryan PB, Panuwet P, Smarr MM, Liu K, et al. 2022. Per- and polyfluoroalkyl substance (PFAS) exposure, maternal metabolomic perturbation, and fetal growth in African American women: a meet-in-the-middle approach. Environ Int 158 :106964, PMID: , 10.1016/j.envint.2021.106964.34735953
73. Szilagyi JT, Avula V, Fry RC. 2020. Perfluoroalkyl substances (PFAS) and their effects on the placenta, pregnancy, and child development: a potential mechanistic role for placental peroxisome proliferator-activated receptors (PPARs). Curr Environ Health Rep 7 (3 ):222–230, PMID: , 10.1007/s40572-020-00279-0.32812200
74. Ku MS, Pan WC, Huang YT, Hsieh WS, Hsu YH, Chen PC, et al. 2022. Associations between prenatal exposure to perfluoroalkyl substances, hypomethylation of MEST imprinted gene and birth outcomes. Environ Pollut 304 :119183, PMID: , 10.1016/j.envpol.2022.119183.35331797
|
PMC010xxxxxx/PMC10016344.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36920445
EHP12209
10.1289/EHP12209
Invited Perspective
Invited Perspective: The Potential of Potential Outcomes in Air Pollution Epidemiology
https://orcid.org/0000-0001-7327-5361
Neophytou Andreas M. 1
1 Department of Environmental and Radiological Health Sciences, Colorado State University, Fort Collins, Colorado, USA
Address correspondence to Andreas Neophytou, 1681 Campus Delivery, Fort Collins, CO 80523-1681 USA. Email: [email protected]
15 3 2023
3 2023
131 3 03130528 9 2022
11 2 2023
14 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The author declares he has no conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11095
==== Body
pmcThe potential outcomes framework1 allows epidemiologists to mathematically define causal effects of interest as a contrast between two potential (or counterfactual) outcomes.2 This, in turn, has allowed the use of observational data in efforts to answer questions relating to hypothetical interventions. The causal interpretation of contrasting potential outcomes relies on assumptions that the potential outcomes framework also allows investigators to explicitly state. Whether or not those assumptions actually hold, however, will typically not be entirely testable or ultimately known. Nevertheless, the appeal of using observational data for causal inference, especially in an area where experimental data are limited, has generated excitement in air pollution epidemiology and has spurred the adoption of methodology from this framework in recent years.3
In one such example, Chen et al. report in this issue of Environmental Health Perspectives on an application of the parametric g-formula to assess the effect of hypothetical interventions on fine particulate matter [particulate matter ≤2.5μm in aerodynamic diameter (PM2.5)] exposures in a Canadian cohort.4 They report findings of reduced mortality outcomes associated with hypothetical reductions in exposure while identifying several advantages of the approach compared with a more traditional analysis approach using a survival framework, the Cox proportional hazards model. The parametric g-formula approach is indeed advantageous because it can address situations of exposure–confounder feedback in time-varying settings and does not suffer from the built-in selection bias of the Cox model.5
However, these are not likely to be major sources of bias in the study by Chen et al.4 Exposure–confounder feedback is not generally a feature in air pollution epidemiology settings and, in the current example, depletion of susceptible individuals leading to crossing of hazards also is not expected to be a major source of bias, with ∼90% of participants still alive at the end of follow-up. Overall, this is confirmed by the sensitivity analysis results fitting a Cox model and yielding qualitatively similar (though not directly comparable) effect estimates as the parametric g-formula. The use of this method is, therefore, not likely to greatly improve upon the internal validity of effect estimates and, as previously stated, we should be cautious about attributing causal interpretations of findings owing simply to the statistical method in any particular study.6,7
The parametric g-formula, nevertheless, is still advantageous in this setting compared with more traditional regression approaches. Unlike the Cox model, which yields target parameters based on the hazard, the parametric g-formula reports findings based on risk (cumulative incidence) and carries advantages over the hazard, such as collapsibility and the increased applicability with respect to public health relevance, especially when focusing on the risk difference.5,8 The approach further yields marginal effect estimates as opposed to conditional. However, in my opinion, the major advantage of the approach is the framing of the parameters of interest in terms of the dynamic interventions considered. Rather than focus target parameters based simply on exposure–response relationships, such as associations for a particular increase in exposure from level A to level B, these hypothetical interventions essentially compare the effect as a contrast of two counterfactual distributions of exposure in the same population. This is far more representative of how an intervention or policy could actually affect exposure and, by extension, health outcomes on the population level in a real-world setting.
Despite this advantage, hypothetical interventions of the type that Chen at al.4 considered do have some perhaps fewer obvious limitations relating to key assumptions required for causal interpretation of these findings. Among these are the assumption of consistency and the “no multiple versions of treatment” assumption, part of Rubin’s stable unit treatment value assumption.9,10 Briefly, the assumptions require that a well-defined intervention is contrasted in the counterfactual effect estimate, and that there do not exist multiple versions of the treatment of interest (here, exposure to PM2.5).
It is arguable that neither holds here. Depending on the intervention or policy change that would lead to a reduction in exposure levels, we may expect different sources of pollution (e.g., transportation, energy, agriculture) to be affected to varying degrees. That, in turn, may affect who experiences a greater change in pollution levels. However, even if we somehow achieved reductions according to some threshold or percentage-based intervention on the individual level, the differing composition of individual-level PM2.5 exposures may mean that the same concentration could have different effects in otherwise perfectly exchangeable individuals.11 This would constitute a violation of “no multiple versions of treatment” and would also lead to bias if unmeasured common causes exist between particle composition (version of treatment) and the outcome.10
Along the same lines, an intervention to reduce PM2.5 will likely lead to differential changes in other pollutants as well, given the common sources many air pollutants share. In that regard, the true intervention effect would be higher than estimated simply by considering PM2.5 alone. As the health burden of air pollution is, in essence, a result of joint effects of multiple pollutants, the solution maximizing potential health benefits should be one of joint interventions. On the other hand, because individual pollutants are regulated separately, it could be argued that reporting of individual effects is still useful for policy purposes. However, those, too, would probably be more accurate if the effect of other pollutants was also taken into account. One application of an exposure mixture approach has been proposed within the g-formula framework,12 although combining some of the work involving multipollutant models and causal inference is an area that requires more applied examples.
Future work leveraging the potential outcomes framework could incorporate knowledge gained from studies on exposure mixtures and source apportionment to help address some of the aforementioned limitations when envisioning hypothetical interventions. Importantly, the potential outcomes framework allows for estimation of causal effects even in the presence of multiple versions of treatment.10 This framework can also be leveraged to assess the effects of less hypothetical, and therefore better defined, interventions as demonstrated in a 2018 study leveraging principal stratification based on potential outcomes to assess the effect of the U.S. Environmental Protection Agency’s National Ambient Air Quality Standards nonattainment designations.13
In conclusion, I applaud Chen et al.4 for their contribution, which brings a more realistic representation of the health effects of potential interventions to reduce air pollution exposures. The adoption of the potential outcomes framework in air pollution epidemiology can be a valuable tool in risk assessment and aid in the interpretability of epidemiological findings in terms of policy, but I would also caution readers to carefully consider the assumptions required for causal inference in each case.
==== Refs
References
1. Rubin DB. 2005. Causal inference using potential outcomes. J Am Stat Assoc 100 (469 ):322–331, 10.1198/016214504000001880.
2. Vanderweele TJ. 2015. Explanation in Causal Inference: Methods of Mediation and Interaction. New York, NY: Oxford University Press.
3. Carone M, Dominici F, Sheppard L. 2020. In pursuit of evidence in air pollution epidemiology: the role of causally driven data science. Epidemiology 31 (1 ):1–6, PMID: , 10.1097/EDE.0000000000001090.31430263
4. Chen C, Chen H, van Donkelaar A, Burnett RT, Martin RV, Chen L, et al. 2023. Using parametric g-computation to estimate the effect of long-term exposure to air pollution on mortality risk and simulate the benefits of hypothetical policies: the Canadian Community Health Survey Cohort (2005 to 2015). Environ Health Perspect 131 (3 ):037010, 10.1289/EHP11095.36920446
5. Hernán MA. 2010. The hazards of hazard ratios. Epidemiology 21 (1 ):13–15, PMID: , 10.1097/EDE.0b013e3181c1ea43.20010207
6. Dominici F, Zigler C. 2017. Best practices for gauging evidence of causality in air pollution epidemiology. Am J Epidemiol 186 (12 ):1303–1309, PMID: , 10.1093/aje/kwx307.29020141
7. Pearce N, Vandenbroucke JP, Lawlor DA. 2019. Causal inference in environmental epidemiology: old and new approaches. Epidemiology 30 (3 ):311–316, PMID: , 10.1097/EDE.0000000000000987.30789434
8. Cole SR, Hudgens MG, Brookhart MA, Westreich D. 2015. Risk. Am J Epidemiol 181 (4 ):246–250, PMID: , 10.1093/aje/kwv001.25660080
9. Rubin DB. 1986. Which ifs have causal answers? J Am Stat Assoc 81 (396 ):961–962, 10.1080/01621459.1986.10478355.
10. VanderWeele TJ, Hernán MA. 2013. Causal inference under multiple versions of treatment. J Causal Inference 1 (1 ):1–20, PMID: , 10.1515/jci-2012-0002.25379365
11. Bell ML, Ebisu K, Leaderer BP, Gent JF, Lee HJ, Koutrakis P, et al. 2014. Associations of PM2.5 constituents and sources with hospital admissions: analysis of four counties in Connecticut and Massachusetts (USA) for persons ≥ 65 years of age. Environ Health Perspect 122 (2 ):138–144, PMID: , 10.1289/ehp.1306656.24213019
12. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 128 (4 ):47004, PMID: , 10.1289/EHP5838.32255670
13. Zigler CM, Choirat C, Dominici F. 2018. Impact of national ambient air quality standards nonattainment designations on particulate pollution and health. Epidemiology 29 (2 ):165–174, PMID: , 10.1097/EDE.0000000000000777.29095246
|
PMC010xxxxxx/PMC10016347.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36920446
EHP11095
10.1289/EHP11095
Research
Using Parametric g-Computation to Estimate the Effect of Long-Term Exposure to Air Pollution on Mortality Risk and Simulate the Benefits of Hypothetical Policies: The Canadian Community Health Survey Cohort (2005 to 2015)
https://orcid.org/0000-0002-0632-946X
Chen Chen 1
Chen Hong 2 3 4 5 *
van Donkelaar Aaron 6
Burnett Richard T. 2
Martin Randall V. 6
Chen Li 2
Tjepkema Michael 7
Kirby-McGregor Megan 8
Li Yi 8
Kaufman Jay S. 8
Benmarhnia Tarik 1 *
1 Scripps Institution of Oceanography, University of California, San Diego, La Jolla, California, USA
2 Environmental Health Science and Research Bureau, Health Canada, Ottawa, Ontario, Canada
3 Public Health Ontario, Toronto, Ontario, Canada
4 ICES, Toronto, Ontario, Canada
5 Dalla Lana School of Public Health, University of Toronto, Toronto, Ontario, Canada
6 Department of Energy, Environmental, and Chemical Engineering, Washington University in St. Louis, St. Louis, Missouri, USA
7 Health Analysis Division, Statistics Canada, Ottawa, Ontario, Canada
8 Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Quebec, Canada
Address correspondence to Chen Chen, 8885 Biological Grade, La Jolla, CA 92037, USA. Email: [email protected]
15 3 2023
3 2023
131 3 03701011 2 2022
05 12 2022
14 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Numerous epidemiological studies have documented the adverse health impact of long-term exposure to fine particulate matter [particulate matter ≤2.5μm in aerodynamic diameter (PM2.5)] on mortality even at relatively low levels. However, methodological challenges remain to consider potential regulatory intervention’s complexity and provide actionable evidence on the predicted benefits of interventions. We propose the parametric g-computation as an alternative analytical approach to such challenges.
Method:
We applied the parametric g-computation to estimate the cumulative risks of nonaccidental death under different hypothetical intervention strategies targeting long-term exposure to PM2.5 in the Canadian Community Health Survey cohort from 2005 to 2015. On both relative and absolute scales, we explored the benefits of hypothetical intervention strategies compared with the natural course that a) set the simulated exposure value at each follow-up year to a threshold value if exposure was above the threshold (8.8 μg/m3, 7.04 μg/m3, 5 μg/m3, and 4 μg/m3), and b) reduced the simulated exposure value by a percentage (5% and 10%) at each follow-up year. We used the 3-y average PM2.5 concentration with 1-y lag at the postal code of respondents’ annual mailing addresses as their long-term exposure to PM2.5. We considered baseline and time-varying confounders, including demographics, behavior characteristics, income level, and neighborhood socioeconomic status. We also included the R syntax for reproducibility and replication.
Results:
All hypothetical intervention strategies explored led to lower 11-y cumulative mortality risks than the estimated value under the natural course without intervention, with the smallest reduction of 0.20 per 1,000 participants (95% CI: 0.06, 0.34) under the threshold of 8.8 μg/m3, and the largest reduction of 3.40 per 1,000 participants (95% CI: −0.23, 7.03) under the relative reduction of 10% per interval. The reductions in cumulative risk, or numbers of deaths that would have been prevented if the intervention was employed instead of maintaining the status quo, increased over time but flattened toward the end of the follow-up period. Estimates among those ≥65 years of age were greater with a similar pattern. Our estimates were robust to different model specifications.
Discussion:
We found evidence that any intervention further reducing the long-term exposure to PM2.5 would reduce the cumulative mortality risk, with greater benefits in the older population, even in a population already exposed to low levels of ambient PM2.5. The parametric g-computation used in this study provides flexibilities in simulating real-world interventions, accommodates time-varying exposure and confounders, and estimates adjusted survival curves with clearer interpretation and more information than a single hazard ratio, making it a valuable analytical alternative in air pollution epidemiological research. https://doi.org/10.1289/EHP11095
Supplemental Material is available online (https://doi.org/10.1289/EHP11095).
* These authors are joint senior authors.
R.V.M. receives grant support currently from the Clean Air Fund and previously from the ClimateWorks Foundation. The other authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Given that collective efforts in previous decades have successfully reduced the level of fine particulate matter [particulate matter ≤2.5μm in aerodynamic diameter (PM2.5)] globally, quantifying the effectiveness of policies that further reduce ambient PM2.5 is becoming particularly important in supporting evidence-based policymaking. Indeed, previous studies found consistent evidence of deleterious associations between long-term exposure to low levels of PM2.5 (e.g., below the current health-based standards or guidelines) and risk of mortality1–6 and morbidity,7–9 suggesting potential reductions in health burden if the PM2.5 level were to be further reduced. Although the evaluation of exposure–response functions in existing studies provides important information in understanding the potential effectiveness of policies, further methodological considerations are required to estimate the potential benefits of realistic interventions.
First, evidence suggested that the risk associated with the changes in acute exposure to PM2.5 could vary with time,10–13 potentially due to changes in chemical compositions of PM2.5, with different toxicity and population susceptibility toward PM2.5.14,15 A similar disparity in toxicity across long-term exposure to PM2.5 components was also observed,16,17 suggesting that such temporal changes could exist in risk associated with long-term exposure to PM2.5. In other words, it is important to use analytical methods flexible enough to incorporate such temporal changes in estimating related health burdens. However, existing studies of health impacts of long-term exposure to PM2.5 generally considered time-fixed exposure and confounders (Table S1 provides a narrative review of recent studies on health impact of long-term exposure to PM2.5 and their methodological considerations). Furthermore, the most widely used estimate for exposure–response function in this field is a single hazard ratio (HR) for the follow-up period estimated with a standard Cox proportional hazards model (Table S1), which is assumed to be constant over time and precludes consideration of temporal changes. Although extension of a Cox proportional hazards model could provide period-specific HRs that incorporate temporal changes,18 recent developments in causal inference literature raise concern about the ambiguity in the causal interpretation of HR and period-specific HRs.19 Specifically, period-specific HRs have a built-in selection bias because susceptible people exposed to higher PM2.5 concentrations are more likely to die early if PM2.5 exposure truly increases risk of mortality and they are removed from the susceptible population in later periods.20 This differential depletion of susceptible individuals over time can lead to artificially diminished or even reversed period-specific HR later in a study even when the cumulative survival is still lower among those exposed to higher PM2.5 concentrations, violating the proportional assumption and hindering interpretation.21
Second, the calculation of the health burden related to long-term exposure to PM2.5 has commonly employed an exposure–response function previously estimated with the static intervention strategy, where a fixed change of exposure value was assigned to the entire population.22 However, the more flexible and realistic dynamic intervention strategy—where the exposure value was assigned based on individuals’ history of covariates, including exposure—is hard to apply when existing exposure–response functions are used.22 Methods capable of incorporating a dynamic intervention strategy to imitate complexities in actual regulatory interventions are needed to provide direct evidence on effectiveness of air pollution control policies.23 To fill this gap in knowledge translation, we propose the parametric g-computation as an analytical alternative in air pollution epidemiological research, a method that could better predict the effectiveness of hypothetical policies while being more flexible in resembling real-world interventions.
G-computation (also known as g-formula) is a generalization of nonparametric standardization developed under the potential outcome framework for causal inference,24 and parametric g-computation is a variation that employs parametric modeling. Under the consistency (i.e., the exposure is defined unambiguously, and all exposed individuals receive the same version of treatment),22,25 exchangeability (i.e., no unmeasured confounding or informative censoring),25 and positivity (i.e., probability of receiving every exposure conditioning on confounders is greater than zero) assumptions,22 and a time-to-event outcome setting, g-computation can provide marginal causal risk estimates at each follow-up time point under hypothetical intervention strategies (i.e., adjusted survival curves) while allowing other population characteristics to be altered according to the intervention.26 Particularly, parametric g-computation excels in estimating adjusted survival curves under dynamic intervention strategies. In other words, g-computation can directly answer causal questions such as, “How many lives could we save if we promulgate a policy that further reduces air pollution to levels lower than the current standard among those whose exposure were above the current standard, compared with maintaining the status quo?” Although parametric g-computation has been widely applied in other fields of epidemiology,27–30 its application in air pollution studies remains limited. Previous applications in this field either focused on a small cohort in occupational settings31–33 or modeled simple air pollution changes on asthmatic outcomes among children (i.e., not considering time-varying confounding or changes in effect estimates over time).34,35
In this study, we aimed to demonstrate the use of parametric g-computation to evaluate the effectiveness of hypothetical intervention strategies targeting long-term exposure to PM2.5 on reducing mortality using a Canadian cohort experiencing low PM2.5 exposure from 2005 to 2015. This analytical alternative can account for previously unaddressed complexities, refine the effect estimates with less restrictive identification conditions, and provide estimates that are more intuitive to policy makers.
Methods
Study Population
We created a retrospective cohort with respondents to the Canadian Community Health Survey (CCHS) from three enrolling cycles in the years of 2000/2001, 2003, and 2005, respectively.36–38 CCHS is a national cross-sectional survey collecting health status, health care utilization, and health determinants information of the Canadian population, covering the population ≥12 years of age in the 10 provinces and the 3 territories. The survey excluded individuals living on reserves and other Aboriginal settlements, full-time members of the Canadian Forces, the institutionalized population, and residents of certain remote regions.
Among CCHS respondents who gave permission to share and link their information with other administrative data sets, we obtained their mailing address history and death records through 31 December 2015 via Statistics Canada’s Social Data Linkage Environment using probabilistic methods based on common identifiers.2,39 We focused on nonaccidental death as outcome (International Classification of Diseases Ninth Revision,40 ICD-9 codes 001–799, and International Classification of Diseases Tenth Revision,41 ICD-10 codes A–R) in this study. To facilitate pooling of results across cycles, we restricted the cohort to participants who were alive on 1 January 2005 and used this date as the start of follow-up for all cycles. We also restricted our cohort to individuals >25 and<80 years of age in 2005, thus all cohort participants were adults and were followed for 11 y or until death. In addition, we dropped respondents who were missing data for covariates, including exposure in 2005. This study was approved by the Health Canada–Public Health Agency of Canada Research Ethics Board.
Exposure Assessment
To estimate respondents’ long-term exposure to PM2.5, we used the ground-level PM2.5 concentrations from V4.NA.02.MAPLE of the Atmospheric Composition Analysis Group of Washington University,42 which covers all of North America below the 70oN latitude. The 0.01°×0.01° (roughly equivalent to 1×1 km2 at the latitudes where most Canadians live) annual estimates of PM2.5 concentrations from 2001 to 2015 were derived using satellite retrievals of aerosol optical depth and chemical transport model simulations and calibrated with ground-based observations using geographically weighted regression.43 The annual estimates of PM2.5 concentrations closely agree with long-term cross-validated ground measurements at fixed-site monitors (n=2,312) across North America (R2=0.70).43 Using the ground-level PM2.5 concentration surfaces described above, we first assigned the annual PM2.5 concentration of the grid cell into which the postal code centroid falls as the postal code–specific annual PM2.5 concentrations. Then we calculated respondents’ annual long-term exposure to PM2.5 as 3-y average postal code–specific PM2.5 concentrations with 1-y lag based on their mailing address history (e.g., a respondent’s long-term exposure to PM2.5 in 2013 is the average of their postal code–specific PM2.5 concentrations in 2010, 2011, and 2012).2 We used the 3-y average with 1-y lag to represent long-term exposure of PM2.5 so that the exposure always preceded the outcome and that the timeframe was consistent with the regulatory review of Canadian Ambient Air Quality Standards for annual PM2.5 exposure.44 This metric of long-term exposure to PM2.5 has been widely used in previous Canada-based studies of long-term health impacts of PM2.5 exposure.2,45,46
Covariates Other than Exposure
In this section we summarize the data sources and meaning of covariates in this study, whereas the covariate selection to control for in our model is discussed in the “Statistical Analysis” section. We used covariates to describe the collection of exposure, time-fixed confounders, and time-varying confounders in this study. Baseline characteristics of respondents were ascertained at the time of enrollment into CCHS via self-report and were processed using the same method as in previous studies,2,45 including sex, age (converted to value in 2005), body mass index (BMI), marital status, immigrant status, visible minority, indigenous status, smoking status, alcohol consumption, consumption of fruits and vegetables, leisure physical activity, working status, and educational attainment (details of variable categorization are listed in Table 1). By using 2005 as the start of the follow-up period for all individuals, we assumed that all baseline characteristics other than age ascertained at the time of enrollment would remain the same through the entire follow-up period.
Table 1 Descriptive statistics for participants of the Canadian Community Health Survey cohort at the start of follow-up (2005) by cycle.
Characteristics Cycle
2000/2001 2003 2005
Cohort size (n) 62,365a 62,380a 66,835a
Nonaccidental deaths [n (%)] 6,475 (10.4) 6,525 (10.5) 6,135 (9.2)
Time-fixed covariates
Age [y (mean±SD)] 52.1±13.4 52.1±14.4 50.9±14.9
Sex (%)
Female 45.2 45.9 46.2
Male 54.8 54.1 53.8
BMI [kg/m2 (%)]
Normal weight (18.5–24.9) 37.5 32.2 32.0
Overweight (25.0–29.9) 36.7 39.8 39.8
Obese 1 (30.0–34.9) 16.4 19.1 18.9
Obese 2 (≥35) 6.8 8.1 8.4
Underweight (<18.5) 2.6 0.8 0.9
Marital status (%)
Married or common-law 65.9 64.3 63.0
Separated, widowed, or divorced 19.6 20.8 20.8
Single 14.5 14.9 16.2
Immigrant status (%)
Immigrant 10.7 11.3 11.6
Time lived in Canada among immigrants [y (mean±SD)] 37.4±13.3 36.8±13.9 35.7±14.1
Nonimmigrant 89.3 88.7 88.4
Visible minority status (%)
Visible minority 5.4 6.3 4.4
Not a visible minority 94.6 93.7 95.6
Indigenous status (%)
Indigenous 1.8 2.3 0b
Nonindigenous 98.2 97.7 100
Smoking status (%)
Never smoker 26.8 27.6 29.0
Occasional smoker 44.5 47.7 47.1
Smoke <10 cigarettes/d 3.8 4.3 4.2
Smoke 11–20 cigarettes/d 6.0 5.6 5.7
Smoke ≥20 cigarettes/d 10.9 9.0 8.6
Former smoker 8.0 5.8 5.4
Alcohol consumption (%)
Never drinker 4.4 4.2 4.1
Occasional drinker 13.1 13.7 13.6
Regular drinker, binging unknown 20.3 18.7 18.2
Regular, non-binge drinker 29.4 31.0 30.4
Regular, binge drinker 26.9 26.7 27.3
Former drinker 5.9 5.7 6.4
Daily consumption of fruits and vegetables (%)
<5 servings/d 64.7 59.6 29.9
5–10 servings/d 32.4 37.0 19.6
≥10 servings/d 2.9 3.4 1.6
Chose to not answerc NA NA 48.9
Employment status (%)
Employed 66.6 62.3 61.8
Not employed 2.6 2.6 2.3
Not in work force 30.8 35.1 35.9
Education (%)
No high school diploma 24.1 22.4 20.6
High school 18.9 18.0 15.4
Any postsecondary 42.0 42.5 46.1
University 15.0 17.1 17.9
Leisure time physical activity (%)
Active 21.2 24.0 23.4
Moderately active 25.4 26.4 26.7
Inactive 53.4 49.6 49.9
Urban form (%)
Active urban core 6.2 7.0 7.0
Transit-reliant suburb 3.9 4.3 4.6
Auto-reliant suburb or no data 26.5 29.4 29.3
Exurb 4.8 5.0 4.8
Non-CMA/CAd 58.6 54.3 54.3
Airshed (%)
Western 12.0 10.7 10.4
Prairie 16.0 14.9 13.5
Western Central 9.1 8.6 7.8
Southern Atlantic 17.2 14.6 17.3
East Central 44.1 49.3 48.9
Northern 1.6 1.9 2.1
Time-varying covariates
Community size (population) [n (%)]
>1,500,000 13.5 14.7 16.9
500,000–1,499,999 10.3 11.9 10.4
100,000–499,999 20.4 21.0 19.7
30,000–99,999 14.7 13.2 12.4
10,000–29,999 7.4 7.0 7.5
Non-CMA/CAd 33.7 32.2 33.1
Annual family income quintile (%)
1st (lowest) 19.0 19.1 19.3
2nd 19.7 19.3 19.1
3rd 19.9 19.8 19.5
4th 20.7 20.4 20.8
5th (highest) 20.7 21.4 21.3
Canadian Marginalization Index—age and labor force [quintile (%)]
1st (lowest marginalization) 14.4 15.3 14.8
2nd 13.5 13.6 13.7
3rd 13.7 13.9 14.1
4th 22.2 20.9 20.8
5th (highest marginalization) 36.2 36.3 36.6
Canadian Marginalization Index—material resources [quintile (%)]
1st (lowest marginalization) 15.5 15.7 15.2
2nd 16.8 16.9 17.0
3rd 20.8 20.8 20.2
4th 18.1 17.8 17.0
5th (highest marginalization) 28.8 28.8 30.6
Canadian Marginalization Index—immigration and visible minority [quintile (%)]
1st (lowest marginalization) 42.6 41.4 41.8
2nd 26.9 26.9 26.5
3rd 17.0 17.3 15.9
4th 8.5 9.0 9.9
5th (highest marginalization) 5.0 5.4 5.9
Canadian Marginalization Index—households and dwellings [quintile (%)]
1st (lowest marginalization) 22.7 21.2 21.7
2nd 28.3 27.7 27.1
3rd 21.0 21.7 21.0
4th 17.3 17.8 17.6
5th (highest marginalization) 10.7 11.6 11.8
Average PM2.5 of previous 3 y (μg/m3) (mean±SD) 6.4±2.2 6.5±2.3 6.5±2.3
Average PM2.5 of previous 3 y (μg/m3) [median (minimum, 25th percentile, 75th percentile, maximum)] 5.9 (1.7, 15.0, 4.6, 7.8) 6.1 (1.7, 4.7, 8.1, 15.0) 6.1 (1.6, 4.6, 8.2, 15.0)
Note: BMI, body mass index; CA, census agglomeration (status); CMA, census metropolitan area; NA, not applicable; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter.
a Rounded to the nearest 5 or 0 in the last digit to protect privacy.
b We did not include the indigenous status indicator in models of cycle 2005.
c Consumption of fruit and vegetable was listed with an additional option in cycle 2005 but not in the other two cycles.
d Not categorized as CMA or CA status and likely in rural area.
We also obtained characteristics of the respondents and their neighborhoods through linkage with administrative data sets using methods similar to those used in previous studies.2,45 Specifically, we obtained the annual income quintile of respondents through linkage with tax data based on common identifiers.45 For person-years with missing annual family income, we imputed them with the nearest prior values and the proportions of missing were 5.21%, 4.97%, and 4.69% for cycles 2000/2001, 2003, and 2005, respectively. We also obtained annual characteristics of neighborhoods through linkage with census data from the nearest census year based on respondents’ mailing address postal codes, including community size at the census metropolitan area level and the Canadian Marginalization Index at the census dissemination-area level. The Canadian Marginalization Index summarizes dissemination area-level socioeconomic status into four dimensions using principal component analysis to reduce the dimensionality of census data: The immigration and visible minority index combines information on the proportion of recent immigrants and proportion of people self-identifying as a visible minority; the households and dwellings index combines information on types and density of residential accommodations and family structure characteristics; the material resources index combines information on access to and attainment of basic material needs; and the age and labor force index combines information on participation in the labor force and the proportion of seniors.47 Last, we obtained airshed (six distinct regions of Canada that cut cross jurisdictional boundaries and show similar air quality characteristics and air movement patterns within each region) to capture large-scale spatial variation,48 and urban form information of respondents’ neighborhoods in 2005 to capture the urbanicity of participants’ residences, through linkage with census data.2
Hypothetical Intervention Strategies
In this study, we explored three types of intervention strategies: a) applying the simulated value of time-varying covariates without any intervention (natural course), b) setting the simulated long-term exposure to the PM2.5 value at each follow-up year to a threshold value if the PM2.5 concentration was higher than the threshold (threshold intervention), and c) reducing the simulated PM2.5 value by a fixed percentage at each interval (i.e., follow-up year; relative reduction intervention). Threshold values explored included the current Canadian Ambient Air Quality Standards for PM2.5 of 8.8 μg/m3, 80% of the current Canadian Ambient Air Quality Standards for PM2.5 (or 7.04 μg/m3), the new World Health Organization (WHO) air quality guideline of 5 μg/m3, and a PM2.5 level that was farther below the WHO guideline (4 μg/m3). The interval-specific relative reduction values explored were 10% and 5% per interval. To avoid extensive model extrapolation, we restricted the relative reduction intervention so that individuals with an exposure <1.8 μg/m3, the background PM2.5 level in Canada,49 would not be further reduced. The first type of intervention strategy represents the predicted covariates based on the observed data without intervening and serves as the reference for other strategies. The second and the third are dynamic intervention strategies that incorporate the impact of intervention on covariates during earlier time points while simulating covariates in later time points.
Statistical Analysis
We applied parametric g-computation with different hypothetical intervention strategies targeting long-term exposure to PM2.5 to understand the benefits of intervention strategies on cumulative risk of nonaccidental death. We conducted g-computation analysis for each enrollment cycle separately and pooled the results across cycles using meta-regression. Briefly, we estimated the cumulative mortality risk at each follow-up year standardized to the distribution of the confounders and long-term exposure to PM2.5 in the study population, with all time-varying covariates (confounders and PM2.5) conditioned on covariates history, with and without intervention on PM2.5 (i.e., adjusted survival curves). Next, we calculated the differences in cumulative mortality risks between the natural course and other intervention strategies on both absolute and relative scales to provide estimates for the benefits of hypothetical intervention strategies compared with maintaining the status quo. We pooled results with fixed-effect meta-regression, which calculates a weighted average of cycle-specific estimates with weights equal to the inverse of the variance using the “meta” package.50
The proof of parametric g-computation are described extensively elsewhere,22(chap21),29 and detailed description of how to implement such an approach in a setting similar to our study was previously published,28 with the available R package for easy implementation.51 However, given that the application of parametric g-computation is limited in air pollution studies, we include a diagram (Figure 1) summarizing the four steps that carry out the g-computation in a time-to-event setting with time-varying exposure and confounders and describe the steps in detail below.
Figure 1. Diagram of g-computation with time-to-event outcome and time-varying covariates. Arrows indicate source of information for the indicated step.
Figure 1 is a flowchart titled Original data set having four steps. Step 1: run a discrete-time hazards model for the outcome. Semi-adjusted pooled logistic model with quadratic function of time for death. Step 2: run regressions for time-varying covariates given measured past. Liner model for exposure and cofounders. Step 3: simulate the new data sets without outcome based on the intervention strategies and standardize on confounders and exposure history. Randomly sample 10,000 participants from the cohort with replacement and create a data set of all subjects without value for all time points up to the time point of interest (for example end of follow-up) for each intervention strategy. Set the baseline values of confounders and exposure to the same as the original data set in the simulated records, then alter the values based on intervention strategies. At each time point after baseline, simulate time-varying exposure and confounders based on their history with models estimated in Step 2, then alter the values based on intervention strategy. Step 3 with simulated data set leads to Step 4. Step 4: calculate cumulative mortality risk at each time point. Estimate differences of risks in absolute and relative scale comparing intervention strategies to natural course.
Steps to Implement Parametric g-Computation
In Step 1, we fitted a pooled logistic model (i.e., discrete-time hazards model) and adjusted for baseline characteristics, time-varying characteristics, quadratic function of year, and interaction between long-term exposure to PM2.5 and categorical year. The pooled logistic model estimated the probability of death during the year conditioning on survival until the start of the year given all covariates (including PM2.5), which allowed the conditional probability of death and its association with PM2.5 to vary over the year. We chose confounders to control for in the outcome model based on substantive knowledge of the relationship between long-term PM2.5 exposure and mortality as summarized in the simplified directed acyclic graph (Figure S1). We include a full list of covariates in Table 1, with specific forms of covariates listed in Table 2. We included both individual socioeconomic status indicators (e.g., education and family income) and community socioeconomic status indicators (e.g., Canadian Marginalization Index for dissemination area) to fully capture the variation in socioeconomic status among cohort participants, which is a major source of residual confounding. We also included individual behavior indicators, such as dietary and exercise patterns, which are strong risk factors for mortality, precede the exposure, and might share common unmeasured causes with the exposure even though they might not directly cause the exposure.52 Of note, in the setting when only time-fixed covariates were used, we could estimate marginal adjusted survival curves directly using outputs from this pooled logistic model by predicting the probability of death standardized to the distributions of covariates under the intervention of interest (e.g., setting the baseline level of exposure to a specific value while keeping all baseline covariates the same as observed for all participants).19,53 However, in our study setting of time-varying covariates and time-to-event outcome, we also needed to model time-varying covariates (including PM2.5) concentration so that we could simulate time-varying covariates at all follow-up years for all cohort participants, especially for periods after participants’ death.28,29
Table 2 Details for covariate formats and model types for both outcome and covariate models in main analysis.
Variable name Type
Independent variable Dependent variable and corresponding model used
Time-fixed covariates
Age in 2005 (y)a Restricted cubic spline function with 5 knots Not predicted
Sex Binary Not predicted
BMI 5 categories Not predicted
Marital status 3 categories Not predicted
Immigrant An indicator for immigrant and an interaction term between the indicator and a continuous variable for years in Canada among immigrants Not predicted
Visible minority Binary Not predicted
Indigenous status Binary Not predicted
Smoking status 6 categories Not predicted
Alcohol consumption 6 categories Not predicted
Daily consumption of fruit and vegetables 4 categories Not predicted
Leisure time physical activity 3 categories Not predicted
Employment status 3 categories Not predicted
Education attainment 4 categories Not predicted
Urban form 5 categories Not predicted
Airshed 6 categories Not predicted
Time-varying covariates
Time Year and quadratic term of yearb Not predicted
Community size Continuous Bounded normalc (1 to 6) and linear regression
Annual family income quintile Continuous Bounded normal (1 to 5) and linear regression
Canadian Marginalization Index
Immigration and visible minority Continuous Bounded normal (1 to 5) and linear regression
Material resources Continuous Bounded normal (1 to 5) and linear regression
Households and dwellings Continuous Bounded normal (1 to 5) and linear regression
Age and labor force Continuous Bounded normal (1 to 5) and linear regression
3-y average PM2.5 concentration with 1-y lag Natural logarithm-transformed Normal with linear regression
Note: BMI, body mass index; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter.
a In subset analysis restricted to cohort participants ≥65 years of age, we used restricted cubic spline function with 3 knots for age.
b Categorical year was used in the interaction terms between time and the exposure.
c Variables with bounded normal category were modeled and simulated by using the standardized value (subtracting the minimum value and dividing by the range) in linear regression. Simulated values that fell outside the observed range were set to the minimum or maximum of the observed range.
In Step 2, we modeled the time-varying covariates (including PM2.5 concentration) using linear regressions while including variables such as the previous-year value of the covariate of interest, baseline characteristics, same-year values of time-varying covariates set to occur before the covariate of interest, and quadratic function of time. The choice of independent variables in covariate models was based on substantive knowledge as summarized in the simplified directed acyclic graph (Figure S1). We summarize the list of all covariates in Table 1 and the specific forms of covariates included in the covariate models in Table 2. We set the sequence of time-varying covariates as community size, income, immigration and visible minority, material resources, households and dwellings, age and labor force, and long-term exposure to PM2.5. Given that previous studies using different cycles of CCHS found a supralinear PM2.5–mortality association,2,45,54,55 we used natural logarithm-transformed long-term exposure to PM2.5 as the independent variable in both the outcome and covariate model in the main analysis.
In Step 3, we simulated new data sets based on the intervention strategies. For each intervention, we randomly sampled 10,000 participants from the cohort with replacement (i.e., Monte Carlo sampling) and created an empty data set of all sampled participants for all follow-up years until the end of the period of interest (normally the end of the follow-up period, as in this study, but extrapolation is possible with extra assumptions). We simulated new data sets for only 10,000 individuals instead of the number of participants in the study cohorts (∼60,000 participants in each cohort) to save computation time, and a similar practice was conducted before with the smaller cohort.29 Next, we assigned the baseline values of all covariates (values of baseline covariates and values of time-varying covariates at the start of the follow-up period) in each simulated data set to the same as its original data set, we then altered the relevant covariate values based on the intervention strategy (e.g., setting the baseline long-term exposure to PM2.5 to 5 μg/m3 if it was higher than 5 μg/m3 in the threshold intervention of 5 μg/m3, but we could include other covariates if needed). Last, we simulated time-varying covariates at each year after baseline based on their history with covariate models estimated in the second step and altered the covariates based on the intervention strategy.
In Step 4, with the simulated data sets and outcome model from the first step, we calculated for each individual the probability of dying during each year, conditioning on surviving to the beginning of this year, standardized to the distribution of the confounders and long-term exposure to PM2.5 under the intervention strategies, regardless of their observed outcomes. Next, we calculated for each individual the cumulative mortality risk at each year as the cumulative sum of the abovementioned conditional probability of mortality times the probability of surviving until the beginning of the time interval. The estimated cumulative mortality risk at each year equals the average of estimates from all individuals for all hypothetical interventions. We also calculated the absolute difference in cumulative mortality risk and the percentage change in cumulative mortality risk with estimated cumulative mortality risk from the natural course as the reference.
In addition, we calculated the 95% confidence intervals (CIs) for all estimates using standard errors from 200 bootstrap iterations.56 In each iteration, we randomly sampled the same number of participants as in the original cohort with replacement and ran the four steps described above to calculate cumulative mortality risks under the intervention strategies. We chose this number of iterations because we were constrained by available computational resources (>1h of computational time for each bootstrap iteration), and the original application of parametric g-computation in time-varying covariates and time-to-event setting used the same number.29 Future studies with more computational resources might consider a larger number of bootstrap iterations.
Sensitivity Analyses
To test the robustness of our results to model misspecification, we considered a number of different model specifications for both outcome and covariate models, including a) reordering the sequence of time-varying covariates in covariate models by moving “age and labor force” to before other dimensions of the Canadian Marginalization Index, moving income to after all dimensions of the Canadian Marginalization Index, and moving PM2.5 to the first place among all covariates; b) extending the extent of history modeled by including the previous 1- and 2-y values of all the time-varying covariates in the covariate models other than just the previous-year value of the covariate of interest; and c) including time-varying covariates other than long-term PM2.5 as categorical in the outcome model and using the multinomial logistic model for them in the covariate model instead of modeling them as continuous with bounds using the linear model (Table 2 presents the details of the model specifications for each time-varying covariate in the main analysis). We also visually evaluated the simulated and observed adjusted survival curves and histories of covariates under no intervention in the main analysis as a heuristic check of model misspecification.27
Next, to facilitate comparison with previous studies, we used long-term PM2.5 concentrations in the original scale in all models as a sensitivity analysis, which assumed the same log-linear PM2.5–mortality association used in other cohorts4,7 instead of the supralinear one supported by previous studies of different cycles of the CCHS cohort.2,45,54,55 In addition, we also ran a Cox proportional hazards model with the same specification as the outcome model in our main analysis except that we included no time variable and used long-term PM2.5 concentrations in the original scale, which assumed a log-linear PM2.5–mortality association.
Last, given that most deaths occurred among older individuals and that age could modify the health impact of long-term exposure to PM2.5, we conducted a subset analysis restricted to cohort participants ≥65 years of age at the time of enrollment. Because it took up to 24 h to run one round of the sensitivity analysis without bootstrapping, we were unable to perform bootstrapping to calculate CIs for sensitivity analyses owing to computational constraints and, therefore, do not present variances for our estimates. We pooled cycle-specific estimates from sensitivity analyses by averaging the estimates in each cycle. All analyses were done in R (version 4.0.5; R Development Core Team) with the “gfoRmula” package.51 The R code to replicate these analyses and a simulated data set are available at the following link: https://github.com/suthlam/cchs_g_computation.git.
Results
We observed 6,475 (10.4%), 6,525 (10.5%), and 6,135 (9.2%) nonaccidental deaths in the 11 y follow-up period starting from 2005 among the three cycles of CCHS cohorts of 62,365, 62,380, and 66,385 participants, respectively (Table 1). The three cycle cohorts were comparable in all descriptive statistics (Table 1). Without any hypothetical intervention, the observed average long-term exposure to PM2.5 in three cycles of our cohort decreased slightly from 6.4±2.2 μg/m3, 6.5±2.3 μg/m3, 6.5±2.3 μg/m3 in 2005 to 5.8±2.0 μg/m3, 6.0±2.0 μg/m3, and 6.0±2.0 μg/m3 in 2015, respectively (Table 1).
All hypothetical intervention strategies explored in this study led to lower 11-y cumulative mortality risks than the estimated value under a natural course without intervention, 102.5 per 1,000 participants (95% CI: 100.3, 104.8). The reductions in 11-y cumulative mortality risks from the natural course were 0.20 per 1,000 participants (95% CI: 0.06, 0.34) under the threshold of 8.8 μg/m3, 0.63 per 1,000 participants (95% CI: 0.18, 1.07) under the threshold of 7.04 μg/m3, 1.87 per 1,000 participants (95% CI: 0.53, 3.21) under the threshold of 5 μg/m3, 3.08 per 1,000 participants (95% CI: 0.85, 5.31) under the threshold of 4 μg/m3, 1.68 per 1,000 participants (95% CI: −0.15, 3.51) under the relative reduction of 5% per interval, and 3.40 per 1,000 participants (95% CI: −0.23, 7.03) under the relative reduction of 10% per interval. Of note, the reduction in 11-y cumulative mortality risks could also be interpreted as the number of deaths that would have been prevented if the intervention was employed instead of maintaining the status quo. Changes in relative scale showed a similar pattern (Table 3). To fulfill the four threshold intervention strategies, averages of 18.7%, 38.3%, 72.0%, and 91.4% of participants experienced change in their natural course exposure every year, respectively, whereas 100% had their exposure changed under the two relative reduction intervention strategies (Table 3). The corresponding reductions in average simulated PM2.5 from the start of follow-up to the end of year 11 ranged from 0.13 to 1.87 μg/m3 for threshold intervention strategies, and from 1.25 to 2.18 μg/m3 for relative reduction intervention strategies (Table 3). Across all strategies, we observed steady expansions in differences of yearly cumulative mortality risks between the natural course and other strategies until the seventh year of follow-up, after which the differences remain constant and shrank during the last year of follow-up (Figure 2, numeric results in Table S2). In the main analysis, we pooled estimates of yearly cumulative mortality risks across cycles using random-effect meta-regression and pooled estimates of differences (absolute and relative scale) in cumulative mortality risks using fixed-effect meta-regression. Cycle-specific results with corresponding I2 values are summarized in Figure S2, with numeric results in Table S3.
Table 3 Summaries of estimated 11-y cumulative mortality risk under different intervention strategies pooled across cycles and differences in estimated risk compared with natural course in relative and absolute scale, and corresponding average simulated PM2.5 and proportion of cohort participants with exposure changed for all intervention strategies.
Intervention strategy 11-y CMR (per 1,000 participants) (95% CI) Difference in 11-y CMR (per 1,000 participants) (95% CI) Percentage change in 11-y CMR (95% CI) Average percentage of participants with exposure changeda Average simulated PM2.5 concentration (μg/m3)a
Natural course 102.5 (100.3, 104.8) Ref Ref 0 5.62
Threshold (μg/m3)
8.8 102.3 (100.1, 104.6) −0.20 (−0.34, −0.06) −0.19 (−0.33, −0.05) 18.7 5.49
7.04 102.0 (99.7, 104.2) −0.63 (−1.07, −0.18) −0.60 (−1.03, −0.17) 38.3 5.21
5 100.9 (98.4, 103.5) −1.87 (−3.21, −0.53) −1.79 (−3.11, −0.48) 72.0 4.42
4 99.8 (96.7, 102.9) −3.08 (−5.31, −0.85) −2.95 (−5.14, −0.77) 91.4 3.75
Percentage reduction per interval (%)
5 101.4 (98.6, 104.2) −1.68 (−3.51, 0.15) −1.61 (−3.40, 0.17) 100 4.37
10 99.8 (95.6, 103.9) −3.40 (−7.03, 0.23) −3.27 (−6.81, 0.28) 100 3.44
Note: CI, confidence interval; CMR, cumulative mortality risk; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; Ref, reference.
a This is the three-cycle average of the mean value across all years.
Figure 2. Differences in yearly cumulative mortality risks pooled across cycles comparing different intervention strategies to natural course, with weights equal to the inverse of variance. Numeric results are presented in Table S2. Note: PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; R90, yearly relative reduction values set at 10% per interval; R95: yearly relative reduction values set at 5% per interval; T4, threshold value set at a PM2.5 level that was further below the WHO guideline (4 μg/m3); T5, threshold value set at the new WHO guideline of 5 μg/m3; T7.04, threshold value set at 80% of the current Canadian Ambient Air Quality Standards for PM2.5 (or 7.04 μg/m3); T8.8, threshold value (reduced to threshold value if above) set at the current Canadian Ambient Air Quality Standards for PM2.5 of 8.8 μg/m3; WHO, World Health Organization.
Figure 2 is a line graph, plotting difference in 11 year cumulative mortality risk (per 1,000 participants), ranging from negative 4 to 0 in unit increments (y-axis) across year since enrollment, ranging from 1 to 11 in increments of 2 (x-axis) for T 8.8, T 7.04, T 5, T 4, R 95, and R 90.
Heuristic checks of model fitting in the main analysis support the robustness of our estimates: a) The cumulative mortality risk estimated by parametric g-computation under the natural course closely tracked the value observed (Figure S3), and b) the observed mean values of all time-varying covariates were similar to those simulated under the natural course over time (Figure S3). Of note, given that cohort participants had no time-varying covariates recorded after their death, whereas we simulated participants’ time-varying covariates for all years, differences between observed and simulated covariates were to be expected, especially later in the study period. Furthermore, sensitivity analyses with different model specifications (i.e., different sequence of time-varying covariate, extent of history modeled, and parametrization of time-varying confounders) resulted in estimates similar to those for the main analysis (Figure 3, numeric results in Table S4).
Figure 3. Differences in 11-y cumulative mortality risks comparing different intervention strategies to natural course for main analysis and sensitivity analyses. Numeric results are presented in Tables 3 and S4. Note: 65+, subset analysis restricted to cohort participants ≥65 years of age; Cat, including time-varying covariates other than long-term PM2.5 as categorical in outcome model and using multinomial logistic model for them in the covariate model; O1, placing Canadian Marginalization Index-age and labor force before the other Canadian Marginalization Index in occurring sequence of time-varying covariate; O2, moving income to after Canadian Marginalization Index in occurring sequence of time-varying covariate; O3, moving PM2.5 to the first in occurring sequence of time-varying covariate; Org, using long-term PM2.5 in original scale in all models; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; R90, yearly relative reduction values set at 10% per interval; R95, yearly relative reduction values set at 5% per interval; T4, threshold value set at a PM2.5 level that was further below the WHO guideline (4 μg/m3); T5, threshold value set at the new WHO guideline of 5 μg/m3; T7.04, threshold value set at 80% of the current Canadian Ambient Air Quality Standards for PM2.5 (or 7.04 μg/m3); T8.8, threshold value (reduced to threshold value if above) set at the current Canadian Ambient Air Quality Standards for PM2.5 of 8.8 μg/m3; TV, adding all time-varying covariates of the previous 1 and 2 y to the covariate model; WHO, World Health Organization.
Figure 3 is a graph, plotting difference in 11 year cumulative mortality risk (per 1,000 participants), ranging from negative 6 to 0 in increments of negative 2 (y-axis) across analysis, ranging as main, O 1, O 2, O 3, T V, Cat, Org, and 65 plus (x-axis) for T 8.8, T 7.04, T 5, T 4, R 95, and R 90.
When assuming a log-linear PM2.5–mortality association in the sensitivity analysis (compared with the supralinear association assumed in main analysis by log-transforming the exposure), reductions in 11-y cumulative mortality risks comparing other intervention strategies to the natural course ranged from 0.01 per 1,000 participants to 1.65 per 1,000 participants, slightly smaller than in the main analysis assuming a supralinear PM2.5–mortality association (log-transformed PM2.5 was used as exposure in modeling) (Table S4). The Cox model assuming a log-linear PM2.5–mortality association found a 15.6% (95% CI: 4.0%, 28.5%) increase in hazard of death per 10-μg/m3 increase in PM2.5. When focusing on cohort participants ≥65 years of age at the start of follow-up, reductions in 11-y cumulative mortality risks comparing other intervention strategies to the natural course ranged from 0.49 per 1,000 participants to 7.07 per 1,000 participants (Table S4), which is larger than was found for the main analysis using the general population at ≥25 years of age.
Discussion
In the present study, we applied the parametric g-computation as an analytical alternative that is particularly valuable for air pollution epidemiological research, especially for evaluating specific intervention strategies. With application in a large Canadian cohort, we demonstrated how to incorporate consideration of complex time structure in the data and how to calculate causally interpretable cumulative risk estimates over the follow-up period (i.e., adjusted survival curves) with parametric g-computation. We described that any intervention further reducing the long-term exposure to PM2.5 would reduce the cumulative mortality risk, even in a region with relatively low levels of ambient PM2.5. Such a reduction in cumulative risk increased over time and flattened toward the end of the follow-up period on both the relative and the absolute scales. The older population also experienced greater benefits from the explored hypothetical intervention strategies than the general population.
Numerous observational studies have found positive associations between long-term exposure to PM2.5 and nonaccidental mortality. A meta-analysis reported a pooled effect estimate of 6% (95% CI: 4%, 8%) increase in hazard of death per 10-μg/m3 increase in PM2.5 (HR-1).5 A recent study from 2000 to 2012 in a similar Canadian cohort found an 11% (95% CI: 4%, 18%) increase in hazard of nonaccidental death per 10-μg/m3 increase in chronic exposure to PM2.5 with a Cox proportional hazards model.2 Our sensitivity analysis using the Cox model without time-varying coefficients found similar numeric results [15.6% (95% CI: 4.0%, 28.5%)]. Although we cannot directly compare our estimates from the main analysis to previous results given the difference in interventions explored, the consistent reductions in cumulative mortality risk over the follow-up period across intervention strategies when compared with the natural course in this study lend further support to previous findings that PM2.5 is detrimental to health, even at levels below current standards. For example, we identified a 0.19% (95% CI: 0.05%, 0.33%) decrease in 11-y cumulative mortality risk comparing the hypothetical intervention strategy with the threshold of 8.8 μg/m3 to natural course, providing evidence of health benefits from policies that further reduce the air pollution level to below the current Canadian standard of 8.8 μg/m3, which is already lower than the 12-μg/m3 standard of the United States explored by previous studies.4,9 To facilitate comparison with previous studies assuming a log-linear PM2.5–mortality association, we included sensitivity analysis using PM2.5 on the normal scale and found reduced cumulative mortality risks in all hypothetical interventions compared with maintaining the status quo, but the numeric values are smaller than those in the main analysis. The observed difference in the numeric values of analysis assuming log-linear association and analysis assuming supralinear association is a combination of difference in how the exposure–response relationship is modeled and how the exposure model performs. However, given the existing evidence from Canadian cohorts and the similarity between the observed survival curve and the estimated survival curve using parametric g-computation under no intervention in the main analysis,2,45,54,55 we have confidence in the validity of results assuming a supralinear association.
More importantly, in this study we demonstrated how to incorporate more flexibilities in simulating real-world interventions with g-computation and provide intuitive estimates for the benefits of such interventions. Taking the hypothetical intervention strategy with the threshold of the current Canadian Ambient Air Quality Standards as an example, the average long-term exposure to PM2.5 in 2005 was ∼6.5 μg/m3, below the standard of 8.8 μg/m3. However, some cohort participants were exposed to PM2.5 levels >8.8 μg/m3 during some years of follow-up, and our hypothetical intervention affected only these subject-years by reducing their exposure to 8.8 μg/m3, representing a three-cycle average of 18.7% of participants across all years. Because the observed PM2.5 levels decreased without any intervention in our study, fewer participants were directly intervened on in later years under threshold intervention strategies, explaining the flattened differences observed in the cumulative risks between intervention strategies in later years. However, all time-varying covariates after the intervention on PM2.5 would change accordingly owing to the intervention on PM2.5, thus influencing future outcomes as well. Such a dynamic intervention strategy incorporated the consideration of people who could be intervened on and is more realistic than the static intervention strategy commonly employed in health burden estimation with traditional exposure–response function, which sets change in exposure at a fixed value for all individuals throughout the period of interest. In addition, although we provided differences only in cumulative risk as compared with the natural course, it is easy to estimate differences between any two hypothetical intervention strategies.
Furthermore, the estimated cumulative risks over the follow-up period by g-computation (i.e., adjusted survival curves) and corresponding comparisons of values between different hypothetical interventions provided clearer causal interpretation and more information than a single HR or period-specific HRs, as generally used in air pollution studies (Table S1). In the context of health impacts from chronic exposure to PM2.5, HR can change over time because the toxicity of PM2.5 (e.g., chemical composition of PM2.5) and the susceptibility of the population to PM2.5 could change over time, whereas the standard Cox model assumed a constant HR and period-specific HR from extensions of the Cox model had a built-in bias that led to ambiguity in causal interpretation.57 On the other hand, the cumulative mortality risks estimated in the present study avoided such ambiguity in interpretation while also demonstrating the change of intervention effect over time.19 In addition, obtaining the casually interpretable absolute differences in cumulative risks between hypothetical intervention strategies over time could be particularly helpful for comparing different scenarios regarding public health benefits.58 Moreover, if policies affecting air pollutants such as PM2.5 could further affect prognostic covariates influencing both future PM2.5 levels and health outcomes (commonly referred to as exposure–confounder feedback), traditional regression methods based on adjustment in a multivariable model would fail and lead to biased estimates for the effect, whereas g-computation is designed particularly to solve this problem.24,26,59 An example of such exposure-confounder feedback is that people might move due to high level of PM2.5 in their current community and subsequently change the characteristics of their community of residence, while the characteristics of their new community also affect the level of PM2.5 and probability of death. Controlling for such community characteristics is necessary for confounding control, but doing so with traditional methods will remove the indirect effect mediated through community characteristics and introduce collider-stratification bias60 if any unmeasured confounder of the community characteristics and death exists.59 However, making the decision to move based on the community level of PM2.5 is unlikely in countries with relatively low PM2.5 levels, such as Canada. Therefore, exposure–confounder feedback is expected to be negligible in our study, but it is possible to be meaningful in countries with higher PM2.5 levels.
This study has a few limitations that need to be acknowledged. First, parametric g-computation can only account for measured confounders and a lack of conditional exchangeability (i.e., residual confounding) might exist due to unmeasured confounders or measurement errors of existing confounders, regardless of our extensive list of confounders considered based on substantive knowledge on risk factors of PM2.5 exposure and death (Figure S1). For example, we assumed many individual behavior, demographic, and socioeconomic variables to be time-invariant (e.g., employment status, BMI) owing to data availability (these variables were only reported once at the time of enrollment), whereas they likely actually changed over the study period. However, we also included time-varying individual income and community demographic and socioeconomic variables in our models, mitigating the concern of residual confounding from these baseline variables. In addition, like other cohort studies of air pollution, we used postal code–level PM2.5 levels as surrogates for individual exposure to PM2.5, which might introduce exposure misclassification.61 Recent studies, however, have shown that such bias may either not bias effect estimates62 or bias these estimates toward the null,63 making our estimates more conservative.
Second, although we explored different model specifications and found similar results in the sensitivity analyses, we cannot rule out the possibility of model misspecification, especially given the fact that parametric g-computation requires correct model specification of both the outcome and covariate models to achieve unbiased results. Notably, McGrath et al. demonstrated that the g-null paradox, a form of model misspecification that was traditionally believed to cause false rejection of null hypothesis under a sharp null effect,64 is unavoidable in parametric g-computation even when the sharp null hypothesis does not hold, and they recommended using more flexible models in analysis.65 However, the magnitude of bias depends on the extent of exposure–confounder feedback and time-varying confounding. In the context of this study, we would expect relatively small exposure–confounder feedback and, thus, less concern over the g-null paradox. In addition, consistent results from sensitivity analysis using more flexible models supported the robustness of our results.
Third, this being an active research field, the existing R package for parametric g-computation does not support features like incorporation of spline functions of time-varying covariates into the model, direct estimation of randomized interventional strategy,66 model fit checking with significance tests, or bias analysis. However, the current package provided enough flexibility for our study to employ flexible models that mitigated the possibility of violating the positivity assumption via model extrapolation. For example, we were able to a) incorporate flexible supralinear PM2.5–mortality association and temporal changes into the conditional probability of mortality in the estimation as supported by previous studies,2,54,55 b) incorporate restricted cubic spline function of baseline age in all models, and c) conduct sensitivity analysis with categorical time-varying confounders. In addition, although not relevant to our cohort given that we had the all-cause mortality as the outcome and no loss to follow-up, right censoring and informative loss to follow-up could be handled by parametric g-computation and the existing R package by simulating data on participants as though they had not been censored.67 It is worth mentioning that other methods could also handle the methodological considerations that g-computation addresses—consideration of complex time structure and reporting of adjusted survival curves—and have been applied in air pollution epidemiological research, including Inverse Probability of Treatment Weighting.6 Furthermore, some recent approaches, such as the targeted maximum likelihood estimation, can also be used to directly evaluate individualized dynamic intervention strategies of continuous exposures and provide doubly robust estimates that are less vulnerable to model misspecification with valid statistical inference when data-adaptive/machine-learning methods are incorporated.68,69
Finally, PM2.5 is a mixture of varying chemical components and toxicity and is generated from different sources of emissions (e.g., traffic, industries, wildfires). In this paper, we focused on PM2.5 without distinguishing the PM2.5 composition or the sources of emissions. This potentially violated the consistency assumption (i.e., no-multiple-versions-of-treatment and all exposed individuals received the same version of treatment). If there is any unmeasured confounder of the “version of treatment” and outcome relationship, the effect estimates could be biased according to a recent simulation study, with magnitude and direction of such bias depending on the strength of confounding.70 In future studies, it would be important to consider the possible differential toxicity of PM2.5 components and define hypothetical interventions targeting different sources of PM2.5 emissions separately.
Conclusion
This study demonstrated the benefits of using parametric g-computation as an analytical alternative for air pollution epidemiological research, especially for evaluating the potential effects of realistic dynamic intervention strategies in the time-to-event setting with time-varying exposure and confounders. With a large Canadian cohort, we calculated causally interpretable cumulative risk estimates over the follow-up period and corresponding benefits compared with maintaining the status quo. We also found that any intervention further reducing the long-term exposure to PM2.5 would reduce the cumulative mortality risk from maintaining the status quo, even in a population already exposed to relatively low levels of ambient PM2.5.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This study was funded by Health Canada (810630). The funder supported the study design, data collection, and analysis in this study.
==== Refs
References
1. Beelen R, Raaschou-Nielsen O, Stafoggia M, Andersen ZJ, Weinmayr G, Hoffmann B, et al. 2014. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre ESCAPE project. Lancet 383 (9919 ):785–795, PMID: , 10.1016/S0140-6736(13)62158-3.24332274
2. Christidis T, Erickson AC, Pappin AJ, Crouse DL, Pinault LL, Weichenthal SA, et al. 2019. Low concentrations of fine particle air pollution and mortality in the Canadian Community Health Survey cohort. Environ Health 18 (1 ):84, PMID: , 10.1186/s12940-019-0518-y.31601202
3. Crouse DL, Peters PA, van Donkelaar A, Goldberg MS, Villeneuve PJ, Brion O, et al. 2012. Risk of nonaccidental and cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate matter: a Canadian national-level cohort study. Environ Health Perspect 120 (5 ):708–714, PMID: , 10.1289/ehp.1104049.22313724
4. Di Q, Wang Y, Zanobetti A, Wang Y, Koutrakis P, Choirat C, et al. 2017. Air pollution and mortality in the Medicare population. N Engl J Med 376 (26 ):2513–2522, PMID: , 10.1056/NEJMoa1702747.28657878
5. Hoek G, Krishnan RM, Beelen R, Peters A, Ostro B, Brunekreef B, et al. 2013. Long-term air pollution exposure and cardio-respiratory mortality: a review. Environ Health 12 (1 ):43, PMID: , 10.1186/1476-069X-12-43.23714370
6. Wu X, Braun D, Schwartz J, Kioumourtzoglou MA, Dominici F. 2020. Evaluating the impact of long-term exposure to fine particulate matter on mortality among the elderly. Sci Adv 6 (29 ):eaba5692, PMID: , 10.1126/sciadv.aba5692.32832626
7. Cesaroni G, Forastiere F, Stafoggia M, Andersen ZJ, Badaloni C, Beelen R, et al. 2014. Long term exposure to ambient air pollution and incidence of acute coronary events: prospective cohort study and meta-analysis in 11 European cohorts from the ESCAPE project. BMJ 348 :f7412, PMID: , 10.1136/bmj.f7412.24452269
8. Danesh Yazdi M, Wang Y, Di Q, Wei Y, Requia WJ, Shi L, et al. 2021. Long-term association of air pollution and hospital admissions among Medicare participants using a doubly robust additive model. Circulation 143 (16 ):1584–1596, PMID: , 10.1161/CIRCULATIONAHA.120.050252.33611922
9. Makar M, Antonelli J, Di Q, Cutler D, Schwartz J, Dominici F. 2017. Estimating the causal effect of low levels of fine particulate matter on hospitalization. Epidemiology 28 (5 ):627–634, PMID: , 10.1097/EDE.0000000000000690.28768298
10. Bi J, D’Souza RR, Rich DQ, Hopke PK, Russell AG, Liu Y, et al. 2020. Temporal changes in short-term associations between cardiorespiratory emergency department visits and PM2.5 in Los Angeles, 2005 to 2016. Environ Res 190 :109967, PMID: , 10.1016/j.envres.2020.109967.32810677
11. Breitner S, Stölzel M, Cyrys J, Pitz M, Wölke G, Kreyling W, et al. 2009. Short-term mortality rates during a decade of improved air quality in Erfurt, Germany. Environ Health Perspect 117 (3 ):448–454, PMID: , 10.1289/ehp.11711.19337521
12. Chen C, Warrington JA, Dominici F, Peng RD, Esty DC, Bobb JF, et al. 2021. Temporal variation in association between short-term exposure to fine particulate matter and hospitalisations in older adults in the USA: a long-term time-series analysis of the US Medicare dataset. Lancet Planet Health 5 (8 ):e534–e541, PMID: , 10.1016/S2542-5196(21)00168-6.34390671
13. Dominici F, Peng RD, Zeger SL, White RH, Samet JM. 2007. Particulate air pollution and mortality in the United States: did the risks change from 1987 to 2000? Am J Epidemiol 166 (8 ):880–888, PMID: , 10.1093/aje/kwm222.17728271
14. Atkinson RW, Mills IC, Walton HA, Anderson HR. 2015. Fine particle components and health—a systematic review and meta-analysis of epidemiological time series studies of daily mortality and hospital admissions. J Expo Sci Environ Epidemiol 25 (2 ):208–214, PMID: , 10.1038/jes.2014.63.25227730
15. Rohr AC, Wyzga RE. 2012. Attributing health effects to individual particulate matter constituents. Atmos Environ 62 :130–152, 10.1016/j.atmosenv.2012.07.036.
16. Wyzga RE, Rohr AC. 2015. Long-term particulate matter exposure: attributing health effects to individual PM components. J Air Waste Manag Assoc 65 (5 ):523–543, PMID: , 10.1080/10962247.2015.1020396.25947312
17. Yang Y, Ruan Z, Wang X, Yang Y, Mason TG, Lin H, et al. 2019. Short-term and long-term exposures to fine particulate matter constituents and health: a systematic review and meta-analysis. Environ Pollut 247 :874–882, PMID: , 10.1016/j.envpol.2018.12.060.30731313
18. Zhang Z, Reinikainen J, Adeleke KA, Pieterse ME, Groothuis-Oudshoorn CGM. 2018. Time-varying covariates and coefficients in Cox regression models. Ann Transl Med 6 (7 ):121, PMID: , 10.21037/atm.2018.02.12.29955581
19. Hernán MA. 2010. The hazards of hazard ratios. Epidemiology 21 (1 ):13–15, PMID: , 10.1097/EDE.0b013e3181c1ea43.20010207
20. Flanders WD, Klein M. 2007. Properties of 2 counterfactual effect definitions of a point exposure. Epidemiology 18 (4 ):453–460, PMID: , 10.1097/01.ede.0000261472.07150.4f.17473709
21. Stovitz SD, Banack HR, Kaufman JS. 2018. ‘Depletion of the susceptibles’ taught through a story, a table and basic arithmetic. BMJ Evid Based Med 23 (5 ):199, PMID: , 10.1136/bmjebm-2018-110972.29982175
22. Hernán MA, Robins JM. 2020. Causal Inference: What If. Boca Raton, FL: CRC Press.
23. Zigler CM, Dominici F. 2014. Point: clarifying policy evidence with potential–outcomes thinking—beyond exposure–response estimation in air pollution epidemiology. Am J Epidemiol 180 (12 ):1133–1140, PMID: , 10.1093/aje/kwu263.25399414
24. Robins J. 1986. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect. Math Model 7 (9–12 ):1393–1512, 10.1016/0270-0255(86)90088-6.
25. Cole SR, Frangakis CE. 2009. The consistency statement in causal inference: a definition or an assumption? Epidemiology 20 (1 ):3–5, PMID: , 10.1097/EDE.0b013e31818ef366.19234395
26. Naimi AI, Cole SR, Kennedy EH. 2016. An introduction to g methods. Int J Epidemiol 46 (2 ):756–762, PMID: , 10.1093/ije/dyw323.28039382
27. Chiu YH, Chavarro JE, Dickerman BA, Manson JE, Mukamal KJ, Rexrode KM, et al. 2021. Estimating the effect of nutritional interventions using observational data: the American Heart Association’s 2020 dietary goals and mortality. Am J Clin Nutr 114 (2 ):690–703, PMID: , 10.1093/ajcn/nqab100.34041538
28. Keil AP, Edwards JK, Richardson DR, Naimi AI, Cole SR. 2014. The parametric g-formula for time-to-event data: intuition with a worked example. Epidemiology 25 (6 ):889–897, PMID: , 10.1097/EDE.0000000000000160.25140837
29. Taubman SL, Robins JM, Mittleman MA, Hernán MA. 2009. Intervening on risk factors for coronary heart disease: an application of the parametric g-formula. Int J Epidemiol 38 (6 ):1599–1611, PMID: , 10.1093/ije/dyp192.19389875
30. Westreich D, Cole SR, Young JG, Palella F, Tien PC, Kingsley L, et al. 2012. The parametric g-formula to estimate the effect of highly active antiretroviral therapy on incident AIDS or death. Stat Med 31 (18 ):2000–2009, PMID: , 10.1002/sim.5316.22495733
31. Keil AP, Buckley JP, Kalkbrenner AE. 2021. Bayesian g-computation for estimating impacts of interventions on exposure mixtures: demonstration with metals from coal-fired power plants and birth weight. Am J Epidemiol 190 (12 ):2647–2657, PMID: , 10.1093/aje/kwab053.33751055
32. Neophytou AM, Costello S, Picciotto S, Brown DM, Attfield MD, Blair A, et al. 2019. Diesel exhaust, respirable dust, and ischemic heart disease: an application of the parametric g-formula. Epidemiology 30 (2 ):177–185, PMID: , 10.1097/EDE.0000000000000954.30489348
33. Neophytou AM, Costello S, Picciotto S, Noth EM, Liu S, Lutzker L, et al. 2019. Accelerated lung function decline in an aluminum manufacturing industry cohort exposed to PM2.5: an application of the parametric g-formula. Occup Environ Med 76 (12 ):888–894, PMID: , 10.1136/oemed-2019-105908.31615860
34. Garcia E, Urman R, Berhane K, McConnell R, Gilliland F. 2019. Effects of policy-driven hypothetical air pollutant interventions on childhood asthma incidence in southern California. Proc Natl Acad Sci USA 116 (32 ):15883–15888, PMID: , 10.1073/pnas.1815678116.31332016
35. Urman R, Garcia E, Berhane K, McConnell R, Gauderman WJ, Gilliland F. 2020. The potential effects of policy-driven air pollution interventions on childhood lung development. Am J Respir Crit Care Med 201 (4 ):438–444, PMID: , 10.1164/rccm.201903-0670OC.31644884
36. Statistics Canada. 2007. Canadian Community Health Survey (CCHS) Cycle 1.1. Last modified 24 October 2007. https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&Id=3359 [accessed 26 September 2021].
37. Statistics Canada. 2007. Canadian Community Health Survey (CCHS) Cycle 2.1. Last modified 24 October 2007. https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&Id=4995 [accessed 26 September 2021].
38. Statistics Canada. 2007. Canadian Community Health Survey (CCHS) Cycle 3.1. Last modified 24 October 2007. https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&Id=22642 [accessed 26 September 2021].
39. Sanmartin C, Decady Y, Trudeau R, Dasylva A, Tjepkema M, Finès P, et al. 2016. Linking the Canadian Community Health Survey and the Canadian Mortality Database: an enhanced data source for the study of mortality. Health Rep 27 (12 ):10–18, PMID: .28002578
40. WHO (World Health Organization). 1978. International Statistical Classification of Diseases, Ninth Revision, Basic Tabulation List with Alphabetic Index. http://apps.who.int/classifications/icd10/browse/2016/en [accessed 8 March 2023].
41. WHO. 2016. International Statistical Classification of Diseases and Related Health Problems, 10th Revision. https://apps.who.int/iris/handle/10665/39473 [accessed 8 March 2023].
42. van Donkelaar A, Martin RV, Li C, Burnett RT. 2019. Regional estimates of chemical composition of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors: v 4.NA.02.MAPLE. Nenodo. Published online 30 January 2019. 10.5281/zenodo.6557778.
43. van Donkelaar A, Martin RV, Li C, Burnett RT. 2019. Regional estimates of chemical composition of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors. Environ Sci Technol 53 (5 ):2595–2611, PMID: , 10.1021/acs.est.8b06392.30698001
44. Canadian Council of Ministers of the Environment. n. d. Canadian Ambient Air Quality Standards. https://ccme.ca/en/air-quality-report#slide-7 [accessed 11 June 2022].
45. Pinault L, Tjepkema M, Crouse DL, Weichenthal S, van Donkelaar A, Martin RV, et al. 2016. Risk estimates of mortality attributed to low concentrations of ambient fine particulate matter in the Canadian Community Health Survey cohort. Environ Health 15 :18, PMID: , 10.1186/s12940-016-0111-6.26864652
46. Pinault LL, Weichenthal S, Crouse DL, Brauer M, Erickson A, van Donkelaar A, et al. 2017. Associations between fine particulate matter and mortality in the 2001 Canadian Census Health and Environment cohort. Environ Res 159 :406–415, PMID: , 10.1016/j.envres.2017.08.037.28850858
47. Ontario Community Health Profiles Partnership. 2020. Canadian Marginalization Index (CAN-Marg). http://www.ontariohealthprofiles.ca/canmargCAN.php [accessed 14 October 2020].
48. Crouse DL, Philip S, van Donkelaar A, Martin RV, Jessiman B, Peters PA, et al. 2016. A new method to jointly estimate the mortality risk of long-term exposure to fine particulate matter and its components. Sci Rep 6 (1 ):18916, PMID: , 10.1038/srep18916.26732864
49. Health Canada. 2021. Health Impacts of Air Pollution in Canada: Estimates of Morbidity and Premature Mortality Outcomes—2021 Report. https://www.canada.ca/en/health-canada/services/publications/healthy-living/2021-health-effects-indoor-air-pollution.html [accessed 17 June 2021].
50. Balduzzi S, Rücker G, Schwarzer G. 2019. How to perform a meta-analysis with R: a practical tutorial. Evid Based Ment Health 22 (4 ):153–160, PMID: , 10.1136/ebmental-2019-300117.31563865
51. McGrath S, Lin V, Zhang Z, Petito LC, Logan RW, Hernán MA, et al. 2020. gfoRmula: an R package for estimating the effects of sustained treatment strategies via the parametric g-formula. Patterns (NY) 1 (3 ):100008, PMID: , 10.1016/j.patter.2020.100008.32656541
52. VanderWeele TJ, Shpitser I. 2011. A new criterion for confounder selection. Biometrics 67 (4 ):1406–1413, PMID: , 10.1111/j.1541-0420.2011.01619.x.21627630
53. Snowden JM, Rose S, Mortimer KM. 2011. Implementation of G-computation on a simulated data set: demonstration of a causal inference technique. Am J Epidemiol 173 (7 ):731–738, PMID: , 10.1093/aje/kwq472.21415029
54. Nasari MM, Szyszkowicz M, Chen H, Crouse D, Turner MC, Jerrett M, et al. 2016. A class of non-linear exposure-response models suitable for health impact assessment applicable to large cohort studies of ambient air pollution. Air Qual Atmos Health 9 (8 ):961–972, PMID: , 10.1007/s11869-016-0398-z.27867428
55. Pappin AJ, Christidis T, Pinault LL, Crouse DL, Brook JR, Erickson A, et al. 2019. Examining the shape of the association between low levels of fine particulate matter and mortality across three cycles of the Canadian Census Health and Environment cohort. Environ Health Perspect 127 (10 ):107008, PMID: , 10.1289/EHP5204.31638837
56. Efron B, Tibshirani RJ. 1993. An Introduction to the Bootstrap. New York, NY: Chapman & Hall/CRC.
57. Stensrud MJ, Hernán MA. 2020. Why test for proportional hazards? JAMA 323 (14 ):1401–1402, PMID: , 10.1001/jama.2020.1267.32167523
58. Schwartz LM, Woloshin S, Dvorin EL, Welch HG. 2006. Ratio measures in leading medical journals: structured review of accessibility of underlying absolute risks. BMJ 333 (7581 ):1248, PMID: , 10.1136/bmj.38985.564317.7C.17060338
59. Daniel RM, Cousens SN, De Stavola BL, Kenward MG, Sterne JAC. 2013. Methods for dealing with time-dependent confounding. Stat Med 32 (9 ):1584–1618, PMID: , 10.1002/sim.5686.23208861
60. Hernán MA, Hernández-Díaz S, Robins JM. 2004. A structural approach to selection bias. Epidemiology 15 (5 ):615–625, PMID: , 10.1097/01.ede.0000135174.63482.43.15308962
61. Sheppard L, Burnett RT, Szpiro AA, Kim SY, Jerrett M, Pope CA III, et al. 2012. Confounding and exposure measurement error in air pollution epidemiology. Air Qual Atmos Health 5 (2 ):203–216, PMID: , 10.1007/s11869-011-0140-9.22662023
62. Richmond-Bryant J, Long TC. 2020. Influence of exposure measurement errors on results from epidemiologic studies of different designs. J Expo Sci Environ Epidemiol 30 (3 ):420–429, PMID: , 10.1038/s41370-019-0164-z.31477780
63. Samoli E, Butland BK, Rodopoulou S, Atkinson RW, Barratt B, Beevers SD, et al. 2020. The impact of measurement error in modeled ambient particles exposures on health effect estimates in multilevel analysis: a simulation study. Environ Epidemiol 4 (3 ):e094, PMID: , 10.1097/EE9.0000000000000094.32656489
64. Robins JM, Wasserman LA. 1997. Estimation of effects of sequential treatments by reparameterizing directed acyclic graphs. Proceedings of the thirteenth conference on uncertainty in artificial intelligence.
65. McGrath S, Young JG, Hernán MA. 2022. Revisiting the g-null paradox. Epidemiology 33 (1 ):114–120, PMID: , 10.1097/EDE.0000000000001431.34711734
66. VanderWeele TJ, Vansteelandt S, Robins JM. 2014. Effect decomposition in the presence of an exposure-induced mediator-outcome confounder. Epidemiology 25 (2 ):300–306, PMID: , 10.1097/EDE.0000000000000034.24487213
67. Keil AP, Edwards JK. 2018. A review of time scale fundamentals in the g-formula and insidious selection bias. Curr Epidemiol Rep 5 (3 ):205–213, PMID: , 10.1007/s40471-018-0153-0.30555772
68. Muñoz ID, van der Laan M. 2012. Population intervention causal effects based on stochastic interventions. Biometrics 68 (2 ):541–549, PMID: , 10.1111/j.1541-0420.2011.01685.x.21977966
69. van der Laan MJ, Petersen ML. 2007. Causal effect models for realistic individualized treatment and intention to treat rules. Int J Biostat 3 (1 ):article 3, PMID: , 10.2202/1557-4679.1022.19122793
70. Weichenthal S, Ripley S, Korsiak J. 2023. Fine particulate air pollution and the “no-multiple-versions-of-treatment” assumption: does particle composition matter for causal inference? Am J Epidemiol 192 (2 ):147–153, PMID: , 10.1093/aje/kwac191.36331277
|
PMC010xxxxxx/PMC10019502.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36927188
EHP12193
10.1289/EHP12193
Invited Perspective
Invited Perspective: Can Eating a Healthy Diet during Pregnancy Attenuate the Obesogenic Effects of Persistent Organic Pollutants?
https://orcid.org/0000-0002-7791-5167
Jaacks Lindsay 1
1 Global Academy of Agriculture and Food Systems, University of Edinburgh, Midlothian, UK
Address correspondence to Lindsay Jaacks. Email: [email protected]
16 3 2023
3 2023
131 3 03130623 9 2022
21 12 2022
17 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The author declares she has no competing interests to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11258
==== Body
pmcThe problem of childhood obesity is substantial. For most of the 20th century, childhood obesity was rare, even in high-income countries.1 However, today obesity affects an estimated 50 million girls and 75 million boys 5–19 years of age around the world.1 The etiology of obesity is complex; it is not always as simple as eating more calories than you expend. Early life exposures to obesogenic synthetic chemicals, particularly persistent organic pollutants (POPs),2 can influence how our bodies metabolize food, acting through epigenetic mechanisms.3 Pregnant women are, therefore, an important population to target to reduce the obesogenic effects of POPs on future generations.4
The persistence of these chemicals means that even after regulations are enacted to restrict or prohibit their use, they can still be detected in the general population. For example, the insecticide dichlorodiphenyltrichloroethane (DDT) was banned in the United States in 1972,5 yet studies have shown that DDT’s primary metabolite, dichlorodiphenyldichloroethene (DDE), continues to be detected in nearly 100% of the biomonitoring subsample of the National Health and Nutrition Examination Survey.6 The pervasiveness of these chemicals also makes it remarkably challenging to reduce exposures through behavior change interventions.7,8
So how can we protect pregnant women and their offspring from potentially obesogenic synthetic chemicals? Nutrition interventions have been proposed as one potential solution, and there is precedent for such interventions in other fields. For example, the COVID-19 pandemic highlighted the role that healthy diets play in reducing susceptibility to viral infection.9 Analogously, healthy diets may reduce susceptibility to the adverse effects of obesogens.10 In addition, some research has shown that certain components of healthy diets, such as fiber,11,12 can bind to POPs and prevent their absorption. On the other hand, high-fat diets may enhance the absorption of POPs.13
In this issue of Environmental Health Perspectives, Cano-Sancho et al. aimed to determine if certain nutrients interact with POPs early in pregnancy and affect risk of obesity in children at 7 years of age.14 To do so, they presented a “comprehensive exploratory framework within a hypothesis-driven context” to screen for two-way interactions between 10 POPs and 17 nutrients. To my knowledge, this is the first study to assess the joint effects of POPs and nutrients on obesity.
After identifying two potential two-way interaction pairs—hexachlorobenzene (HCB) and vitamin B12, and perfluorooctane sulfonate (PFOS) and β-cryptoxanthin (an antioxidant)—the authors characterized each pair’s interactive relationships using multivariable regression. They found that the increased risk of childhood obesity associated with HCB exposure was greatest in women with the highest blood levels of vitamin B12; in other words, a synergistic effect. In contrast, the increased risk of childhood obesity associated with PFOS exposure was attenuated in women with the highest blood levels of β-cryptoxanthin, although the effect was smaller and less consistent.
The lack of studies on nutrient–pollutant interactions is surprising given that it was posited in Environmental Health Perspectives >50y ago that nutritional status plays an important role in susceptibility to toxicity.15 How can we make more substantive progress? To start, we need more training and research to be conducted by multidisciplinary teams that combine epidemiologists, toxicologists, nutritionists, data scientists, and clinicians, among others. Given that pregnant women’s chemical mixture exposures and their underlying nutritional status will vary widely across populations, it is important that these multidisciplinary teams conduct studies in a diversity of settings around the world.
One of the key limitations of the study by Cano-Sancho et al.14 is the small sample size, which, as they pointed out, increased their false negative discovery rate. Efforts to collate data across birth cohorts and conduct pooled analyses (i.e., a mega-analysis) could help address issues of small sample size. In Europe alone, a recent systematic review identified 111 birth cohorts.16 The Environmental Health Risks in European Birth Cohorts project, which launched in 2009, reviewed environmental exposure and health data available in 37 European birth cohorts.17 At the time of that review, 20 of the 37 cohorts were assessed for metals, 18 for pesticides, 19 for POPs, and 18 for other chemicals.17 More recently, the Environment and Child Health International Birth Cohort Group demonstrated the feasibility of harmonizing blood lead levels across five cohorts.18,19 Two studies published in 2016 combined results of four birth cohorts in the United States that evaluated biomarkers of organophosphorus pesticide exposure.20,21
Can eating a healthy diet during pregnancy attenuate the obesogenic effects of POPs? It is too soon to say conclusively one way or the other. In their study, Cano-Sancho et al.14 suggest “maybe antioxidants can.” However, much more research is needed, and we should proceed with caution in making clinical recommendations. This is particularly true as relates to supplements, given the observed increased risk of childhood obesity related to HCB exposure in women with the highest intakes of vitamin B12, which was related to supplement intake.14 Nonetheless, existing recommendations for pregnant and lactating women, such as the Dietary Guidelines for Americans, 2020–2025,22 already advise a diet rich in fresh fruits and vegetables (an important source of antioxidants), whole grains, low-fat or fat-free dairy products, and high-protein foods, including lean meats, chicken, eggs, seafood, beans, lentils, nuts, seeds, and tofu. Therefore, continuing to recommend they follow these guidelines may have co-benefits by reducing the consequences of obesogen exposure.
==== Refs
References
1. Di Cesare M, Sorić M, Bovet P, Miranda JJ, Bhutta Z, Stevens GA, et al. 2019. The epidemiological burden of obesity in childhood: a worldwide epidemic requiring urgent action. BMC Med 17 (1 ):212, PMID: , 10.1186/s12916-019-1449-8.31760948
2. Stratakis N, Rock S, La Merrill MA, Saez M, Robinson O, Fecht D, et al. 2022. Prenatal exposure to persistent organic pollutants and childhood obesity: a systematic review and meta-analysis of human studies. Obes Rev 23 (suppl 1 ):e13383, PMID: , 10.1111/obr.13383.34766696
3. Heindel JJ, Howard S, Agay-Shay K, Arrebola JP, Audouze K, Babin PJ, et al. 2022. Obesity II: establishing causal links between chemical exposures and obesity. Biochem Pharmacol 199 :115015, PMID: , 10.1016/j.bcp.2022.115015.35395240
4. Grandjean P, Abdennebi‐Najar L, Barouki R, Cranor CF, Etzel RA, Gee D, et al. 2019. Timescales of developmental toxicity impacting on research and needs for intervention. Basic Clin Pharmacol Toxicol 125 (suppl 3 ):70–80, PMID: , 10.1111/bcpt.13162.30387920
5. U.S. EPA (U.S. Environmental Protection Agency). 2022. DDT—A Brief History and Status. Last updated 21 A pril 2 022. https://www.epa.gov/ingredients-used-pesticide-products/ddt-brief-history-and-status [accessed 21 December 2022].
6. Li M, Wang R, Su C, Li J, Wu Z. 2022. Temporal trends of exposure to organochlorine pesticides in the United States: a population study from 2005 to 2016. Int J Environ Res Public Health 19 (7 ):3862, PMID: , 10.3390/ijerph19073862.35409545
7. Sathyanarayana S, Alcedo G, Saelens BE, Zhou C, Dills RL, Yu J, et al. 2013. Unexpected results in a randomized dietary trial to reduce phthalate and bisphenol A exposures. J Expo Sci Environ Epidemiol 23 (4 ):378–384, PMID: , 10.1038/jes.2013.9.23443238
8. Galloway TS, Baglin N, Lee BP, Kocur AL, Shepherd MH, Steele AM, et al. 2018. An engaged research study to assess the effect of a ‘real-world’ dietary intervention on urinary bisphenol A (BPA) levels in teenagers. BMJ Open 8 (2 ):e018742, PMID: , 10.1136/bmjopen-2017-018742.29431133
9. Merino J, Joshi AD, Nguyen LH, Leeming ER, Mazidi M, Drew DA, et al. 2021. Diet quality and risk and severity of COVID-19: a prospective cohort study. Gut 70 (11 ):2096–2104, PMID: , 10.1136/gutjnl-2021-325353.34489306
10. Sargis RM, Heindel JJ, Padmanabhan V. 2019. Interventions to address environmental metabolism-disrupting chemicals: changing the narrative to empower action to restore metabolic health. Front Endocrinol (Lausanne) 10 :33, PMID: , 10.3389/fendo.2019.00033.30778334
11. Zhan J, Liang Y, Liu D, Ma X, Li P, Zhai W, et al. 2019. Pectin reduces environmental pollutant-induced obesity in mice through regulating gut microbiota: a case study of p,p′-DDE. Environ Int 130 :104861, PMID: , 10.1016/j.envint.2019.05.055.31195221
12. Aozasa O, Ohta S, Nakao T, Miyata H, Nomura T. 2001. Enhancement in fecal excretion of dioxin isomer in mice by several dietary fibers. Chemosphere 45 (2 ):195–200, PMID: , 10.1016/S0045-6535(00)00557-9.11572611
13. Dulfer WJ, Groten JP, Govers HA. 1996. Effect of fatty acids and the aqueous diffusion barrier on the uptake and transport of polychlorinated biphenyls in Caco-2 cells. J Lipid Res 37 (5 ):950–961, PMID: , 10.1016/S0022-2275(20)42006-1.8725148
14. Cano-Sancho G, Warembourg C, Güil N, Stratakis N, Lertxundi A, Irizar A, et al. 2023. Nutritional modulation of associations between prenatal exposure to persistent organic pollutants and childhood obesity: a prospective cohort study. Environ Health Perspect 131 (3 ):037011, 10.1289/EHP11258.36927187
15. Mahaffey KR, Vanderveen JE. 1979. Nutrient-toxicant interactions: susceptible populations. Environ Health Perspect 29 :81–87, PMID: , 10.1289/ehp.792981.510246
16. Pansieri C, Pandolfini C, Clavenna A, Choonara I, Bonati M. 2020. An inventory of European birth cohorts. Int J Environ Res Public Health 17 (9 ):3071, PMID: , 10.3390/ijerph17093071.32354101
17. Gehring U, Casas M, Brunekreef B, Bergström A, Bonde JP, Botton J, et al. 2013. Environmental exposure assessment in European birth cohorts: results from the ENRIECO project. Environ Health 12 :8, PMID: , 10.1186/1476-069X-12-8.23343014
18. Etzel R, Charles MA, Dellarco M, Gajeski K, Jöckel K, Hirschfeld S, et al. 2014. Harmonizing biomarker measurements in longitudinal studies of children’s health and the environment. Biomonitoring 1 :97223266.
19. Nakayama SF, Espina C, Kamijima M, Magnus P, Charles MA, Zhang J, et al. 2019. Benefits of cooperation among large-scale cohort studies and human biomonitoring projects in environmental health research: an exercise in blood lead analysis of the Environment and Child Health International Birth Cohort Group. Int J Hyg Environ Health 222 (8 ):1059–1067, PMID: , 10.1016/j.ijheh.2019.07.005.31327570
20. Harley KG, Engel SM, Vedar MG, Eskenazi B, Whyatt RM, Lanphear BP, et al. 2016. Prenatal exposure to organophosphorous pesticides and fetal growth: pooled results from four longitudinal birth cohort studies. Environ Health Perspect 124 (7 ):1084–1092, PMID: , 10.1289/ehp.1409362.26685281
21. Engel SM, Bradman A, Wolff MS, Rauh VA, Harley KG, Yang JH, et al. 2016. Prenatal organophosphorus pesticide exposure and child neurodevelopment at 24 months: an analysis of four birth cohorts. Environ Health Perspect 124 (6 ):822–830, PMID: , 10.1289/ehp.1409474.26418669
22. U.S. Department of Agriculture. 2020. Dietary Guidelines for Americans, 2020–2025. Washington DC: U.S. Department of Agriculture. https://www.dietaryguidelines.gov/sites/default/files/2021-03/Dietary_Guidelines_for_Americans-2020-2025.pdf [accessed 8 March 2023].
|
PMC010xxxxxx/PMC10019508.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36927187
EHP11258
10.1289/EHP11258
Research
Nutritional Modulation of Associations between Prenatal Exposure to Persistent Organic Pollutants and Childhood Obesity: A Prospective Cohort Study
Cano-Sancho German 1
Warembourg Charline 2 3 4 5
Güil Nuria 2 3 4
Stratakis Nikos 2
Lertxundi Aitana 4 6 7
Irizar Amaia 4 6 7
Llop Sabrina 4 8
Lopez-Espinosa Maria-Jose 4 8 9
Basagaña Xavier 2 3 4
González Juan Ramon 2 3 4
Coumoul Xavier 10
Fernández-Barrés Sílvia 2 3 4
Antignac Jean-Philippe 1
Vrijheid Martine 2 3 4
https://orcid.org/0000-0002-2112-6740
Casas Maribel 2 3 4
1 Laboratory for the Study of Residues and Contaminants in Foods (LABERCA), Oniris, Institut national de la recherche agronomique (INRAE), Nantes, France
2 Barcelona Institute for Global Health (ISGlobal), Barcelona, Spain
3 Pompeu Fabra University, Barcelona, Spain
4 Spanish Consortium for Research on Epidemiology and Public Health (CIBERESP), Madrid, Spain
5 Institut de recherche en santé, environnement et travail (IRSET), Ecole des hautes études en santé publique (EHESP), Unité Mixte de Recherche (UMR) 1085 Institut national de la santé et de la recherche médicale (INSERM), Université de Rennes, Rennes, France
6 Biodonostia, Unidad de Epidemiologia Ambiental y Desarrollo Infantil, San Sebastian, Gipuzkoa, Spain
7 Facultad de Medicina, Universidad del País Vasco/Euskal Herriko Unibertsitatea, Leioa, Bizkaia, Spain
8 Epidemiology and Environmental Health Joint Research Unit, Foundation for the Promotion of Health and Biomedical Research in the Valencian Region, Foundation for the Promotion of Health and Biomedical Research in the Valencian Community (FISABIO)–Public Health, FISABIO–Universitat Jaume I–Universitat de València, Valencia, Valencia, Spain
9 Faculty of Nursing and Chiropody, University of Valencia, Valencia, Valencia, Spain
10 Institut national de la santé et de la recherche médicale (INSERM) UMR-S1124, Université de Paris, Paris, France
Address correspondence to Maribel Casas, Barcelona Institute for Global Health (ISGlobal)–Campus MAR Barcelona Biomedical Research Park (PRBB) Doctor Aiguader, 88 08003 Barcelona, Spain. Email: [email protected]
16 3 2023
3 2023
131 3 03701115 3 2022
10 2 2023
17 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Prenatal exposure to persistent organic pollutants (POPs) may contribute to the development of childhood obesity and metabolic disorders. However, little is known about whether the maternal nutritional status during pregnancy can modulate these associations.
Objectives:
The main objective was to characterize the joint associations and interactions between prenatal levels of POPs and nutrients on childhood obesity.
Methods:
We used data from to the Spanish INfancia y Medio Ambiente–Environment and Childhood (INMA) birth cohort, on POPs and nutritional biomarkers measured in maternal blood collected at the first trimester of pregnancy and child anthropometric measurements at 7 years of age. Six organochlorine compounds (OCs) [dichlorodiphenyldichloroethylene, hexachlorobenzene (HCB), β-hexachlorocyclohexane (β-HCH) and polychlorinated biphenyls 138, 153, 180] and four per- and polyfluoroalkyl substances (PFAS) were measured. Nutrients included vitamins (D, B12, and folate), polyunsaturated fatty acids (PUFAs), and dietary carotenoids. Two POPs–nutrients mixtures data sets were established: a) OCs, PFAS, vitamins, and carotenoids (n=660), and b) OCs, PUFAs, and vitamins (n=558). Joint associations of mixtures on obesity were characterized using Bayesian kernel machine regression (BKMR). Relative importance of biomarkers and two-way interactions were identified using gradient boosting machine, hierarchical group lasso regularization, and BKMR. Interactions were further characterized using multivariate regression models in the multiplicative and additive scale.
Results:
Forty percent of children had overweight or obesity. We observed a positive overall joint association of both POPs–nutrients mixtures on overweight/obesity risk, with HCB and vitamin B12 the biomarkers contributing the most. Recurrent interactions were found between HCB and vitamin B12 across screening models. Relative risk for a natural log increase of HCB was 1.31 (95% CI: 1.11, 1.54, pInteraction=0.02) in the tertile 2 of vitamin B12 and in the additive scale a relative excess risk due to interaction of 0.11 (95% CI: 0.02, 0.20) was found. Interaction between perfluorooctane sulfonate and β-cryptoxanthin suggested a protective effect of the antioxidant on overweight/obesity risk.
Conclusion:
These results support that maternal nutritional status may modulate the effect of prenatal exposure to POPs on childhood overweight/obesity. These findings may help to develop a biological hypothesis for future toxicological studies and to better interpret inconsistent findings in epidemiological studies. https://doi.org/10.1289/EHP11258
Supplemental Material is available online (https://doi.org/10.1289/EHP11258).
The authors declare that they have no conflict of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Growing evidence supports that environment insults and nutrition during the early stages of development may influence subsequent health during childhood, including obesity and metabolic diseases.1,2 Childhood obesity remains a public health priority owing to its high prevalence, associated risk of comorbidities, and high societal costs,3,4 and has been linked to a growing list of environmental factors including synthetic chemicals.5,6 Persistent organic pollutants (POPs) represent a vast family of chemicals characterized by their hydrophobicity, stability, and capacity to bioaccumulate across the trophic chains and are widespread in the fatty tissues of populations across the globe.7–9 Some POPs, such as the pesticide dichlorodiphenyltrichloroethane (p,p′-DDT), its main metabolite dichlorodiphenyldichloroethane (p,p′-DDE), or hexachlorobenzene (HCB), have been associated with obesity or metabolic disruption in human prospective studies and this has been supported by several experimental studies.10–14
Diet is a major pathway of exposure to many POPs,15 but it can also modulate the effect of toxicants by means of bioactive nutrients,16 thus becoming a source of heterogeneity or confounding bias in environmental epidemiology if not properly addressed.17 Bioactive lipids and lipophilic nutrients are specially relevant for POPs because their shared mechanisms of uptake, transport, and metabolism, and targeting similar molecular pathways, resulting in a large potential to confound, jointly affect, or interact in multiple health outcomes.18 This may be the case of carotenoids, lipophilic vitamins or ω3 poly unsaturated fatty acids (PUFAs) when assessing the obesogenicity of POPs, owing to their reported in vitro bioactivity on lipogenic and lipolytic pathways.19–21 The interactive effect of POPs with ω3-PUFAs has been observed in experimental studies, counterbalancing their effects on insulin resistance prevention.19–21
Previous studies within the Spanish longitudinal INfancia y Medio Ambiente–Environment and Childhood (INMA) birth cohort have reported positive associations between prenatal exposure to organochlorine compounds (OCs) with higher offspring obesity risk,5,14,22,23 but mild or null associations for per- and polyfluoroalkyl substances (PFAS).24 Although some studies have assessed the joint associations of prenatal mixtures of toxic metals and nutritional factors on child outcomes,25 to the best of our knowledge, no studies have attempted to assess the joint associations of POPs and nutrients on overweight/obesity to date. Thus, built on the hypothesis that the health effects of POPs may be modulated by bioactive lipids and lipophilic nutrients,18 this study extends the previous INMA work to characterize the joint effect of prenatal POPs and lipophilic micronutrients and/or PUFAs on childhood obesity risk. Considering the high-dimensional nature of interaction evaluations, the present study develops a comprehensive exploratory framework within a hypothesis-driven context, intended to answer major questions in epidemiological mixture analyses.26–29 More specifically, those questions included a) “What is the association of prenatal POPs and nutrients as a mixture on childhood overweight/obesity risk?”, and b) “Are there interactions between biomarkers of nutritional status in the associations between POPs and overweight/obesity risk?”
Methods
Study Population
Data from the Spanish INMA birth cohort study were used for the present analysis, extensively described elsewhere.30 Briefly, the INMA study is a network of population-based birth cohorts conducted in seven different regions from Spain, four of them (Asturias, Gipuzkoa, Sabadell, and Valencia) sharing the recruitment timeframe and protocol to ensure harmonization. For the present study, we considered three of those regions (Gipuzkoa, Sabadell, and Valencia) owing to the data availability, representing a total of 2,150 pregnant women recruited at the first trimester of pregnancy (weeks 10–13 of gestation) from 2003 to 2008. To be eligible, women must be at least 16 y old and present singleton pregnancy, no communication barrier, need no reproductive assistance, and be giving birth in the reference hospital.30 Follow-up visits were conducted when children were 1–1.5, 4, 7, and 11 y old, yet the study is still ongoing. When the children were 7 years of age, 1,394 mother–child pairs participated in the follow-up. The study was approved by the ethical committees of the centers involved in the study (listed in the “Acknowledgments” section). Written informed consent was obtained from the parents of all children. In the present analysis, we included mother–child pairs with information on blood biomarkers of POPs (OCs and PFAS) exposure and nutrient intake (vitamins, PUFAs, and carotenoids) during pregnancy and child anthropometric measurements at 7 years of age (Figure S1). A total of 1,241 mothers had information on OCs and vitamins and child obesity outcomes (Figure S1). However, because not all women had available information of all POPs and nutrients, we generated two consolidated data sets for the mixture analysis: the ANTIOX data set with OCs, PFAS, vitamins, and carotenoids (n=660 with 30 variables) and the PUFA data set consisting of OCs, vitamins, and PUFAs (n=558 with 14 variables), see details in Table 1. Among both data sets there was an overlap of 241 mother–child pairs. The list of compounds included within each data set is detailed in Table 1.
Table 1 Distribution [median (25th–75th percentile)] of persistent organic pollutants and nutrients for each data set measured in blood from mothers from INMA Guipuzkoa, Sabadell, and Valencia cohorts (Spain).
Maternal biomarkers (units) ANTIOX
N=660 PUFA
N=558
Organochlorine compounds (ng/g lipid)
HCB 55.2 (31.3–91.0) 39.0 (29.2–78.7)
β-HCH 23.6 (6.71–35.1) 30.5 (15.3–38.4)
p,p′-DDE 134.0 (85.5–234) 122.0 (79.7–202)
PCB138 33.5 (23.5–45.6) 25.5 (16.0–37.6)
PCB153 52.5 (37.3–71.6) 42.9 (29.0–61.8)
PCB180 39.2 (26.7–55.6) 30.2 (20.2–43.1)
Per- and polyfluoroalkyl substances (ng/mL)
PFOA 2.10 (1.47–2.91) —
PFOS 5.88 (4.49–7.66) —
PFHxS 0.483 (0.37–0.66) —
PFNA 0.582 (0.43–0.76) —
Vitamins
B12 (pmol/L) 397 (310–519) 341 (271–428)
D (mmol/L) 30.3 (22.9–37.6) 29.7 (21.2–37.5)
Folate (mmol/L) 17.9 (12.7–25.7) 14.2 (9.70–23.1)
Polyunsaturated fatty acids (% of fatty acids)
LA — 30.9 (25.2–34.2)
ALA — 0.3 (0.2–0.3)
EPA — 0.3 (0.2–0.5)
DHA — 2.9 (2.3–4.1)
AA — 8.0 (6.8–9.6)
Carotenoids (μmol/L)
γ-Tocopherol 1.49 (1.24–1.87) —
α-Tocopherol 30.9 (26.1–35.6) —
β-Cryptoxanthin 0.17 (0.111–0.25) —
α-Carotene 0.11 (0.07–0.18) —
β-Carotene 0.278 (0.17–0.43) —
Lutein 0.22 (0.17–0.28) —
Lycopene 0.408 (0.23–0.77) —
Zeaxanthin 0.06 (0.05–0.08) —
Retinol 1.95 (1.52–2.55) —
Note: The recruitment of mothers and blood sampling was conducted during the first trimester of pregnancy between 2003 and 2008. —, not applicable; AA, arachidonic acid; ALA, α-linolenic acid; β-HCH, β-hexachlorocyclohexane; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; Folate, Folic acid; HCB, hexachlorobenzene; INMA, INfancia y Medio Ambiente–Environment and Childhood (birth cohort); LA, linoleic acid; PCB138, 2,2′,3,4,4′,5′-hexachlorobiphenyl; PCB153, 2,2′,4,4′,5,5′-hexachlorobiphenyl; PCB180, 2,2′,3,4,4′,5,5′-heptachlorobiphenyl; PFHXS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; p,p′-DDE, dichlorodiphenyldichloroethylene.
Prenatal POPs Determination
Blood samples from mothers were collected at recruitment at the end of the first trimester of gestation (weeks 10–13 of gestation), aliquoted into 1.5mL cryotubes, and stored at −80°C until their analysis. Concentrations of HCB, β-hexachlorocyclohexane (β-HCH), p,p′-DDE, and polychlorinated biphenyl (PCB) congeners 138, 153, and 180 were determined in serum samples using gas chromatography (GC) equipped coupled to electron capture detector or mass spectrometer, as previously described.14,31–33 Concentrations of OCs were adjusted to total serum lipids calculated with the reduced equation using cholesterol and triglycerides determined enzymatically.34 Concentrations of perfluorooctanoate (PFOA), perfluorooctane sulfonate (PFOS), perfluorohexane sulfonate (PFHxS), and perfluorononanoate (PFNA) were determined in plasma samples by column-switching liquid chromatography coupled with tandem mass spectrometry at the Institute for Occupational Medicine, Rhine-Westphalia Technical University (RWTH) Aachen University (Aachen, Germany), as described previously.24,35 Limits of detection (LODs) ranged between 0.01 and 0.07 ng/mL for OCs and between 0.05 and 0.20 ng/mL for PFAS (Excel Table S1). The coefficients of variation for all PFAS and OC compounds were <15%.24,31 The selected congeners were detected in all samples included in the present study.
Nutritional Biomarkers Determination
Biomarkers of vitamins, PUFAs, and carotenoids were determined in maternal blood using validated analytical methods described elsewhere.36–38 Briefly, serum levels of vitamin B12 and folate were measured using a commercially available radioassay.38 Levels of 25-hydroxyvitamin D3 were measured in plasma by high-performance liquid chromatography (HPLC) using a BIO-RAD kit (Ref. 1956527) (BIO-RAD Laboratories GmbH) as measure of vitamin D status.37 Concentrations of long-chain PUFAs were determined in maternal plasma using fast-GC.36 The levels of carotenoids (α- and β-tocopherol, β-cryptoxanthin, α- and β-carotene, lutein, lycopene, zeaxanthin, and retinol) were measured in serum using HPLC with diode array detection and ultraviolet detection at 292 nm in the case of α-tocopherol.38 All samples included in the study showed levels above the LODs detailed elsewhere (Excel Table S1).36–38 The biomarkers showed coefficients of variation <10% in interassays and 5% in intra-assays.
Childhood Obesity Outcomes
Trained nurses measured the weight and height of the children at the follow-up visit at 7 years of age [mean±standard deviation (SD)=7.7±0.23; ANTIOX data set], using standard protocols. Age- and sex-specific body mass index (BMI) z-scores (zBMI) were calculated using the World Health Organization Growth Reference for children 5–19 years of age.39 Overweight and obesity were defined as the proportion of children with zBMI values of >1 SD and >2 SDs, respectively.39,40
Covariates
Information on sociodemographic (age, parity, education) and lifestyle characteristics (e.g., smoking, diet) of the mothers was collected by questionnaires administered to mothers during the first trimester of pregnancy. Measured maternal height and weight reported by the mother at the first trimester visit was used to calculate prepregnancy BMI (in kilograms per meter squared). Data regarding the maternal health status during pregnancy was collected directly from clinical records.
Confounding variables were selected on the basis of a priori knowledge supported by published literature on established determinants of maternal levels of POPs, nutritional status/diet, and childhood obesity risk40–43 and are depicted in a directed acyclic graph (Figure S2). The models for zBMI were thus adjusted for a minimal set of maternal variables at enrollment including prepregnancy BMI (in kilograms per meter squared), age of the mother (in years), education level (primary or without education, secondary, university), region of residence (Gipuzkoa, Sabadell, Valencia) or at delivery [e.g., smoking during pregnancy (nonsmoking, any smoking during pregnancy)]. The models for overweight/obesity were further adjusted for child’s sex assigned at birth (female, male) and detailed child age at outcome assessment (in months). Other potential confounders were also tested using the change in estimate method [i.e., change in β or relative risk (RR)≥10%], including parity (categorical: 0, 1, ≥2), maternal physical activity during the first trimester (metabolic equivalent of tasks), supplementation of vitamin B12 during the first trimester (low, medium, and high based on tertiles), adherence to the Mediterranean diet during the first trimester (high, medium, and low adherence; as described by Fernández-Barrés et al.41) energy-adjusted dietary intake of key food groups [fish, meat and fruits (in grams per day)] or gestational diabetes risk (diabetes diagnosed previous to pregnancy, gestational diabetes, and none/low risk). In the end, none of those variables were finally retained because of the low impact on our model estimates (i.e., <5% modification of coefficients) and because some of these variables can also be determinants of exposures (e.g., dietary intake of fish) or mediators (e.g., gestational diabetes) instead of confounders.
Data Analysis
The multistep workflow for data analysis is illustrated in Figure S3, covering major questions about the effect of mixtures: Step 1, an exploratory analysis to identify the correlations between POPs and nutrients; Step 2, a preliminary characterization of associations between individual biomarkers and obesity outcomes without accounting for the rest of biomarkers; Step 3, a ranking of biomarker importance accounting for the co-exposure in multipollutant models; Step 4, an estimation of the joint associations between chemical mixtures and obesity outcomes; Step 5, a screening of two-way interactions to select suspected pairs; and Step 6, a refined characterization of those interactions using conventional regression methods to facilitate the interpretation in terms of risk estimation. To this end, we applied a battery of complementary algorithms developed to integrate multiple correlated exposure variables to identify joint effects and interactions.26,29
Data preprocessing and unsupervised exploratory analysis.
The distribution profiles of POPs and nutrients were explored to characterize the skewness and to identify extreme values, left-censored data, and missing data. Biomarkers with detection frequencies <75% (p,p′-DDT and PCB118) were removed from the analysis. Participants with nonmeasured exposure data were not considered in the statistical analysis (n=153–836; Figure S1). A multiple multivariate imputation procedure was applied to covariate variables with some missing data following the workflow described elsewhere for data missing at random.44 Batches of 15 variables were considered in the multivariable imputation models using the R package “mice,” developed specifically for each data set. Preprocessing of continuous data included natural log-transformation and scaling to the standard deviation (SD) to improve model fit, if necessary. Exploratory analysis included Spearman’s rank correlation analysis to identify correlation patterns between pairs of biomarkers and to support the biomarker grouping in the Bayesian kernel machine regression (BKMR) (see the section “Biomarker importance from multipollutant models” below) and the interpretation of results.
Single-biomarker outcome associations.
Multivariate linear and Poisson regression (MLR) with robust variance were used to characterize the associations between individual biomarkers (POPs and nutrients) and continuous (zBMI) or binary outcomes (normal weight vs. overweight and obese combined), respectively. Confounding variables were included in the model as covariates, allowing a flexible estimation of marginal effects. Modified Poisson regression with robust variance was computed using the sandwich approach, which is considered to provide unbiased estimates of risk ratios under potential model misspecification.45 The method was implemented in R software using the coeftest function with “sandwich” and “lmtest” packages.
Biomarker importance from multipollutant models.
Three different statistical methods to examine multipollutant associations were selected based on their complementarity to characterize linear and nonlinear associations, with the capacity to manage confounding data and identify potential interactions. These models also allow the characterization of associations for individual exposures and the identification of interactions (see the section “Joint effects of mixtures”), thus providing a measure of their relative variable importance within each model to rank them while accounting for the rest of biomarkers. A summary of the main features from each model supporting the complementarity is displayed in Excel Table S2. Group Lasso Interaction Network (Glinternet) is a flexible regularization algorithm designed to identify pairwise interactions in regression models imposing the group lasso (L1) penalties with strong hierarchy.46 Thus, if an interaction coefficient is estimated to be nonzero, then its two associated main associations also have nonzero estimated coefficients, controlled by the parameter λ. Optimal λ can be selected by cross-validation to identify an adequate bias-variance trade-off. In the present study, 10-fold cross-validation was used to identify the λ exhibiting the lowest mean squared error, computed with the R package “glinternet.” To increase the robustness of findings, the models were fitted to 100 bootstrap samples, as described elsewhere.47 This feature allowed the measurement of coefficient variability and the frequency of interaction detection across bootstrap samples. Averaged model coefficients were used as variable importance scores, considering all of them were at the same scale. The approach is intuitive, may handle a large number of independent variables (e.g., up to 105) and their interactions, and is computationally efficient and the results are straightforward to interpret. The method may be limited to characterizing nonlinear associations, and the built-in package does not allow forcing confounding variables out of model penalties.
Gradient boosting machine (GBM) is one of the first approaches proposed to evaluate the joint associations of environmental exposures and their interactions. The use of boosting machines combines a large number of simple regression tree models throughout an iterative process of simpler models’ combination (boosting) to improve the overall model fit. We implemented the method using the R packages “gbm” and “dismo.”48 The gbm.step function was computed using the input parameters setup 0.001 for the learning rate and 4 for the tree complexity, the 0.8 bag fraction at 0.8, and a 10‐fold cross‐validation. The learning rate allows the minimization of the loss function; the tree complexity (interaction depth) controls the number of nodes in the trees, hence the potential interactions; and the bag fraction defines the subsample of data to propose the next step in the tree. To obtain robust estimates, we replicated the model 100 times and extracted the most frequently detected variables (>50%) and their interactions across the replicates. The relative contribution of biomarkers to the overall model fit were used as a variable importance metric. Strengths of GBM includes the capacity to detect nonlinear associations and high-order interactions at a moderate computational cost; however, the interpretability is often challenging, requiring a second modeling step.
BKMR. The BKMR framework is a flexible nonparametric approach that allows the estimation of the overall effect estimate of multiple correlated exposures accounting for confounding variables.49 The method was implemented with the R package “bkmr” using 10,000 iterations.50 All variables were included in the model using the variable selection mode, which allows the computation of posterior inclusion probabilities (PIPs) to support the selection of the most relevant variables and to rank the variables according to their probability to be included in the model as an approximate measure of variable importance. Important assets of BKMR compared with the previous methods include the unique capacity to measure the joint effect of the mixture and that the model structure specifically accounts for confounding variables. Although the method can efficiently account for complex interactions, the identification process involves graphical inspection that can become sometimes conflicting.
Joint effects of mixtures.
The joint effect of the mixture composed by those POPs and nutrients with higher PIPs was characterized using BKMR with hierarchical variable selection, as described above and using the OverallRiskSummaries function from the “bkmr” R package. The summary estimate displayed the overall effect of biomarkers as the comparison of the predicted outcome when all biomarkers are fixed at given percentiles with the predicted outcome and all biomarkers are fixed at the 10th percentile.
Screening of two-way interactions.
In this step, the two-way interactions included in the multipollutant models were identified. With Glinternet, two-way interactions were identified through the inherent strong hierarchical fitting process, together with those main effects likely to be nonzero. In the case of GBM, as other tree-based models, interactions between predictors are inherently included, given that some of the regression trees used are likely to be asymmetric (thus inducing interactions between variables), where the response of one variable depends on the others higher in the tree. The presence of interactions was assessed using the gbm.interactions routine from the “dismo” package. Briefly, the interactions were identified if departures of model predictions for linear combinations of pair of variables was elucidated.48 For BKMR, two-way interactions were graphically explored using the cross-section plots depicting the exposure–response function for a given exposure when another exposure was fixed at the 25th, 50th, or 75th percentile, fixing the rest of exposures to the median.
To develop a priority list of interactions for a detailed regression analysis, we first selected those chemicals comprising a matching pair of POP and nutrient. In addition, for each model we ranked the pairs based on each specific metric of variable importance as a measure of interaction strength: We used the model contribution for GBM, the standardized model coefficients for Glinternet, and the visual trends of graphical cross-sections for BKMR. For GBM and Glinternet, including bootstrapping, a frequency detection threshold was defined based on the number of interactions identified (e.g., 50%, a large number of interactions; 25%, a low number of interactions). To establish a rank of most influential interactions across three models, we considered the top ranked pairs in at least one method (stronger interactions) or those with lower ranking positions (e.g., weaker interactions) but identified in at least two approaches or obesity outcomes (i.e., continuous or binary).
Characterization of interactions for selected pairs.
In the latter step, we aimed to characterize the joint effect of pairs biomarkers (i.e., POPs and nutrients) identified in the previous screening phase to facilitate interpretation. We first built the generalized additive models (GAMs) including the interaction product term for the selected pairs of variables, considering the POPs in continuous scale and nutrients in categorical scale (tertiles), adjusted for the abovementioned covariates. Interaction plots were inspected to characterize the shape of the exposure–response functions and refine the regression models. Unlike distributions of nutrients broken into tertiles for a better interpretation of interactions, POPs were categorized in quartiles if their departure from linearity was graphically visualized. Risk estimates of POPs were then evaluated across the different tertiles of nutrients using linear or robust Poisson regression models for zBMI or overweight/obesity risk, respectively, as detailed above. To formally evaluate the departures from additive joint effects, we also estimated the relative excess risk of overweight/obesity due to interaction (RERI) with the regression models’ coefficients from Poisson regression.28,51 The additive interaction is present when the RERI [95% confidence interval (CI)] is >0 if positive or <0 if negative. All analysis were conducted with R software (version 4.0.2; R Development Core Team).
Results
Unsupervised Exploratory Analysis
The prevalence of overweight, including obesity, was 43% and 41% in the ANTIOX and PUFA data sets, respectively (Table 2). The respective mean age of children ranged from 7.7 to 7.3 y, and there was an equal proportion of girls and boys in both data sets (Table 2). Mothers had a mean age of 30 y at recruitment with a presence of ∼30% of smokers at some time during pregnancy. Levels of POPs and biomarkers of nutritional status were globally similar between data sets (Table 1) or when compared with those reported in the entire cohort before subsetting (Excel Table S1), yet median concentrations of organochlorine pesticides (OCPs) were slightly higher in the ANTIOX data set compared with the other data sets. The correlation analysis showed high positive correlations (ρ>0.5) between POPs within families (Figure 1A,B; numeric values in Excel Table S3 and S4), and mild or no correlation between families. In turn, p,p′-DDE was not correlated with PFAS and moderately correlated (ρ= 0.3–0.2) with PCBs, HCB, and β-HCH. Regarding the fatty acids, arachidonic acid (AA) and docosahexaenoic acid (DHA), both showed positive correlations with PCBs and negative correlation with β-HCH (ρ= 0.3 and −0.1, respectively). Similar but weaker association profiles were shown for eicosapentaenoic acid (Figure 1B, Excel Table S4), and linoleic acid (LA) was negatively correlated with PCBs and the rest of the PUFAs. Most OCs and PFAS congeners were not correlated with dietary carotenoids or vitamins, with few exceptions. For instance PFHxS showed mild positive correlations with most nutrients (ρ= 0.1–0.2), whereas p,p′-DDE was negatively associated with retinol (ρ= −0.3) (Figure 1A, Excel Table S3).
Table 2 Characteristics [mean±SD or n (%)] of mothers and children from INMA Guipuzkoa, Sabadell, and Valencia cohorts (Spain) included in each data set.
Characteristics ANTIOX
N=660 PUFA
N=558
Maternal characteristics
Age at recruitment (y) 31.0±3.8 30.8±3.9
Prepregnancy maternal BMI (kg/m2) 23.4±4.0 23.7±4.2
Parity
0 370 (56) 311 (56)
1 250 (38) 214 (38)
≥2 40 (6) 32 (6)
Missing 0 1
Education
Primary or without education 143 (22) 121 (22)
Secondary 251 (38) 228 (41)
University 265 (40) 209 (37)
Missing 1 0
Region of residence
Gipuzkoa 272 (41) 141 (25)
Sabadell — 150 (27)
Valencia 388 (59) 267 (48)
Adherence to Mediterranean diet
Low adherence 282 (43) 219 (39)
Medium adherence 169 (26) 169 (30)
High adherence 204 (31) 167 (30)
Missing 5 3
Smoking during pregnancy
No 445 (67) 393 (70)
Yes 206 (31) 157 (28)
Missing 9 8
Child characteristics
Sex
Female 325 (49) 277 (50)
Male 335 (51) 281 (50)
Child age (y) 7.7±0.2 7.3±0.6
Child BMI WHO class at 7 y of age
Thinness (<−SD2; <3rd percentile) 4 (1) 3 (1)
Normal (–SD2 to SD1; 3rd–85th percentile) 371 (56) 325 (58)
Overweight (SD1 to SD2; 85th–97th percentile) 160 (24) 135 (24)
Obesity (>SD2; ≥97th percentile) 125 (19) 95 (17)
Note: The recruitment of mothers was conducted during the first trimester of pregnancy between 2003 and 2008. —, not applicable; BMI, body mass index; INMA, INfancia y Medio Ambiente–Environment and Childhood (birth cohort); SD, standard deviation; WHO, World Health Organization.
Figure 1. Spearman’s correlation plots depicting the association strength between POPs and nutrients in (A) the ANTIOX data set (n=660) and (B) the PUFA data set (n=558). Details of chemical abbreviations are provided in Table 1 and numerical values in Excel Tables S3 and S4. Note: AA, arachidonic acid; ALA, α-linolenic acid; b_HCH, β-hexachlorocyclohexane; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; HCB, hexachlorobenzene; LA, linoleic acid; PCB138, 2,2′,3,4,4′,5′-hexachlorobiphenyl; PCB153, 2,2′,4,4′,5,5′-hexachlorobiphenyl; PCB180, 2,2′,3,4,4′,5,5′-heptachlorobiphenyl; PFHXS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; POPs, persistent organic pollutants; p,p′-DDE, dichlorodiphenyldichloroethylene; VIT_B12, vitamin B12; VIT_D, vitamin D.
Figure 1A is a Spearman’s correlation plot, plotting V I T underscore D, V I T underscore B 12, P F H X S, P F O S, P F N A, P F O A, b underscore H C H, H C B, p p D D E, P C B 153, P C B 138, P C B 180, Zeaxanthin, Lutein, b underscore cryptoxanthin, b underscore carotene, a underscore carotene, Lycopene, F O LAT E, a underscore tocopherol, and g underscore tocopherol (y-axis) across association strength, ranging from negative 1 to 1 in increments of 0.2 (x-axis) for V I T underscore B 12, P F H X S, P F O S, P F N A, P F O A, b underscore H C H, H C B, p p D D E, P C B 153, P C B 138, P C B 180, Zeaxanthin, Lutein, b underscore cryptoxanthin, b underscore carotene, a underscore carotene, Lycopene, F O LAT E, a underscore tocopherol, g underscore tocopherol, and Retinol. Figure 1B is a Spearman’s correlation plot, plotting P C B 153, P C B 138, P C B 180, V I T underscore B 12, V I T underscore D, D H A, A A, E P A, F O L A T E, b underscore H C H, H C B, p p D D E, and A L A (y-axis) across association strength, ranging from negative 1 to 1 in increments of 0.2 (x-axis) for P C B 138, P C B 180, V I T underscore B 12, V I T underscore D, D H A, A A, E P A, F O L A T E, b underscore H C H, H C B, p p D D E, A L A, and L A.
Single-Biomarkers Outcome Associations
The contributions and associations of individual biomarkers with overweight/obesity risk are summarized in Figure 2 (ANTIOX data set) and Figure S4 (PUFA data set), whereas the estimates for zBMI scores can be found in Figures 2 (ANTIOX data set) and 4 (PUFA data set), and numeric results are reported in Excel Table S5. Single-pollutant models showed consistent statistically significant associations of prenatal exposure to HCB and β-HCH with obesity outcomes in children [e.g., adjusted RR=1.18 (95% CI: 1.07, 1.31) per 1-SD increase in the natural log of HCB (ANTIOX dataset); Figure 3]. PFNA was positively associated with overweight/obesity risk [e.g., RR= 1.10 (95% CI: 1.01, 1.20) per 1-SD increase in in the natural log of PFNA, p=0.03; Figure 3] but not with zBMI (Figures S3 and S5). Null associations were found for the rest of POPs, nonetheless PCB153, PCB180, and PFOA were positively associated with overweight/obesity risk but at limit of conventional statistical significance (p=0.06–0.10; Excel Table S5). Estimates for most nutrients were also null, with the exception of β-cryptoxanthin, zeaxanthin, or α-tocopherol, which were positively associated with zBMI, whereas DHA showed a negative association.
Figure 2. Associations between persistent organic pollutants and nutrients with childhood overweight/obesity within the ANTIOX data set (n=660). (A) Forest plot depicting the associations between individual prenatal exposures (log scaled) and risk of childhood overweight/obesity. Summary estimates from single-biomarker models based on multivariate robust Poisson regression are depicted by adjusted relative risks (RRs) and respective 95% confidence intervals (CIs). Variable importance plots depict the rank of variables based on their relative importance in multipollutant models using the (B) absolute coefficients for Glinternet, (C) model contribution for gradient boosting machine (GBM) regression, and (D) posterior inclusion probabilities (PIPs) for Bayesian kernel machine regression (BKMR). All models were adjusted for maternal age, prepregnancy body mass index, smoking during pregnancy, region of residence, education, child sex, and age. Details of chemical abbreviations are provided in Table 1 and numerical values in Excel Tables S5 and S8. Note: b_HCH, β-hexachlorocyclohexane; HCB, hexachlorobenzene; PCB138, 2,2′,3,4,4′,5′-hexachlorobiphenyl; PCB153, 2,2′,4,4′,5,5′-hexachlorobiphenyl; PCB180, 2,2′,3,4,4′,5,5′-heptachlorobiphenyl; PFHXS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; p,p′-DDE, dichlorodiphenyldichloroethylene; VIT_B12, vitamin B12; VIT_D, vitamin D.
Figures 2A to 2D are forest plots. Figure 2A is titled Individual regression models, plotting Lycopene, b underscore carotene, retinol, a underscore carotene, lutein, P F H X S, P F O S, V I T underscore D, F O L A T E, P C B 138, V I T underscore B 12, g underscore tocopherol, p p D D E, zeaxanthin, b underscore cryptoxanthin, P C B 180, P F O A, P C B 153, P F N A, a underscore tocopherol, b underscore H C H, and H C B (y-axis) across Relative risk (95 percent confidence interval), ranging from 0.9 to 1.5 in increments of 0.2 (x-axis). Figures 2B, 2C, and 2D are titled Glinternet, G B M, and B K M R under Multipollutant models, plotting Zeaxanthin, b underscore carotene, P C B 153, a underscore carotene, Lutein, P C B 180, F O L A T E, P C B 138, lycopene, P P H X S, g underscore tocopherol, retinol, p p D D E, V I T underscore D, V I T underscore B 12, P F O S, P F O A, b underscore cryptoxanthin, P F N A, b underscore H C H, a underscore tocopherol, and H C B; P C B 180, b underscore carotene, zeaxanthin, retinol, P C B 153, lycopene, a underscore carotene, b underscore cryptoxanthin, V I T underscore D, lutein, g underscore tocopherol, F O L A T E, P F H X S, a underscore tocopherol, P F O S, p p D D E, P C B 138, P F O A, P F N A, b underscore H C H, H C B, and V I T underscore B 12; and P C B 138, p p D D E, P C B 180, P C B 153, zeaxanthin, a underscore carotene, retinol, b underscore carotene, g underscore tocopherol, P F O A, P F H X S, lycopene, lutein, V I T underscore D, b underscore cryptoxanthin, b underscore H C H, P F N A, a underscore tocopherol, F O L A T E, V I T underscore B 12, P F O S, and H C B (y-axis) across Absolute coefficients, ranging from 0.0 to 0.3 in increments of 0.1; Contribution, ranging from 0 to 9 in increments of 3; and P I P, ranging from 0.00 to 1.00 in increments of 0.25 (x-axis), respectively.
Figure 3. Overall effect estimates from Bayesian kernel machine regression (BKMR) on the association between mixtures of POPs and nutrients and (A) childhood overweight/obesity risk and (B) body mass index z-score (zBMI) for the mixtures of chemicals selected with the hierarchical procedure from ANTIOX data set (gray) and the PUFA data set (black). Details of most relevant chemicals in the mixtures are depicted in Figures S10–S13. Graphs show the difference in the effect estimates when all exposures are at a particular quantile compared with when all are at the 10th quantile as reference. All models were adjusted for maternal age, prepregnancy body mass index, smoking during pregnancy, and region of residence and education; child sex and age were also included in overweight/obesity models. Numerical values can be found in Excel Table S9.
Figures 3A and 3B are error bar graphs titled overweight or obesity and zBMI, plotting lowercase h (lowercase z), ranging from 0.0 to 1.5 in increments of 0.5 (y-axis) across effect exposures, ranging from 0.25 to 0.75 in increments of 0.25 (x-axis), respectively, for P U F A and A N T I O X.
Biomarker Importance from Multipollutant Models
Variable selection methods (Glinternet, GBM, and BKMR) consistently indicated that HCB and vitamin B12 had a high relative importance in the multipollutant models of overweight/obesity risk (Figures 2 and S5) and zBMI (Figures S4 and S6), numeric values are displayed in Excel Tables S6–S11. In turn, β-HCH, PFNA, PFOA, and PFOS also scored high, but the ordering was less consistent across data sets, models, and obesity outcomes. Similarly, the ordering of PUFAs in the importance ranking was less consistent between obesity outcomes. For instance, AA was identified among the most contributing variables on overweight/obesity risk (Figure S5), whereas DHA, LA, and α-linolenic acid (ALA) appeared among the most important contributors on zBMI models (Figure S6). Among the carotenoids, β-cryptoxanthin was the most important biomarker across models and outcomes, followed by α- and γ-tocopherol. In turn, folate showed one of the most inconsistent behaviors across data sets, models, and obesity outcomes, being among the top five PIPs from BKMR in one overweight/obesity model (ANTIOX data set; Figure 2) but scoring low in the rest of the models.
Joint Associations of Mixtures
The profile of most relevant chemicals in the mixtures is depicted by the conditional PIPs discussed in the section “Biomarker Importance from Multipollutant Models.” The hierarchical variable selection approach, built on groups of biomarker nature (i.e., OCs, PFAS, vitamin, PUFA, and antioxidant), resulted in a balanced composition of top ranked biomarkers with representatives from each family. The overall joint association of the mixture was positive, mostly linear, and significant for overweight/obesity risk (Figure 3A, Excel Table S9) and zBMI (Figure 3B, Excel Table S9). Details of most relevant biomarkers in the mixtures can be found in Figures S11–S14.
Screening of Two-Way Interactions
The screening of interactions relied in three computational methods (Glinternet, GBM, and BKMR) with complementary features to fit the data and identify interactions. Therefore, and to increase the robustness of findings, we selected interactions with the largest strength present in at least one single model or weaker strength but selected by multiple models. Results from each model and outcome are summarized in tables and interaction network plots depicting the bivariate interactions in the ANTIOX data set (Excel Tables S10–S12 and Figures S7 and S8) and the PUFA data set (Excel Tables S10–S12 and Figures S9 and S10). Specific interactions outputs from the GBM model (bootstrap interaction contributions) and Glinternet model (bootstrap interaction coefficients) can be found in Figures S15 and S16 for overweight/obesity risk and in Figures S17 and S18 for zBMI, with the related numerical values in Excel Tables S10 and S11. The frequency of interactions and the variability of strength attributed to each interaction across bootstrap samples is depicted with the raw shaded points and summarized with the means depicted with the white dot.
Overall, we observed larger uncertainty for interaction coefficients in Glinternet models than model contributions from the GBM models. Detection frequency of interactions across bootstrap samples was higher among the findings from PUFA data set (threshold setup 50%) than ANTIOX data set (threshold setup 25%). The detailed graphical output of the bivariate cross-section from BKMR can be found in Figures S11–S14. For the present study, we focused on interactions between POPs and nutrients; however, lists of POP–POP and nutrient–nutrient interactions were also automatically identified from the GBM and Glinternet models and are represented in the interaction network plots (Figures S7–S10). In general, interactions between POPs and nutrients were weak and inconsistent across models, with the presence of more, but weaker, interactions within the ANTIOX data set, whereas in the PUFA data set, we found fewer, but stronger, interactions.
To conduct a more refined analysis, 10 interactions were discerned based on their largest strength/contribution in models or because they were frequently detected across models (highlighted in bold in Excel Table S12), as detailed in the prioritization criteria. For instance, the interaction between HCB and vitamin B12 showed the largest contribution in the GBM models for overweight/obesity risk but that was also graphically suggested in the BKMR model (Figure S11). Similarly, but to lesser extent, the interaction between β-HCH and vitamin B12 was found in the ANTIOX data set (Excel Table S12). The pesticide β-HCH appeared to interact with folate, recurrently detected by GBM in the ANTIOX data set and by Glinternet in the PUFA data set (Excel Table S12). In turn, β-HCH also showed frequently detected interactions with ALA and LA (Excel Table S12). The interaction between PCB138 and LA also exhibited the highest coefficient and largest detection rates in the Glinternet model for overweight/obesity risk. The interactions between PFOA and vitamin B12, and between PFOS and γ-tocopherol, β-cryptoxanthin, and retinol, were also considered of priority interest based on detection frequency and/or strength (Excel Table S12).
Characterization of Interactions for Selected Pairs of POPs and Nutrients
For the selected 10 interactions, we further conducted a regression analysis using GAM models supported by graphical summaries (interaction plots) to identify nonlinearity and the direction of interactions. This visualization allowed, for instance, the identification of the potential synergic effect of HCB and vitamin B12 (Figure 4A) and the potential protective effect of β-cryptoxanthin on PFOS (Figure 4B, with further details in Excel Table S13). To further characterize the impact of the identified interactions between OCs and vitamins, we used the data set with complete data on OCs and vitamins (n=1,241), as depicted in Figure S1.
Figure 4. Interaction plots on the associations between (A) hexachlorobenzene (HCB, log increase) and tertiles of vitamin B12 and between (B) perfluorooctane sulfonate (PFOS, log increase) and tertiles of β-cryptoxanthin on overweight/obesity risk. Tertile 1 (T1) is depicted by a solid line; tertile 2 (T2) by a long-dashed line, and tertile 3 (T3), by a dotted line. All models were adjusted for maternal age, prepregnancy body mass index, smoking during pregnancy, region of residence, education, and child sex and age. Coefficient details can be found in Excel Table S13.
Figures 4A and 4B are interaction plots, plotting overweight or obesity, ranging from negative 3.0 to 0.5 in increments of 0.5 and negative 2.5 to 0.5 in increments of 0.5 (y-axis) across H C B, ranging from 1 to 7 in unit increments and P F O S, ranging from 0.0 to 3.0 in increments of 0.5 (x-axis) for vitamin B 12 T 1, vitamin B 12 T 2, vitamin B 12 T 3; and b underscore cryptoxanthin T 1, b underscore cryptoxanthin T 2, b underscore cryptoxanthin T 3, respectively.
The detailed regression analysis confirmed those effects: The associations between HCB and childhood overweight/obesity risk and zBMI strengthened at higher levels of vitamin B12 (Table 3). For instance, the associations between HCB and overweight/obesity risk at tertile 2 of vitamin B12 showed an RR=1.31 (95% CI: 1.11, 1.54), whereas at tertile 1, the associations were close to the null [RR=0.99 (95% CI: 0.85, 1.14, pInt=0.02)]. Those trends were also observed in the additive scale, with corresponding RERIs=0.11 (95% CI: 0.02, 0.20) for tertile 2 and 0.12 (95% CI: 0.03, 0.21) for tertile 3. A similar, but weaker, trend was also observed for the interaction between β-HCH and vitamin B12 (Excel Table S14) and between β-HCH and folate (Table 3). In the latter case, despite being statistically not significant at the multiplicative scale, a synergism was suggested in the additive scale between β-HCH and tertile 2 folate with an RERI=0.11 (95% CI: 0.01, 0.21).
Table 3 Summary estimates for the associations between selected POPs across tertiles of nutrients with obesity outcomes from Poisson/lineal regression models with a cross-product interaction term.
POP Nutrient Tertile Overweight/obesity zBMI
RR 95% CI pInt RERI 95% CI β 95% CI pInt
HCB (log)a Vitamin B12 (pmol/L) <302 0.99 (0.85, 1.14) 0.022 Ref −0.07 (−0.21, 0.07) 0.002
302–339 1.31 (1.11, 1.54) 0.11 (0.02, 0.20) 0.26 (0.12, 0.40)
>339 1.21 (1.06, 1.37) 0.12 (0.03, 0.21) 0.16 (0.05, 0.28)
β-HCH (log)a Folate (mmol/dL) <11.4 1.00 (0.88, 1.15) 0.260 Ref 0.00 (−0.12, 0.12) 0.110
11.4–19.0 1.16 (1.02, 1.32) 0.11 (0.01, 0.21) 0.16 (0.06, 0.27)
>19.0 1.13 (1.02, 1.25) 0.09 (−0.02, 0.20) 0.12 (0.03, 0.21)
PFOSb
Q2 Retinol (mmol/L) <1.7 0.90 (0.59, 1.38) 0.069 Ref −0.34 (−0.80, 0.13) 0.076
Q3 0.77 (0.48, 1.22) Ref −0.26 (−0.73, 0.20)
Q4 0.97 (0.63, 1.49) Ref −0.13 (−0.59, 0.33)
PFOS
Q2 Retinol (mmol/L) 1.7–2.3 1.20 (0.83, 1.74) 0.12 (−0.23, 0.81) 0.37 (−0.05, 0.78)
Q3 1.09 (0.74, 1.61) 0.32 (−0.19, 0.84) 0.26 (−0.17, 0.69)
Q4 1.16 (0.79, 1.69) 0.18 (−0.36, 0.72) 0.23 (−0.19, 0.64)
PFOS
Q2 Retinol (mmol/L) >2.3 0.62 (0.35, 1.10) 0.12 (−0.80, 0.38) −0.30 (−0.72, 0.12)
Q3 1.42 (0.96, 2.11) 0.58 (0.11, 1.05) 0.35 (−0.06, 0.76)
Q4 1.10 (0.69, 1.74) 0.11 (−0.45, 0.67) 0.07 (−0.35, 0.49)
PFOSb
Q2 β-cryptoxanthin (mmol/L) <0.1 1.32 (0.82, 2.11) 0.244 Ref 0.11 (−0.30, 0.52) 0.367
Q3 1.34 (0.85, 2.13) Ref 0.38 (−0.07, 0.82)
Q4 1.59 (1.02, 2.50) Ref 0.26 (−0.17, 0.69)
PFOS
Q2 β-cryptoxanthin (μmol/L) 0.1–0.2 0.79 (0.49, 1.28) 0.12 (−1.48, 0.30) −0.02 (−0.45, 0.41)
Q3 1.05 (0.70, 1.58) −0.28 (−1.09, 0.54) 0.12 (−0.31, 0.55)
Q4 1.12 (0.74, 1.69) −0.44 (−1.32, 0.45) 0.28 (−0.14, 0.70)
PFOS
Q2 β-cryptoxanthin (mmol/L) >0.2 0.75 (0.50, 1.11) 0.12 (−1.75, 0.20) −0.32 (−0.77, 0.12)
Q3 0.86 (0.60, 1.23) −0.59 (−1.50, 0.32) −0.19 (−0.61, 0.24)
Q4 0.71 (0.47, 1.07) −1.12 (−2.19, −0.05) −0.35 (−0.79, 0.09)
PCB138 (log)c Linoleic acid (percentage fatty acids) <27.6 0.96 (0.75, 1.24) 0.297 Ref −0.09 (−0.35, 0.16) 0.372
27.6–33.4 1.10 (0.86, 1.41) 0.10 (−0.13, 0.33) 0.03 (−0.21, 0.27)
>33.4 1.25 (0.99, 1.58) 0.18 (0.02, 0.33) 0.14 (−0.08, 0.36)
Note: Additive interactions on overweight/obesity risk are depicted by the RERI and respective 95% 95% CI. All models were adjusted for maternal age, prepregnancy body mass index, smoking during pregnancy, education and region of residence, in addition models on overweight/obesity risk were further adjusted for child sex and age. Detailed levels of POP quartiles are displayed in Excel Table S15. Details of chemical abbreviations are provided in Table 1. CI, confidence interval; HCB, hexachlorobenzene; pInt, p-Value from interaction testing PCB138, 2,2′,3,4,4′,5′-hexachlorobiphenyl; PFOS, perfluorooctane sulfonate; POP, persistent organic pollutant; Q, quartile; Ref, reference; RERI, relative excess risk due to interaction; RR, adjusted relative risks; zBMI, child body mass index z-score; β-HCH, β-hexachlorocyclohexane.
a Population sample size n=1,241 (see details in the study flowchart in Figure S1).
b Population sample size n=660 (ANTIOX data set).
c Population sample size n=558 (PUFA data set).
Synergistic interactions were also noticed between PFOS and retinol on the associations with overweight/risk and zBMI, yet at the limit of statistical significance (pInt=0.069) at the multiplicative scale (Table 3). For example, at the highest concentration of retinol, the RR of overweight/obesity of PFOS was 1.42 (95% CI: 0.96, 2.11) for quartile 3 vs. 1. Likewise, in the additive scale the estimates supported the synergism with an RERI=0.58 (95% CI: 0.11, 1.05) for the same contrast. Similar associations were also observed between PFOS and γ-tocopherol, with an RERI=0.60 (95% CI: 0.06, 1.15) at the highest levels of PFOS and γ-tocopherol.
In the opposite direction, a negative interaction was observed between PFOS and β-cryptoxanthin. In this case, the associations between PFOS and overweight/obesity or zBMI, were substantially higher at the lowest tertile of β-cryptoxanthin, reaching an RR=1.59 (95% CI: 1.02, 2.05) for quartile 4 vs. 1 of PFOS. Although the interaction was not statistically significant in the multiplicative scale (pInt>0.1), the RERIs supported antagonisms in the additive scale (Table 3). Finally, the synergistic interactions suggested between POPs and PUFAs appeared to be mostly weak and nonstatistically significant in both scales (Excel Tables S14 and S15) with the exception of PCB138 and LA, which showed an RERI=0.18 (95% CI: 0.02, 0.33) in the third tertile.
Discussion
In the present study, we attempted to develop and apply a comprehensive statistical framework to evaluate the mixture effect of prenatal exposure to POPs and nutrients on childhood overweight/obesity. This approach, applied to the population-based birth cohort study INMA, confirmed findings from previous studies in this cohort, highlighting the role of prenatal exposure to HCB5,14,22 and β-HCH22 on childhood obesity risk and providing evidence of a positive joint effect of the mixture of POPs and nutrients. Screening for interactions using advanced approaches highlighted a number of potential combinations. Conventional regression models allowed translating those interactions into more meaningful effect estimates in terms of inferential interpretation. Among the 10 POPs–nutrient interactions retained in the screening step, HCB–vitamin B12 was the most consistent across models and outcomes, and the regression models suggested a potential synergistic effect. Interaction between PFOS and β-cryptoxanthin suggested a protective effect of the antioxidant on overweight/obesity risk, with a higher risk associated with PFOS exposure being observed only at the lower concentrations of β-cryptoxanthin.
Despite the rise of multipollutant modeling approaches, few studies have considered mixtures of biomarkers others than pollutants.52 In fact, some dietary nutrients have the potential to counterbalance the effects of environmental pollutants, highlighting the interest of accounting for them in the mixture model.18,53 The mixture analysis allowed the identification of an unexpected synergistic effect of vitamin B12, strengthening the associations of HCB. Vitamin B12 is an essential hydrophilic vitamin found mainly in animal food products, with dairy products and meat the major contributors and exhibiting short half-lives in the body.54 Despite the lack of international consensus on the optimal levels during pregnancy, there is some agreement that the range between 220 and 850 pmol/L would be adequate.55 Thus, considering that median levels of vitamin B12 within our population were 350 pmol/L (interquartile range: 279–435 pmol/L) most of our study population would fall within that optimal range, with only 12 participants exceeding the upper threshold. Considering that sources of vitamin B12 in humans are exogenous, the blood levels may be determined either by the dietary intake or as result of their metabolism. There is the possibility that vitamin B12 may be confounding the true effect of some other unmeasured concomitant nutrients or specific food habits associated with HCB56 or childhood obesity41,57; however, adjustment for meat intake did not modify our results, consistent with previous findings.42,43 Our secondary analysis showed that vitamin B12 levels were statistically associated with vitamin supplementation and meat intake, hence supporting the role of intake in the vitamin status, which has been also reported in the same cohort.58 Mechanisms supporting the joint effect of OCs and vitamin B12 or folate on overweight risk can involve epigenetic programming.59,60 A maternal intake of methyl-group donors (e.g., folates, vitamin B12) could also alter the DNA methylation profiles of an offspring’s metabolic genes.61 In turn, HCB and vitamin B12 can both individually impact the metabolic programing of adipocytes during differentiation or their DNA methylation profiles.62,63 Indeed, vitamin B12 plays a crucial role in humans as a cofactor of methionine synthase, which is actively involved in methionine biosynthesis via the remethylation of total homocysteine. Interestingly, two enzymes involved in S-adenosyl methionine synthesis (phosphatidylethanolamine N-methyltransferase and glycine N-methyltransferase), are transcriptional targets of the aryl hydrocarbon receptor,64 which is activated by HCB.65 Thus, we may hypothesize that a maternal intake of methyl-group donors (i.e., vitamin B12) together with a higher HCB exposure contribute to an increased lipogenesis. Considering the active research to establish more accurate recommendations and thresholds of vitamin supplementation during pregnancy,66 future studies should consider the concomitant presence of environmental pollutants.
The protective effect of β-cryptoxanthin on the association between PFOS and childhood obesity also deserve attention. β-Cryptoxanthin is a naturally occurring carotenoid found in many foods of plant and animal origin (e.g., oranges, apples, egg yolk). It is closely related to β-carotene and has antioxidant properties. Conversely, PFOS is known for its prooxidative activity67 and increase of adipogenesis in vitro,68 but epidemiological studies are globally inconsistent. A study in the same INMA cohort showed mild or null associations,24 whereas other previous studies have generally shown inconsistent findings,69 and some authors have even proposed PFOS as an anti-obesogen.70 For the first time, we were able to observe a higher risk of obesity among children exposed to higher levels of PFOS and lower levels of this antioxidant during gestation. Current evidence with adult women has shown that concentrations of carotenoids, including β-cryptoxanthin, are inversely associated with BMI and waist circumference, with major effect modification by exposure to toxicants, such as by smoking.71 In experimental studies, β-cryptoxanthin exerted an anti-obesogenic effect, reducing the body fat of mice and increasing the expression of uncoupling protein 1 (UCP1) in adipose tissue via the retinoic acid receptor (RAR).72 In turn, PFOA and PFOS have also been shown to activate UCP1 in brown adipose tissue, which can modulate the food intake and body weight70; however, our findings suggest the presence of other potential mechanisms to explain the obesogenic effects of PFOS. For instance, PFOS and PFOA are activators of the peroxisome proliferator-activated receptor-alpha (PPAR-α) in human cells.73 PPAR-α and RAR share a common dimerization partner, the retinoid X receptor (RXR).74 Hence, the activation of both receptors (PPAR-α by PFOS) and (RAR by β-cryptoxanthin) could lead to a competitive effect toward this RXR partner.75
Biomonitoring studies during perinatal periods supports the fact that women are exposed to multiple environmental chemicals during pregnancy and lactation periods, critical windows for offspring development.53,76 This exposure paradigm has stimulated the increasing interest in characterizing the joint effect of environmental chemicals during pregnancy on an offspring’s health outcomes, raising the development and implementation of statistical approaches to address mixture related questions.29,52 Although the list of algorithms and statistical methods has been growing during the last few years, no one specific method consistently outperforms the others as assessed in simulation studies.26,29,77 Therefore, we conceived an approach combining multiple models to strengthen the robustness of our findings supported by the specific features from each algorithm. The method selection included Glinternet, GBM, and BKMR, based on previous literature supporting their relatively high statistical performance and capacity to characterize the joint associations of correlated variables accounting for their interactions.26,78 A simulation study showed that BKMR, in the case of nonmonotonic exposure–response relationships, may outperform penalized regression methods that assume linearity,29 a group of methods which includes the Glinternet model. Identifying relevant components of the mixtures remains a major question in terms of public health but also in terms of regulatory decision-making. Statistically, this is commonly accomplished by using variable selection, a process that becomes especially challenging as correlation between variables increases.79 Although data-driven approaches such as BKMR may improve the predictive performance of models, those may fail to attribute the true effect to the right candidates within the correlated cluster.27 This fact, together with the different nature of variable selection method (e.g., Glinternet and GBM), may help to explain some inconsistencies in the variable importance rankings between models. In any case, the variable selection process prevents selection of highly correlated variables in the model and, thus, reduces the risk of amplification bias expected in chemical mixtures.80 An alternative way to leverage this issue would be to use a priori toxicological knowledge to inform the variable selection process, especially in exploratory contexts where improving predictive performance falls out of scope. Selection of relevant interactions follows a process similar to that of the main effects, based on likelihood penalization in the case of Glinternet or spike-and-slab priors in the case of BKMR.46,49 We noticed that detection of interactions becomes specially challenging and inconsistent between models when the interaction is weak,81 something supported by the fact that low-powered studies are prone to false positive detection.82 Thus, the agreement criteria across screening methods could be a solution to attenuate the risk of false positives, as observed in the case of the strongest interaction between HCB and vitamin B12. To increase the robustness of findings from the Glinternet and GBM models, we applied a bootstrapping approach with 100 replicates, allowing the identification of interactions in terms of frequency of detection and relative contribution or strength. An additional factor affecting those inconsistencies may also be related to the differences on chemical distributions and related mixture structures between data sets. For instance, we noticed slightly higher median levels of OCPs in the ANTIOX data set compared with the rest of the data sets (Excel Table S1), consequently impacting the data set specific risk estimates. The findings also highlighted the presence of other interactions (i.e., pollutant–pollutant or nutrient–nutrient) not discussed in the present manuscript, but which may help to illustrate the complex interplay of chemicals within the mixture.
The present study should be considered with caution in light of some study limitations. First, the sample size was relatively small (n=558–1,241) for the exploration and characterization of interactions and might have resulted in low power to detect interacting effects and thus could be prone to false negatives. Second, we applied a data-driven approach to explore potential interactions with biological meaning. The high correlation between some pollutants and the current lack of congener-specific knowledge about their obesogenic potential, may increase the risk of exposure misclassification, thus attributing the interactive effect to the wrong chemical within the clusters of highly correlated variables. Current in vitro and in vivo studies about obesogenic effects of POPs are relatively limited to a few congeners, which in turn, can be highly correlated in biological matrices. For simplicity and due to the limited sample size, we focused the study to characterize two-way interactions; however, higher-order interactions cannot be neglected, either between pollutants and nutrients but also, with other individual characteristics such as maternal smoking or child’s sex, as previously observed.83 We may have also failed to accurately measure vitamin levels representative of all pregnancy given that these nutrient biomarkers reflect the current intakes or their relatively short timeframes due to their short half-lives, which range from several minutes in the case of vitamin B12 (∼7 min) to days (26–76 d) for carotenoids and even longer for folate.84,85 Nevertheless, we believe that measurement error would be homogeneously distributed across the study population; hence, it is expected that it would lead to null findings instead of spurious associations. In addition, the biomarkers used have been positively associated with established predictors from food frequency questionnaires administered to pregnant women.38 In the case of POPs, a single spot blood measurement is often considered indicative of long-term exposure given their long elimination half-lives as also shown in the INMA cohort.35,86 The statistical framework was developed assuming interactions between nutrients and POPs; however, the exploratory nature of this study did not have the ambition of elucidating the detailed underlying causal structure to determine, for instance, whether nutrients were actually effect modifiers.87 In this regard, additional causal structures could also be considered in the light of potential toxicokinetic effects of the considered nutrients on the biomarker levels of POPs. For instance, it could be plausible that reported anti-obesogenic effects of certain carotenoids may interfere with the circulating levels of POPs,19 hence contributing to measurement error and amplification bias. In addition, residual confounding cannot be completely ruled out, especially in the light of false negatives, despite the sensitivity analysis showing that our findings are robust to an extensive list of potential confounders reflecting dietary habits and lifestyle. The exploration of interactions is an emerging and active field of methodological research, and other novel approaches accommodating the complexities of pollutant data sets could be considered in future studies.88,89 Finally, the biological interpretation of statistical interactions requires some attention, in part because of the different implication of interaction scales or the definitions used in different fields.90 Following current recommendations, we reported the interactions in multiplicative and additive scale,28 and we believe that our findings may help to develop biological hypothesis for future toxicological studies and better interpret inconsistent findings in epidemiological studies.
In summary, the present study supports the hypothesis that nutritional status during pregnancy can modify the effect of environmental pollutants on child health. Specifically, we found that high levels of vitamin B12 may strengthen the associations between prenatal exposure to HCB and childhood obesity. In the opposite direction, the dietary antioxidant β-cryptoxanthin might have a protective effect against the obesogenic effects of PFOS. Our findings suggest that independent models may fail to identify weak interactions between pollutants and nutrients; therefore, combining complementary models may be a more powerful approach to consider. In the light of the public health implications of these findings, further observational and experimental research will be required for confirmation of these findings and gaining insight into the complex interplay between pollutants and nutrients during pregnancy on the metabolic programing of the offspring. As highlighted, these interactions may uncover subpopulations at risk for specific chemicals under regulatory policies, but they also support the need for more accurate nutritional guidelines during pregnancy.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank all study participants for their generous collaboration.
This project has received funding from the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement no. 874583, the ATHLETE project, and no. 825712, the OBERON project. This publication reflects only the authors’ view and the European Commission is not responsible for any use that may be made of the information it contains. INMA-Sabadell: This study was funded by grants from Instituto de Salud Carlos III (Red INMA G03/176; CB06/02/0041; PI041436; PI081151 incl. FEDER funds), CIBERESP, Generalitat de Catalunya-CIRIT 1999SGR 00241, Generalitat de Catalunya-AGAUR 2009 SGR 501, and Fundació La Marató de TV3 (090430). M.C. holds a Miguel Servet fellowship (CP16/00128) funded by Instituto de Salud Carlos III and co-funded by European Social Fund “Investing in your future.” N.S. has received funding from the Ministry of Science and Innovation and State Research Agency through the Centro de Excelencia Severo Ochoa 2019–2023 program (CEX2018-000806-S) and from IJC2020-043630-I financed by MCIN/AEI/10.13039/501100011033 and the European Union NextGenerationEU/PRTR. We also acknowledge support from the Generalitat de Catalunya through the CERCA Program. INMA-Gipuzkoa: This study was funded by grants from Instituto de Salud Carlos III (FIS-PI13/02187 and FIS-PI18/01142 incl. FEDER funds), CIBERESP, Department of Health of the Basque Government (2015111065), and the Provincial Government of Gipuzkoa (DFG15/221) and annual agreements with the municipalities of the study area (Zumarraga, Urretxu, Legazpi, Azkoitia y Azpeitia y Beasain). INMA-Valencia: This study was funded by grants from the UE (FP7-ENV-2011 cod 282957 and HEALTH.2010.2.4.5-1), Spain: ISCIII (G03/176; FIS-FEDER: PI11/01007, PI11/02591, PI11/02038, PI12/00610, PI13/1944, PI13/2032, PI14/00891, PI14/01687, PI16/1288, and PI17/00663; Miguel Servet-FEDER MS11/00178, MS15/00025, and MSII16/00051), Generalitat Valenciana: FISABIO (UGP 15-230, UGP-15-244, and UGP-15-249), and the Alicia Koplowitz Foundation 2017.
==== Refs
References
1. Gluckman PD, Hanson MA. 2004. Living with the past: evolution, development, and patterns of disease. Science 305 (5691 ):1733–1736, PMID: , 10.1126/science.1095292.15375258
2. Inadera H. 2013. Developmental origins of obesity and type 2 diabetes: molecular aspects and role of chemicals. Environ Health Prev Med 18 (3 ):185–197, PMID: , 10.1007/s12199-013-0328-8.23382021
3. Lin X, Li H. 2021. Obesity: epidemiology, pathophysiology, and therapeutics. Front Endocrinol (Lausanne) 12 :706978, PMID: , 10.3389/fendo.2021.706978.34552557
4. NCD-RisC (NCD Risk Factor Collaboration). 2017. Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of 2416 population-based measurement studies in 1289 million children, adolescents, and adults. Lancet 390 (10113 ):2627–2642, PMID: , 10.1016/S0140-6736(17)32129-3.29029897
5. Güil-Oumrait N, Valvi D, Garcia-Esteban R, Guxens M, Sunyer J, Torrent M, et al. 2021. Prenatal exposure to persistent organic pollutants and markers of obesity and cardiometabolic risk in Spanish adolescents. Environ Int 151 :106469, PMID: , 10.1016/j.envint.2021.106469.33711537
6. Legler J, Fletcher T, Govarts E, Porta M, Blumberg B, Heindel JJ, et al. 2015. Obesity, diabetes, and associated costs of exposure to endocrine-disrupting chemicals in the European Union. J Clin Endocrinol Metab 100 (4 ):1278–1288, PMID: , 10.1210/jc.2014-4326.25742518
7. Jones KC, de Voogt P. 1999. Persistent organic pollutants (POPs): state of the science. Environ Pollut 100 (1–3 ):209–221, PMID: , 10.1016/S0269-7491(99)00098-6.15093119
8. UNEP (UN Environment Programme). 2017. Stockholm Convention Global Monitoring Plan on Persistent Organic Pollutants Second Monitoring Report. UNEP/POPS/COP.8/INF/38. http://chm.pops.int/Portals/0/download.aspx?d=UNEP-POPS-COP.8-INF-38.English.pdf [accessed 21 October 2020].
9. WHO (World Health Organization). 2010. Persistent Organic Pollutants: Impact on Child Health. Geneva, Switzerland: World Health Organization. https://apps.who.int/iris/rest/bitstreams/53004/retrieve [accessed 13 September 2020].
10. Iszatt N, Stigum H, Verner MA, White RA, Govarts E, Murinova LP, et al. 2015. Prenatal and postnatal exposure to persistent organic pollutants and infant growth: a pooled analysis of seven European birth cohorts. Environ Health Perspect 123 (7 ):730–736, PMID: , 10.1289/ehp.1308005.25742056
11. Nadal A, Quesada I, Tudurí E, Nogueiras R, Alonso-Magdalena P. 2017. Endocrine-disrupting chemicals and the regulation of energy balance. Nat Rev Endocrinol 13 (9 ):536–546, PMID: , 10.1038/nrendo.2017.51.28524168
12. Ren XM, Kuo Y, Blumberg B. 2020. Agrochemicals and obesity. Mol Cell Endocrinol 515 :110926, PMID: , 10.1016/j.mce.2020.110926.32619583
13. Stratakis N, Rock S, La Merrill MA, Saez M, Robinson O, Fecht D, et al. 2022. Prenatal exposure to persistent organic pollutants and childhood obesity: a systematic review and meta-analysis of human studies. Obes Rev 23 (suppl 1 ):e13383, PMID: , 10.1111/obr.13383.34766696
14. Valvi D, Mendez MA, Martinez D, Grimalt JO, Torrent M, Sunyer J, et al. 2012. Prenatal concentrations of polychlorinated biphenyls, DDE, and DDT and overweight in children: a prospective birth cohort study. Environ Health Perspect 120 (3 ):451–457, PMID: , 10.1289/ehp.1103862.22027556
15. Papadopoulou E, Haug LS, Sakhi AK, Andrusaityte S, Basagaña X, Brantsaeter AL, et al. 2019. Diet as a source of exposure to environmental contaminants for pregnant women and children from six European countries. Environ Health Perspect 127 (10 ):107005, PMID: , 10.1289/EHP5324.31617753
16. Hennig B, Ormsbee L, McClain CJ, Watkins BA, Blumberg B, Bachas LG, et al. 2012. Nutrition can modulate the toxicity of environmental pollutants: implications in risk assessment and human health. Environ Health Perspect 120 (6 ):771–774, PMID: , 10.1289/ehp.1104712.22357258
17. Choi AL, Cordier S, Weihe P, Grandjean P. 2008. Negative confounding in the evaluation of toxicity: the case of methylmercury in fish and seafood. Crit Rev Toxicol 38 (10 ):877–893, PMID: , 10.1080/10408440802273164.19012089
18. Cano-Sancho G, Casas M. 2021. Interactions between environmental pollutants and dietary nutrients: current evidence and implications in epidemiological research. J Epidemiol Community Health 75 (2 ):108–113, PMID: , 10.1136/jech-2020-213789.33023970
19. Bonet ML, Canas JA, Ribot J, Palou A. 2016. Carotenoids in adipose tissue biology and obesity. Subcell Biochem 79 :377–414, PMID: , 10.1007/978-3-319-39126-7_15.27485231
20. Warnke I, Goralczyk R, Fuhrer E, Schwager J. 2011. Dietary constituents reduce lipid accumulation in murine C3H10 T1/2 adipocytes: a novel fluorescent method to quantify fat droplets. Nutr Metab (Lond) 8 (1 ):30, PMID: , 10.1186/1743-7075-8-30.21569430
21. Warnke I, Jocken JWE, Schoop R, Toepfer C, Goralczyk R, Schwager J. 2016. Combinations of bio-active dietary constituents affect human white adipocyte function in-vitro. Nutr Metab (Lond) 13 (1 ):84, PMID: , 10.1186/s12986-016-0143-5.27895698
22. Agay-Shay K, Martinez D, Valvi D, Garcia-Esteban R, Basagaña X, Robinson O, et al. 2015. Exposure to endocrine-disrupting chemicals during pregnancy and weight at 7 years of age: a multi-pollutant approach. Environ Health Perspect 123 (10 ):1030–1037, PMID: , 10.1289/ehp.1409049.25956007
23. Valvi D, Mendez MA, Garcia-Esteban R, Ballester F, Ibarluzea J, Goñi F, et al. 2014. Prenatal exposure to persistent organic pollutants and rapid weight gain and overweight in infancy. Obesity (Silver Spring) 22 (2 ):488–496, PMID: , 10.1002/oby.20603.23963708
24. Manzano-Salgado CB, Casas M, Lopez-Espinosa MJ, Ballester F, Iñiguez C, Martinez D, et al. 2017. Prenatal exposure to perfluoroalkyl substances and birth outcomes in a Spanish birth cohort. Environ Int 108 :278–284, PMID: , 10.1016/j.envint.2017.09.006.28917208
25. Zhang X, Chiu YHM, Kannan S, Cowell W, Deng W, Coull BA, et al. 2022. Joint associations among prenatal metal mixtures and nutritional factors on birth weight z-score: evidence from an urban U.S. population. Environ Res 208 :112675, PMID: , 10.1016/j.envres.2022.112675.34995543
26. Barrera-Gómez J, Agier L, Portengen L, Chadeau-Hyam M, Giorgis-Allemand L, Siroux V, et al. 2017. A systematic comparison of statistical methods to detect interactions in exposome–health associations. Environ Health 16 (1 ):74, PMID: , 10.1186/s12940-017-0277-6.28709428
27. Braun JM, Gennings C, Hauser R, Webster TF. 2016. What can epidemiological studies tell us about the impact of chemical mixtures on human health? Environ Health Perspect 124 (1 ):A6–A9, PMID: , 10.1289/ehp.1510569.26720830
28. Knol MJ, VanderWeele TJ. 2012. Recommendations for presenting analyses of effect modification and interaction. Int J Epidemiol 41 (2 ):514–520, PMID: , 10.1093/ije/dyr218.22253321
29. Lazarevic N, Knibbs LD, Sly PD, Barnett AG. 2020. Performance of variable and function selection methods for estimating the nonlinear health effects of correlated chemical mixtures: a simulation study. Stat Med 39 (27 ):3947–3967, PMID: , 10.1002/sim.8701.32940933
30. Guxens M, Ballester F, Espada M, Fernández MF, Grimalt JO, Ibarluzea J, et al. 2012. Cohort profile: the INMA—INfancia y Medio Ambiente—(Environment and Childhood) Project. Int J Epidemiol 41 (4 ):930–940, PMID: , 10.1093/ije/dyr054.21471022
31. Goñi F, López R, Etxeandia A, Millán E, Amiano P. 2007. High throughput method for the determination of organochlorine pesticides and polychlorinated biphenyls in human serum. J Chromatogr B Analyt Technol Biomed Life Sci 852 (1–2 ):15–21, PMID: , 10.1016/j.jchromb.2006.12.049.17267299
32. Grimalt JO, Howsam M, Carrizo D, Otero R, de Marchi MR, Vizcaino E. 2010. Integrated analysis of halogenated organic pollutants in sub-millilitre volumes of venous and umbilical cord blood sera. Anal Bioanal Chem 396 (6 ):2265–2272, PMID: , 10.1007/s00216-010-3460-y.20135307
33. Mendez MA, Garcia-Esteban R, Guxens M, Vrijheid M, Kogevinas M, Goñi F, et al. 2011. Prenatal organochlorine compound exposure, rapid weight gain, and overweight in infancy. Environ Health Perspect 119 (2 ):272–278, PMID: , 10.1289/ehp.1002169.20923745
34. Phillips DL, Pirkle JL, Burse VW, Bernert JT Jr, Henderson LO, Needham LL. 1989. Chlorinated hydrocarbon levels in human serum: effects of fasting and feeding. Arch Environ Contam Toxicol 18 (4 ):495–500, PMID: , 10.1007/BF01055015.2505694
35. Manzano-Salgado CB, Casas M, Lopez-Espinosa MJ, Ballester F, Basterrechea M, Grimalt JO, et al. 2015. Transfer of perfluoroalkyl substances from mother to fetus in a Spanish birth cohort. Environ Res 142 :471–478, PMID: , 10.1016/j.envresenvres.2015.07.020.26257032
36. Montes R, Chisaguano AM, Castellote AI, Morales E, Sunyer J, López-Sabater MC. 2013. Fatty-acid composition of maternal and umbilical cord plasma and early childhood atopic eczema in a Spanish cohort. Eur J Clin Nutr 67 (6 ):658–663, PMID: , 10.1038/ejcn.2013.68.23549201
37. Morales E, Rodriguez A, Valvi D, Iñiguez C, Esplugues A, Vioque J, et al. 2015. Deficit of vitamin D in pregnancy and growth and overweight in the offspring. Int J Obes (Lond) 39 (1 ):61–68, PMID: , 10.1038/ijo.2014.165.25189178
38. Vioque J, Navarrete-Muñoz EM, Gimenez-Monzó D, García-de-la-Hera M, Granado F, Young IS, et al. 2013. Reproducibility and validity of a food frequency questionnaire among pregnant women in a Mediterranean area. Nutr J 12 :26, PMID: , 10.1186/1475-2891-12-26.23421854
39. de Onis M, Onyango AW, Borghi E, Siyam A, Nishida C, Siekmann J. 2007. Development of a WHO growth reference for school-aged children and adolescents. Bull World Health Organ 85 (9 ):660–667, PMID: , 10.2471/BLT.07.043497.18026621
40. Vrijheid M, Fossati S, Maitre L, Márquez S, Roumeliotaki T, Agier L, et al. 2020. Early-life environmental exposures and childhood obesity: an exposome-wide approach. Environ Health Perspect 128 (6 ):67009, PMID: , 10.1289/EHP5975.32579081
41. Fernández-Barrés S, Romaguera D, Valvi D, Martínez D, Vioque J, Navarrete-Muñoz EM, et al. 2016. Mediterranean dietary pattern in pregnant women and offspring risk of overweight and abdominal obesity in early childhood: the INMA birth cohort study. Pediatr Obes 11 (6 ):491–499, PMID: , 10.1111/ijpo.12092.26763767
42. Llop S, Ballester F, Vizcaino E, Murcia M, Lopez-Espinosa MJ, Rebagliato M, et al. 2010. Concentrations and determinants of organochlorine levels among pregnant women in Eastern Spain. Sci Total Environ 408 (23 ):5758–5767, PMID: , 10.1016/j.scitotenv.2010.07.085.20832846
43. Ibarluzea J, Alvarez-Pedrerol M, Guxens M, Marina LS, Basterrechea M, Lertxundi A, et al. 2011. Sociodemographic, reproductive and dietary predictors of organochlorine compounds levels in pregnant women in Spain. Chemosphere 82 (1 ):114–120, PMID: , 10.1016/j.chemosphere.2010.09.051.20965545
44. van Buuren S, Groothuis-Oudshoorn K. 2011. mice: multivariate imputation by chained equations in R. J Stat Softw 45 (3 ):1–67, 10.18637/jss.v045.i03.
45. Chen L, Bao BY, Chang WC, Ho JYP, Cheng BH, Wang CL, et al. 2018. Short androgen receptor poly-glutamine-promoted endometrial cancer is associated with benzo[a]pyrene-mediated aryl hydrocarbon receptor activation. J Cell Mol Med 22 (1 ):46–56, PMID: , 10.1111/jcmm.13291.28782227
46. Lim M, Hastie T. 2015. Learning interactions via hierarchical group-lasso regularization. J Comput Graph St at 24 (3 ):627–654, PMID: , 10.1080/10618600.2014.938812.26759522
47. Havard A, Straka P, Sara G, Lujic S, Tran DT, Jorm LR. 2019. Identifying patients using antidepressants for the treatment of depression: a predictive algorithm for use in pharmaceutical and medical claims data. Pharmacoepidemiol Drug Saf 28 (3 ):354–361, PMID: , 10.1002/pds.4739.30680859
48. Elith J, Leathwick JR, Hastie T. 2008. A working guide to boosted regression trees. J Anim Ecol 77 (4 ):802–813, PMID: , 10.1111/j.1365-2656.2008.01390.x.18397250
49. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16 (3 ):493–508, PMID: , 10.1093/biostatistics/kxu058.25532525
50. Bobb JF, Claus Henn B, Valeri L, Coull BA. 2018. Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression. Environ Health 17 (1 ):67, PMID: , 10.1186/s12940-018-0413-y.30126431
51. Knol MJ, van der Tweel I, Grobbee DE, Numans ME, Geerlings MI. 2007. Estimating interaction on an additive scale between continuous determinants in a logistic regression model. Int J Epidemiol 36 (5 ):1111–1118, PMID: , 10.1093/ije/dym157.17726040
52. Lazarevic N, Barnett AG, Sly PD, Knibbs LD. 2019. Statistical methodology in studies of prenatal exposure to mixtures of endocrine-disrupting chemicals: a review of existing approaches and new alternatives. Environ Health Perspect 127 (2 ):26001, PMID: , 10.1289/EHP2207.30720337
53. Cano-Sancho G, Alexandre-Gouabau MC, Moyon T, Royer AL, Guitton Y, Billard H, et al. 2020. Simultaneous exploration of nutrients and pollutants in human milk and their impact on preterm infant growth: an integrative cross-platform approach. Environ Res 182 :109018, PMID: , 10.1016/j.envres.2019.109018.31863943
54. Obeid R, Heil SG, Verhoeven MMA, van den Heuvel EGHM, de Groot LCGM, Eussen SJPM. 2019. Vitamin B12 intake from animal foods, biomarkers, and health aspects. Front Nutr 6 :93, PMID: , 10.3389/fnut.2019.00093.31316992
55. Sukumar N, Adaikalakoteswari A, Venkataraman H, Maheswaran H, Saravanan P. 2016. Vitamin B12 status in women of childbearing age in the UK and its relationship with national nutrient intake guidelines: results from two National Diet and Nutrition Surveys. BMJ Open 6 (8 ):e011247, PMID: , 10.1136/bmjopen-2016-011247.27519920
56. Gasull M, Bosch de Basea M, Puigdomènech E, Pumarega J, Porta M. 2011. Empirical analyses of the influence of diet on human concentrations of persistent organic pollutants: a systematic review of all studies conducted in Spain. Environ Int 37 (7 ):1226–1235, PMID: , 10.1016/j.envint.2011.05.008.21683445
57. Chen LW, Aubert AM, Shivappa N, Bernard JY, Mensink-Bout SM, Geraghty AA, et al. 2021. Maternal dietary quality, inflammatory potential and childhood adiposity: an individual participant data pooled analysis of seven European cohorts in the ALPHABET consortium. BMC Med 19 (1 ):33, PMID: , 10.1186/s12916-021-01908-7.33612114
58. Navarrete-Muñoz EM, Valera-Gran D, García de la Hera M, Gimenez-Monzo D, Morales E, Julvez J, et al. 2015. Use of high doses of folic acid supplements in pregnant women in Spain: an INMA cohort study. BMJ Open 5 (11 ):e009202, PMID: , 10.1136/bmjopen-2015-009202.26603248
59. McKay JA, Groom A, Potter C, Coneyworth LJ, Ford D, Mathers JC, et al. 2012. Genetic and non-genetic influences during pregnancy on infant global and site specific DNA methylation: role for folate gene variants and vitamin B12. PLoS One 7 (3 ):e33290, PMID: , 10.1371/journal.pone.0033290.22479380
60. Ouidir M, Mendola P, Buck Louis GM, Kannan K, Zhang C, Tekola-Ayele F. 2020. Concentrations of persistent organic pollutants in maternal plasma and epigenome-wide placental DNA methylation. Clin Epigenetics 12 (1 ):103, PMID: , 10.1186/s13148-020-00894-6.32653021
61. Pauwels S, Ghosh M, Duca RC, Bekaert B, Freson K, Huybrechts I, et al. 2017. Maternal intake of methyl-group donors affects DNA methylation of metabolic genes in infants. Clin Epigenetics 9 (1 ):16, PMID: , 10.1186/s13148-017-0321-y.28191262
62. Bastos Sales L, Kamstra JH, Cenijn PH, van Rijt LS, Hamers T, Legler J. 2013. Effects of endocrine disrupting chemicals on in vitro global DNA methylation and adipocyte differentiation. Toxicol In Vitro 27 (6 ):1634–1643, PMID: , 10.1016/j.tiv.2013.04.005.23603478
63. Green CR, Wallace M, Divakaruni AS, Phillips SA, Murphy AN, Ciaraldi TP, et al. 2016. Branched-chain amino acid catabolism fuels adipocyte differentiation and lipogenesis. Nat Chem Biol 12 (1 ):15–21, PMID: , 10.1038/nchembio.1961.26571352
64. Kim YC, Seok S, Byun S, Kong B, Zhang Y, Guo G, et al. 2018. AhR and SHP regulate phosphatidylcholine and S-adenosylmethionine levels in the one-carbon cycle. Nat Commun 9 (1 ):540, PMID: , 10.1038/s41467-018-03060-y.29416063
65. Chiappini F, Ceballos L, Olivares C, Bastón JI, Miret N, Pontillo C, et al. 2022. Endocrine disruptor hexachlorobenzene induces cell migration and invasion, and enhances aromatase expression levels in human endometrial stromal cells. Food Chem Toxicol 162 :112867, PMID: , 10.1016/j.fct.2022.112867.35181438
66. Maruvada P, Stover PJ, Mason JB, Bailey RL, Davis CD, Field MS, et al. 2020. Knowledge gaps in understanding the metabolic and clinical effects of excess folates/folic acid: a summary, and perspectives, from an NIH workshop. Am J Clin Nutr 112 (5 ):1390–1403, PMID: , 10.1093/ajcn/nqaa259.33022704
67. Chen N, Li J, Li D, Yang Y, He D. 2014. Chronic exposure to perfluorooctane sulfonate induces behavior defects and neurotoxicity through oxidative damages, in vivo and in vitro. PLoS One 9 (11 ):e113453, PMID: , 10.1371/journal.pone.0113453.25412474
68. Modaresi SMS, Wei W, Emily M, DaSilva NA, Slitt AL. 2022. Per- and polyfluoroalkyl substances (PFAS) augment adipogenesis and shift the proteome in murine 3T3-L1 adipocytes. Toxicology 465 :153044, PMID: , 10.1016/j.tox.2021.153044.34800597
69. Lee YJ, Jung HW, Kim HY, Choi YJ, Lee YA. 2021. Early-life exposure to per- and poly-fluorinated alkyl substances and growth, adiposity, and puberty in children: a systematic review. Front Endocrinol (Lausanne) 12 :683297, PMID: , 10.3389/fendo.2021.683297.34566884
70. Di Gregorio I, Busiello RA, Burgos Aceves MA, Lepretti M, Paolella G, Lionetti L. 2019. Environmental pollutants effect on brown adipose tissue. Front Physiol 9 :1891, PMID: , 10.3389/fphys.2018.01891.30687113
71. Kabat GC, Heo M, Ochs-Balcom HM, LeBoff MS, Mossavar-Rahmani Y, Adams-Campbell LL, et al. 2016. Longitudinal association of measures of adiposity with serum antioxidant concentrations in postmenopausal women. Eur J Clin Nutr 70 (1 ):47–53, PMID: , 10.1038/ejcn.2015.74.26014267
72. Hara H, Takahashi H, Mohri S, Murakami H, Kawarasaki S, Iwase M, et al. 2019. β-Cryptoxanthin induces UCP-1 expression via a RAR pathway in adipose tissue. J Agric Food Chem 67 (38 ):10595–10603, PMID: , 10.1021/acs.jafc.9b01930.31475817
73. Bjork JA, Butenhoff JL, Wallace KB. 2011. Multiplicity of nuclear receptor activation by PFOA and PFOS in primary human and rodent hepatocytes. Toxicology 288 (1–3 ):8–17, PMID: , 10.1016/j.tox.2011.06.012.21723365
74. Zhang R, Wang Y, Li R, Chen G. 2015. Transcriptional factors mediating retinoic acid signals in the control of energy metabolism. Int J Mol Sci 16 (6 ):14210–14244, PMID: , 10.3390/ijms160614210.26110391
75. Mounien L, Tourniaire F, Landrier JF. 2019. Anti-obesity effect of carotenoids: direct impact on adipose tissue and adipose tissue-driven indirect effects. Nutrients 11 (7 ):1562, PMID: , 10.3390/nu11071562.31373317
76. Haug LS, Sakhi AK, Cequier E, Casas M, Maitre L, Basagana X, et al. 2018. In-utero and childhood chemical exposome in six European mother–child cohorts. Environ Int 121 (pt 1 ):751–763, PMID: , 10.1016/j.envint.2018.09.056.30326459
77. Agier L, Portengen L, Chadeau-Hyam M, Basagaña X, Giorgis-Allemand L, Siroux V, et al. 2016. A systematic comparison of linear regression-based statistical methods to assess exposome–health associations. Environ Health Perspect 124 (12 ):1848–1856, PMID: , 10.1289/EHP172.27219331
78. Lampa E, Lind L, Lind PM, Bornefalk-Hermansson A. 2014. The identification of complex interactions in epidemiology and toxicology: a simulation study of boosted regression trees. Environ Health 13 :57, PMID: , 10.1186/1476-069X-13-57.24993424
79. Lenters V, Vermeulen R, Portengen L. 2018. Performance of variable selection methods for assessing the health effects of correlated exposures in case–control studies. Occup Environ Med 75 (7 ):522–529, PMID: , 10.1136/oemed-2016-104231.28947495
80. Weisskopf MG, Seals RM, Webster TF. 2018. Bias amplification in epidemiologic analysis of exposure to mixtures. Environ Health Perspect 126 (4 ):047003, PMID: , 10.1289/EHP2450.29624292
81. Brookes ST, Whitely E, Egger M, Smith GD, Mulheran PA, Peters TJ. 2004. Subgroup analyses in randomized trials: risks of subgroup-specific analyses; power and sample size for the interaction test. J Clin Epidemiol 57 (3 ):229–236, PMID: , 10.1016/j.jclinepi.2003.08.009.15066682
82. Christley RM. 2010. Power and error: increased risk of false positive results in underpowered studies. Open Epidemiol J 3 (1 ):16–19, 10.2174/1874297101003010016.
83. Casas M, Nieuwenhuijsen M, Martínez D, Ballester F, Basagaña X, Basterrechea M, et al. 2015. Prenatal exposure to PCB-153, p,p′-DDE and birth outcomes in 9000 mother–child pairs: exposure–response relationship and effect modifiers. Environ Int 74 :23–31, PMID: , 10.1016/j.envint.2014.09.013.25314142
84. Burri BJ, Neidlinger TR, Clifford AJ. 2001. Serum carotenoid depletion follows first-order kinetics in healthy adult women fed naturally low carotenoid diets. J Nutr 131 (8 ):2096–2100, PMID: , 10.1093/jn/131.8.2096.11481400
85. Gregory JF III, Williamson J, Liao JF, Bailey LB, Toth JP. 1998. Kinetic model of folate metabolism in nonpregnant women consuming [2H2]folic acid: isotopic labeling of urinary folate and the catabolite para-acetamidobenzoylglutamate indicates slow, intake-dependent, turnover of folate pools. J Nutr 128 (11 ):1896–1906, PMID: , 10.1093/jn/128.11.1896.9808640
86. Lopez-Espinosa MJ, Murcia M, Iñiguez C, Vizcaino E, Costa O, Fernández-Somoano A, et al. 2016. Organochlorine compounds and ultrasound measurements of fetal growth in the INMA cohort (Spain). Environ Health Perspect 124 (1 ):157–163, PMID: , 10.1289/ehp.1408907.26058084
87. VanderWeele TJ. 2009. On the distinction between interaction and effect modification. Epidemiology 20 (6 ):863–871, PMID: , 10.1097/EDE.0b013e3181ba333c.19806059
88. Ferrari F, Dunson DB. 2021. Bayesian factor analysis for inference on interactions. J Am Stat Assoc 116 (535 ):1521–1532, PMID: , 10.1080/01621459.2020.1745813.34898761
89. Ferrari F, Dunson DB. 2020. Identifying main effects and interactions among exposures using Gaussian processes. Ann Appl Stat 14 (4 ):1743–1758, PMID: , 10.1214/20-AOAS1363.34630816
90. Howard GJ, Webster TF. 2013. Contrasting theories of interaction in epidemiology and toxicology. Environ Health Perspect 121 (1 ):1–6, PMID: , 10.1289/ehp.1205889.23014866
|
PMC010xxxxxx/PMC10022655.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
EHP12915
10.1289/EHP12915
Erratum
Erratum: Current Breast Milk PFAS Levels in the United States and Canada: After All This Time, Why Don’t We Know More?
LaKind Judy S.
Verner Marc-André
Rogers Rachel D.
Goeden Helen
Naiman Daniel Q.
Marchitti Satori A.
Lehmann Geniece M.
Hines Erin P.
https://orcid.org/0000-0002-8956-398X
Fenton Suzanne E.
17 3 2023
3 2023
17 3 2023
131 3 03900117 2 2023
23 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Environ Health Perspect 130(2):025002 (2022), https://doi.org/10.1289/EHP10359
==== Body
pmcDuring the paper editing process, two values in Table 2 were mistakenly modified. Milk:serum concentration ratios from the “Kärrman et al. 2007” study should be 0.01 for PFOS and 0.01 PFNA. This error affects values in Table 2 only. All calculations were performed with the correct values.
The authors regret the error.
Table 2 Published milk:serum concentration ratios for PFOA, PFOS, PFHxS, and PFNA.
Study Country PFOA PFOS PFHxS PFNA
Cariou et al. 2015 France 0.038 (n=10) 0.011 (n=19) 0.012 (n=8) —
Kärrman et al. 2007 Sweden 0.12 (n=1) 0.01 (n=12) 0.02 (n=12) 0.01 (n=2)
Kim et al. 2011 Korea 0.025 (n=17) 0.011 (n=17) 0.008 (n=17) —
Liu et al. 2011 China 0.11 (n=50) 0.02 (n=50) — 0.05 (n=50)
Mean 0.0733 0.0130 0.0133 0.030
Note: n=number of samples with measured milk and serum concentrations within individual studies. —, no data; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonate.
|
PMC010xxxxxx/PMC10032041.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36946585
EHP11371
10.1289/EHP11371
Research Letter
Associations among Historical Neighborhood Disinvestment, Hazardous Air Pollutants, and Current Adult Asthma Prevalence
https://orcid.org/0000-0002-4884-1621
Campbell Erin J. 1 2
Sims Kendra D. 3
Hill Elaine L. 2
Willis Mary D. 1
1 Department of Epidemiology, School of Public Health, Boston University, Boston, Massachusetts, USA
2 Department of Public Health Sciences, School of Medicine and Dentistry, University of Rochester, Rochester, New York, USA
3 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA
Address correspondence to Mary D. Willis. Email: [email protected]
22 3 2023
3 2023
131 3 03770206 4 2022
27 1 2023
02 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors report no competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Structural racism manifests through systematic neighborhood disinvestment in local infrastructure where racial and ethnic minorities reside,1 leading to a disproportionate burden of environmental pollution and associated respiratory health hazards.2 A former federal agency, the Home Owners’ Loan Corporation (HOLC, established in 1933), codified neighborhood disinvestment across the United States. Using a grading system, HOLC appraisers assessed perceived mortgage security risk of a neighborhood’s residents defaulting on their home loans.3 Riskier grades were consistently assigned to neighborhoods with high concentrations of racial and ethnic minorities, whom financial institutions perceived as “undesirable” recipients of home loans.3 The legacy of this system is often referred to as “redlining,” a term based on the red color assigned to the highest-risk neighborhoods on maps used by agencies and organizations involved in real estate laws and lending. We hypothesized that neighborhoods with historically (c. 1933) higher-risk HOLC grades have worse contemporary respiratory health than neighborhoods with lower-risk HOLC grades. Current residents of historically disinvested neighborhoods may experience an increased prevalence of asthma through past land-use patterns that led to worse environmental quality, such as less green space or higher air pollution.4,5
In this study, we integrated four public data sources to evaluate national and city-by-city associations among historical neighborhood disinvestment, modern hazardous air pollutant levels, and current asthma prevalence.
Methods
To ascertain historic neighborhood disinvestment, we aligned HOLC grade polygons from the Mapping Inequality database (https://dsl.richmond.edu/panorama/redlining) with modern census tracts; we overlaid the polygons and assigned a census tract to the least risky overlapping HOLC district grade. Grades ranged from A (lowest risk) to D (highest risk).6 This method conservatively assumes that any overlap with a lower-risk HOLC grade would yield some degree of long-term benefits for the neighborhood. A- and B-graded tracts were collapsed in the reference group. HOLC grades in Savannah, Georgia, Duluth, Minnesota, and Bronx, New York, contained a grade of E (n=4 tracts total), which we reclassified as D for our analyses.
We used the U.S. Environmental Protection Agency’s Respiratory Hazard Index (RHI) from the 2014 National Air Toxics Assessment, a census tract–level estimate of risk that air toxic concentrations would cause adverse respiratory health outcomes (https://www.epa.gov/national-air-toxics-assessment). An RHI >1 indicates that the ambient average concentration of the air toxics is greater than the concentration associated with adverse respiratory health outcomes.
To evaluate asthma prevalence, we used self-reported adult (≥18 y old) asthma prevalence from the 2017 Behavioral Risk Factor Surveillance System; this estimate is aggregated to the census tract level in the Centers for Disease Control and Prevention 500 Cities database (https://www.cdc.gov/places/about/500-cities-2016-2019/index.html). This measure is the weighted number of respondents that answered “yes” to both of the following questions: “Have you ever been told by a doctor, nurse, or other health professional that you have asthma?” and “Do you still have asthma?”
We examined census tract–level data from the American Community Survey 5-y estimates (2012–2017; https://www.census.gov/programs-surveys/acs). Our study included all census tracts in the 500 Cities database that overlapped with the HOLC maps (n=11,874 census tracts in 148 cities) (https://www.nhgis.org/). When matching the data sets, 49 cities were in the HOLC data set but not the 500 Cities database, and 352 cities were in the 500 Cities database but not HOLC.
We conducted a) an overall nationwide analysis (n=148 cities), including descriptive statistics, associations between HOLC grades and mean RHI, and associations between HOLC grades and self-reported adult asthma prevalence; and b) city-by-city analyses (n=53 cities) that spatially examined the heterogeneity of these outcomes.
Nationally, we calculated the mean percent difference in the RHI by HOLC grade. In linear regression models for adult asthma prevalence, we adjusted for population age (proportion over 65 y, proportion under 18 y), and sex (proportion male). Results are reported as the increased adult asthma prevalence of D-graded tracts in comparison with A- and B-graded tracts and C-graded tracts in comparison with A- and B-graded tracts.
We conducted the same analyses by city, with separate regression analyses, corresponding to the total number of cities in the data set. Incorporating the restriction of city boundaries, city-by-city analyses mirrored the nationwide analysis. The mean percent difference revealed the average difference in RHI between D-graded and A- or B-graded census tracts only. To ensure sufficient sample size, we included cities with: a) ≥50 census tracts that overlapped with historic HOLC maps, and b) ≥1 census tract assigned A or B HOLC grade (n=9,031 tracts in 53 cities).
Analyses and map figure were generated using R (version 4.0.2; R Development Core Team).
Results
We present neighborhood context characteristics by HOLC grade (Table 1). For RHI, we observed a mean 7.7% difference for C-graded census tracts and 17.0% for D-graded census tracts in comparison with the A- and B-grades nationwide. In adjusted estimates for adult asthma, we observed a 0.52% [95% confidence interval (CI): 0.47, 0.57] increased prevalence for C-graded census tracts and 0.62% (95% CI: 0.55, 0.69) increased prevalence for D-graded census tracts, in comparison with A- and B-graded census tracts.
Table 1 Descriptive statistics of neighborhood context characteristics by historical Home Owners’ Loan Corporation (HOLC) grades (n=148 cities in the United States).
HOLC Grade
A B C D
n (tracts) 1,754 3,202 4,755 2,063
Respiratory hazard indexa
Minimum 0.20 0.18 0.20 0.20
25th percentile 0.37 0.39 0.42 0.49
Mean 0.48 (0.13) 0.51 (0.15) 0.54 (0.16) 0.59 (0.15)
75th percentile 0.56 0.62 0.65 0.71
Maximum 0.92 1.12 4.06 1.62
Asthma prevalenceb
Minimum 6.0 5.9 5.8 5.7
25th percentile 8.2 8.7 8.9 8.8
Mean 9.5 (1.7) 10.2 (1.9) 10.5 (2.0) 10.5 (2.1)
75th percentile 10.4 11.4 11.9 11.9
Maximum 16.2 17.4 17.4 17.8
Demographicsc
Population, mean, thousands 3,643 (1,481) 3,302 (1,662) 3,464 (1,641) 3,378 (1,770)
Population density, people/km2 3.2 (5.9) 6.5 (9.5) 6.9 (9.6) 8.6 (10.0)
Percentage over 65 y 15 (6.8) 13 (6.1) 11 (5.7) 11 (5.8)
Percentage under 18 y 19 (7.0) 21 (7.7) 22 (8.3) 22 (9.8)
Percentage male 48 (4.6) 49 (4.6) 49 (5.1) 49 (6.4)
Percentage White 69 (26) 55 (29) 46 (29) 43 (29)
Percentage Black or African American 19 (26) 27 (31) 31 (33) 34 (33)
Percentage American Indian or Alaska Native 1 (1) 1 (1) 1 (1) 1 (1)
Percentage Asian 5 (7.5) 7 (11) 8 (13) 8 (12)
Percentage Hispanic or Latino (of any race) 12 (15) 20 (24) 27 (27) 28 (28)
Socioeconomicc
Percentage with bachelor’s degree or higher 48 (24) 33 (22) 25 (20) 29 (29)
Percentage of whom poverty is determined 15 (12) 21 (13) 25 (14) 27 (16)
Median household income, 2017 dollars 37,707 (18,811) 29,001 (14,192) 25,261 (11,961) 27,363 (16,706)
Housingc
Percentage detached unit 56 (29) 43 (30) 38 (30) 28 (30)
Percentage housing with 2+ units 37 (28) 47 (30) 53 (30) 60 (32)
Percentage owner-occupied unit 57 (21) 46 (21) 41 (21) 34 (22)
Percentage of structures built pre-1940 33 (23) 37 (23) 36 (24) 33 (24)
Median gross rent, 2017 dollars 1,161 (514) 1,103 (420) 1,071 (375) 1,101 (503)
Median home value, 2017 dollars 375,609 (386,389) 316,674 (304,776) 291,647 (243,740) 351,717 (301,012)
Note: Standard deviations are presented in parentheses.
a Derived from Environmental Protection Agency National Air Toxics Assessment (2014).
b Derived from Centers for Disease Control and Prevention 500 Cities Database (2017): percentage of population ≥18 y old who self-report an asthma diagnosis.
c Derived from Census Bureau American Community Survey (2017).
In city-by-city analyses, we observed substantial heterogeneity (Figure 1). San Francisco, California, provides an example of the association between HOLC grade and both RHI and asthma prevalence; in comparison with A- or B-graded tracts, we observed mean 24.65% difference in RHI in D-graded tracts. In comparison with A- or B-graded tracts, the prevalence of asthma was 0.58% (95% CI: 0.30, 0.85) higher. To highlight the heterogeneity of results on the city level, the prevalence of asthma in D-graded census tracts in Atlanta, Georgia, was 2.66% (95% CI: 1.81, 3.51) higher than in A- or B-graded census tracts, and we observed a mean difference of 3.35% in the RHI of D-graded tracts in comparison with the RHI of A- or B-graded tracts.
Figure 1. City-by-city mean percent differences in respiratory hazard index (RHI) and associations with asthma prevalence by historical Home Owners’ Loan Corporation (HOLC) grade. Panel (A) shows a map of the results of city-specific mean percent differences in respiratory hazard index (RHI) between D-graded census tracts and A- and B-graded census tracts. Panel (B) shows a map of the results of a city-specific regression for the association between Home Owners’ Loan Corporation (HOLC) grade and adult asthma prevalence. Displayed coefficients are comparing D-graded census tracts to A- and B-graded census tracts, with 95% confidence intervals in parenthesis. Adjusted linear regressions contained city (categorical variable for each city), age distribution (proportion over 65 y of age, proportion under 18 y of age), and sex distribution (proportion male). The coefficients can be interpreted as the increased asthma prevalence of D-graded tracts compared to A- and B-graded tracts. In both maps, the cities with the 5 largest and 5 smallest magnitude results are bolded. Maps are generated in R using “rnaturalearth” and “maps” packages.
Figure 1 is a set of two maps of the United States. Panel A, titled mean percent difference in the respiratory hazard index by the home owners’ loan corporation grade open parenthesis uppercase d graded tracts relative to uppercase a and uppercase b graded tracts closed parenthesis, depicts latitude and longitude positions of the cities: 4.23 for Seattle, 3.20 for Spokane, 22.40 for Chicago, 12.26 for Minneapolis, 6.15 for Kansas City, 6.25 for Detroit, 25.63 for Buffalo, 0.73 for Rochester New York, 11.81 for Syracuse, 19.31 for Boston, 5.93 for Portland, 8.45 for Saint Louis, 16.45 for Cleveland, 3.95 for Manhattan, 24.65 for San Francisco, 6.17 for Oakland, negative 0.98 for Salt Lake City, 1.18 for Denver, negative 2.25 for Gary, 3.24 for Grand Rapids, 5.24 for Indianapolis, 4.53 for Toledo, 4.66 for Columbus, 2.99 for Bronx, 11.92 for Queens, negative 2.56 for San Jose, negative 3.42 for Tulsa, 5.35 for Saint Paul, 2.63 for Louisville, negative 11.14 for Dayton, 12.64 for Pittsburgh, 5.14 for Staten Island, negative 1.11 for Brooklyn, 1.77 for Los Angeles, 6.62 for San Diego, 7.30 for Oklahoma City, 3.11 for Dallas, 5.09 for Memphis, 11.80 for Baltimore, 10.68 for Philadelphia, negative 3.32 for Richmond, negative 0.95 for Norfolk, negative 0.27 for Fort Worth, negative 0.35 for Austin, negative 1.43 for Birmingham, 3.35 for Atlanta, 5.70 for Nashville, negative 3.24 for San Antonio, 0.28 for Houston, 4.77 for New Orleans. 8.34 for Tampa, 6.90 for Saint Petersburg, and 28.41 for Miami. The following cities are highlighted: San Francisco, San Jose, Chicago, Tulsa, San Antonio, Buffalo, Dayton, Richmond, Miami, Boston. A scale depicts the kilometer ranges from 0 to 1000. Panel B, titled associations between the home owners’ loan corporation grades and adult asthma prevalence open parenthesis uppercase d graded tracts relative to uppercase a and uppercase b graded tracts closed parenthesis, depicts the latitude and longitude positions of the cities: 0.64 open parenthesis 0.24, 1.03 closed parenthesis for Seattle, 0.89 open parenthesis negative 0.01, 1.77 closed parenthesis for Spokane, negative 0.33 open parenthesis negative 0.89, 0.23 closed parenthesis for Detroit, 0.62 open parenthesis negative 0.15, 1.39 closed parenthesis for Rochester New York, 0.47 open parenthesis 0.22, 0.72 closed parenthesis for Manhattan, 1.08 open parenthesis 0.42, 1.73 closed parenthesis for Portland, 0.38 open parenthesis negative 0.25, 1.01 closed parenthesis for Salt Lake City, 0.38 open parenthesis negative 0.29, 1.05 closed parenthesis for Minneapolis, 0.12 open parenthesis negative 1.60, 1.84 closed parenthesis for Grand Rapids, 0.70 open parenthesis negative 0.93, 2.32 closed parenthesis for Syracuse, 0.87 open parenthesis 0.58, 1.17 closed parenthesis for Bronx, 0.91 open parenthesis negative 0.07, 1.89 closed parenthesis for Columbus, 0.77 open parenthesis negative 0.22, 1.77 closed parenthesis for Gary, 0.76 open parenthesis 0.44, 1.08 closed parenthesis for Chicago, 0.74 open parenthesis 0.26, 1.23 closed parenthesis for Saint Paul, 0.62 open parenthesis negative 1.34, 2.58 closed parenthesis for Buffalo, 0.35 open parenthesis negative 0.22, 0.93 closed parenthesis for Boston, 0.45 open parenthesis 0.12, 0.79 closed parenthesis for Oakland, 0.17 open parenthesis negative 0.30, 0.63 closed parenthesis for Denver, 0.71 open parenthesis 0.07, 1.89 closed parenthesis for Kansas City, 1.41 open parenthesis 0.57, 2.25 closed parenthesis for Indianapolis, 1.14 open parenthesis negative 1.67, 3.96 closed parenthesis for Toledo, 1.72 open parenthesis 1.17, 2.26 closed parenthesis for Cleveland, 0.58 open parenthesis 0.34, 0.83 closed parenthesis for Brooklyn, 0.58 open parenthesis 0.30, 0.85 closed parenthesis for San Francisco, negative 0.12 open parenthesis negative 0.52, 0.27 closed parenthesis for San Jose, 0.81 open parenthesis 0.03, 1.60 closed parenthesis for Tulsa, 0.96 open parenthesis 0.02, 1.90 closed parenthesis for Saint Louis, 1.79 open parenthesis 0.66, 2.93 closed parenthesis for Louisville, 1.13 open parenthesis 0.03, 2.24 closed parenthesis for Dayton, 1.05 open parenthesis 0.26, 1.84 closed parenthesis for Pittsburgh, negative 0.05 open parenthesis negative 0.51, 0.39 closed parenthesis for Staten Island, negative 0.04 open parenthesis negative 0.21, 0.12 closed parenthesis for Los Angeles, 0.33 open parenthesis negative 0.43, 1.10 closed parenthesis for Oklahoma City, 1.14 open parenthesis 0.25, 2.03 closed parenthesis for Memphis, 0.39 open parenthesis negative 0.19, 0.96 closed parenthesis for Baltimore, 1.45 open parenthesis 0.46, 2.45 closed parenthesis for Richmond, 0.41 open parenthesis 0.05, 0.78 closed parenthesis for Philadelphia, 1.39 open parenthesis 0.94, 1.84 closed parenthesis for Queens, 0.14 open parenthesis negative 0.09, 0.38 closed parenthesis for San Diego, negative 0.55 open parenthesis negative 1.23, 0.13 closed parenthesis for Fort Worth, 1.28 open parenthesis 0.12, 2.44 closed p.
Discussion
We observed increased respiratory hazards and higher self-reported adult asthma prevalence in historically disinvested neighborhoods. The variation across cities in this association indicates the importance of local context. This heterogeneity is largely consistent with existing literature,3,7 because HOLC risk grade assignment varied substantially by metropolitan area. Because HOLC maps are only one example of a changing body of policies and practices from the New Deal era, differences in present day health from city to city are likely influenced by local or regional policies in intervening post-HOLC decades.8
A previous study examined associations between HOLC risk grades and asthma-related respiratory outcomes, finding rates of asthma-related emergency room visits were 2.4 times higher in D-grade than in A-grade census tracts in California.9 Another study examined HOLC grades and asthma prevalence in select U.S. cities; as in our study, their results indicated heterogenous associations.10 Our results expand on these analyses by comparing the magnitude of associations in cities across the United States. We find that the influence of a racist historical policy related to neighborhood disinvestment on modern-day hazardous air pollutants and self-reported adult asthma prevalence varies substantially across metropolitan areas.
D-graded census tracts were associated with elevated levels of hazardous pollutants or asthma prevalence in a majority (90%, n=133) of cities we examined. Due to the heterogeneous city-by-city findings, we did not control for modern sociodemographic characteristics (e.g., household income) that may mediate the association between historical HOLC grades and self-reported adult asthma prevalence.
Future policies to promote urban respiratory health should incorporate flexible strategies to mitigate inequitable air quality downstream from historic neighborhood disinvestment.
Acknowledgments
The authors would like to thank J. Ciminelli (University of Rochester) for input on an earlier draft of the manuscript and E. Volkin (University of Rochester) for excellent research assistance and feedback on an earlier draft of the manuscript.
E.J.C. and M.D.W. developed the study design and conducted the literature review. E.J.C. was responsible for writing the code, leading the data analysis, and drafting the manuscript. K.D.S., E.L.H., and M.D.W. provided detailed input on the conceptual framework and analytic decisions. All authors contributed to interpretation of findings, writing, and editing of final manuscript.
All data are from public sources that are referenced in the manuscript. Code for data processing and analysis is available in a GitHub repository (https://github.com/erincampbell234/Disinvestment-Asthma).
==== Refs
References
1. Greer J. 2013. The Home Owners’ Loan Corporation and the development of the residential security maps. J Urban Hist 39 (2 ):275–296, 10.1177/0096144212436724.
2. Bailey ZD, Feldman JM, Bassett MT. 2021. How structural racism works—racist policies as a root cause of U.S. racial health inequities. N Engl J Med 384 (8 ):768–773, PMID: , 10.1056/NEJMms2025396.33326717
3. Michney TM, Winling L. 2020. New perspectives on new deal housing policy: explicating and mapping HOLC loans to African Americans. J Urban Hist 46 (1 ):150–180, 10.1177/0096144218819429.
4. Lane HM, Morello-Frosch R, Marshall JD, Apte JS. 2022. Historical redlining is associated with present-day air pollution disparities in U.S. cities. Environ Sci Technol Lett 9 (4 ):345–350, PMID: , 10.1021/acs.estlett.1c01012.35434171
5. Nardone A, Rudolph KE, Morello-Frosch R, Casey JA. 2021. Redlines and greenspace: the relationship between historical redlining and 2010 greenspace across the United States. Environ Health Perspect 129 (1 ):17006, PMID: , 10.1289/EHP7495.33502254
6. Nelson RK, Winling L, Marciano R, Connoly N. 2020. Mapping Inequality. https://dsl.richmond.edu/panorama/redlining/ [accessed 23 October 2020].
7. Xu W. 2021. Legacies of institutionalized redlining: a comparison between speculative and implemented mortgage risk maps in Chicago, Illinois. Hous Policy Debate 0 (0 ):1–26.
8. Graetz N, Esposito M. 2022. Historical redlining and contemporary racial disparities in neighborhood life expectancy. Soc Forces soac114 . 10.1093/sf/soac114.
9. Nardone A, Casey JA, Morello-Frosch R, Mujahid M, Balmes JR, Thakur N. 2020. Associations between historical residential redlining and current age-adjusted rates of emergency department visits due to asthma across eight cities in California: an ecological study. Lancet Planet Health 4 (1 ):e24–e31, PMID: , 10.1016/S2542-5196(19)30241-4.31999951
10. Nardone A, Chiang J, Corburn J. 2020. Historic redlining and urban health today in U.S. cities. Environ Justice 13 (4 ):109–119, 10.1089/env.2020.0011.
|
PMC010xxxxxx/PMC10032045.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36946580
EHP10757
10.1289/EHP10757
Research
The Health Effects of 72 Hours of Simulated Wind Turbine Infrasound: A Double-Blind Randomized Crossover Study in Noise-Sensitive, Healthy Adults
https://orcid.org/0000-0002-9014-1397
Marshall Nathaniel S. 1 2 *
Cho Garry 1 *
Toelle Brett G. 1 2
Tonin Renzo 1 3
Bartlett Delwyn J. 1 2
D’Rozario Angela L. 1 4
Evans Carla A. 1
Cowie Christine T. 1 5 6
Janev Oliver 1
Whitfeld Christopher R. 1
Glozier Nick 2 7
Walker Bruce E. 8
Killick Roo 1
Welgampola Miriam S. 2 7
Phillips Craig L. 1 9
Marks Guy B. 1 5 6
Grunstein Ronald R. 1 2 7
1 Woolcock Institute for Medical Research, Sydney, New South Wales, Australia
2 Faculty of Medicine and Health, University of Sydney, Sydney, New South Wales, Australia
3 Renzo Tonin Associates, Sydney, Australia (Retired)
4 School of Psychology, Faculty of Science, University of Sydney, Sydney, New South Wales, Australia
5 Ingham Institute for Applied Medical Research, Liverpool, New South Wales, Australia
6 South West Sydney Clinical School, University of New South Wales, Sydney, New South Wales, Australia
7 Royal Prince Alfred Hospital, Camperdown, Sydney, New South Wales, Australia
8 Channel Islands Acoustics, Santa Barbara, California, USA (Retired)
9 School of Medicine, Macquarie University, Sydney, New South Wales, Australia
Address correspondence to Nathaniel S. Marshall, The Woolcock Institute of Medical Research, Macquarie University, NSW, Australia. Email: [email protected]
22 3 2023
3 2023
131 3 03701208 12 2021
09 2 2023
21 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Large electricity-generating wind turbines emit both audible sound and inaudible infrasound at very low frequencies that are outside of the normal human range of hearing. Sufferers of wind turbine syndrome (WTS) have attributed their ill-health and particularly their sleep disturbance to the signature pattern of infrasound. Critics have argued that these symptoms are psychological in origin and are attributable to nocebo effects.
Objectives:
We aimed to test the effects of 72 h of infrasound (1.6–20 Hz at a sound level of ∼90 dB pk re 20μPa, simulating a wind turbine infrasound signature) exposure on human physiology, particularly sleep.
Methods:
We conducted a randomized double-blind triple-arm crossover laboratory-based study of 72 h exposure with a >10-d washout conducted in a noise-insulated sleep laboratory in the style of a studio apartment. The exposures were infrasound (∼90 dB pk), sham infrasound (same speakers not generating infrasound), and traffic noise exposure [active control; at a sound pressure level of 40–50 dB LAeq,night and 70 dB LAFmax transient maxima, night (2200 to 0700 hours)]. The following physiological and psychological measures and systems were tested for their sensitivity to infrasound: wake after sleep onset (WASO; primary outcome) and other measures of sleep physiology, wake electroencephalography, WTS symptoms, cardiovascular physiology, and neurobehavioral performance.
Results:
We randomized 37 noise-sensitive but otherwise healthy adults (18–72 years of age; 51% female) into the study before a COVID19-related public health order forced the study to close. WASO was not affected by infrasound compared with sham infrasound (−1.36 min; 95% CI: −6.60, 3.88, p=0.60) but was worsened by the active control traffic exposure compared with sham by 6.07 min (95% CI: 0.75, 11.39, p=0.02). Infrasound did not worsen any subjective or objective measures used.
Discussion:
Our findings did not support the idea that infrasound causes WTS. High level, but inaudible, infrasound did not appear to perturb any physiological or psychological measure tested in these study participants. https://doi.org/10.1289/EHP10757
Supplemental Material is available online (https://doi.org/10.1289/EHP10757).
* These authors contributed equally to this work.
All of the authors have superannuation accounts which are compulsory in Australia and these accounts may contain investments in both traditional and renewable energy, including wind turbines. R.T. is the founding principal of Renzo Tonin Associates who have previously worked as consultants for the NSW Department of Planning on several wind farms in NSW, Australia. None of the investigators have any other pecuniary interest or academic conflicts of interest in the outcomes of this study.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Large electricity-generating wind turbines emit both audible sound and inaudible infrasound at very low frequencies that are outside of the normal human range of hearing. There have been several studies published recently on the effects, or lack thereof, of audible wind turbine sound on sleep1–4 including a meta-analysis incorporating several of these studies with wake after sleep onset (WASO) as core outcome.5 Despite these previous studies generally finding no effect of audible wind turbine noise (WTN) on WASO, other observers6–8 have proposed that people who live in the vicinity of wind turbines suffer from wind turbine syndrome (WTS described in this case series9) with dizziness, sleep disturbance, and other symptoms. The causes of this syndrome have been the subject of substantial international controversy.10–13 Proponents have contended that the symptoms that compose this syndrome are caused by low frequency subaudible infrasound generated by wind turbines.14 Critics have argued that these symptoms are psychological in origin and are attributable to nocebo effects.9,15–18 The Australian National Health and Medical Research Council (NHMRC) Wind Farms and Human Health Reference Group concluded that the available evidence was not sufficient to establish which, if either, of these explanations is correct.10 One of the reasons for this is because there had been to our knowledge no robustly designed, double-blind, controlled, and adequately powered studies of the health effects of exposure to sustained infrasound of the type emitted by a wind turbine. This absence of evidence has the potential to complicate initiatives to decarbonize electricity generation due to community concerns about the possibility for this measurable, but inaudible, infrasound to affect their health.
Our principal hypothesis was that exposure to infrasound in healthy individuals, at a level of ∼90 dB pk re 20μPa compared with the sham infrasound, increases WASO—a measure of sleep disturbance—and worsens other measures of sleep quality, mood, WTS symptoms, and other electrophysiological measures. In addition, as a positive control, we also tested whether audible traffic noise, a mixture of road (motorbike, truck, car) and aircraft noise (at a sound level of 40–50 dB LAeq,night and 70 dB LAFmax transient maxima) had an adverse impact on these same outcomes, when compared with sham infrasound.
Methods
Study Design, Setting, and Ethics
This was a randomized, double-blind, three-way crossover study (equal ratios of all possible sequences) in noise-sensitive participants exposed during three noncontiguous 72-h periods to a) wind turbine simulated infrasound at ∼90 dB peak, with reference to 20μPa in air (test exposure); b) no added sound (sham, negative control, speakers not generating infrasound); and c) traffic noise [positive control mixture of road and air traffic with background levels of 40–50 dB LAeq, at night (2200–0700 hours) and 70 dB LAFmax transient maxima]. The metrics LAeq and LAFmax are common measures used to indicate the noise level of traffic noise but are not applicable to infrasound, which has a frequency range of <20 Hz. Although any number of metrics could be used to measure the amplitude of infrasound, in this study the use of dB pk is adopted, corresponding to the peak amplitude of the infrasound waveform, to enable comparison with the dB pk level measured at a working wind farm, as described in more detail in the section “Experimental Exposures (Infrasound, Sham, and Traffic).”
The study was conducted exclusively at the Woolcock Institute of Medical Research, Glebe, New South Wales, Australia, in our sleep laboratory which is shielded from external sound. The ambient underlying noise level in the sleep laboratory in the nighttime period (i.e., without the generated traffic noise) was ∼39 dB LAeq,night with the air-conditioning operating in the room. Air-conditioning was not turned off during the experiment to prevent ambient elevated temperature disturbing sleep. The ambient underlying level of infrasound was 80–85 dB pk predominantly in the frequencies of <1 Hz (i.e., below the frequency range of the simulated infrasound).
During each test period starting around noon, the participants were subjected to one of the three noise conditions continuously for 72 h (including 3 normal nocturnal sleep periods) without leaving the testing setting (bedroom with ensuite approximating a studio apartment). The testing protocol within each 72-h period is described in Figure S1. Each noise condition was separated by a >10-d washout period during which participants lived normally outside of the laboratory environment.
This study was registered in the Australian and New Zealand Clinical Trials Registry (ACTRN12617000001392) before the first participant was randomized. A full copy of the final study protocol followed throughout the study without amendment and was lodged in the registry before the first participant was randomized after it was approved by the Sydney Local Health District Ethics Committee at The Royal Prince Alfred Hospital (protocol nos. X16-0073 and HREC/16/RPAH/91) and was performed in accordance with the Declaration of Helsinki, the 2006 Australian Clinical Trials Handbook Version 1.0, and the guidelines of the NHMRC for human research.19 Participants also gave written informed consent. Participants were offered an AUD1,000 payment upon completion (∼AUD4.55/h in the laboratory) and the study paid for all transportation and meal costs.
Randomization Sequence, Allocation Concealment, and Blinding
The randomization sequence was a simple 1:1:1 ratio with no blocking that was computer generated by an investigator (C.T.C.) who played no role in participant selection or data collection and never met any participants or undertook statistical analysis. The sequence was secured in our password-protected research database and was only accessible by that investigator and the acoustic and data engineers who played no role in the screening for eligibility or the decision to randomize participants. The principal investigator (PI) and lead study coordinator assessed eligibility and consent before the PI irrevocably recorded the decision to randomize against the participants’ screening number and this was recorded by the study database.
Participants and study staff were blinded to the infrasound and sham infrasound exposures (given that the infrasound is inaudible). Study investigators and outcome assessors/processors were blinded to study allocation by using engineering staff who were solely responsible for delivering the exposures based on the randomization schedule and who never disclosed to any other study staff what exposure had been used, would be used, or was occurring. Infrasound or sham generation were controlled from a locked box kept outside the participants’ room only accessible by the engineers. Engineering staff did not meet the participants. The audible positive control (loud traffic noise) by its nature cannot be subject to either participant or investigator blinding and was an open-label exposure.
We asked all of the participants and staff during our informal conversations with them whether they could sense that infrasound was playing. None of the participants said they could sense infrasound was playing. None of the staff who processed the electrophysiology measurements [including polysomnography (PSG)] or were in the presence of the participants reported being able to tell whether infrasound or sham infrasound was being delivered. Statistical analyses were undertaken by investigators (N.S.M. and G.C.) who were also unable to tell which condition was infrasound. Exposures were labeled numerically (1 vs. 2 vs. 3) in the database rather than descriptively (traffic vs. infrasound vs. sham) until after the principal analyses were completed and presented to the chief investigators.
Participants
We undertook a two-stage screening process (online and clinical screening). Online screening selected for adults who were fluent in English, ≥18 y old, and who were noise sensitive (21-Question Weinstein Noise Sensitivity Scale, with a score above the reported mean for the measurement of >58),20 did not have severe insomnia (7-question Insomnia Severity Index of ≤18),21 or any other detectable medical or psychiatric morbidity (including caffeine, alcohol, tobacco, or hypnotic dependence) that would preclude residing in a controlled environment for 72 h. We also excluded women who reported being pregnant or breastfeeding owing to the unknown risks to unborn children and babies. To benchmark the sample for mood and sleepiness levels, we also measured the Depression Anxiety and Stress Scale (DASS-21) and the Epworth Sleepiness Scale (ESS).22,23 We measured participants’ attitudes to windfarms (“How concerned are you about the health effects of infrasound generated from wind farms?”) with a Likert scale (0–6, with 0 being completely unconcerned to 6, extremely concerned); we planned to test this in the context of a main effect of infrasound as a potential explanatory variable (Figure S2).
Online-screened eligible participants were invited to proceed to the clinical screening stage. The clinical screening involved sending participants an Actiwatch 2 (Philips Respironics) using light and movement measurements to determine at-home sleep/wake patterns which were then clinically interpreted by a sleep psychologist (D.J.B.). Instructions were provided on how to use the watch and to wear it for a minimum period of 7 nights. Participants were excluded if that actigraphy recording showed abnormal sleep patterns, such as unusually short/late/early and shift work and/or transmeridian travel within 2 wk preceding randomization. Eligible participants were then invited to attend an in-person screening at the Woolcock Institute of Medical Research for audiological testing, psychological review and, if relevant, sign informed consent documents. Psychological screening was conducted by a psychologist (D.J.B.) who interviewed participants and reviewed their psychological history and questionnaire data [including assessments of claustrophobia (the Claustrophobia Questionnaire or CLQ)]24 and determined whether they were likely to tolerate 72 h in a studio apartment style accommodation.
Audiological screening was conducted to exclude any participants who had an existing hearing loss. Only those with normal audiometry in the opinion of the audiologist and neurotologist (M.S.W.) were included. As per a standard audiological examination, otoscopy, tympanometry, and pure tone audiometry (PTA) were conducted.
Otoscopy was conducted to check for the presence of excessive wax or anything that could obstruct probe measurements for tympanometry. Tympanometry provides quantitative information on the function of structures and the presence of fluid in the middle ear and was done to exclude those with underlying middle ear pathology that may impact upon sound transmission to the inner ear. Tympanograms were recorded using a Madsen OTOflex 100 (Natus Medical) system with a 226-Hz probe tone.25 Static compliance (in millimho), middle ear pressure (in decapascals) and ear canal volume (in cubic centimeters) were recorded. Only those with clinically normal middle ear function as indicated by equipment norms (static compliance of ≥0.3 mmho and middle ear pressure of between –100 and 100 daPa) were included in the study.
PTA was conducted using a personal computer-based audiometer (Oscilla USB-350B; Inmedico A/F), and thresholds were obtained following standard clinical practice using the Hughson-Westlake procedure.26 Thresholds were recorded at 250, 500, 1,000, 2,000, 4,000, and 8,000 Hz for air conduction (AC) and at 500, 1,000, 2,000, and 4,000 Hz for bone conduction (BC); 125 Hz AC thresholds were not tested owing to the risk of vibrotactile stimulation, which could affect the reliability of the results. Only participants with AC and BC thresholds at ≤20 dB Hearing Level (HL) (i.e., clinically normal hearing) were included in the study.
Experimental Exposures (Infrasound, Sham, and Traffic)
The infrasound attributable to wind turbines was simulated digitally using a trapezoidal-shaped waveform with 16 harmonics in the frequency range of 0.8–20 Hz at a sound level of ∼90 dB pk re 20μPa (measurable but inaudible to all participants).16 This infrasound level is higher than what has been recorded both inside and outside a dwelling where people have previously reported WTS from exposure experienced at 1,100 ft (335 meters) from a wind turbine located at the Shirley Wind Farm, Wisconsin, USA.14,27 The Shirley wind project has eight Nordex100 wind turbines. The simulated wind turbine infrasound was generated by a Teensy microprocessor fitted with an SGTL5000 audio processor and the signal fed to a purpose-built Direct Current (DC)-coupled class D amplifier and four 18-in JBL subwoofer loudspeakers in four fully sealed timber enclosures faced with heavy protective mesh so that the participants could not observe the speakers in operation. The simulated wind turbine infrasound comprised sinusoidal harmonics in the frequency range specified with monotonically decreasing amplitude and selected phase shift, resulting in a trapezoidal waveform as observed in field measurements (Figure S3).5 The sham infrasound exposure involved use of the same equipment but with the loudspeakers wired in antiphase so that the cones moved by an equivalent amount but did not generate infrasound.
Because of the limitation in the physical size and type of loudspeakers, the 0.8-Hz fundamental frequency could not be generated at the required sound level and so the frequencies generated commenced at the second harmonic at 1.6 Hz. The peak sound level was nonetheless maintained as specified.
A mixture of road (motorcycles, trucks, cars) and aircraft noise (referred to herein as traffic noise) was generated via a Bose audio sound system and audio digital player at a sound level of 40–50 dB LAeq,night and 70 dB LAFmax (loud enough to disrupt normal conversation). The traffic noise was timed specifically to disrupt the last 3 h of sleep as a positive control condition to increase WASO. The six sound files that constituted this exposure are available as supplemental audio files 1–6 upon request to the authors. The counts of sound events labeled airplane, motorbike, truck, or car can be found in Figure S4 for the full 72 h. Traffic noise events occurred approximately three to four times an hour during daylight and early sleep but, between ∼0400 and 0700 hours, the traffic noise events occurred closer to 15 events an hour (for a person with a normal wake time about 0700 hours). The events in the last 3 h of sleep were clustered noise events occurring one after the other. A typical noise event outside of the time 0400–0700 hours would be experienced as a truck going past. But between the 0400–0700 hours, the experience would be six trucks going past one after the other.
The infrasound sound levels were measured continuously above the pillow of each participant using a GRAS 46AZ one-half inch low frequency microphone and preamplifier set connected to the microprocessor. Software was used to enable the overall dB pk sound level and narrow band frequency analysis to be measured continuously and stored at 1-s intervals for later processing. The sound levels of the traffic noise were measured using an NTi XL2 sound level meter (having a frequency bandwidth of 10 Hz–20 kHz) with its microphone placed adjacent to the infrasound microphone. There was no attempt to address noise from the subjects such as snoring, talking in their sleep, or turning over in bed because it was considered that this extraneous noise would not contribute to the LAeq,night, which is measured over the whole nighttime period. A review of the sound levels was made after each session to identify any unusual extraneous noise. The equipment was calibrated before and after each test period with a Bruel & Kjaer 4231 sound level calibrator to confirm there was no drift in calibration. After the experiment, an acoustic engineer (O.J.) compared the exposure received to the exposure allocated by the randomized sequence to check for correct allocation and also for any equipment malfunction or performance issues.
Neurophysiological Measurements of Sleep and Wake
The primary and secondary outcomes were neurophysiological measurements of sleep and wake. Participants were measured using PSG at three time points per visit (1 sleep study per night). PSG was measured using Alice 6 LDX and Sleepware G3 (Philips Respironics) at electroencephalographic (EEG) derivations (F3-A2, Fz-A2, F4-A1, C3-A2, Cz-A2, C4-A1, Pz-A2, O1-A2, Oz-A2, O2-A1), left and right electrooculogram (EOG), chin electromyogram (EMG), and lead II electrocardiogram (ECG). Additional PSG sensors, such as nasal cannula and thermistor (to measure airflow), thoracic and abdomen respiratory bands, and leg EMG were measured during the first night of each visit. For nights 2 and 3, only EEG, EOG, ECG, and Chin EMG were measured. The PSG was manually scored by a sleep technician blinded to noise conditions (except traffic) according to the American Academy of Sleep Medicine guideline that was operational at the time when the study started (version 2.2).28
To reduce laboratory-induced variability in sleep behavior, individualized habitual sleep and wake times were calculated based on the average sleep and wake times from the 7 d of actigraphy measured during screening. Participants’ sleep opportunity period in the laboratory protocol was calculated using their habitual sleep and wake times with an additional 30-min windows on each end. Participants could then elect a set time inside those 30-min windows to be their go-to-sleep and wake-up times that they would then follow for the duration of the protocol.
Our primary outcome, WASO, was measured in minutes and was calculated from the first epoch of any sleep detected on the polysomnogram until the last epoch of any sleep detected on the PSG or the participant’s individually calculated habitual wake-up time, whichever happened last. Sleep technicians and staff were instructed to keep in-room disruption (e.g., fixing signals or physiological sensors) to a minimum to reduce artificially induced wake impacting the WASO calculation. All other PSG outcomes were scored according to the guideline and analyzed in the order in which they were listed in the preregistration.
EEG microstructure (power spectral density and spindle density) was analyzed by importing the PSG as a European data format (EDF) and sleep stage file into an in-house–built program.29–31 Preprocessing consisted of manual artifact detection and visual signal quality checks prior to exporting the EDF file. If judged that >25% of the primary EEG channel (Cz-A2) was artifact, we substituted EEG channel (C3-A2). Where both EEG channels were deemed nonvalid, PSG recordings were excluded from analysis. The program calculates power for five frequency bands [beta (β)frequency band=15–32 Hz; sigma (σ)frequency band=12–15 Hz; alpha (α) frequency band=8–12 Hz; theta (θ) frequency band=4.5–8 Hz; delta (δ) frequency band=0.5–4.5 Hz] in both rapid eye movement stage (REM) and non-REM sleep. The output of this program provides the “power” describing the density of each predefined EEG frequency bands during the PSG recording. This program also automatically detects the number of sleep spindles per minute during non-REM sleep (11–16 Hz) and fast and slow frequency spindles (fast 14–16 Hz; slow 11–13 Hz). Sleep spindles are a distinctive feature of non-REM sleep that are thought to play a role in memory and sleep stability.
Additional Secondary and Tertiary Outcome Measures
Neurocognitive battery and affective scales.
The battery of tests was repeated 12 times per visit and therefore a maximum of 36 times per participant. The exact sequence and timing of the battery are provided in Figure S1 and Table S2.
VAS.
Participants were asked to plot on a 100-mm scale/line to indicate the momentary severity of symptoms reported in WTS (see Figure S5 for an example). The distance from the left end of the line to the participants’ response was recorded in millimeters. The 19 symptoms measured were: Headaches, Ringing in the ear, Itchy Skin, Blurred Vision, Dizziness, Racing Heart, Nausea, Tiredness, Feeling Faint, Sleepiness, Difficulty Concentrating, Difficulty Remembering, Fatigue, Irritability, Muscle Spasms, Disruption while falling asleep, Awakening from Sleep, and Anxiety. We also measured an additional visual analog scale (VAS) asking how annoying the noise is right now (Figure S5). The statistical processing of the 20 VAS is described in the “Statistical Analyses” section.
Karolinska Drowsiness Test.
Participants were asked to look at a dot directly in front of them at eye level (∼1m away) for 2.5 min followed by 2.5 min with eyes closed and, last, another 2.5 min with their eyes opened. EEG was simultaneously measured (Alice Respironics G3 Sleepware). The alpha density in the eyes-open and the eyes-closed states, as measured in the EEG evaluates the level of wakefulness in participants during the course of the test. EEG data was exported into an EDF that was processed using an in-house analysis program.29,30
N-back.
The N-back test is a computerized test which involved the participant monitoring a series of stimuli (letters) and requires them to respond whenever a stimulus was presented in the same location as the one presented two trials previously. The task requires monitoring, updating, and manipulation of memorized information and is, therefore, assumed to place great demand on a number of key processes within working memory. The total number of correct responses out of 48 trials is the outcome. Before unblinding, we excluded all scores <10 correct as evidence that a participant had not followed instructions.
Tower of London.
This is a neuropsychological test of planning, executive/spatial function, and working memory.32,33 The participant is instructed to move three colored balls on three pegs from an initial state to a goal state in the minimum number of moves necessary. Each trial contains multiple problems of increasing difficulty, requiring the participant to set more subgoals to reach the illustrated goal and requires preplanning. Each trial has a prespecified number of moves that the participant must aim to complete the trial within. The variables analyzed in this task are the mean of planning + execution time and the grand mean number of errors across all puzzles attempted.
Psychomotor vigilance task.
This test is a measurement of simple reaction time (RT) over a 10-min period using an independent hand-held box with an light emitting diode (LED) display and two buttons that can be depressed using your thumbs. Participants are instructed to respond to a red LED display counting up in milliseconds that is stopped by pressing the right button as fast as possible. Stimulus delivery is randomly spaced between 2 and 10 s after the last one. An overall mean RT over the 10 min of the test period is generated (mean RT) and reciprocated to normalize the distribution.34,35
End-of-visit questionnaire.
Participants were asked four computerized questionnaires before exiting the laboratory environment at the end of each visit. These questionnaire, which are described in Table S1 are the Insomnia Severity Index (modified to a 3-day version), Kessler-10, Warwick—Edinburgh Mental Wellbeing Scale, Depression Anxiety Scale, and Stress Scale DASS-21).
Cardiovascular and Blood Measurements
Twenty-four–hour blood pressure.
Participants had ambulatory blood pressure measured over a 24-h period each visit using the Oscar2 device (SunTech Medical). Blood pressure measurements were taken at the brachial location, using a blood pressure cuff sized according to each participant’s arm. This procedure occurred from the morning of the second day to the third day of each visit. Blood pressure readings occurred at half-hourly intervals while participants were awake and hourly when participants were asleep (50 measurements per visit). Participants were instructed to keep their arm relaxed during readings and were notified there would be two inflations of the cuff (arterial and central blood pressure). Data was recorded and downloaded using provided programs by SunTech Medical and exported into comma-separated files for SAS import and statistical analysis.
Endothelial function test.
This procedure was undertaken once per visit at approximately noon after the last night and ∼2h after the completion of noise exposure. In this 15-min test, participants were required to rest in a supine position on a bed while vascular tone was measured at the index fingers of each hand using the EndoPAT device (Itamar Medical). Vascular tone was measured at a baseline resting state at both fingers for 5 min. Following the reading at rest, a blood pressure cuff was inflated for 5 min on one arm to a pressure of 80 mmHg above their resting blood systolic pressure or 200 mmHg, whichever occurred first. A baseline blood pressure measurement was taken prior to the test to determine the pressure of cuff inflation. Following the occlusion period, the blood pressure cuff was fully deflated to allow measurement of endothelial mediated vasodilation (endothelial function). The cuff then remained deflated for another 5 min of rest.
Blood-based cardiovascular markers.
A fasted blood sample was collected on the final morning of each visit. Due to infection and safety procedures surrounding venipuncture, participants were briefly taken away from the noise exposure to perform blood draws in a separate room. Blood samples were commercially processed using standard assays by a local pathology company (Laverty Pathology, Sydney, Australia). Four serum variables were prespecified as of interest and were entered from the pathology reports: cortisol [a potential measure of stress; limit of detection (LOD) was 5.5 nmol/L], highly sensitive C-reactive protein (HsCRP; a measure of inflammation; LOD of 0.1mg/L), glucose and insulin (measures of metabolic homeostasis that might be perturbed by sleep disruption; the glucose LOD was 0.1 mmol/L and the insulin LOD was 0.2 mU/L). When a value was reported as being below the level of detection for any of these variables, we entered that value as being at that level (except HsCRP). Only HsCRP was ever reported with a lower than detectible level result (see the “Results” section).
Pulse wave velocity.
This procedure was undertaken once per visit at ∼1000 hours the morning after the second night and while in the presence of the speakers. This 10-min test required participants to be supine on the bed with a baseline rest period of 5 min prior to measurements using the SphygmaCor-XCEL (AtCor medical). A blood pressure cuff was attached around the participants’ thigh, and their carotid pulse was located using a high-fidelity pressure tonometer at the neck on the same side of the body. The carotid and femoral pulse waves were then simultaneously acquired from the thigh cuff and the tonometer. Pulse wave velocity was calculated based upon the distance between the middle of the thigh cuff and the site of the carotid pulse using established techniques.36 Recordings were deemed technically acceptable if pulse waves could be simultaneously acquired for at least 10 s.
Twenty-four–hour urinary catecholamines.
On the morning of the third day, participants were provided with a 24-h urine collection container and asked to void their bladder immediately prior to urine collection commencement. All urinary output was then collected in the container until the following morning. Urinary catecholamines were commercially processed using standard assays used by a local pathology company (Laverty Pathology, Sydney, Australia). The pathology service assayed these urine samples for creatinine, noradrenaline, adrenaline, and dopamine. Noradrenaline, adrenaline, and dopamine levels were divided by the amount of creatinine for statistical analyses. The minimum levels of detectable noradrenaline and adrenaline values are ∼14.0 nmol/L, creatinine is detectable at 2 mmol/L, and dopamine is detectable above levels of ∼200 nmol/L. Values at the LOD indicated in the pathology report were entered in as being at that point (e.g., <38 nmol/24h of adrenaline was data entered as 38).
Statistical Analyses
In our prestudy planning, we calculated that in a crossover study design, a sample size of 38 participants would provide ∼85% power to detect a difference in the primary end point, WASO, of 15 min between infrasound and sham infrasound (Cohen’s d=0.5). Our analysis code is available in the University of Sydney’s data repository and can be accessed on request of the authors. Repeated linear mixed models were performed for the primary and secondary end points in SAS (version 9.4; SAS Institute, Inc.), including all randomized participants in the groups they were randomized to and using the least squares means procedure to address missing data. For the primary and secondary outcomes, Participant Identification Numbers were coded as random effects, and we used the exposure received, the order it was received, the night of exposure and interactions between the exposure by night, and night by order as fixed effects. We regarded p<0.05 as statistically significant, and exposure effects were tested using the least squares means option. Tertiary outcomes were analyzed using the same model without the night effect or any of its interactions but, instead, included a time-point term indicating how far through the sequence the test was made. Blood pressure had 50 repeated measurements per visit and the repeated daytime test battery data, including symptom VAS scores, had 12 repeated measurements per visit.
Because a large proportion of VAS scores were 0/100 and to reduce the number of end points, a post hoc decision was made to use principal components factor analysis (proc factor). We used squared multiple correlations of each variable with the 19 other variables (including the noise annoyance scale) as the prior communality estimates. We selected five factors to be extracted based on an examination of the scree plot and evaluation of the interpretability of the resultant rotated factors. Varimax orthogonal rotation was implemented after extraction. The five factors were labeled according to the variables that loaded on them, as follows: Fatigue, Irritability, Nausea and Dizziness, Sleep Disturbance, and Tinnitus. Scores for each factor for each subject on each of 36 occasions of measurement were calculated. These factor scores had 3 added to them and then were log-transformed to approximate a normal distribution. The log-transformed factor scores were used as dependent variables in five mixed effects regression models.
Results
Between 19 April 2017 and 23 March 2020, we randomized 37 participants (see Table 1 for demographic information), 2 of whom withdrew themselves from the study after one exposure (see the “Adverse Events” section) and 1 of whom completed infrasound and sham infrasound exposures, but not the positive control exposure because our facility was closed by a COVID-related public health directive (5 missed exposure visits in total). Within conditions data collection was interrupted for 2 participants on night 1 (power cut) and night 2 (a staff scheduling error) that meant we had to send those participants home for 1 night.
Table 1 Participant demographics in the laboratory-based three-arm crossover study of 72 h of exposure to simulated wind turbine infrasound, sham infrasound, and traffic noise.
Variable Descriptive statistics Range
Sex, n
Females 19
Males 18
Age [y (mean±SD)] 32±12 18–72
Weinstein Noise Sensitivity Score (0–126 points) (mean±SD) 83.0±13.4 64–108
Insomnia Severity Scale (0–28 points) (mean±SD) 5.3±4.4 0–16
DASS-2123—Depression (0–21 points) (mean±SD) 3.7±4.3 0–16
DASS-2123—Anxiety (0–21 points) (mean±SD) 2.5±3.5 0–12
Epworth Sleepiness Scale (0–24 points) (mean±SD) 4.7±3.4 0–13
Attitude toward windfarm (0–6) [median] 3 0–6, including 7 participants with scores >3
Note: There were no data missing for any of these variables; baseline data was complete in all participants. Weinstein scores had to be >58 to be eligible (approximately the median value in community-dwelling people, rather than a cutoff value indicative of a person being noise sensitive) and an Insomnia Severity Index (ISI) of ≤18. ISI scores <8 indicate no insomnia; <15, subthreshold insomnia; and <22, moderate severity insomnia. Participants had to be at least 18 years of age. No other metrics reported were subject to inclusion/exclusion for a participant to be eligible. Scores <10 on the DASS Depression scale indicate no depression, scores between 10 and 13 indicate mild depression, and scores between 14 and 20 indicate moderate depression. Scores <8 on the DASS Anxiety scale indicate no anxiety, scores between 8 and 9 mild anxiety, and scores between 10 and 14 moderate anxiety. Scores >10 on the Epworth Sleepiness Scale indicate clinically significant daytime sleepiness, and scores >16 severe daytime sleepiness (see the “Participants” section in the “Methods” section for more detail). Attitude toward windfarm was scored from 0 (completely unconcerned) to 6 (extremely concerned). DASS, Depression Anxiety and Stress Scale; SD, standard deviation.
The lead authors (N.S.M. and G.C.) asked all of the staff and all of the participants whether they were able to differentiate in any way between infrasound and sham infrasound. None of them were able to. The sound engineer checked whether participants were exposed to the condition that they were scheduled to be exposed to in the randomization sequence. For one participant, the first and third visit, traffic and infrasound respectively were inadvertently switched compared with the randomization schedule due to human error. All randomized participants were analyzed for primary and secondary outcomes. Analysis was by intention-to-treat, that is, according to the randomly assigned exposure condition. The model estimated effects of infrasound and traffic noise compared with sham infrasound for all outcomes are shown in Table 2. Because there was no main effect of infrasound on sleep, we did not test for an interaction with prestudy attitudes to the health effects of windfarms. Five of 62 outcomes listed in Table 2 were found to be significantly different in the infrasound exposure (8.1% compared with the false positive rate of 5%): blood pressure [−2.1 mmHg; 95% confidence interval (CI): −2.9, −1.2]; insulin (−1.7 mU/L; 95% CI: −3.3, −0.2); percentage REM (% REM) sleep (1.5%; 95% CI: 0.3, 2.7); Warwick–Edinburgh Mental Wellbeing Scale (1.9 points; 95% CI: 0.2, 3.5); and change in power in the alpha frequency in the eyes-closed condition (−2.00 μV2; 95% CI: −3.95, −0.04). Figure 1 shows the estimated effect of infrasound, sham infrasound, and traffic noise on the primary outcome (WASO) after 3 nights of each exposure, indicating the lack of effect of infrasound on WASO compared with sham and the effect of traffic noise on the first 2 nights. There was no evidence of a first-night effect whereby participants’ WASO were worse on the first night they spent in the laboratory (p=0.63). Figure 2 shows the estimated effects on other standard electrophysiological measures of human sleep quality, including sleep onset latency (Figure 2A), sleep stage shifts (Figure 2B), the arousal index (Figure 2C), and the distribution of the five sleep stages (Wake, N1, N2, N3, and REM; Figure 2D) for each of the three exposures. Figure 3 shows the quantitative analysis of the EEG at Cz (top middle of the scalp) in the five frequency bands during sleep (delta, theta, alpha, sigma, beta) for each exposure in non-REM sleep (Figure 3A), REM sleep (Figure 3B), and the sleep spindle analysis (Figure 3C), again showing the lack of perturbed human sleep quality/continuity that could be attributable to infrasound.
Table 2 Effects of infrasound and traffic noise on outcome measures in 37 participants in the in the laboratory-based three-arm crossover study of 72 h of exposure to simulated wind turbine infrasound, sham infrasound, and traffic noise (n=37).
Variable Units of measurement Sham mean±SEM; actual/potential obs Infrasound mean±SEM; actual/potential obs Traffic mean±SEM; actual/potential obs Difference (95% CI), p-value infrasound vs. sham Difference (95% CI), p-value traffic vs. sham
Wake after sleep onset Min 24.49±3.17; 107/111 23.13±3.16; 108/111 30.56±3.21; 101/111 −1.36 (−6.60, 3.88), 0.601 6.07 (0.75, 11.39), 0.02
Electrophysiological measurements of sleep (secondary outcomes)
Sleep onset latency Min 10.75±2.3; 107/111 13.22±2.2; 108/111 13.76±2.3; 101/111 2.48 (−2.87, 7.82), 0.36 3.0 (−2.4, 8.4), 0.27
Arousal index Events/h 8.79±0.58; 107/111 9.08±0.58; 108/111 9.13±0.59; 101/111 0.29 (−0.58, 1.16), 0.51 0.34 (−0.54, 1.21), 0.45
Sleep stage % Wake % of TST 8.7±0.9; 107/111 8.6±0.9; 108/111 10.0±0.9; 101/111 −0.1 (−1.7, 1.6), 0.95 1.3 (−0.3, 3.0), 0.11
Sleep stage % N1 % of TST 6.3±0.4; 107/111 6.5±0.4; 108/111 6.9±0.4; 101/111 0.2 (−0.3, 0.7), 0.49 0.6 (0.1, 1.1), 0.015
Sleep stage % N2 % of TST 40.6±1.0; 107/111 39.4±1.0; 108/111 41.6±1.0; 101/111 −1.2 (−3.0, 0.6), 0.21 1.0 (−0.8, 2.9), 0.28
Sleep stage % N3 % of TST 23.1±0.8; 107/111 22.4±0.8; 108/111 20.8±0.8; 101/111 −0.7 (−1.9, 0.4), 0.21 −2.4 (−3.5, −1.2), 0.0001
Sleep stage % REM % of TST 21.3±0.6; 107/111 22.8±0.6; 108/111 20.8±0.6; 101/111 1.5 (0.3, 2.7), 0.017 −0.5 (−1.7, 0.7), 0.42
Sleep stage shifts Count 119.7±4.8; 107/111 116.2±4.8; 108/111 123.9±4.8; 101/111 −3.5 (−9.5, 2.4), 0.24 4.2 (−1.8, 10.2), 0.17
Delta power—NREM μV2 481.68±35.85; 106/111 487.28±35.81; 107/111 467.10±36.02; 100/111 5.61 (−27.04, 38.25), 0.74 −14.58 (−47.67, 18.52), 0.39
Theta power—NREM μV2 42.61±2.53; 105/111 41.93±2.53; 107/111 41.99±2.53; 100/111 −0.68 (−2.36, 0.99), 0.42 −0.62 (−2.31, 1.08), 0.47
Alpha power—NREM μV2 20.47±1.18; 105/111 20.54±1.18; 107/111 20.04±1.18; 100/111 0.07 (−0.82, 0.96), 0.87 −0.43 (−1.33, 0.47), 0.35
Sigma power—NREM μV2 12.55±1.11; 105/111 12.40±1.11; 107/111 12.54±1.12; 100/111 −0.16 (−0.75, 0.44), 0.60 −0.01 (−0.61, 0.59), 0.97
Beta power—NREM μV2 10.95±1.10; 105/111 8.99±1.09; 107/111 11.06±1.13; 100/111 −1.96 (−4.92, 1.01), 0.20 0.11 (−2.90, 3.12), 0.94
Delta power—REM μV2 113.28±6.79; 105/111 118.20±6.75; 107/111 109.07±6.91; 99/111 4.92 (−6.17, 16.02), 0.38 −4.21 (−15.57, 7.14), 0.47
Theta power—REM μV2 25.74±1.78; 105/111 25.74±1.78; 107/111 25.14±1.78; 99/111 0.00 (−1.14, 1.15), 1.00 −0.60 (−1.76, 0.57), 0.31
Alpha power—REM μV2 11.32±0.68; 105/111 11.28±0.68; 107/111 11.08±0.68; 99/111 −0.04 (−0.54, 0.46), 0.87 −0.24 (−0.75, 0.26), 0.34
Sigma power—REM μV2 3.82±0.24; 105/111 3.90±0.24; 107/111 3.51±0.24; 99/111 0.08 (−0.31, 0.47), 0.69 −0.31 (−0.71, 0.08), 0.12
Beta power—REM μV2 9.57±0.92; 105/111 9.91±0.90; 107/111 10.29±0.93; 99/111 0.34 (−1.89, 2.57), 0.77 0.72 (−1.56, 3.00), 0.54
Spindles in NREM Count 723.0±57.5; 106/111 711.6±57.5; 107/111 725.4±57.7; 101/111 −11.5 (−53.9, 31.0), 0.60 2.4 (−40.5, 45.3), 0.91
Average Spindle density in NREM Count/min 2.43±0.19; 106/111 2.43±0.19; 107/111 2.45±0.19; 101/111 0.01 (−0.12, 0.14), 0.91 0.03 (−0.10, 0.16), 0.69
Average fast spindle density in NREM Count/min 1.45±0.17; 106/111 1.46±0.17; 107/111 1.47±0.17; 101/111 0.02 (−0.09, 0.12), 0.73 0.03 (−0.08, 0.13), 0.59
Average slow spindle density in NREM Count/min 0.98±0.14; 106/111 0.97±0.14; 107/111 0.98±0.14; 101/111 −0.01 (−0.10, 0.08), 0.81 −0.00 (−0.10, 0.09), 0.96
Cardiovascular measurements
Pulse transit time ms 67.7±2.1; 33/37 65.9±2.1; 36/37 67.7±2.2; 30/37 −1.8 (−6.1, 2.5), 0.41 −0.0 (−4.5, 4.5), 0.99
Pulse wave velocity m/s 7.55±0.35; 33/37 8.05±0.34; 36/37 7.41±0.37; 30/37 0.50 (−0.26, 1.25), 0.20 −0.14 (−0.93, 0.66), 0.73
Reactive hyperemia (endothelial function test) Index 0.78±0.10; 35/37 1.78±0.10; 32/37 1.70±0.10; 33/37 0.00 (−0.24, 0.24), 0.97 −0.08 (−0.32, 0.16), 0.51
Blood pressure—systolic (24-h) mmHg 113.9±1.8; 1,212/1,850 111.9±1.8; 1,246/1,850 112.6±1.8; 1,196/1,850 −2.1 (−2.9, −1.2), <0.0001 −1.4 ( −2.2, −0.5), 0.002
Blood pressure—diastolic (24-h) mmHg 66.5±1.3; 1,212/1,850 65.8±1.3; 1,246/1,850 66.0±1.3; 1,196/1,850 −0.6 (−1.3, 0.1), 0.07 −0.5 (−1.2, 0.3), 0.19
Heart rate bpm 67.1±1.3; 1,212/1,850 67.1±1.3; 1,246/1,850 66.4±1.3; 1,196/1,850 0.1 (−0.6, 0.7), 0.80 −0.6 (−1.3, 0.1), 0.07
Blood and urine measurements
Serum cortisol nmol/L 413±30; 28/37 391±32; 23/37 382±30; 28/37 −23 (−74, 28), 0.38 −31 (−80, 17), 0.20
Glucose mmol/L 4.9±0.1; 28/37 4.9±0.1; 24/37 4.9±0.1; 29/37 0.0 (−0.1, 0.1), 0.88 0.0 (−0.1, 0.1), 0.88
Insulin mU/L 9.5±1.0; 28/37 7.7±1.0; 24/37 10.0±1.0; 29/37 −1.7 (−3.3, −0.2), 0.025 0.6 (−0.9, 2.0), 0.44
HsCRP mg/L 2.1±0.6; 27/37 1.9±0.7; 20/37 2.1±0.6; 25/37 −0.1 (−1.7, 1.4), 0.87 0.1 (−1.4, 1.5). 0.93
Urinary creatinine mmol/24 h 11.3±0.7; 31/37 11.6±0.7; 32/37 11.3±0.7; 32/37 0.3 (−0.8, 1.4), 0.54 0.0 (−1.0, 1.1), 0.95
Urinary noradrenaline (adjusted)a nmol/24 h/creatinine 19.3±2.2; 31/37 19.4±2.2; 32/37 23.4±2.2; 32/37 0.1 (−4.3, 4.4), 0.98 4.1 (−0.2, 8.4), 0.06
Urinary adrenaline (adjusted)a nmol/24 h/creatinine 3.8±0.3; 31/37 3.7±0.3; 32/37 4.0±0.3; 32/37 −0.1 (−0.8, 0.6), 0.81 0.2 (−0.5, 0.9), 0.49
Urinary dopamine (adjusted)a nmol/24 h/creatinine 151.4±10.8; 31/37 160.2±10.8; 32/37 160.1±10.8; 32/37 8.8 (−7.3, 25.0), 0.28 8.7 (−7.2, 24.7), 0.28
Electrophysiological measurements of wake (derived from the Karolinska Drowsiness Test)
Eyes open—delta μV2 53.53±4.35; 392/444 58.51±4.32; 403/444 64.34±4.42; 379/444 4.98 (−3.62, 13.58), 0.26 10.81 (2.06, 19.55), 0.016
Eyes open—theta μV2 21.61±2.48; 392/444 22.50±2.47; 403/444 22.68±2.48; 379/444 0.89 (−0.16, 1.94), 0.095 1.07 (0.01, 2.13), 0.048
Eyes open—alpha μV2 24.97±2.93; 392/444 24.57±2.93; 403/444 24.48±2.94; 379/444 −0.41 (−1.44, 0.63), 0.44 −0.50 (−1.55, 0.55), 0.35
Eyes open—sigma μV2 6.10±0.59; 392/444 6.05±0.59; 403/444 5.91±0.59; 379/444 −0.04 (−0.25, 0.17), 0.69 −0.18 (−0.40, 0.03), 0.09
Eyes open—beta μV2 18.12±1.31; 392/444 18.09±1.30; 403/444 16.44±1.31; 379/444 −0.02 (−1.33, 1.29), 0.97 −1.67 (−3.00, −0.34), 0.014
Eyes closed—delta μV2 49.04±5.63; 391/444 59.28±5.56; 403/444 50.96±5.74; 378/444 10.24 (−2.46, 22.95), 0.11 1.93 (−11.01, 14.86), 0.77
Eyes closed—theta μV2 26.71±3.03; 391/444 28.10±3.03; 403/444 27.27±3.04; 378/444 1.39 (−0.61, 3.39), 0.17 0.56 (−1.47, 2.59), 0.59
Eyes closed—alpha μV2 45.36±4.62; 391/444 43.36±4.62; 403/444 44.33±4.63; 378/444 −2.00 (−3.95, −0.04), 0.045 −1.03 (−3.01, 0.96), 0.31
Eyes closed—sigma μV2 6.93±0.64; 391/444 6.99±0.64; 403/444 6.79±0.64; 378/444 0.07 (−0.22, 0.35), 0.65 −0.14 (−0.43, 0.15), 0.33
Eyes closed—beta μV2 19.54±1.36; 391/444 19.87±1.36; 403/444 18.43±1.37; 378/444 0.33 (−1.13, 1.78), 0.66 −1.11 (−2.59, 0.37), 0.14
Affective symptoms of WTS (based on a 5-factor PCA of the original 20 visual analog scales)
Factor 1: Fatigue Log-standardized 0.45±0.01; 404/444 0.46±0.01; 403/444 0.48±0.01; 375/444 0.01 (−0.02, 0.05), 0.41 0.03 (−0.00, 0.07), 0.09
Factor 2: Irritability Log-standardized 0.46±0.01; 404/444 0.46±0.01; 403/444 0.48±0.01; 375/444 0.01 (−0.02, 0.04), 0.60 0.02 (−0.01, 0.05), 0.16
Factor 3: Nausea and Dizziness Log-standardized 0.47±0.01; 404/444 0.47±0.01; 403/444 0.46±0.01; 375/444 0.00 (−0.02, 0.03), 0.87 −0.02 (−0.04, 0.01), 0.19
Factor 4: Sleep disturbance Log-standardized 0.46±0.01; 404/444 0.46±0.01; 403/444 0.48±0.01; 375/444 0.00 (−0.02, 0.02), 0.96 0.02 (0.01, 0.04), 0.01
Factor 5: Tinnitus Log-standardized 0.46±0.01; 404/444 0.46±0.01; 403/444 0.49±0.01; 375/444 −0.00 (−0.03, 0.02), 0.79 0.03 (−0.00, 0.05), 0.09
Behavioral performance measures
PVT reciprocal mean reaction time 1/s 3.78±0.08; 407/444 3.78±0.08; 412/444 3.79±0.08; 390/444 0.00 (−0.05, 0.05), 0.99 0.01 (−0.04, 0.06), 0.83
Tower of London grand mean execution test s 10.4±0.6; 409/444 10.5±0.6; 414/444 10.2±0.6; 384/444 0.1 (−0.4, 0.7), 0.64 −0.1 (−0.7, 0.4), 0.62
Tower of London grand number of errors % 5.7±1.0; 409/444 5.9±1.0; 414/444 6.7±1.0; 384/444 0.2 (−0.9, 1.2), 0.75 0.9 (−0.1, 2.0), 0.08
N-back—total number correct Count 40.4±1.6; 403/444 40.9±1.6; 411/444 41.5±1.6; 379/444 0.5 (−0.2, 1.2), 0.20 1.1 (0.3, 1.8), 0.005
End-of-visit questionnaire
Insomnia Severity Index (3-d modified) Points 5.1±0.8; 106/111 4.8±0.8; 106/111 7.2±0.8; 106/111 −0.2 (−1.4, 0.9), 0.68 2.1 (1.0, 3.3), <0.001
Warwick–Edinburgh Mental Wellbeing Scale Points 53.6±1.1; 106/111 55.4±1.1; 106/111 53.6±1.1; 106/111 1.8 (0.2, 3.5), 0.03 −0.0 (−1.7, 1.7), 0.99
Kessler-10 Points 13.9±0.6; 106/111 13.9±0.6; 106/111 13.9±0.6; 106/111 0.0 (−0.8, 0.8), 1.00 0.1 (−0.8, 0.9), 0.89
Depression Anxiety Stress Scale–Depression Points 3.4±1.0; 106/111 3.3±1.0; 106/111 4.2±1.0; 106/111 −0.1 (−1.5, 1.3), 0.88 0.8 (−0.6, 2.2), 0.24
Depression Anxiety Stress Scale—Anxiety Points 2.4±0.5; 106/111 2.5±0.5; 106/111 3.0±0.5; 106/111 0.1 (−0.8, 1.0), 0.83 0.6 (−0.3, 1.5) 0.22
Depression Anxiety Stress Scale—Stress Points 5.8±1.0; 106/111 5.1±1.0; 106/111 6.0±1.0; 106/111 −0.8 (−2.1, 0.5), 0.23 0.2 (−1.1, 1.5), 0.76
Note: Actual/potential obs gives the amount of observed and recorded data being used by the mixed model compared with the potential number of observations (i.e., complete data) so readers can gauge how much missing data for each of the conditions the model is having to deal with. %, Percentage; alpha frequency band, 8–12 Hz; beta frequency band, 15–32 Hz; CI, confidence interval; delta frequency band, 0.5–4.5 Hz; HsCRP, highly sensitive C-reactive protein; N1, non-REM sleep stage 1; N2, non-REM sleep stage 2; N3, non-REM sleep stage 3; NREM, non-REM sleep; obs, observation; PCA, principal component analysis; PVT, psychomotor vigilance task (simple reaction time task); REM, rapid eye movement sleep; SEM, standard error of the mean; sigma frequency band, 12–15 Hz; theta frequency band, 4.5–8 Hz; TST, total sleep time; WTS, wind turbine syndrome.
a Unit of measurement is the variable measured nmol/24h divided (adjusted) by the simultaneously measured creatinine (mmol/24h).
Figure 1. Effect estimates of infrasound and traffic on wake after sleep onset (WASO) over 3 nights in the laboratory-based study of the three-arm crossover study of 72 h of exposure to simulated wind turbine infrasound, sham infrasound, and traffic noise. The mixed model estimates of the effect of infrasound and traffic noise on electrophysiologically measured human WASO. The primary outcome of this study was WASO as a measure of the effects of noise on sleep perturbation. WASO is the amount of time spent awake between sleep onset and final wake-up time. We measured sleep for 3 nights under each of the three exposures [infrasound in blue squares, sham infrasound in red triangles, and traffic noise in black circles; −1.36 min difference between infrasound and sham infrasound (95% CI: −6.60, 3.88, p=0.601); Table 2]. Error bars indicate the 95% CIs. Effects estimates are derived from mixed models of repeated measures where the participants were classed as random effects and exposure (3 levels), the order the exposure was received (1, 2, 3), the night of exposure (1, 2, 3), and interactions between the exposure by night and night by order as fixed effects. The least squares means procedure was used to address missing data. The exact numerical values for the estimated means and 95% CIs can be found in Table S3. Note: CI, confidence interval.
Figure 1 is an error bar graph, plotting W A S O (minute), ranging from 0 to 50 in increments of 10 (y-axis) across night, ranging from 1 to 3 in unit increments (x-axis) for infrasound, sham infrasound, and traffic.
Figure 2. Effect estimates of infrasound and traffic on measures of sleep quality over 3 nights in the laboratory-based study of the three-arm crossover study of 72 h of exposure to simulated wind turbine infrasound, sham infrasound, and traffic noise. Infrasound is represented in blue squares, sham infrasound in red triangles, and traffic noise in black circles. Error bars are 95% CI. Effects estimates are derived from mixed models of repeated measures where the participants were classed as random effects and exposure (3 levels), the order the exposure was received (1, 2, 3), the night of exposure (1, 2, 3), and interactions between the exposure by night and night by order as fixed effects. The least squares means procedure was used to address missing data. The exact numerical values for the estimated means and 95% CIs can be found in Table S3. (A) Sleep onset latency is the amount of time taken to fall asleep [2.48 min difference between infrasound and sham (95% CI: −2.87, 7.82, p=0.36); Table 2]. (B) Number of sleep stage shifts is a measure of sleep stability [−3.5 shifts difference between infrasound and sham (95% CI: −9.5, 2.4, p=0.24)]. (C) Proportions of the sleep period scored as each of the traditional sleep stages plotted together to test whether infrasound causes perturbation to sleep depth. Numerical values above each plot are the p-values for the difference between infrasound and sham infrasound. p-Values above the stacked columns are comparing infrasound to sham infrasound. (D) Arousal index is the number of cortical arousals detected during each hour of sleep as a measure of sleep quality [0.29 events/h difference between infrasound and sham (95% CI: −0.58, 1.16, p=0.51)]. Note: %, percentage; 1, non-REM sleep stage 1; 2, non-REM sleep stage 2; 3, non-REM sleep stage 3; CI, confidence interval; REM, rapid eye movement stage; SEM, standard error of the mean; W, wake.
Figures 2A, 2B, and 2D are error bar graphs, plotting sleep onset latency (minute), ranging from 0 to 30 in increments of 5; number of stage shifts, ranging from 90 to 160 in increments of 10; and arousals (events per hour), respectively, ranging from 6 to 13 in unit increments (y-axis) across night, ranging from 1 to 3 in unit increments. Figure 2C is an error bar graph, plotting percentage of sleep opportunity, ranging from 0 to 50 in increments of 10 (y-axis) across Sleep stage, ranging from wake, N 1, N 2, N 3, rapid eye movement (x-axis). All for infrasound, sham infrasound, and traffic.
Figure 3. Effect estimates of infrasound and traffic on quantitative measures of electroencephalography during sleep in the laboratory-based study of the three-arm crossover study of 72 h of exposure to simulated wind turbine infrasound, sham infrasound, and traffic noise. Infrasound is represented in blue squares, sham infrasound in red triangles, and traffic noise in black circles. Absolute power derived from overnight electrophysiology transformed into five frequency bands as a measure of cortical activity in (A) NREM and (B) REM (delta δ=0.5–4.5 Hz, theta θ=4.5–8 Hz, alpha α=8–12 Hz sigma σ=12–15 Hz, beta β=15–32 Hz). Sleep spindle density in NREM overall and density of fast and slow spindles in NREM. (C) Fast spindle=13–16 Hz and slow spindle=11–13 Hz. Numerical values above each plot are the p-values for the difference between infrasound and sham infrasound. Effects estimates are derived from mixed models of repeated measures where the participants were classed as random effects and exposure (3 levels), the order the exposure was received (1, 2, 3), the night of exposure (1, 2, 3), and interactions between the exposure by night and night by order as fixed effects. The least squares means procedure was used to address missing data. The exact numerical values for the estimated means and 95% CIs can be found in Table S3. Point estimates are indicated graphically by the shapes and 95% CIs indicated by the bars. Note: CI, confidence interval; NREM, non-REM sleep; REM, rapid eye movement sleep.
Figures 3A, 3B, and 3C are error bar graphs, plotting absolute power (microvolt squared), ranging from 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript; absolute power (microvolt squared), ranging from 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, and 10 begin superscript 3 end superscript; spindles (Number per minute), ranging from 0 to 4 in increments of 0.5 (y-axis) across electroencephalographic frequency band, ranging as lowercase delta, lowercase theta, lowercase alpha, lowercase sigma, and lowercase beta; electroencephalographic frequency band, ranging as lowercase delta, lowercase theta, lowercase alpha, lowercase sigma, and lowercase beta; overall, fast, slow (x-axis), respectively, for infrasound, sham infrasound, and traffic.
Adverse Events
There were no serious adverse events (unplanned hospitalizations or deaths). There were four adverse events in participants, who we describe in this section as participants A, B, C, and D. Participant A completed only one visit while being exposed to infrasound. They complained that the tips of their hair were being made brittle by the EEG paste and declined further contact and participation. We found no pattern of complaints in their data that would match WTS after reviewing all their objective and subjective data after unblinding. Participant B completed only one visit while being exposed to sham infrasound. They complained of social isolation and did not enjoy being in a windowless room and then declined further contact and participation. Participant C completed all visits and during the first visit was exposed to traffic noise and had a mild asthma attack possibly triggered by a bushfire event in Sydney that weekend. This participant was immediately seen by a respiratory physician in our clinic to update their asthma management before they left the laboratory. Participant D completed all visits but emailed the study coordinator to report feeling on edge in the immediate days after returning home from the second visit (infrasound exposure). We then reviewed their data where they did not report having, or later retrospectively report having, that experience in the laboratory environment.
Data Quality
Our quality assurance for our blood pressure/heart rate data indicated that other than the missing data at a participant level (where 2 visits were missed because of equipment availability and a staff scheduling accident plus the 5 missed visits from people dropping out and the final COVID shutdown plus 1 collection missed because of missing equipment=2+5+1 visits ×50=400 missing values) that the missing data were randomly distributed among different times of the day and night but that the data loss proportion was greater than the ideal proportions used clinically (i.e., having >85% complete data).37 Our missing data rate (1,523 missing values) after allowance for the people in whom it was not collected (i.e., 400 missing measurements) was 71%, which is not as good as the 85% target. These missing data were due to equipment slippage and the failure of the blood pressure cuff to inflate properly.
Urinary analyses were part of the protocol that people could opt out of and some participants declined to consent. Table 2 reports we missed 16 of the potential urine samples we could theoretically have collected, and this is accounted for by five missed visits (5), plus two people opting out of urine in all visits (6), plus four people opting out of urine on one visit (4; one of whom was actually the final COVID-affected participant and we could not collect their sample because of the start of the Public Health Order that shut down research in the state and our study’s data collection phase), plus one missed sample (1) in conjunction with the staffing scheduling error (5+6+4+1=16). Two values for noradrenaline and 48 values for adrenaline were at the lower detectable limit.
HsCRP data were missing for 39/111 potential measurements. These are accounted for by five missed visits (5), the one visit cut short at the end by the public health order (1), plus one person opting out of all blood collection (3), plus one person opting out of blood on one visit (1), plus two people opting out of blood collection on two visits (4), 8 samples returned a value at or below the LODs and were entered as being missing data (0.1mg/L; n=8), and 17 samples were missing owing to insufficient blood volume collection to send for testing. There were no samples missing because of pathology service errors or a failed sample in transit (5+1+3+1+4+8+17+0=39).
Cortisol data were missing for 32/111 potential measurements. These are accounted for by five missed visits (5), the one visit cut short at the end by the public health order (1), plus one person opting out of all blood collection (3), plus one person opting out of blood on one visit (1), plus two people opting out of blood collection on two visits (4), and 18 samples were missing owing to insufficient blood volume collection to send for testing. There were no samples missing because of pathology service errors or a failed sample in transit or being below the LODs (5+1+3+1+4+18+0=32).
Glucose and insulin data were missing for 30/111 potential measurements. These are accounted for by five missed visits (5), the one visit cut short at the end by the public health order (1), plus one person opting out of all blood collection (3), plus one person opting out of blood on one visit (1), plus two people opting out of blood collection on two visits (4), and 16 samples were missing owing to insufficient blood volume collection to send for testing. There were no samples missing because of pathology service errors or a failed sample in transit or being below the LODs (5+1+3+1+4+16+0=30).
Discussion
This study found that 72 h of the simulated wind turbine infrasound (∼90 dB pk re 20μPa) in controlled laboratory conditions did not worsen any measure of sleep quality compared with the same speakers being present but not generating infrasound (sham infrasound). The positive control condition, audible traffic noise (sound level of 40–50 dB LAeq,night and 70 LAFmax transient maxima) worsened the primary measure of sleep quality, WASO, by ∼6 min compared with the sham infrasound. This effect was evident on the first 2 nights where the magnitude of the effect is similar to a previous report.38 The lack of an effect on the third night of traffic noise might have been caused by the build-up of homeostatic sleep pressure caused by the sleep disturbance on the first 2 nights.38,39 Furthermore, none of the staff or participants involved in the study reported being able to distinguish the infrasound condition from the sham infrasound and none of the participants displayed objective or subjective features consistent with WTS. From our list of 22 secondary outcome measures of sleep electrophysiology only one (% REM sleep) was significantly different in the infrasound exposure compared with sham. Furthermore, the effect estimate was in the opposite direction to that which was hypothesized and was not offset by changes in other sleep stages. As far as we can tell, this is the first study to investigate the effects on human sleep of simulated wind turbine infrasound in double-blind conditions.
We also measured the effects of infrasound on a wide range of nonsleep-related physiological and psychological measures. Four of the 39 tertiary outcomes we analyzed demonstrated an estimated effect of infrasound. These very small differences were not systematically in an adverse direction. Hence, we believe these effects have been detected by chance. We note that the design of the study, with many repeated measurements for each participant, meant there was a very high level of statistical power. Three of the measures (systolic blood pressure, insulin, and the Warwick–Edinburgh Mental Wellbeing Scale) improved by a small amount in association with exposure to infrasound compared with sham. In addition, the amount of alpha power during the eyes-closed condition of the Karolinska Drowsiness Test changed by a very small amount (2 μV2/s, Cohens d<0.1) in association with infrasound, and it is not clear whether this direction of effect is helpful or harmful. We conclude that these findings suggest the absence of detectable health effects of infrasound on humans in our study.
People who suffer from WTS report that their symptoms begin quickly when they are exposed to infrasound from wind turbines and are then sustained.9,40 Our scientifically robust study provides evidence to address this claim. The Australian NHMRC report10 that gave rise to our study made note of this “absence of evidence” rather than concluding an “evidence of absence” owing to the lack of any laboratory-controlled double-blind experiments of sufficient duration and intensity to hypothetically induce WTS in a human. Our study attempted to address this absence of evidence by rigorously simulating wind turbine infrasound in the frequency range of 1.6–20 Hz at a sound level of ∼90 dB pk re 20μPa (measurable but inaudible) and comparing it in a double-blind randomized exposure study to a sham infrasound exposure using the same equipment but with the speakers wired in antiphase so that they did not generate infrasound. Furthermore, all previous studies we are aware of testing wind turbine infrasound (not audible WTN) in double-blind conditions have exposed people to 7.5 or 23 min of infrasound and not 72 h as we have here.15,16,18,41
Our original sample size goal was 40 people, with an assumption that 2 people would drop out during the study. We actually randomized and analyzed data from 37 people and had 2 people drop out after completing only one of the planned three exposure periods before COVID forced us to close the study. The last participant, who was still in the study on the day it was closed, had completed both infrasound and sham infrasound conditions; therefore, we have not considered this person as having dropped out. The power of the study remains strong owing to the triple-arm crossover design coupled with multiple repeated measurements of numerous physiological and psychological outcomes. For instance, the primary and secondary outcomes were measured nine times per person, which allowed us to estimate the difference between exposures for the primary outcome (WASO) within a span of ∼9.5 min for the full width of the 95% confidence limit. The greatest number of repeated measures was for blood pressure where we collected ∼150 blood pressure measurements across the three exposure per person. The width of the 95% confidence limits for the difference between exposures in 24-h systolic blood pressure measurement was only 1.7 mmHg. This helps improve our confidence that there are no meaningful adverse health effects of this specific formulation of infrasound in these humans. In addition, we found no evidence for a first-night effect in our primary outcome where humans often have a poor sleep in their first night in an unfamiliar laboratory environment compared with their subsequent stays.
The study has some limitations. It is possible that, despite the application of a noise-sensitivity eligibility criterion and also having seven participants who were somewhat concerned or concerned about the health effects of infrasound generated by wind turbines, we inadvertently recruited a group of people who were not sensitive to the effects of infrasound. It remains possible that some humans are sensitive to infrasound. This hypothesis has been tested in a short term study in double-blind conditions in people who said that wind turbines make them feel ill. Those participants were unable to reliably detect infrasound and did not physiologically react to it.40 All of our quantitative electrophysiology measurements of the brain were taken from a single channel, so it remains possible that infrasound could have effects on other brain regions. It is also possible that our outcome measures lacked sensitivity for detecting adverse health effects of noise and the lack of effect of traffic noise on a number of secondary and tertiary outcome measures could be evidence for that lack of sensitivity. The background noise level in the sleeping environment created by the air-conditioning system was measured at about 39 dB LAeq, which is above the recommended levels of the World Health Organization night noise guidelines of 35 dB LAeq through the night and is louder than background recordings in sleep laboratories internationally that have undertaken noise exposure studies.42 This might have caused some sleep perturbation in some people that might have biased our results toward a null effect. However, we do not think that this would have been a powerful biasing effect. Our reasoning is that the sleep recordings in the people in this study when they were in the sham condition were in a very healthy range compared with reference values.43 In addition, the sound levels in the two rooms used are experienced as a white noise background that is sometimes used by people as a sleep aid at considerably louder levels, albeit with a poor evidence base to be able to tell whether it is helpful or harmful.44 Nevertheless, a background level of 39 dB LAeq is somewhat louder than optimal for a control condition in a study of noise effects on sleep. However, on balance, we felt it was important to have a comfortable environment for the participants given many individuals in our country would have home air-conditioning or fans. Although the sham and traffic conditions might appear to be similar using averaged overnight sound pressure levels (39 dB LAeq,night in the sham condition vs. 40–50 dB LAeq,night in the traffic noise condition) they are actually quite different because the traffic noise includes infrequent, abrupt 70-dB LAFmax noise peaks designed to awaken humans (Figure S4). Presumably, if we had employed these noise maxima peaks more often we would have disrupted sleep to a greater degree. Our data completeness for our primary and secondary outcomes was high and most data losses were at random and of small numbers, which can be dealt with by the statistical techniques we used. Our analyses of some tertiary outcome variables, including the blood-derived variables and the urinary catecholamines and also the heart rate and blood pressure data, should be interpreted with some caution given that they did suffer from higher than expected missing data proportions.37
Conclusion
Our study found no evidence that 72 h of exposure to a sound level of ∼90 dB pk re 20μPa of simulated wind turbine infrasound in double-blind conditions perturbed any physiological or psychological variable. None of the 36 people exposed to infrasound developed what could be described as WTS. Our study is unique because it measured the effects of infrasound alone on sleep. This study suggests that the infrasound component of WTN is unlikely to be a cause of ill-health or sleep disruption, although this observation should be independently replicated.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The contributions of the authors were as follows: experiment design and application for funding (N.S.M., B.G.T., R.T., D.J.B., C.T.C., N.G., M.S.W., C.L.P., G.B.M., and R.R.G.), principal investigators (N.S.M. and R.R.G.), protocol development (N.S.M., G.C., and R.K.), lead study coordinator (G.C.), medical oversight (G.B.M. and R.R.G.), psychological screening (D.J.B.), audiological and neurological screening (C.R.W. and M.S.W.), staff coordination (G.C., B.G.T., and C.A.E.), grant budget control (B.G.T.), noise simulation speaker construction and acoustic engineering quality assurance and quality control (R.T. and O.J.), electroencephalography data quality and processing oversight (G.C. and A.L.D.), statistical analyses (N.S.M. and G.C.), manuscript drafting (N.S.M. and G.C.), and manuscript planning committee (N.S.M., B.G.T., R.T., G.B.M., and R.R.G.).
We acknowledge the people who collected or processed data: C. Berry, N. Hurst, E. Argaet, L. Fratturo, S. Theocharous, S. Haffar, B. Zhang, S. Shulka, M. Nguyen, I. Wood, J. Nguyen, H. Smith, N. Charlwood, C. Low, K. Kremerskothen, V. Fuchsova, B. Tiik, A. Bertolin, S. Yuhedran, A. Radowiecka, J. Ponahajba, E. Olszewska, M. Bronisz, C. Lorilla, M. Allado, N. Mutombe, H. Linn, M. Balenzuela, I. Basic, C. Geha, B. Pan, D. Jia, R. Wallis, K. O’Doherty, and I. Juria. We also acknowledge the members of the data safety monitoring board (DSMB): S. Loughran, G. Hamilton, S. O’Leary, and F. Garden (DSMB statistician), as well as the data engineers, G. Unger and Z. Zhou, and graphic design assistance, R. Wassing.
The funder played no role in the conduct of the study or the decision to publish.
Trial Registration: ANZCTR (ACTRN12617000001392).
Funding for this study was granted by the National Health and Medical Research Council of Australia (NHMRC) through the Targeted Call for Research into Wind Farms and Human health, APP1113615.
==== Refs
References
1. Ageborg Morsing J, Smith M, Ögren M, Thorsson P, Pedersen E, Forssén J, et al. 2018. Wind turbine noise and sleep: pilot studies on the influence of noise characteristics. Int J Environ Res Public Health 15 (11 ):2573, PMID: , 10.3390/ijerph15112573.30453618
2. Lechat B, Scott H, Decup F, Hansen KL, Micic G, Dunbar C, et al. 2022. Environmental noise-induced cardiovascular responses during sleep. Sleep 45 (3 ):zsab302, PMID: , 10.1093/sleep/zsab302.34965303
3. Michaud DS, Keith SE, Guay M, Voicescu S, Denning A, McNamee JP. 2021. Sleep actigraphy time-synchronized with wind turbine output. Sleep 44 (9 ):zsab070, PMID: , 10.1093/sleep/zsab070.33768252
4. Smith MG, Ögren M, Thorsson P, Hussain-Alkhateeb L, Pedersen E, Forssén J, et al. 2020. A laboratory study on the effects of wind turbine noise on sleep: results of the polysomnographic WiTNES study. Sleep 43 (9 ):zsaa046, PMID: , 10.1093/sleep/zsaa046.32211778
5. Liebich T, Lack L, Hansen K, Zajamšek B, Lovato N, Catcheside P, et al. 2021. A systematic review and meta‐analysis of wind turbine noise effects on sleep using validated objective and subjective sleep assessments. J Sleep Res 30 (4 ):e13228, PMID: , 10.1111/jsr.13228.33179850
6. Nissenbaum MA, Aramini JJ, Hanning CD. 2012. Effects of industrial wind turbine noise on sleep and health. Noise Health 14 (60 ):237–243, PMID: , 10.4103/1463-1741.102961.23117539
7. Salt AN, Kaltenbach JA. 2011. Infrasound from wind turbines could affect humans. Bull Sci Technol Soc 31 (4 ):296–302, 10.1177/0270467611412555.
8. Schomer PD, Erdreich J, Pamidighantam PK, Boyle JH. 2015. A theory to explain some physiological effects of the infrasonic emissions at some wind farm sites. J Acoust Soc Am 137 (3 ):1356–1365, PMID: , 10.1121/1.4913775.25786948
9. Pierpont N. 2009. Wind Turbine Syndrome: a Report on a Natural Experiment. https://www.windturbinesyndrome.com/wind-turbine-syndrome/ [accessed 14 March 2022].
10. NHMRC (National Health and Medical Research Council). 2015. Information Paper: Evidence on Wind Farms and Human Health. https://www.nhmrc.gov.au/about-us/publications/nhmrc-information-paper-evidence-wind-farms-and-human-health [accessed 14 March 2022].
11. Council of Canadian Academies. 2015. Understanding the Evidence: Wind Turbine Noise. The Expert Panel on Wind Turbine Noise and Human Health. https://cca-reports.ca/reports/understanding-the-evidence-wind-turbine-noise/ [accessed 14 March 2022].
12. Knopper LD, Ollson CA. 2011. Health effects and wind turbines: a review of the literature. Environ Health 10 :78, PMID: , 10.1186/1476-069X-10-78.21914211
13. McCunney RJ, Mundt KA, Colby WD, Dobie R, Kaliski K, Blais M. 2014. Wind turbines and health: a critical review of the scientific literature. J Occup Environ Med 56 (11 ):e108–e130, PMID: , 10.1097/JOM.0000000000000313.25376420
14. Tonin R. 2018. A review of wind turbine-generated infrasound: source, measurement and effect on health. Acoust Aust 46 (1 ):69–86, 10.1007/s40857-017-0098-3.
15. Chapman S, Crichton F. 2017. Wind Turbine Syndrome: a Communicated Disease. https://open.sydneyuniversitypress.com.au/9781743324967.html [accessed 14 March 2022].
16. Tonin R, Brett J, Colagiuri B. 2016. The effect of infrasound and negative expectations to adverse pathological symptoms from wind farms. J Low Freq Noise V A 35 (1 ):77–90, 10.1177/0263092316628257.
17. Rubin GF, Burns M, Wessely S. 2014. Possible psychological mechanisms for “wind turbine syndrome”. On the windmills of your mind. Noise Health 16 (69 ):116–122, PMID: , 10.4103/1463-1741.132099.24804716
18. Crichton F, Dodd G, Schmid G, Gamble G, Petrie KJ. 2014. Can expectations produce symptoms from infrasound associated with wind turbines? Health Psychol 33 (4 ):360–364, PMID: , 10.1037/a0031760.23477573
19. NHMRC. 2007. National Statement on Ethical Conduct in Human Research (Updated 2018). http://www.nhmrc.gov.au/guidelines/publications/e72 [accessed March 14 2022].
20. Ekehammar B, Dornic S. 1990. Weinstein’s Noise Sensitivity Scale: reliability and construct validity. Percept Mot Skills 70 (1 ):129–130, PMID: , 10.2466/pms.1990.70.1.129.2326111
21. Morin CM, Belleville G, Belanger L, Ivers H. 2011. The Insomnia Severity Index: psychometric indicators to detect insomnia cases and evaluate treatment response. Sleep 34 (5 ):601–608, PMID: , 10.1093/sleep/34.5.601.21532953
22. Johns MW. 1991. A new method for measuring daytime sleepiness: the Epworth Sleepiness Scale. Sleep 14 (6 ):540–545, PMID: , 10.1093/sleep/14.6.540.1798888
23. Lovibond SH, Lovidbond PF. 1995. Manual for the Depression Anxiety Stress Scales. 2nd ed. Sydney, New South Wales, Australia: Psychology Foundation.
24. Radomsky AS, Rachman S, Thordarson DS, McIsaac HK, Teachman BA. 2001. The Claustrophobia Questionnaire. J Anxiety Disord 15 (4 ):287–297, PMID: , 10.1016/s0887-6185(01)00064-0.11474815
25. Onusko EM. 2004. Tympanometry. Am Fam Physician 70 (9 ):1713–1720, PMID: .15554489
26. Carhart R, Jerger JF. 1959. Preferred method for clinical determination of pure-tone thresholds. J Speech Hear Disord 24 (4 ):330–345, 10.1044/jshd.2404.330.
27. Walker B, Hessler GF, Hessler DM, Rand R, Schomer P. 2012. A Cooperative Measurement Survey and Analysis of Low Frequency and Infrasound at the Shirley Wind Farm in Brown County, Wisconsin. Report No. 122412-1. https://apps.psc.wi.gov/ERF/ERFview/viewdoc.aspx?docid=178200 [accessed 14 March 2022].
28. Berry RB, Brooks R, Gamaldo CE, Harding SM, Lloyd RM, Marcus CL, et al. 2015. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, Version 2.2. Darien, IL: American Academy of Sleep Medicine.
29. D’Rozario AL, Cross NE, Vakulin A, Bartlett DJ, Wong KKH, Wang D, et al. 2017. Quantitative electroencephalogram measures in adult obstructive sleep apnea—potential biomarkers of neurobehavioural functioning. Sleep Med Rev 36 :29–42, PMID: , 10.1016/j.smrv.2016.10.003.28385478
30. Wang D, Bai XX, Williams SC, Hua SC, Kim JW, Marshall NS, et al. 2015. Modafinil increases awake EEG activation and improves performance in obstructive sleep apnea during continuous positive airway pressure withdrawal. Sleep 38 (8 ):1297–1303, PMID: , 10.5665/sleep.4912.26158894
31. Vakulin A, D’Rozario A, Kim JW, Watson B, Cross N, Wang D, et al. 2016. Quantitative sleep EEG and polysomnographic predictors of driving simulator performance in obstructive sleep apnea. Clin Neurophysiol 127 (2 ):1428–1435, PMID: , 10.1016/j.clinph.2015.09.004.26480833
32. Phillips L, Gilhooly K, Logie R, Della Sala S, Wynn V. 2003. Age, working memory, and the Tower of London task. Eur J Cogn Psychol 15 (2 ):291–312, 10.1080/09541440244000148.
33. Shallice T. 1982. Specific impairments of planning. Philos Trans R Soc Lond B Biol Sci 298 (1089 ):199–209, PMID: , 10.1098/rstb.1982.0082.6125971
34. Dorrian J, Rogers NL, Dinges DF. 2004. Psychomotor vigilance performance: neurocognitive assay sensitive to sleep loss. In: Sleep Deprivation. Clinical Issues, Pharmacology, and Sleep Loss Effects. Kushida CA, ed. 1st ed. Boca Raton, FL: CRC Press, 39–70.
35. Dinges DF, Powell JW. 1985. Microcomputer analyses of performance on a portable, simple visual RT task during sustained operations. Behav Res Methods Instrum Comput 17 (6 ):652–655, 10.3758/BF03200977.
36. Butlin M, Qasem A. 2017. Large artery stiffness assessment using SphygmoCor technology. Pulse (Basel) 4 (4 ):180–192, PMID: , 10.1159/000452448.28229053
37. McGrath BP. 2002. Ambulatory blood pressure monitoring. Med J Aust 176 (12 ):558–592, PMID: , 10.5694/j.1326-5377.2002.tb04590.x.12120602
38. Borbély AA. 1982. A two process model of sleep regulation. Hum Neurobiol 1 (3 ):195–204, PMID: .7185792
39. Basner M, Müller U, Elmenhorst EM. 2011. Single and combined effects of air, road, and rail traffic noise on sleep and recuperation. Sleep 34 (1 ):11–23, PMID: , 10.1093/sleep/34.1.11.21203365
40. Chapman S, St George A, Waller K, Cakic V. 2013. The pattern of complaints about Australian wind farms does not match the establishment and distribution of turbines: support for the psychogenic, ‘communicated disease’ hypothesis. PLoS One 8 (10 ):e76584, PMID: , 10.1371/journal.pone.0076584.24146893
41. Maijala PP, Kurki I, Vainio L, Pakarinen S, Kuuramo C, Lukander K, et al. 2021. Annoyance, perception, and physiological effects of wind turbine infrasound. J Acoust Soc Am 149 (4 ):2238, PMID: , 10.1121/10.0003509.33940893
42. World Health Organization. 2009. Night Noise Guidelines for Europe. https://apps.who.int/iris/handle/10665/326486 [accessed 3 November 2022].
43. Mitterling T, Högl B, Schönwald SV, Hackner H, Gabelia D, Biermayr M, et al. 2015. Sleep and respiration in 100 healthy Caucasian sleepers—a polysomnographic study according to American Academy of Sleep Medicine standards. SLEEP 38 (6 ):867–875, PMID: , 10.5665/sleep.4730.25515109
44. Riedy SM, Smith MG, Rocha S, Basner M. 2021. Noise as a sleep aid: a systematic review. Sleep Med Rev 55 :101385, PMID: , 10.1016/j.smrv.2020.101385.33007706
|
PMC010xxxxxx/PMC10032369.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36947409
EHP12772
10.1289/EHP12772
Science Selection
Precursor to Dengue: Projecting Effects of Climate Change on Mosquito Density in Southeast Asia
https://orcid.org/0000-0001-6525-8502
Osuolale Olayinka
22 3 2023
3 2023
131 3 03400219 1 2023
08 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Closeup of three mosquito larvae in water.
==== Body
pmcChanges in local temperature and precipitation patterns are affecting the geographic range of disease vectors such as the Aedes mosquitoes that spread dengue, Zika, and chikungunya.1 Effectively preparing for a warmer future requires understanding the extent to which climate change could affect the future distribution of vectors, with possible consequences for infectious disease transmission. In a new study published in Environmental Health Perspectives, a team of researchers from Cambodia, Lao People’s Democratic Republic (PDR), France, and New Caledonia projected climate-related changes in the densities of Ae. aegypti and Ae. albopictus mosquitoes in Southeast Asia.2
Changes in local temperature and precipitation patterns are affecting the geographic range and population densities of disease vectors such as Aedes aegypti (shown here) and Ae. albopictus mosquitoes, which spread dengue, Zika, and chikungunya. Image: © Cacio Murilo/stock.adobe.com.
Closeup of three mosquito larvae in water
Global incidence of one mosquito-borne disease, dengue, has risen over the past two decades. Infections can range from asymptomatic to life-threatening.3 In 2000, just over 500,000 cases were reported to the World Health Organization, compared with 5.2 million reported infections in 2019.4 However, actual numbers could be many times higher, given that most cases are not severe enough to require medical care.5 Although dengue is endemic in dozens of countries, Asia bore approximately 70% of the global burden as of 2010.4
The authors modeled seasonal variations of densities of Ae. aegypti and Ae. albopictus under current weather patterns and future climate scenarios (based on greenhouse gas emissions) over four periods: 2021–2040, 2041–2060, 2061–2080, and 2081–2100. The study area included the countries of Myanmar, Lao PDR, Thailand, Cambodia, and Vietnam, where dengue outbreaks occur frequently.6 They used a compartmental process-based model that was validated using density data from entomological surveys in the cities of Phnom Penh (Cambodia) and Vientiane (Lao PDR).
Model estimates indicated that densities of both species would increase during the region’s coolest months (December–February) under all climate scenarios. The warmest months (April–June) saw elevated population densities of Ae. aegypti throughout most of the region, whereas Ae. albopictus showed marked increases and decreases in densities in the northern and southern parts of the study area, respectively, for all years and climate scenarios. Model projections also suggested that Aedes densities are more likely to be affected by changes in temperature than precipitation.
Kristie Ebi, a Professor of Global Health and of Environmental and Occupational Health Sciences at the University of Washington, explains that process-based models are built from equations describing the dynamics of disease origins. “They generally simulate the impact of weather variables on the health outcomes of interest using equations describing disease burdens at daily or weekly time steps,” she says.
In this case, the authors modeled mosquito densities without factoring in disease incidence or transmission. To do this, they simulated density variations of mosquitoes’ different life stages based on development and mortality rates and their established dependency on temperature and precipitation. Future vector densities were modeled using climate projections based on nine climate models and four scenarios of greenhouse gas emissions. Ebi, who was not involved in the study, says the model could apply to other Aedes-borne diseases such as Zika and chikungunya.
Transmission of mosquito-borne diseases is a highly complex process. “For example, as temperature increases, the rate of virus development within the mosquito is increased, so it can take less time for the mosquito to be infectious, increasing transmission,” says Simon Hales, a research professor in the Department of Public Health at the University of Otago, Wellington. “But the mortality rate of the mosquito also increases above a certain temperature, which reduces transmission.”
Although vector density is an important determinant of disease transmission,7 Hales notes that very low density could be sufficient to maintain transmission if other conditions are right, “so a linear relationship with vector density is not expected.” Hales was not involved in the new work.
Lead author Lucas Bonnin, a postdoctoral associate with the French National Research Institute for Sustainable Development working in Nouméa, New Caledonia, points to the model’s usefulness for exploring the complexities of disease transmission. “The outputs of the model could be used to further investigate how vector density impacts disease dynamics,” he says, “and to predict how climate change would affect those dynamics.”
Olayinka Osuolale, PhD, is an environmental health and microbiologist researcher and a senior lecturer at Elizade University, Ilara Mokin, Nigeria.
==== Refs
References
1. Ryan SJ, Carlson CJ, Mordecai EA, Johnson LR. 2019. Global expansion and redistribution of Aedes-borne virus transmission risk with climate change. PLOS Negl Trop Dis 13 (3 ):e0007213, PMID: , 10.1371/journal.pntd.0007213.30921321
2. Bonnin L, Tran A, Herbreteau V, Marcombe S, Boyer S, Mangeas M, et al. 2022. Predicting the effects of climate change on dengue vector densities in Southeast Asia through process-based modeling. Environ Health Perspect 130 (12 ):127002, PMID: , 10.1289/EHP11068.36473499
3. WHO (World Health Organization). 2009. Dengue: Guidelines for Diagnosis, Treatment, Prevention and Control. Geneva, Switzerland: WHO.
4. WHO. 2022. Dengue and severe dengue. [Website.] Updated 10 January 2022. https://www.who.int/news-room/fact-sheets/detail/dengue-and-severe-dengue [accessed 17 February 2023].
5. Bhatt S, Gething PW, Brady OJ, Messina JP, Farlow AW, Moyes CL, et al. 2013. The global distribution and burden of dengue. Nature 496 (7446 ):504–507, PMID: , 10.1038/nature12060.23563266
6. Tian N, Zheng JX, Guo ZY, Li LH, Xia S, Lv S, et al. 2022. Dengue incidence trends and its burden in major endemic regions from 1990 to 2019. Trop Med Infect Dis 7 (8 ):180, PMID: , 10.3390/tropicalmed7080180.36006272
7. Bowman LR, Runge-Ranzinger S, McCall PJ. 2014. Assessing the relationship between vector indices and dengue transmission: a systematic review of the evidence. PLoS Negl Trop Dis 8 (5 ):e2848, PMID: , 10.1371/journal.pntd.0002848.24810901
|
PMC010xxxxxx/PMC10038136.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36961446
EHP12092
10.1289/EHP12092
Focus
Breaking New Ground: Space Agencies and Epidemiologists Partner Up on Particulates
Averett Nancy
24 3 2023
3 2023
131 3 03200101 9 2022
18 11 2022
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Aerial view of Johannesburg showing brown layer of air pollution
==== Body
pmcWhen cold air traps pollutants over Johannesburg on winter afternoons, a brown haze can form, with particulate matter levels that exceed South Africa’s ambient air quality standards.1 In mid-June 2022 (early winter in South Africa), an inversion also magnified a foul smell that left one resident tweeting, “Johannesburg is a toxic cesspool. It now smells like rotten eggs.”2 Although the gas responsible for that smell, hydrogen sulfide, is not regulated in South Africa, its noxious odor was a reminder that levels of other pollutants might also be high.3
Smog blankets Johannesburg on a sunny day in 2019. The capital city is located in South Africa’s Highveld—a high-altitude inland area that is the country’s industrial heartland.3 Image: © iStock.com/Rich Townsend.
Aerial view of Johannesburg showing brown layer of air pollution
It is hard to capture the spatial and temporal variations in air pollution at specific spots in Johannesburg. The region’s sparse monitoring system measures the total mass of particulate matter in the air but not the composition or size of specific particles. “We really don’t know exactly what’s in it,” says Kristy Langerman, an atmospheric scientist at the University of Johannesburg. “But I can speculate and say it’s from traffic as well as some domestic burning and dust.”
In the coming years, Langerman and other air pollution experts may get some answers, thanks to an upcoming U.S. National Aeronautics and Space Administration (NASA) satellite mission, which has become a collaboration with the Italian Space Agency (ASI). The mission will deploy the Multi-Angle Imager for Aerosols (MAIA) instrument to capture information about the size and composition of particulates in the atmosphere. MAIA is anticipated to launch in late 2024, according to sources interviewed for this story.
Nothing Straightforward about Data Collection
One of those sources, David Diner, is a senior research scientist at NASA’s Jet Propulsion Laboratory and principal investigator on the MAIA project. As NASA’s first competitively selected Earth-observing mission to include epidemiologists on the scientific team, Diner says the project will help support studies of health effects from short-term, long-term, and gestational exposures to various air pollutants.
For some localities, the information MAIA will provide could be revolutionary. Studies are showing that fine particulate matter (PM2.5) may cause detrimental health effects at levels lower than originally thought.4,5 The World Health Organization (WHO) has dropped its annual average PM2.5 exposure guideline from 10 μg/m3 to 5 μg/m3.6 In the United States, the Biden administration convened a scientific panel in 2022 to discuss lowering the U.S. annual average exposure standard for PM2.5 from 12 μg/m3 to as low as 8 μg/m3.7
According to several epidemiologists working with MAIA—such as Beate Ritz from the University of California, Los Angeles—public health officials in low- and middle-income countries (LMICs) might not have as much influence as those who advise the U.S. president or the WHO. “Many government leaders will say, ‘What you see in the West does not apply to us; our situation is different,’” says Ritz. “But with MAIA’s data, hopefully, health researchers in those countries will be able to finally pressure their governments into action.”
Pros and Cons of Satellite Data
More than two decades ago, Diner helped design the Multi-angle Imaging SpectroRadiometer (MISR), which monitors how aerosols and clouds affect climate.8 MISR was designed to view Earth from nine fixed angles and in four spectral bands (blue, green, red, and near-infrared), with a spatial resolution of a few hundred meters. Its data offer a detailed picture of how the abundance of atmospheric particulate matter varies in space and time, among other outputs.9 MAIA will add to that by using a specialized digital camera that can be pointed at selected view angles and record light in 14 spectral bands.
MISR images over India in 2016. Left, the vertical viewing camera captured the Himalayas, which tend to concentrate pollution to the south. New Delhi was under such thick haze that aerosol optical depths, right, were not calculated because the algorithm classified the haze as a cloud.33 Image: Courtesy of MISR Team, NASA/GSFC/LaRC/JPL.
Side-by-side visual and digital satellite images of air pollution over India
Diner explains that the additional spectral bands will yield greater insight into particle composition. For instance, the shortest (ultraviolet) bands are sensitive to particles containing certain types of minerals and organic matter; the visible and near-infrared spectra are primarily sensitive to fine particles. Shortwave infrared bands will provide information about coarser particles, such as dust grains and volcanic ash. Additional bands measure polarization, providing another means of understanding particle properties.
“We’re basically looking at scattered sunlight and using the characteristics of that sunlight—its variation with wavelength and angle—to infer the physical and optical properties of the particles that are scattering the light,” says Diner. The shape of particles—whether spherical, like liquid droplets, or irregular, like dust grains—affects how they scatter sunlight, he explains. “All of these factors together make this multi-angle method, in conjunction with the multispectral and polarimetric measurements, pretty powerful for helping to identify specific pollutants.”
Nevertheless, it is tricky to translate satellite data into accurate estimates of what humans breathe at ground level. “Satellite-based PM2.5 data are not direct measurements,” explains Meredith Fowlie, an energy economist at the University of California, Berkeley. “They are noisy estimates, and it requires a lot of scientific modeling, assumptions, and statistical modeling to construct the [ground-level] PM measures.”
Joel D. Kaufman, a professor of environmental health at the University of Washington and editor-in-chief of Environmental Health Perspectives, takes it a step further. “Even when the satellite estimates can get pollution concentrations correct at a 1×1km resolution—which is much better than no data at all—they are missing the finer scale of important pollutant gradients that exist over tens of meters from important sources,” he says. “Sources like roadways and waste incineration can be important at this scale.”
MAIA will acquire multiple-angle imagery of selected target areas both along the orbit path as well as across it. Image: Courtesy of NASA/Jet Propulsion Laboratory, California Institute of Technology.
Graphic of different camera angles as MAIA moves over southwest North America
Diner and the MAIA team acknowledge the trade-offs between satellite and surface-based approaches. “With certain pollutants, the spatial variability is quite significant, and so a straight interpolation could, itself, lead to substantial errors,” he says. But there are simply not enough ground-based monitors in many parts of the world—especially in LMICs—to determine what residents are exposed to on a daily basis, he explains.
Target Areas
The mission will target 11 primary areas—regions surrounding Los Angeles, Atlanta, Boston, Barcelona, Rome/Bologna, Addis Ababa, Johannesburg/Pretoria, Tel Aviv/Haifa, Delhi, Beijing, and Taipei/Kaohsiung—and approximately two dozen secondary sites.10 The group will use data from both satellite- and ground-based instruments to generate pollution maps. But not every target city has a network of speciation monitors for determining chemical species and amounts of collected particles.
MAIA researchers will use the speciation monitors already in place in the primary target areas in the United States, Spain, and Italy, Diner says. And thanks to a grassroots program called the Surface Particulate Matter Network (SPARTAN), which works to put air pollution monitors in international locations, three other cities (Pretoria, Tel Aviv, and Beijing) received speciation monitors before the MAIA project started.11 Additional monitors are now sited in Johannesburg, Addis Ababa, Haifa, Delhi, Taipei, and Kaohsiung. The MAIA team is putting aerosol mass and optical depth monitors (another type of filter-based sampler) in several other primary target cities.
“The result will be that each primary target area will have at least two speciation monitors capable of collecting particulate matter on filters, which are then analyzed in the lab for sulfates, nitrates, elemental carbon, organic carbon, and dust,” Diner says. Aethalometers, which measure airborne black carbon concentrations in real time, have been placed in Addis Ababa, Delhi, and Beijing. Such ground-based measurements are essential to MAIA’s approach for converting the satellite measurements into reliable estimates of pollution concentrations. The U.S. Department of State has assisted with the shipment and operation of many of these sensors, and the U.S. Agency for International Development is supporting the ground-based measurements and chemical analyses in MAIA’s primary target areas in Africa.
The MAIA team is supporting other creative solutions to the scarcity of monitors. For instance, Addis Ababa is the primary target area with the fewest air quality monitors, so MAIA researchers have installed a network of PurpleAir monitors—low-cost sensors that measure total PM2.5 and provide citizen scientist and community groups with measures of local air quality.12
Amid visible smog, children play in the street in Addis Ababa on 3 February 2019, a car-free day promoted to help reduce air pollution. MAIA will bring new ground-based monitors to the city to complement satellite data. Image: © Eduardo Soteras/AFP via Getty Images.
Kids playing soccer in an empty city street on hazy day
Chemical transport modeling for each target area will factor in emissions from both anthropogenic and natural sources, the temporal and spatial distribution of those emissions, and their chemical reactions. Such modeling can account for wind movement, trapping by cold air, and other local weather phenomena.13 The resulting estimates will help fill in spatial and temporal gaps in the satellite coverage.14 “It’s really three points on the triangle,” Diner says. “The satellite, the surface monitors, and the model [are] all working together.”
The newly minted NASA-ASI collaboration means that more of Italy will potentially be studied than had been proposed in the original MAIA mission. Massimo Stafoggia, an epidemiologist with the Lazio Regional Health Service in Rome and long-time MAIA team member, says the central part of the country, including Rome and the Po Valley—which, due to traffic, industry, and geography, has the highest levels of PM2.5 in all of Europe—will remain one of the mission’s target areas. But, he says, ASI researchers now hope to study some secondary target areas in the north and in the south—including Taranto, home to one of Europe’s largest steel factories.
“There is a large population living very, very close to that plant,” Stafoggia says, “and because of that, it’s a very hot topic in that area.” He is involved in several longitudinal population-based air pollution studies, including some where investigators have tried to estimate the components of PM2.5 that residents are breathing in certain cities.15 “But,” he says, “I am pretty sure that the data made available [from] MAIA will be much more accurate.”
Collaborating with Epidemiologists
Diner says he has learned a great deal from working with the team’s epidemiologists to design the mission so its data will best support research on health effects. “When we were choosing our target areas, I would have thought to look only at the most polluted places,” he says. “But the epidemiologists said, ‘No, we want to look at clean places, too, because when you look at a 10-μg/m3 change in an aerosol or particulate matter concentration where it’s relatively clean, it’s a much bigger change than in a place that’s very polluted.’ So they want to see the whole range.” Diner says MAIA will fly over each area at least three times per week.
MAIA team members who specialize in remote sensing helped Diner select the spectral bands. The team epidemiologists weighed in on other constraints. For instance, if there are two targets that could be observed in one area as the satellite passes overhead, which one would they prioritize?
Yang Liu, who studies air pollution at Emory University’s Rollins School of Public Health, leads the group that is integrating instrument data with atmospheric chemistry and spatial statistical models. The operational algorithm they create will translate the satellite signals into concentrations of the same pollutants collected by the speciation monitors, at a resolution of 1 km2.16,17
Epidemiologists will compare those maps with local records to study health effects associated with PM2.5 exposure in the short and long terms, as well as during pregnancy.18 An estimated 4.1 million deaths globally are attributed to PM2.5,19 and investigators have shown links between exposure to fine particles and “everything from simple asthma exacerbation20 to heart attacks21 to cancer,22” says Liu. “In recent years, it has been linked to cognitive changes23 and birth defects.24 It has such a broad spectrum of impacts.”
Birth outcomes are among the health effects that MAIA researchers will study. Preeclampsia and low birthweight are some of the pregnancy and childbirth concerns already linked to air pollution.27 Image: © iStock.com/FatCamera.
Smiling woman holding a baby who is looking up at her
Short-term exposures (daily to monthly) are perhaps the easiest to study. Liu says investigators in the Atlanta region will use MAIA data to generate pollution estimates, then compare those to emergency department visits for problems such as asthma attacks and cardiovascular disease. Long-term (multiyear) exposures will be compared with hospital or physician office visits for cardiovascular disease. Pregnancy-related exposures will be studied by comparing hospital or physician office visits for preeclampsia, low-birthweight infants, and other pregnancy and childbirth concerns with PM2.5 levels in the preceding 12 months. Given that adverse cardiovascular, respiratory, and birth outcomes have already been linked to air pollution,25–27 these studies can confirm whether MAIA findings regarding PM2.5 constituents are indeed valid for health effects studies in places without ground-based monitors.
Electronic medical records simplify health studies, and Liu will have little trouble getting digital emergency department records in Georgia. However, not all hospitals in LMICs are set up to keep digital records.28 In South Africa, for instance, public hospitals have struggled to adopt them,29 so MAIA researcher Janine Wichmann, an epidemiologist at the University of Pretoria, will use records from private hospitals. This will skew the cohort toward more affluent patients, she says, “but at least, perhaps, one can then compare it to, say, European or U.S. studies.”
For pregnancy studies, MAIA researchers will need access to birth certificates. Ritz says MAIA-affiliated researchers in countries without electronic recordkeeping, such as India and Ethiopia, will have to gather paper birth certificates. “You have to have special outreach efforts,” she says.
Keeping It Local
Wichmann and her colleagues are pleased that the MAIA team has made partnering with local researchers a priority. “Parachute science is my pet peeve,” says Rebecca Garland, an atmospheric scientist at the University of Pretoria, referring to studies conducted by outside researchers without involving local expertise or sharing findings. “Parachute scientists don’t engage with local researchers or stakeholders, so their work, and thus their results, aren’t always relevant or needed. Even if it is cool science, if it is not a priority of the local stakeholders it won’t have any impact. As these [parachute] researchers are working alone, they don’t know the actual gaps or priorities.”
Once the MAIA data are available, Garland, Langerman, and Wichmann hope to see a more complete picture of how different pollutants move through the province at different times of the year. This will be important for establishing policies that protect health, Garland says. Many countries around the world have no air quality standards at all.30 South Africa adopted air quality limits for eight priority pollutants in 2004 and began setting up a network of air monitors.31 That was a good start, says Wichmann, but more still needs to be done—South Africa’s air quality standards are less stringent than WHO guidelines. For instance, the WHO guideline for sulfur dioxide is an average of 40 μg/m3 over 24 hours, compared with the South African standard of 125 μg/m3. “On the vast majority of days, the WHO standard is being exceeded [in the province],” says Wichmann. “But according to South African officials, there’s no pollution problem.”
SPARTAN monitors are improving ground-based air monitor coverage in the MAIA primary target areas. Here, Siyabonga Simelane, a PhD student at the University of Johannesburg, stands with a new monitor installed by SPARTAN in the capital city (visible in the background). Image: Courtesy Kristy Langerman/University of Johannesburg.
Smiling African man on rooftop by air monitoring equipment
Wichmann is dubious that any amount of data will trigger South African regulatory agencies to rethink the nation’s air quality standards. The country’s economy is struggling, she says, and politicians tend to think that the economy should take precedence over environmental standards.
Another epidemiologist on the MAIA project, Bart Ostro, from the University of California, Davis, offers a more optimistic view. He has seen officials change their outlook on air pollution once they are presented with analysis of local data. For example, in the 1990s, Ostro helped health researchers in Chile and Thailand determine that the particulate matter they were measuring was associated with a multitude of health effects.32
“I was told later that [this research] had had a very important role in awakening scientists and politicians to the air pollution issue in these countries,” Ostro says. “Conducting studies in people’s own countries can really be a wake-up call.”
Nancy Averett writes about science and the environment from Cincinnati, OH. Her work has been published in Scientific American, Discover, Audubon, Sierra, and a variety of other publications.
==== Refs
References
1. Republic of South Africa Department of Environment, Forestry and Fisheries. 2020. The Draft Second Generation Air Quality Management Plan for Vaal Triangle Airshed Priority Area. https://cer.org.za/wp-content/uploads/2005/09/Draft-Second-Generation-Air-Quality-Management-Plan-For-Vaal-Triangle.pdf [accessed 14 February 2023].
2. Zimbology. 2022. Twitter.com, 9 June 2022. https://twitter.com/zimbology/status/1534895989466845191 [accessed 14 February 2023].
3. Garland R. 2022. The air in the South African Highveld cities smells foul in the winter: here’s why. The Conversation, Environment + Energy section. 19 June 2022. https://theconversation.com/the-air-in-south-african-highveld-cities-smells-foul-in-the-winter-heres-why-184884 [accessed 14 February 2023].
4. Strak M, Weinmayr G, Rodopoulou S, Chen J, de Hoogh K, Andersen ZJ, et al. 2021. Long term exposure to low level air pollution and mortality in eight European cohorts within the ELAPSE project: pooled analysis. BMJ 374 :n1904, PMID: , 10.1136/bmj.n1904.34470785
5. Johnson M, Brook JR, Brook RD, Oiamo TH, Luginaah I, Peters PA, et al. 2020. Traffic-related air pollution and carotid plaque burden in a Canadian city with low-level ambient pollution. J Am Heart Assoc 9 (7 ):e013400, PMID: , 10.1161/JAHA.119.013400.32237976
6. World Health Organization. 2021. WHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide. https://apps.who.int/iris/handle/10665/345329 [accessed 14 February 2023].
7. Clean Air Scientific Advisory Committee. 2022. Clean Air Scientific Advisory Committee, to Michael S. Regan, Administrator, U.S. Environmental Protection Agency. CASAC review of the EPA’s Policy Assessment for the Reconsideration of the National Ambient Air Quality Standards for Particulate Matter (External Review Draft –October 2021). EPA-CASAC-22-002. 18 March 2022. https://casac.epa.gov/ords/sab/f?p=105:12:8943953574498:::12f [accessed 14 February 2023].
8. Jet Propulsion Laboratory, California Institute of Technology, National Aeronautics and Space Administration. 2023. MISR: mission. [Website.] https://misr.jpl.nasa.gov/mission/ [accessed 14 February 2023].
9. Diner DJ, Beckert JC, Reilly TH, Bruegge CJ, Conel JE, Kahn R, et al. 1998. Multi-angle imaging SpectroRadiometer (MISR) description and experiment overview. IEEE Trans Geosci Rem Sens 36 (4 ):1072–1087, 10.1109/36.700992.
10. JPL (Jet Propulsion Laboratory, California Institute of Technology, National Aeronautics and Space Administration). 2023. The MAIA investigation. Primary target areas. [Website.] https://maia.jpl.nasa.gov/investigation/#target_areas [accessed 14 February 2023].
11. SPARTAN Network. 2022. SPARTAN: a global particulate matter network. [Website.] https://www.spartan-network.org/ [accessed 14 February 2023].
12. Morawska L, Thai PK, Liu X, Asumadu-Sakyi A, Ayoko G, Bartonova A, et al. 2018. Applications of low-cost sensing technologies for air quality monitoring and exposure assessment: how far have they gone? Environ Int 116 :286–299, PMID: , 10.1016/j.envint.2018.04.018.29704807
13. Matthias V, Arndt JA, Aulinger A, Bieser J, Denier van der Gon H, Kranenburg R, et al. 2018. Modeling emissions for three-dimensional atmospheric chemistry transport models. J Air Waste Manag Assoc 68 (8 ):763–800, PMID: , 10.1080/10962247.2018.1424057.29364776
14. Zhang H, Wang J, Janechek N, Zhou M, Ge C, Castro Garcia L, et al. 2021. Development of UI-WRF-Chem for MAIA satellite mission: case demonstration. In: AGU Fall Meeting 2021. 13–17 December 2021. New Orleans, LA, A15C-1634. https://ui.adsabs.harvard.edu/abs/2021AGUFM.A15C1634Z/abstract [accessed 14 February 2023].
15. Rodopoulou S, Stafoggia M, Chen J, de Hoogh K, Bauwelinck M, Mehta AJ, et al. 2022. Long-term exposure to fine particle elemental components and mortality in Europe: results from six European administrative cohorts within the ELAPSE project. Sci Total Environ 809 :152205, PMID: , 10.1016/j.scitotenv.2021.152205.34890671
16. Emory University. 2020. Emory scientists working with NASA to map air pollution and its impact on health. [Press release.] 8 April 2020. https://news.emory.edu/stories/2020/04/nasa_maia_yang_liu/index.html [accessed 14 February 2023].
17. Yang CE, Fu JS, Liu Y, Dong X, Liu Y. 2022. Projections of future wildfires impacts on air pollutants and air toxics in a changing climate over the western United States. Environ Pollut 304 :119213, PMID: , 10.1016/j.envpol.2022.119213.35351594
18. JPL. 2023. Science Objectives. [Website.] https://maia.jpl.nasa.gov/science-objectives/ [accessed 14 February 2023].
19. Wu Y, Song P, Lin S, Peng L, Li Y, Deng Y, et al. 2021. Global burden of respiratory diseases attributable to ambient particulate matter pollution: findings from the Global Burden of Disease Study 2019. Front Public Health 9 :740800, PMID: , 10.3389/fpubh.2021.740800.34888281
20. Anenberg SC, Henze DK, Tinney V, Kinney PL, Raich W, Fann N, et al. 2018. Estimates of the global burden of ambient PM2.5, ozone, and NO2 on asthma incidence and emergency room visits. Environ Health Perspect 126 (10 ):107004, PMID: , 10.1289/EHP3766.30392403
21. Liang F, Liu F, Huang K, Yang X, Li J, Xiao Q, et al. 2020. Long-term exposure to fine particulate matter and cardiovascular disease in China. J Am Coll Cardiol 75 (7 ):707–717, PMID: , 10.1016/j.jacc.2019.12.031.32081278
22. Wang N, Mengersen K, Tong S, Kimlin M, Zhou M, Liu Y, et al. 2020. County-level variation in the long-term association between PM2.5 and lung cancer mortality in China. Sci Total Environ 738 :140195, PMID: , 10.1016/j.scitotenv.2020.140195.32806350
23. Weuve J, Bennett EE, Ranker L, Gianattasio KZ, Pedde M, Adar SD, et al. 2021. Exposure to air pollution in relation to risk of dementia and related outcomes: an updated systematic review of the epidemiological literature. Environ Health Perspect 129 (9 ):96001, PMID: , 10.1289/EHP8716.34558969
24. Girguis MS, Strickland MJ, Hu X, Liu Y, Bartell SM, Vieira VM. 2016. Maternal exposure to traffic-related air pollution and birth defects in Massachusetts. Environ Res 146 :1–9, PMID: , 10.1016/j.envres.2015.12.010.26705853
25. U.S. EPA (U.S. Environmental Protection Agency). 2019. Integrated Science Assessment (ISA) for Particulate Matter (Final Report, Dec 2019). EPA/600/R-19/188. https://cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid=347534 [accessed 14 February 2023].
26. Yang Y, Ruan Z, Wang X, Yang Y, Mason TG, Lin H, et al. 2019. Short-term and long-term exposures to fine particulate matter constituents and health: a systematic review and meta-analysis. Environ Pollut 247 :874–882, PMID: , 10.1016/j.envpol.2018.12.060.30731313
27. Klepac P, Locatelli I, Korošec S, Künzli N, Kukec A. 2018. Ambient air pollution and pregnancy outcomes: a comprehensive review and identification of environmental public health challenges. Environ Res 167 :144–159, PMID: , 10.1016/j.envres.2018.07.008.30014896
28. Ngugi P, Were MC, Babic A. 2018. Facilitators and barriers of electronic medical records systems implementation in low resource settings: a holistic view. Stud Health Technol Inform 251 :187–190, PMID: , 10.3233/978-1-61499-880-8-187.29968634
29. Msomi M, Kalusopa T, Luthuli LP. 2021. Change management in the implementation of electronic health records (EHR) systems at Inkosi Albert Luthuli Central Hospital, South Africa. S Afr J Libr Inf Sci 87 (2 ):1–10, 10.7553/87-2-2107.
30. United Nations Environment Programme. 2021. One in three countries in the world lack any legally mandated standards for outdoor air quality. [Press release.] 2 September 2021. https://www.unep.org/news-and-stories/press-release/one-three-countries-world-lack-any-legally-mandated-standards [accessed 14 February 2023].
31. Republic of South Africa Department of Department of Planning, Monitoring & Evaluation. 2009. Chapter 3: air quality standards and objectives. In: State of Air Report 2005: A Report on the State of the Air in South Africa. https://evaluations.dpme.gov.za/evaluations/102 [accessed 21 February 2023].
32. Ostro B, Sanchez JM, Aranda C, Eskeland GS. 1996. Air pollution and mortality: results from a study of Santiago, Chile. J Expo Anal Environ Epidemiol 6 (1 ):97–114, PMID: .8777376
33. JPL. 2016. PIA21100: severe air pollution in New Delhi view by NASA’s MISR. https://photojournal.jpl.nasa.gov/catalog/PIA21100 [accessed 14 February 2023].
|
PMC010xxxxxx/PMC10044337.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36975774
EHP12535
10.1289/EHP12535
Invited Perspective
Invited Perspective: Environmental Chemical-Sensing AHR Remains an Enigmatic Key Player in Toxicology
https://orcid.org/0000-0002-1992-4145
Kaplan Barbara L.F. 1
Lawrence B. Paige 2
1 Center for Environmental Health Sciences, Department of Comparative Biomedical Sciences, College of Veterinary Medicine, Mississippi State University, Mississippi, USA
2 Department of Environmental Medicine, University of Rochester School of Medicine and Dentistry, Rochester, New York, USA
Address correspondence to Barbara L.F. Kaplan, Center for Environmental Health Sciences, Department of Comparative Biomedical Sciences, 240 Wise Center Dr., Mississippi State University, Mississippi State, MS 39762 USA. Email: [email protected]
28 3 2023
3 2023
131 3 03130705 12 2022
03 1 2023
09 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors are supported in part by the following grants from the U.S. National Institutes of Health, National Institute of Environmental Health Sciences: R15ES027650 (B.L.F.K.), P30ES01247 (B.P.L.), and R01ES030300 (B.P.L.). Both authors declare they have no potential or actual conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11580
==== Body
pmcOver the last several years, the aryl hydrocarbon receptor (AHR) has gone from being characterized as not only a receptor that modulates cellular responses to myriad external environmental changes but also as one that regulates intricate aspects of immune homeostasis in response to endogenously produced ligands. Indeed, several endogenous AHR ligands have now been identified, including chemicals derived from tryptophan, phytochemicals, or commensal microbiota.1,2 Although AHR might play many roles, no doubt it remains a critical receptor in modulating immune responses within the context of exposure to environmental chemicals.
In this issue of Environmental Health Perspectives, Liu et al. provide a detailed characterization of immune effects of fine particulate matter [PM with an aerodynamic diameter of ≤2.5μm (PM2.5)] obtained from atmospheric monitoring stations in Taiwan.3 This PM2.5 contained various polycyclic aromatic hydrocarbons (PAHs), including indeno[1,2,3-cd]pyrene (IP).4 They observed that intratracheal administration of PM2.5 or IP alone exacerbated pathology and modulated immune responses by using a mouse model for house dust mite (HDM)-mediated asthma.3 Importantly, they used mass cytometry to determine that both PM2.5 and IP exposure increased the percentage and number of TCRβ+CD4+CXCR5+Bcl-6+PD-1hi T follicular helper (Tfh) cells in the lung-draining lymph nodes. In addition, the enhancement in Tfh cells was dependent on AHR expression in T cells, at least for PM2.5. Tfh cells are CD4+ T cells that play a critical role in inducing isotype-switched antibody responses in germinal centers (GCs).5 Therefore, it was important that they also demonstrated that IP enhanced levels of HDM-specific immunoglobulin E (IgE) and IgG in serum.3 Previously, they had demonstrated that IP-mediated exacerbation in a mouse ovalbumin-induced model of allergic lung inflammation was attenuated with the AHR antagonist CH223191.4
Interestingly, the observation that IP, as an AHR ligand, increased the percentage and number of Tfh cells contradicts another recent paper, in which other AHR ligands suppressed Tfh cell percentage and number in mice infected with influenza A virus.6 In that paper, the environmental contaminant 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), the tryptophan derivative 2-(1′H-indole-3′-carbonyl)-thiazole-4-carboxylic acid methyl ester (ITE), and the endogenous ligand kynurenic acid (KYNA) all suppressed CD4+CD44+CXCR5+PD-1hi Tfh in infected mice.6 Using TCDD, Houser and Lawrence also showed in lung-draining lymph nodes that AHR activation significantly suppressed Tfh cell number as early as day 3 postinfection, that the GC B cell number directly correlated with Tfh cell number and, further, that AHR activation suppressed GC B cells. Consistent with these findings, AHR activation with ligands, including TCDD and ITE, also reduced virus-specific IgG levels in plasma.6
The juxtaposition of the results from these papers displays a common theme in the AHR field: Different AHR ligands produce different, and sometimes opposite, effects in the immune system (Figure 1). The reasons for this remain uncertain, but they could be due to ligand affinity, ligand bioavailability or metabolism, strength of response, or differences in the immune stimulus. Indeed, the two juxtaposed studies used different disease models, with Houser and Lawrence6 using a replicating pathogen (influenza virus), and Liu et al.3 using a representative allergen (HDM).
Figure 1. Relationship between AHR ligand affinity, metabolism, and biological effect. The bubble plot depicts the relative biological or toxicological consequences of agonists that bind AHR with either greater or lesser affinity and that are metabolized either more or less rapidly. The size of the bubble denotes relative effect. Note: AHR, aryl hydrocarbon receptor; FICZ, 6-formylindolo(3,2-b)carbazole; IP, indeno[1,2,3-cd]pyrene; ITE, (1′H-indole-3′-carbonyl)-thiazole-4-carboxylic acid methyl ester; KynA, kynurenic acid; PAHs, polyaromatic hydrocarbons; PCBs, polychlorinated biphenyls; TCDD, 2,3,7,8-tetrachlorodibenzo-p-dioxin.
Figure 1 is a bubble chart, plotting signal duration, ranging from transient to enduring (y-axis) across interaction with aryl hydrocarbon receptor, ranging from low affinity to high affinity (x-axis) for 6-formylindolo(3,2-b)carbazole, indeno[1,2,3-cd]pyrene, (1′ H-indole-3′-carbonyl)-thiazole-4-carboxylic acid methyl ester, kynurenic acid, polyaromatic hydrocarbons, polychlorinated biphenyls, and 2,3,7,8-tetrachlorodibenzo-p-dioxin.
However, even in the same model system, two different AHR ligands can produce opposite effects. For instance, during influenza A virus infection in which TCDD suppressed Tfh cell percentage and number, the tryptophan derivative 6-formylindolo(3,2-b)carbazole (FICZ)—which, like TCDD, binds to AHR with high affinity—elevated Tfh cell percentage and number in the lung-draining lymph nodes.7 In addition to this contrasting effect on Tfh cells in influenza virus–infected mice,7 opposite effects of AHR ligands were observed in the experimental autoimmune encephalomyelitis model of multiple sclerosis, in which TCDD suppressed but FICZ exacerbated disease.8,9 Thus, ligand source might not necessarily be an ideal predictor of whether the outcome will be beneficial or detrimental.
The demonstration that two environmentally relevant ligands exhibit opposite effects on the Tfh cell population (i.e., IP enhanced and TCDD suppressed)3,6 could also be explained by differences in AHR affinity, ligand metabolism, or bioavailability. IP was determined to have a half maximal effective concentration (EC50) for induction of ethoxyresorufin-O-deethylase, an enzymatic measure of AHR-responsive cytochrome P4501A1, of 5.2×10−1 μM, which is 10,000 times less effective than TCDD.10 Unlike TCDD, which is relatively resistant to metabolism,11–13 there were several hydroxylated metabolites of IP identified in vitro using rat liver enzyme homogenates.14 Further support that metabolism of AHR ligands can influence immune effects, such as CD4+CD44hiCXCR5+PD-1+ Tfh populations in influenza virus infection was shown by comparing TCDD and FICZ in mice.7 Specifically, oral TCDD (administered once) suppressed the percentage and number of Tfh cells in the lung-draining lymph nodes, whereas oral FICZ (administered once per day for 8 d) had no effect. However, in cytochrome P 450, family 1, subfamily A, polypeptide 1 (Cyp1a1−/−) mice infected with influenza virus, both TCDD and FICZ suppressed the number of Tfh cells.7 This indicates that attenuating the clearance of FICZ in vivo produced an effect that was not observed in the presence of normal metabolism. Thus, the metabolism of ligands, regardless of whether the ligands are endogenous or exogenous, likely influences their effects in immune cells.
Another issue that cannot be ignored, regardless of the source of AHR ligand being studied, is the fact that exposure to a single chemical is unlikely. In their paper, Liu et al. acknowledged this critical fact by showing that exposure to IP, both on its own and in a PM2.5 mixture, exacerbated the HDM-induced Tfh populations.3 In addition to drilling down from PM2.5 to PAH to a specific moiety, the leveraging of in vitro results strengthened the in vivo findings. Specifically, Liu et al. differentiated Tfh cells in vitro and connected altered Tfh cells to changes in gene expression.3 In particular, the associated increase in Il4 and Il21 gene expression was found to be due, in part, to AHR binding to xenobiotic response elements in the promoters of both Il4 and Il21. Importantly, Tfh cells producing interleukin-4 (IL-4) is necessary for IgE class switch15–17 and, as noted, IP increased HDM-mediated IgE.3
Overall, the work by Liu et al.3 advances the field of immunotoxicology through the use of mass cytometry to identify a specific immune cell population that was sensitive to regulation by an environmental contaminant via AHR. Further, although this is likely not the sole consequence of AHR activation by IP, or by PAHs in general, the authors identified a specific role that AHR played in regulating Il4 and Il21 gene expression.3 This work provides a compelling and complete story about how the environmentally relevant AHR ligand IP-induced respiratory disease in vivo identified that exacerbated disease directly correlated with an increased Tfh cell population and IgE production and then showed that AHR was involved in IP-induced Il4 production in vitro, something that is necessary for IgE production.
It will be exciting to see where future work leads, including further investigation into the mechanisms by which IP alone, and in combination with other contaminants, alters immune responses. For instance, it was recently suggested by using in silico modeling that IP would bind to toll-like receptor 4 with high affinity.18 This is an intriguing finding, considering that Liu et al.3 found that IP exacerbated airway disease. Hopefully, future research will also identify the mechanistic explanation as to why different AHR ligands sometimes cause similar effects but at other times can cause different, and even completely divergent, consequences. Understanding how differences in AHR affinity, ligand metabolism, and other characteristics influence immune function will be critical to parsing and predicting the immunological consequences of exposure to the myriad ligands of this fascinating environment-sensing transcription factor.
==== Refs
References
1. Lin L, Dai Y, Xia Y. 2022. An overview of aryl hydrocarbon receptor ligands in the last two decades (2002–2022): a medicinal chemistry perspective. Eur J Med Chem 244 :114845, PMID: , 10.1016/j.ejmech.2022.114845.36274276
2. Rannug A. 2022. 6-Formylindolo[3,2-b]carbazole, a potent ligand for the aryl hydrocarbon receptor produced both endogenously and by microorganisms, can either promote or restrain inflammatory responses. Front Toxicol 4 :775010, PMID: , 10.3389/ftox.2022.775010.35295226
3. Liu KY, Gao Y, Xiao W, Fu J, Huang S, Han X, et al. 2023. Multidimensional analysis of lung lymph nodes in a mouse model of allergic lung inflammation following PM2.5 and indeno[1,2,3-cd]pyrene exposure. Env Health Perspect 131 (3 ):037014, 10.1289/EHP11580.36975775
4. Wong TH, Lee CL, Su HH, Lee CL, Wu CC, Wang CC, et al. 2018. A prominent air pollutant, indeno[1,2,3-cd]pyrene, enhances allergic lung inflammation via aryl hydrocarbon receptor. Sci Rep 8 (1 ):5198, PMID: , 10.1038/s41598-018-23542-9.29581487
5. Ribeiro F, Perucha E, Graca L. 2022. T follicular cells: the regulators of germinal center homeostasis. Immunol Lett 244 :1–11, PMID: , 10.1016/j.imlet.2022.02.008.35227735
6. Houser CL, Lawrence BP. 2022. The aryl hydrocarbon receptor modulates T follicular helper cell responses to influenza virus infection in mice. J Immunol 208 (10 ):2319–2330, PMID: , 10.4049/jimmunol.2100936.35444027
7. Boule LA, Burke CG, Jin GB, Lawrence BP. 2018. Aryl hydrocarbon receptor signaling modulates antiviral immune responses: ligand metabolism rather than chemical source is the stronger predictor of outcome. Sci Rep 8 (1 ):1826, PMID: , 10.1038/s41598-018-20197-4.29379138
8. Quintana FJ, Basso AS, Iglesias AH, Korn T, Farez MF, Bettelli E, et al. 2008. Control of Treg and TH17 cell differentiation by the aryl hydrocarbon receptor. Nature 453 (7191 ):65–71, PMID: , 10.1038/nature06880.18362915
9. Veldhoen M, Hirota K, Westendorf AM, Buer J, Dumoutier L, Renauld JC, et al. 2008. The aryl hydrocarbon receptor links TH17-cell-mediated autoimmunity to environmental toxins. Nature 453 (7191 ):106–109, PMID: , 10.1038/nature06881.18362914
10. Bosveld ATC, de Bie PA, van den Brink NW, Jongepier H, Klomp AV. 2002. In vitro EROD induction equivalency factors for the 10 PAHs generally monitored in risk assessment studies in the Netherlands. Chemosphere 49 (1 ):75–83, PMID: , 10.1016/S0045-6535(02)00161-3.12243332
11. Gasiewicz TA, Geiger LE, Rucci G, Neal RA. 1983. Distribution, excretion, and metabolism of 2,3,7,8-tetrachlorodibenzo-p-dioxin in C57BL/6J, DBA/2J, and B6D2F1/J mice. Drug Metab Dispos 11 (5 ):397–403, PMID: .6138222
12. Olson JR. 1986. Metabolism and disposition of 2,3,7,8-tetrachlorodibenzo-p-dioxin in guinea pigs. Toxicol Appl Pharmacol 85 (2 ):263–273, PMID: , 10.1016/0041-008x(86)90121-3.3764914
13. Sorg O, Zennegg M, Schmid P, Fedosyuk R, Valikhnovskyi R, Gaide O, et al. 2009. 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) poisoning in Victor Yushchenko: identification and measurement of TCDD metabolites. Lancet 374 (9696 ):1179–1185, PMID: , 10.1016/S0140-6736(09)60912-0.19660807
14. Rice JE, Coleman DT, Hosted TJ Jr, LaVoie EJ, McCaustland DJ, Wiley JC Jr. 1985. Identification of mutagenic metabolites of indeno[1,2,3-cd]pyrene formed in vitro with rat liver enzymes. Cancer Res 45 (11 pt 1 ):5421–5425, PMID: .4053016
15. Liang HE, Reinhardt RL, Bando JK, Sullivan BM, Ho IC, Locksley RM. 2012. Divergent expression patterns of IL-4 and IL-13 define unique functions in allergic immunity. Nat Immunol 13 (1 ):58–66, PMID: , 10.1038/ni.2182.22138715
16. Meli AP, Fontés G, Leung Soo C, King IL. 2017. T follicular helper cell-derived IL-4 is required for IgE production during intestinal helminth infection. J Immunol 199 (1 ):244–252, PMID: , 10.4049/jimmunol.1700141.28533444
17. Noble A, Zhao J. 2016. Follicular helper T cells are responsible for IgE responses to Der p 1 following house dust mite sensitization in mice. Clin Exp Allergy 46 (8 ):1075–1082, PMID: , 10.1111/cea.12750.27138589
18. Cabral MB, Dela Cruz CJ, Sato Y, Oyong G, Rempillo O, Galvez MC, et al. 2022. In silico approach in the evaluation of pro-inflammatory potential of polycyclic aromatic hydrocarbons and volatile organic compounds through binding affinity to the human toll-like receptor 4. Int J Environ Res Public Health 19 (14 ):8360, PMID: , 10.3390/ijerph19148360.35886213
|
PMC010xxxxxx/PMC10044348.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36975775
EHP11580
10.1289/EHP11580
Research
Multidimensional Analysis of Lung Lymph Nodes in a Mouse Model of Allergic Lung Inflammation following PM2.5 and Indeno[1,2,3-cd]pyrene Exposure
Liu Kwei-Yan 1 2 3 *
https://orcid.org/0000-0002-1639-7909
Gao Yajing 2 3 *
Xiao Wenfeng 2 3 *
Fu Jinrong 2 4 *
Huang Saihua 2 3
Han Xiao 2 3
Hsu Shih-Hsien 5
Xiao Xiaojun 6
Huang Shau-Ku 1 6 7 8
https://orcid.org/0000-0002-6050-4951
Zhou Yufeng 2 3
1 Department of Respirology & Allergy, Third Affiliated Hospital of Shenzhen University, Shenzhen, China
2 Institute of Pediatrics, Children’s Hospital of Fudan University, National Children’s Medical Center, and the Shanghai Key Laboratory of Medical Epigenetics, International Co-laboratory of Medical Epigenetics and Metabolism, Ministry of Science and Technology, Institutes of Biomedical Sciences, Fudan University, Shanghai, China
3 National Health Commission (NHC) Key Laboratory of Neonatal Diseases, Fudan University, Shanghai, China
4 Department of General Medicine, Children’s Hospital of Fudan University, Shanghai, China
5 Graduate Institute of Medicine, College of Medicine, Kaohsiung Medical University, Kaohsiung, Taiwan
6 Institute of Allergy and Immunology, School of Medicine, Shenzhen University, Shenzhen, China
7 National Institute of Environmental Health Sciences, National Health Research Institutes, Taiwan
8 Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA
Address correspondence to Yufeng Zhou, 399 Wanyuan Rd., Minhang, Shanghai 201102, China. Telephone: 86-21-64932907. Email: [email protected]. And, Shau-Ku Huang, National Institute of Environmental Health Sciences, National Health Research Institutes, No. 35, Keyan Rd., Zhunan, 35053 Miaoli County, Taiwan. Email: [email protected]
28 3 2023
3 2023
131 3 03701419 5 2022
23 1 2023
09 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Ambient particulate matter with an aerodynamic diameter of ≤2.5 μm (PM2.5) is suggested to act as an adjuvant for allergen-mediated sensitization and recent evidence suggests the importance of T follicular helper (Tfh) cells in allergic diseases. However, the impact of PM2.5 exposure and its absorbed polycyclic aromatic hydrocarbon (PAHs) on Tfh cells and humoral immunity remains unknown.
Objectives:
We aimed to explore the impact of environmental PM2.5 and indeno[1,2,3-cd]pyrene (IP), a prominent PAH, as a model, on Tfh cells and the subsequent pulmonary allergic responses.
Methods:
PM2.5- or IP-mediated remodeling of cellular composition in lung lymph nodes (LNs) was determined by mass cytometry in a house dust mite (HDM)-induced mouse allergic lung inflammation model. The differentiation and function of Tfh cells in vitro were analyzed by flow cytometry, quantitative reverse transcription polymerase chain reaction, enzyme-linked immunosorbent assay, chromatin immunoprecipitation, immunoprecipitation, and western blot analyses.
Results:
Mice exposed to PM2.5 during the HDM sensitization period demonstrated immune cell population shifts in lung LNs as compared with those sensitized with HDM alone, with a greater number of differentiated Tfh2 cells, enhanced allergen-induced immunoglobulin E (IgE) response and pulmonary inflammation. Similarly enhanced phenotypes were also found in mice exposed to IP and sensitized with HDM. Further, IP administration was found to induce interleukin-21 (Il21) and Il4 expression and enhance Tfh2 cell differentiation in vitro, a finding which was abrogated in aryl hydrocarbon receptor (AhR)-deficient CD4+ T cells. Moreover, we showed that IP exposure increased the interaction of AhR and cellular musculoaponeurotic fibrosarcoma (c-Maf) and its occupancy on the Il21 and Il4 promoters in differentiated Tfh2 cells.
Discussion:
These findings suggest that the PM2.5 (IP)–AhR–c-Maf axis in Tfh2 cells was important in allergen sensitization and lung inflammation, thus adding a new dimension in the understanding of Tfh2 cell differentiation and function and providing a basis for establishing the environment–disease causal relationship. https://doi.org/10.1289/EHP11580
Supplemental Material is available online (https://doi.org/10.1289/EHP11580).
* These authors contributed equally to this work.
The authors declare that they have no relevant conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
T helper 2 (Th2) cells have long been considered to mainly regulate the pathogenic manifestations of allergic asthma, such as immunoglobulin E (IgE)–mediated sensitization, airway hyperresponsiveness, and eosinophil infiltration.1 However, more recent work has demonstrated that interleukin 4-positive (IL-4+) T follicular helper (Tfh) cells, not Th2 cells, are required for IgE production.2–5 Tfh cells are identified by their unique phenotype, which includes high expression levels of several markers, such as CXC-chemokine receptor 5 (CXCR5), inducible T-cell co-stimulator (ICOS), programmed cell death protein 1 (PD1), B-cell lymphoma 6 (Bcl-6, a transcriptional repressor), and IL-21 (a cytokine).6 Tfh cells are the primary helper T-cell subset responsible for directing the affinity, longevity, and isotype of antibodies produced by B cells.7,8 Owing to the fundamental role of Tfh cells in adaptive immunity, the stringent control of their production and function is critically important.
Tfh cell differentiation requires the cytokine milieu,9 that is, IL-6 initiates Tfh cell development by activating the transcriptional factor signal transducer and activator of transcription 3 (Stat3) to induce transcriptional expression of Bcl-6 and to repress type I interferon (IFN) signaling.10,11 IL-21 is also critical for Tfh cell differentiation by inducing Bcl-6 and CXCR5 expression in an autocrine manner.12,13 According to their cytokine and transcription factor profile, Tfh cells are classified into three subsets, namely, Tfh1, Tfh2, and Tfh13.14 Tfh2 cells are distinguished from others by their ability to produce abundant quantities of IL-4.15 A distinctive interleukin-4 (Il4) enhancer locus is bound by basic leucine zipper transcriptional factor activating transcription factor (ATF)–like (BATF) in Tfh2 cells, which is distinct from the regulatory element bound by GATA binding protein 3 (GATA3).16 The Tfh2-derived IL-4 cytokine is critical for IgE induction, whereas IL-21 is critical for IgG1 induction.
It has been demonstrated that ambient particulate matter (PM) with an aerodynamic diameter of ≤2.5μm (PM2.5) acts as an adjuvant for IgE-mediated allergen sensitization in both animal and human studies.17–19 However, the causal relationship and mechanisms have not been sufficiently established. Polycyclic aromatic hydrocarbons (PAHs) are a group of environmental pollutants mainly generated from the pyrolysis of organic matter and, in Asia, are mainly from traffic exhaust and industrial emissions.20 These pollutants, which are abundantly adsorbed into PM2.5, generate reactive oxygen species and induce the oxidative stress associated with inflammation.21 Recent studies have reported that PAHs are released from burning organic particles and interact with the intracellular aryl hydrocarbon receptor (AhR) to influence immune responses, highlighting the potential role of the PM2.5–PAH–AhR axis in allergic diseases.22,23 We previously reported that PM2.5 and its associated PAHs disturb the balance of T helper 17 (Th17)/regulatory T (Treg) cells to aggravate allergic lung inflammation through AhR activation in a mouse model of asthma.24 However, whether and how the PM2.5–PAH–AhR axis affects antigen sensitization and humoral immune responses remain unclear. Based on a transcriptomics analysis of in vitro differentiated Tfh cells, increased expression of AhR has been reported during Tfh cell differentiation, suggesting the potential role of the AhR–PAH axis in regulating Tfh function and subsequent humoral immune responses, but the governing mechanisms are unknown.25
To this end, we employed a multidimensional approach, including single-cell mass cytometry, to explore the causal relationship between PM2.5 exposure, remodeling of lymph node (LN) cells, humoral immunity, and allergen-induced pulmonary inflammation in a mouse model.
Materials and Methods
Mice
All mice used were on the C57BL/6 background. AhR-null mice (AhR KO; B6.129-Ahrtm1Bra/J) and AhRloxp [B6.129(FVB)-Ahrtm3.1Bra/J] mice were purchased from the Jackson Laboratory. CD4-Cre mice [Tg (Cd4-cre)1Cwi/BfluJ] were obtained from Shanghai Model Organisms. AhRloxp mice were crossed with CD4-Cre transgenic mice to obtain Ahrfl/flCD4-Cre mice (CD4ΔAhR) with Ahr deficiency in CD4+ T cells. C57BL/6 mice were obtained from the SLAC Laboratory Animal Co. All animal experiments complied with the relevant laws and institutional guidelines as overseen by the Animal Studies Committee of the Children’s Hospital of Fudan University (Shanghai, China). All the animals were housed, bred, and maintained under specific pathogen-free conditions with a temperature of 22±2°C under a 12-h light/12-h dark cycle and 40%–70% relative humidity. All mice received sterile food and filtered water. A total of 101 mice at 6–8 wk of age were used in this study. A total of 11 mice [phosphate-buffered saline (PBS) group n=3; house dust mite (HDM) group n=4; HDM plus PM2.5 group n=4] were used for mass cytometry analysis. A total of 90 mice were used for the HDM-induced mouse allergic lung inflammation model in Figure 3 (n=10 for each of the three groups, including PBS, HDM, and HDM plus IP) and Figure 4 (n=10 for each of the six groups, including wild-type PBS [(WT_PBS), WT_HDM, WT_HDM plus PM2.5, CD4ΔAhR_PBS, CD4ΔAhR_HDM, and CD4ΔAhR_HDM plus PM2.5]. All mice were sacrificed in the automated carbon dioxide (CO2) delivery system, and their lung-draining LNs and bronchoalveolar lavage fluids (BALFs) were individually harvested for subsequent flow cytometry analysis, and lung tissues were harvested and sectioned for hematoxylin and eosin (H&E) and periodic acid–Schiff (PAS) staining.
PM2.5 and PAHs
PM2.5 samples were continuously collected from June 2019 to December 2020 using Quartz fiber filters (1851-090; Whatman) and an air sampler (ZR-3930; Qingdao Junray Intelligent Instruments) on the open-air rooftop of the fourth floor of a research building of the Children’s Hospital of Fudan University. Filter papers were replaced every 2 wk. The collected filter papers absorbing PM2.5 were cut into small pieces, immersed in distilled water overnight at 4°C, and then sonicated (Skymon) for 30 min on three occasions. The PM2.5 turbid liquid after sonication was filtered through sterile gauze, lyophilized to measure the weight, and stored at −20°C. The composition analysis of PM2.5 (Figure S1A) was performed by China Certification & Inspection Group (CCIC) Physical and Chemical Testing Co., Ltd. In brief, ion chromatography was used for the detection of water-soluble cations and anions in PM2.5 samples. Inductively coupled mass spectrometry (MS) was used for the determination of metal elements. A thermal–optical method was used for the determination of organic carbon and elemental carbon. For the composition analysis by CCIC, the PAHs in PM2.5 (Figure S1B) were extracted, purified, and then analyzed by gas chromatography-MS (GC-MS; GC7890A-5975C; Agilent). PM2.5 samples were extracted with 120mL acetone/dichloromethane (1:1, vol/vol) in a Soxhlet extraction apparatus for 20 h to obtain PAHs data. After the completion of extraction, an internal standard (consisting of Acenaphthene-D10, Chrysene-D12, Naphthalene-D8, Phenanthrene-D10, and Pyrene-D12) was injected into the organic reagent to calibrate the concentration of each PAHs compound. Next, the filter was cut into pieces, and copper powder and anhydrous sodium sulfate were added to remove any contaminants. The extract was condensed with a rotary evaporator. Then, amorphous sodium sulfate, silica gel, and alumina were used to separate the aliphatic hydrocarbons and PAHs by the column. The elute solvent consisted of 15mL hexane and 70mL hexane/dichloromethane (3:7; vol/vol). The solvent was exchanged to hexane under vacuum afterward. The extracts were concentrated to 1mL under a slow nitrogen stream in a 40°C water bath. The extracts were transferred into GC bottles with lids and stored at −20°C. The samples were analyzed by GC-MS (5975C; Agilent). One microliter of the sample was injected into a capillary column in splitless mode. The selected ion monitoring mode was used to collect the data. Helium was used as the carrier gas at a flow rate of 1.0mL/min. The GC oven of the temperature program was 80°C for 1 min at the first stage, then it was increased to 235°C at a rate of 10°C/min. Next, it was increased to 300°C at a rate of 4°C/min and held for 4 min. The MS was operated in an electron impact model at 70 eV. For the animal experiment, the lyophilized PM2.5 (6.25μg/μL) was dissolved in PBS and stored at −20°C. Indeno[1,2,3-cd]pyrene (IP; ERI-001) was purchased from Sigma-Aldrich.
HDM-Induced Mouse Allergic Lung Inflammation Model
The HDM-induced mouse allergic lung inflammation model was established, as described previously,26 with slight modifications. Briefly, 6- to 8-wk-old female C57BL/6 mice (Shanghai SLAC Experimental Animals) were exposed to light anesthesia (isoflurane). The mice were then suspended by their front incisors, and their tongues were gently extended to their lower mandibles. The different allergen solutions (HDM, HDM plus PM2.5, or HDM plus IP) for different groups were delivered into the hypopharynx in 50-μL aliquots. The stock solutions were as follows: HDM (1μg/μL; Cat# XPB91D3A2.5; Greer Laboratories) and PM2.5 (6.25μg/μL) were dissolved in PBS, and the stock solution of IP (10 mM) was dissolved in dimethyl sulfoxide (DMSO) and further diluted with PBS for the working concentration (6.25μM). Mice in the HDM group were sensitized by intratracheal instillation of 10μg of HDM dissolved in 50μL of PBS per mouse on days 0, 1, and 2 and challenged on days 9, 10, 11, and 12 with the same amount of HDM. Mice in the HDM plus PM2.5 group or the IP group were sensitized by intratracheal instillation of 10μg of HDM mixed with PM2.5 (250μg/mouse) or IP (5μM/mouse) in a final volume of 50μL per mouse on days 0, 1, and 2 and challenged on days 9, 10, 11, and 12 with 10μg of HDM only. Mice in the control group received 50μL of PBS during the sensitization and the challenge phases. In addition, 0.05% equal concentrations of DMSO were used as the vehicle control in PBS and HDM group to make the comparison with of HDM plus IP group.
Mass Cytometry and Data Analysis
Mass cytometry and data analysis were performed by Zhejiang Puluoting Health Tech Co., Ltd. Briefly, mice were sacrificed in an automated CO2 delivery system on day 14, then the lung LNs of each mouse were harvested (PBS group n=3; HDM group n=4; HDM plus PM2.5 group n=4). LNs were ground with a sterile syringe piston in Dulbecco’s Modified Eagle Medium (DMEM) containing 10% fetal bovine serum (FBS), and filtered through a 70-μm cell strainer (Cat# 352350; BD Biosciences) to obtain a single-cell suspension. Then the cells were washed with PBS containing 0.5% bovine serum albumin (BSA) and centrifuged at 400×g for 5 min at 4°C. Subsequently, 3×106 cells per sample were resuspended in 100 μL of Cell-ID Cisplatin-194Pt (250 nM; Cat# 201194; Fluidigm) solution for 5 min on ice, then the cells were washed twice by adding 1mL of fluorescence-activated cell sorting (FACS) buffer (1×PBS containing 2% FBS) and centrifuged at 400×g for 5 min at 4°C. Then, to block Fc receptors, the cells were incubated with an antimouse CD16/32 monoclonal antibody (mAb) (clone 93; Cat# 101301; 1.0μg mAb/106 cells; Biolegend) in a 100 -μL volume for 10 min on ice before staining with a cocktail of metal-labeled mAbs against cell surface molecules. Cells were then treated with Fixation/Permeabilization Buffer (eBioscience) and further stained with an intracellular antibody cocktail. The cells were then incubated with Cell-ID Intercalator (191/193Ir; Cat# 201192A; Fluidigm) to discriminate single nucleated cells from doublets. The antibodies conjugated with isotopically pure elements are listed in Table S1. Then cells were washed twice with 2mL of deionized water and centrifuged at 800×g for 5 min at 4°C. Finally, the cells were resuspended with deionized water, added to 20% EQ beads (Cat# 201078; Fluidigm), and injected into the mass cytometer (Helios; Fluidigm). All instruments were evaluated to ensure performance at or above the minimum Helios system specifications for calibration. Following the instrument tuning and beads sensitivity test, the system was preconditioned with deionized water. A minimum of 300,000 events for lung LNs were acquired per file at a typical acquisition rate of 300 events/s. Prior to data analysis, the data of each sample were de-barcoded from raw data using a doublet-filtering scheme with unique mass-tagged barcodes. All data files were normalized and manually gated by FlowJo software (version 10.8.1, BD Biosciences) to exclude debris, dead cells, and doublets, resulting in live, single CD45+ immune cells. X-shift was applied to automatically identify distinct immune cell subsets. Dimensionality reduction algorithm t-stochastic neighbor embedding (t-SNE) was performed to visualize the high-dimensional data in two dimensions, showing the distribution of each cluster and marker expression, as well as the difference among each group or different samples. Heatmaps were generated according to the median value for the assigned markers in clusters.
Lung Pathology
The mouse lung tissues were collected individually from each of the experimental groups, as noted above, then perfused and fixed in 4% neutral-buffered formalin. Paraffin-embedded lung sections (5μm) were stained with H&E or PAS according to the manufacturer’s protocols (Wuhan Servicebio Technology). The severity of peribronchial inflammation was scored with H&E staining as follows: 0, normal; 1, few cells; 2, a ring of inflammatory cells 1-cell layer deep; 3, a ring of inflammatory cells 2- to 4-cells deep; and 4, a ring of inflammatory cells 4-cells deep. The severity of peribronchial inflammation was scored with PAS staining as follows: 0, PAS-positive cells≤5%; 1, 5%<PAS-positive cells≤25%; 2, 25%<PAS-positive cells≤50%; 3, 50%<PAS-positive cells≤75%; and 4, PAS-positive cells>75%.24 For each mouse, three light fields were randomly selected and counted from one slide, and the mean score was calculated.
T Cell Isolation and Differentiation in Vitro
Splenic CD4-naïve T cells were isolated from 6-wk-old female C57BL/6 mice using a Naïve CD4+ T Cell Isolation kit (Cat# 130-104-453; Miltenyi Biotec). Approximately 2×105 cells in 200μL were seeded into 96-well plates precoated with 1μg/mL of anti-CD28 (37.51; Cat# 553295; BD Biosciences) and 1μg/mL of anti-CD3e (145-2C11; Cat# 553058; BD Biosciences). For cells cultured under Tfh cell–skewing conditions with some modifications, as described previously,27 the medium was supplemented with anti-interferon-gamma (anti-IFN-γ; XMG1.2; 15μg/mL; Cat# BE0055; Bioxcell), anti-IL-4 (11B11; 15μg/mL; Cat# BE0045; Bioxcell), anti-IL-2 (JES6-5H4; 15μg/mL; Cat#BE0042; Bioxcell), and IL-6 (15 ng/mL; Cat#216-16; Peprotech). Immature T helper (Th0) cells were cultured under the same conditions without IL-6. IP at the specified concentration (30 nM or 300 nM) and an AhR inhibitor (CH223191, 10μM) was added to the medium 6 h after initial cell plating. IP and CH223191 were dissolved in DMSO. The stock solution was further diluted with PBS for the working solution. A 0.01% equal concentration of DMSO was used in the control group as the vehicle control. Cells were cultured for 6 d before analysis.
Lentivirus Packaging and Infection
pLKO.1-c-Maf short hairpin RNA (shRNA; lentivirus plasmid; Cat# TRCN0000208001; RNAi Core Facility), pCMV-ΔR8.91 (packaging plasmid; Cat# C6-6-1; RNAi Core Facility), and pMD.G (envelope plasmid; Cat# C6-6-1; RNAi Core Facility) were transfected into 293T cells (ATCC) using TransIT-LT1 (Cat# MIR2300; Mirus Bio). The lentiviral supernatant was harvested 48 h after transfection. Splenic CD4-naïve T cells were isolated from three 6-wk-old female C57BL/6 mice using a Naïve CD4+ T Cell Isolation kit. Naïve CD4+ T cells were stimulated with the plate-bound anti-CD3e (5μg/mL) and anti-CD28 (5μg/mL) under neutral conditions (10μg/mL anti-IFN-γ and 10μg/mL anti-IL4) for 20 h. Lentiviral supernatant was added to the cell culture medium in the presence of polybrene (4μg/mL) with centrifugation (1,800 rpm at 32°C) for 90 min, and the medium was exchanged with fresh medium 4 h after infection. One day after infection, the cells were cultured under Tfh cell–skewing conditions for 6 d before analysis.
Quantitative Reverse Transcription Polymerase Chain Reaction
Total RNA was extracted from cells using TRIzol Reagent (Invitrogen), and the concentration and quality (OD260/280 between 1.9 and 2.1 is qualified) of RNA were determined by NanoDrop (Thermo Fisher). Single-stranded complementary DNA (cDNA) was synthesized using the PrimeScript II first Strand cDNA Synthesis Kit (Takara). Quantitative reverse transcription polymerase chain reaction (RT-qPCR) was performed with SYBR Premix Ex Taq II (Takara) using the Roche 480 Real Time PCR System (denature: 1 cycle for 95°C, 30 s; PCR: 40 cycles for 95°C, 5 s and 60°C, 30 s; cooling: 1 cycle for 50°C, 30 s). The mRNA expression levels of target genes were normalized to β-actin using the 2−ΔΔCt method. The PCR primers used are provided in Table S2.
Antibodies and Flow Cytometry
BALF was harvested by two consecutive flushes of the lung with 0.8mL of ice-cold PBS. The total cell numbers were counted with default count settings using a Countess II Automated Cell Counter (Thermo Fisher). Differential cell counts in BALF were obtained by flow cytometry, as previously described. BALF cells were stained with the following antibodies (all of which were from eBioscience): PE anti-Siglec-F (1RNM44N; Cat# 12-1702-80; 0.2μg mAb/1×106 cells), fluorescein isothiocyanate anti-alpha-mac3 (M1/70; Cat# 85-11-0112-82; 0.2μg mAb/1×106 cells), allophycocyanin (APC) anti-Gr-1 (RB6-8C5; Cat# 85-94-5931-71; 0.2μg mAb/1×106 cells), Percp-Cy5.5 anti-CD3e (145-2C11; Cat# 85-15-0031-63, 0.1μg mAb/1×106 cells), and Percp-Cy5.5 anti-CD19 (MB19-1; Cat# 85-15-0191-81; 0.1μg mAb/1×106 cells). The cells were then washed twice by adding 1mL of FACS buffer (PBS containing 2% FBS) and centrifuged at 400×g for 5 min at 4°C. Finally, the cells were resuspended with 300 μL of FACS buffer, and 1×104 cells per sample were collected and analyzed on a BD FACS Canto II flow cytometer (BD Biosciences). Data were analyzed by FlowJo software (version 10.8.1, BD Biosciences). Eosinophils were determined as side scatter (SSC)highSiglecF+Mac-3− cells, and alveolar macrophages cells were gated as SSChighSiglecF+Mac-3+ cells. Granulocytes were gated as SSChigh Gr-1+ cells, and lymphocytes were gated as forward scatter (FSC)low/SSClow CD3e+ or CD19+. For analysis of Tfh cells in lung LNs, the LNs were harvested, ground with a sterile syringe piston in DMEM containing 10% FBS, and filtered through a 70-μm cell strainer (Cat# 352350; BD Biosciences) to obtain a single-cell suspension. Then the cells were washed with PBS containing 2% FBS and centrifuged at 400×g for 5 min at 4°C. Subsequently, 5×106 cells per sample were resuspended in 100μL of an antibody cocktail comprising anti-CD4-FITC (H129.19; Cat# 100540; 0.01μg mAb/1×106 cells; Biolegend), anti-CD19-Percp-cy5.5 (1D3/CD19; Cat# 115530; 0.01μg mAb/1×106 cells; Biolegend), anti-ICOS-PE (15F9; Cat# 107706; 0.02μg mAb/1×106 cells; Biolegend), anti-CXCR5-biotin (2G8; Cat# 551960; 0.1μg mAb/1×106 cells; BD Biosciences), and anti-PD1-PE-cy7 (RMP1-30; Cat# 109110; 0.02μg mAb/1×106 cells; Biolegend) at 4°C for 30 min in the dark. The cells were then washed and stained with streptavidin-BV421 (Cat# 405225; 0.1μg mAb/1×106 cells; Biolegend) in 100μL of solution at 4°C for 30 min in the dark. For transcription factor staining, the cells were washed and fixed with the Foxp3 Fixation/Permeabilization buffer set (Cat# 20201221; eBioscience) at 4°C overnight, according to the manufacturer’s instructions. One day later, the cells were washed, resuspended, and stained with anti-Foxp3-APC (FJK-16s; Cat# 20201221; 0.04μg mAb/1×106 cells; eBioscience) in 100μL of solution at room temperature for 30 min. The cells were washed with 1mL 1×Perm/Wash buffer and centrifuged at 600×g for 5 min at 4°C for the first time and again washed with 1mL of FACS buffer (PBS containing 2% FBS), centrifuged at 600×g for 5 min at 4°C. Then the cells were resuspended in 300μL of FACS buffer. Finally, 1×105 cells per sample were collected and analyzed on a BD FACS Canto II flow cytometer (BD Biosciences). For analysis of in vitro Tfh cell differentiation, the cells were first stained with the surface markers mentioned above, followed by intracellular staining with antimouse Bcl-6-PerCP-eflour 710 (BCL-DWM; Cat# 46-5453-82; 0.02μg mAb/1×106 cells; eBioscience) or PerCP-eflour 710-conjugated rat IgG2a isotype control (eBR2a; Cat# 46-4321-80; 0.02μg mAb/1×106 cells; eBioscience) using the Foxp3 Fixation/Permeabilization buffer set (Cat# 20201221; eBioscience). Finally, >1×104 cells per sample were collected and analyzed on a BD FACS Canto II flow cytometer (BD Biosciences). Data were analyzed by FlowJo software (version 10.8.1, BD Biosciences).
Enzyme-Linked Immunosorbent Assay
Levels of total serum IgE and IgG1 were measured by a mouse IgE uncoated enzyme-linked immunosorbent assay (ELISA) kit (Cat# 88-50460-22; Invitrogen) and IgG1 uncoated ELISA kit (Cat# 88-50410-22; Invitrogen) according to the manufacturer’s instructions. Briefly, 96-well plates were coated with antimouse IgE or IgG1. After blocking, the diluted serum samples were added into the wells and incubated for 2 h at room temperature. After incubation, the plates were washed with phosphate buffered solution with 0.1% tween-20 (PBST), and the antimouse IgE- or IgG1-conjugated biotin was added and incubated for 1 h at room temperature. The plate was then washed and incubated with streptavidin–horseradish peroxidase (HRP) for 30 min at room temperature in the dark. Finally, tetramethylbenzidine substrate was added, and the reaction was stopped by sulfuric acid. The absorbance at 450 nm was measured using a plate reader. For the HDM-specific IgG1 ELISA, plates were coated with HDM (50μg/mL) followed by secondary reagents, as mentioned above. For the measurement of HDM-specific IgE levels, serum IgG was first depleted with Protein G magnetic beads (Beyotime).28 Similarly, plates were coated with HDM (50μg/mL), and, after blocking, the serum sample with IgG depletion was added and incubated at 4°C overnight. Biotin-conjugated antimouse IgE was added followed by the addition of streptavidin-HRP and tetramethylbenzidine substrate. Finally, the absorbance was read at 450 nm using a plate reader.
Chromatin Immunoprecipitation
The chromatin immunoprecipitation (ChIP) assay was performed using the Pierce Magnetic ChIP Kit (Cat# 26157; Thermo Fisher) according to the manufacturer’s instructions. Briefly, cells were fixed with 1% formaldehyde for 10 min and quenched with 0.125 M glycine for 5 min. The chromatin was fragmented by micrococcal nuclease digestion. The chromatin was immunoprecipitated with an anti-AhR antibody (dilution 1:50; Cat# SA-210; Biomol) or anti-Maf antibody (dilution 1:50; Cat# GTX129420; GeneTex) overnight at 4°C. The antibody–chromatin complexes were then pulled down by ChIP-Grade Protein A/G Magnetic Beads and eluted with the elution buffer provided in the ChIP Kit. After incubation at 65°C for crosslink reversal, the samples were digested by RNase A and proteinase K. The immunoprecipitated DNA was collected using a DNA Clean-Up Column. DNA was quantified by qPCR using SYBR Premix Ex Taq II (Takara). The PCR primers used are provided in Table S3.
Immunoprecipitation and Western Blotting
Cell lysis was performed using Pierce IP Lysis Buffer (Cat# 87787; Thermo Fisher), and the sample was then subjected to ultrasonic treatment (Cat# JP-010T; Skymen), followed by centrifugation at 16,000×g for 15 min at 4°C. The protein concentration of harvested supernatants was quantified by the Bicinchoninic Acid (BCA) Assay Kit (Cat# T9300A; Takara), and 500μg of protein in a 500-μL volume was incubated overnight at 4°C with anti-AhR antibody (5 μg; Cat# SA-210; Biomol) or IgG antibody (1U6H0; 5 μg; Cat# MA5-42729; Thermo Fisher) with gentle rotation. Protein A/G Magnetic Beads (Cat# 88802-3; Thermo Fisher) were first washed twice with PBS and then incubated with the previously described protein–antibody complexes for 2 h at 4°C with gentle rotation. Then the protein A/G magnetic beads were collected and washed three times with ice-cold PBS containing protease phosphatase inhibitor (Cat# 78440; Thermo Fisher) by magnetic frame. After being washed, the immunocomplexes were boiled with 60 -μL of 2× Laemmli sample buffer (Cat# 1610747; Bio-Rad) in 95°C for 5 min, and a 25-μL sample was loaded into a 12% sodium dodecyl sulfate–polyacrylamide gel electrophoresis gel and subjected to electrophoresis at 100 V for ∼90 min. Next, the proteins separated in the gel were transferred to polyvinylidene fluoride membranes (Cat# 88520; Thermo Fisher) at 0.18 A for 1.5 h. The membranes were then blocked in 5% BSA (Cat# ST023; Beyotime) for 1 h, followed by incubation with primary antibodies overnight at 4°C. After being washed with 0.1% Tween-20–containing tris buffered saline solutions (TBST) three times, the membranes were incubated with HRP-conjugated goat-antirabbit or goat-antimouse antibody (1:5,000; Cell Signaling Technology) for 1 h. After being washed again with TBST three times, the bands were finally imaged using the chemiluminescent HRP substrate (Thermo Fisher) and a Molecular Imager ChemiDoc XRS+ Imaging System (Bio-Rad). The primary antibodies used were as follows: anti-c-Maf antibody (1:1,000; Cat# GTX129420; GeneTex), anti-AhR antibody (1:1,000), and anti-β-actin antibody (AC-15; 1:5,000; Cat# A5441; Sigma-Aldrich).
Statistical Analyses
All statistical analyses were performed using GraphPad Prism software (version 9.0). Unless otherwise stated, all statistical analyses were carried out through unpaired t-tests or one-way analysis of variance (ANOVA) followed by post hoc Bonferroni’s multiple comparisons test multiple comparison tests, as appropriate. Data were considered statistically significant when p<0.05 and presented as means±standard errors of the mean (SEMs).
Results
Analysis of the Immune Cell Populations in Lung LNs and Measure of Allergen-Specific Antibody Production and Lung Inflammation in Mice Exposure to PM2.5
To investigate the impact of PM2.5 exposure on a well-established mouse model of airway inflammation, female C57BL/6 mice were intratracheally sensitized with saline control PBS, HDM, or HDM plus PM2.5 daily from day 0 to day 2. The PBS group was challenged intratracheally with PBS; the HDM group and the HDM plus PM2.5 group were challenged intratracheally with HDM from day 9 to day 12 (Figure 1A). Compared with the PBS and HDM groups, mice in the HDM plus PM2.5 group showed significantly aggravated HDM-induced pulmonary inflammatory response, as evidenced by the recruitment of inflammatory cells in the lungs, dense peribronchial infiltrates, and goblet cell hyperplasia (Figure S2A–E), as well as by higher serum levels of total IgE, total IgG1, HDM-specific IgE, and HDM-specific IgG1 antibodies (Figure S2F–I).
Figure 1. Mass cytometry analysis of the immune cell composition in lung LNs. (A) Schematic representation of the house dust mite (HDM)-induced mouse allergic lung inflammation model. Mice in the HDM group were sensitized by intratracheal inhalation of 10μg of HDM dissolved in 50μL of PBS per mouse on days 0, 1, and 2 and challenged on days 9, 10, 11, and 12 with the same amount of HDM. Mice in the HDM plus PM2.5 group were sensitized by intratracheal inhalation of 10μg of HDM mixed with PM2.5 (250μg/mouse) in a final volume of 50μL per mouse on days 0, 1, and 2 and challenged on days 9, 10, 11, and 12 with 10μg of HDM only. The PBS group mice received 50μL of PBS during the sensitization and the challenge phases. (B) CD45+ leukocytes in lung LNs were analyzed by mass cytometry (CyTOF; n=3–4 samples per group) using a panel of 41 antibodies on day 14 (D14). t-SNE plot of CD45+ compartments overlaid with color-coded clusters. An equal number of CD45+ compartments in the lung LNs from the PBS, HDM, and HDM plus PM2.5 groups was compared in the t-SNE plots. (C) Total cell number of lung LNs from each group (numerical values are shown in Excel Table S9). (D) Relative frequencies (numerical values are shown in Excel Table S10) and (E) cell numbers (numerical values are shown in Excel Table S11) of the indicated immune cell subsets among each group. (F) t-SNE plot of CD45+ leukocytes overlaid with the expression of IgE among the PBS, HDM, and HDM plus PM2.5 groups. The colors are coded according to the relative expression of the indicated IgE markers. Red indicates high expression, and blue indicates low expression. The percentages of (G) IgE-positive B cells (CD19+IgE+CD138−) (numerical values are shown in Excel Table S12) and (H) activated IgE-positive B cells (CD19+IgE+IgM+IgD+CD138−) (numerical values are shown in Excel Table S13) were manually gated by FlowJo software (version 10.8.1, BD Biosciences). Data are shown as the means±SEMs. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 according to ANOVA followed by Bonferroni’s multiple comparisons test. Note: ANOVA, analysis of variance; Ig, immunoglobulin; i.t., intratracheally; LN, lymph node; PBS, phosphate-buffered saline; PM2.5, ambient particulate matter with an aerodynamic diameter of ≤2.5μm; SEM, standard error of the mean; t-SNE, t-stochastic neighbor embedding.
Figure 1A is a schematic illustration, depicting the house dust mite-induced mouse allergic lung inflammation model. The mice in the house dust mite group were sensitized by intratracheal inhalation of 10 micrograms of the house dust mite dissolved in 50 microliters of P B S per mouse on Days 0, 1, and 2, and challenged on Days 9, 10, 11, and 12 with the same amount of the house dust mite. The mice in the house dust mite plus particulate matter begin subscript 2.5 end subscript group were sensitized by intratracheal inhalation of 10 micrograms of house dust mite mixed with particulate matter begin subscript 2.5 end subscript (250 micrograms per mouse) in a final volume of 50 microliters per mouse on Days 0, 1, and 2 and challenged on Days 9, 10, 11, and 12 with 10 micrograms of house dust mite only. P B S group mice received 50 microliters P B S during the sensitization and challenge phases. i.t., intratracheally. Figure 1B is three two-dimensional T S N E plots, displaying the cell composition of lung lymph nodes from P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript under T S N E 1 and one row, namely, T S N E 2 for C D 4 positive T cells, C D 8 positive T cells, C D 4 negative C D 8 negative T cells, B cells, and Myeloid cells. Figures 1C, 1G, and 1H are graphs, plotting Cell numbers (10 caret 6), ranging from 0 to 15 in increments of 5; percentage of C D 45 positive cells, ranging from 0 to 60 in increments of 20; and percentage of C D 45 positive cells, ranging from 0 to 40 in increments of 10 (y-axis) across P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript (x-axis) for total cells, C D 19 positive immunoglobulin E positive C D 138 negative cells, and C D 19 positive immunoglobulin E positive C D 138 negative immunoglobulin M positive immunoglobulin D positive cells. Figures 1D and 1E are graphs, plotting percentage of total C D 45 positive cells, ranging from 0 to 100 in increments of 20 and cell number (10 caret 4), ranging from 0 to 1,000 in increments of 200 (y-axis) across C D 4 positive T cells, C D 8 positive T cells, C D 4 negative C D 8 negative cells, B cells, and Myeloid cells (x-axis) for P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript. Figure 1F is three two-dimensional T S N E plots, displaying the cell composition of lung lymph nodes from P B S, H D M, and H D M plus particulate matter begin sub-script 2.5 end subscript under T S N E 1 and one row, namely, T S N E 2 for immunoglobulin E. A color scale depicts the value ranges from 0.0 to 1.5 in increments of 0.5.
To examine whether PM2.5 exposure affects lung LN cell composition, LN cells were used for single-cell mass cytometry analysis with a panel of antibodies targeting lineage markers and cytokines associated with adaptive and innate immune systems. For the unsupervised detection of cell subsets, we applied the X-shift clustering algorithm, which divides the cells into phenotypically distinct clusters based on optimal k-nearest neighbor (k-NN) values. We also examined surface marker staining intensity on t-SNE dimension–reduced space, which visualizes the high-dimensional data. The t-SNE plot generated from the merged data represents a map of CD45+ cells present in lung LNs. Based on lineage markers, we assessed the locations within the t-SNE plot of the following five major subsets: CD4+ T cells, CD8+ T cells, CD4−CD8− T cells, CD19+ B cells, and myeloid cells (Figure 1B).
As expected, exposure to HDM resulted in higher numbers of B cells compared with the control mice (PBS group). The HDM plus PM2.5 group showed an even greater proportion and number of B cells compared with the PBS and HDM groups (Figure 1B–E). In addition, the t-SNE results showed that the expression of IgE in the HDM plus PM2.5 group was significantly higher than that in the HDM or PBS group (Figure 1F). The percentages of IgE+ B cells (CD19+IgE+CD138−) and active IgE+ B cells (CD19+IgE+IgM+IgD+CD138−) were also higher following PM2.5 exposure (Figures 1G,H and S3).
Because dendritic cells (DCs) are critical in initiating an immune response, we manually gated CD11c+ myeloid cells (Figure S4A). Based on the t-SNE plot and marker density, we defined the following six clusters: migratory DC-1 (MHCIIhiCD11chiPD-L1+PD-L2+CD11b+CCR7hi), migratory DC-2 (MHCIIhiCD11cintPD-L1intPD-L2intCD11b+CCR7hi), plasmacytoid DCs (pDCs; MHCIIloCD11cintB220+CD11b−), conventional DC1 (cDC1s; MHCIIhiCD11chiB220−CD8a+CD11bloCCR7−), conventional DC2 (cDC2s; MHCIIintCD11chiB220−CD4+CD8a−CD11b+CCR7−), and macrophages (CD11cintCD11bhiF4/80hi; Figure S4B and Excel Table S14). An additional two clusters were identified as expressing low levels of major histocompatibility complex II (MHCII) or CD11c (Figure S4B).
Overall, the frequency of the migratory DC-1 subset was significantly higher in both the HDM and HDM plus PM2.5 groups compared with the PBS group, whereas the migratory DC-2 subset was lower (Figure S4C–E). Further, our results showed that the frequency of cDC2s was significantly higher in the HDM and HDM plus PM2.5 groups compared with the PBS group, whereas no significant differences were observed between the HDM and HDM plus PM2.5 groups (Figure S4C,F). The frequency of the CD11clow subset was decreased in the HDM plus PM2.5 group compared with the PBS group (Figure S4C,G). In addition, the frequencies of the MHCIIlow, pDC, and cDC1 subsets were similar among the groups (Figure S4C,H–J). Interestingly, we noted that a relatively minor cell population, defined as macrophages (CD11cintCD11bhiF4/80hi), was significantly higher in the HDM plus PM2.5 group compared with the PBS and HDM groups (Figure S4C,K).
We next gated CD45+TCRβ+CD4−CD8+ cells as CD8+ T cells and divided CD8+ T cells into the following four subsets by t-SNE analysis: naïve CD8+ T cells (CD44−CD62L+), central memory CD8+ T cells (Tcm, CD44+CD62L+), effector CD8+ T-cell/effector memory CD8+ T cells (Tem, CD44+CD62L−), and preeffector-like CD8+ T cells (CD44−CD62L−; Figure S5A,B and Excel Table S23). The HDM and HDM plus PM2.5 groups showed a significant decrease in the proportion of naïve CD8+ T cells, but an increase in preeffector-like CD8+ T cells and CD8+ Tem cells compared with the PBS group, and no difference was found between the HDM plus PM2.5 group and the HDM group during the sensitization phase. In addition, there was no difference in CD8+ Tcm cells among the PBS, HDM, and HDM plus PM2.5 groups (Figure S5C,D).
Mass Cytometry Analysis of the CD4+ T-Cell Population in Lung LNs
Because CD4+ T-cell subsets play crucial roles in the regulation of humoral immunity, we sorted the CD4+ T cells and eliminated unrelated cell lineage markers for the unsupervised detection of CD4+ T-cell subsets. We evaluated the locations within the t-SNE plot of the following five different subsets: naïve CD4+ T cells (TCRβ+CD4+CD62L+), effector CD4+ T cells (TCRβ+CD4+CD62L−), Treg cells (TCRβ+CD4+CD25+Foxp3+), T follicular regulatory (Tfr) cells (TCRβ+CD4+CXCR5+Foxp3+), and Tfh cells (TCRβ+CD4+CXCR5+Bcl-6+) (Figure 2A,B and Excel Table S25). To further investigate whether PM2.5 preferentially altered a specific CD4+ T-cell subset, we determined the proportion and number of each subset among the PBS, HDM, and HDM plus PM2.5 groups (Figure 2A). Naïve CD4+ T cells were present at lower cell numbers and proportions in the HDM and HDM plus PM2.5 groups compared with the PBS group (Figure 2C,D). Although the numbers and proportions of effector CD4+ T cells, Treg cells, and Tfr cells were different between the PBS group and the HDM group, their numbers and proportions were not significantly different between the HDM group and the HDM plus PM2.5 group (Figure 2E–J). Mice exposed to HDM exhibited a higher proportion and number of Tfh cells compared with those seen in mice exposed to PBS alone. The HDM plus PM2.5 group had a significantly higher proportion and number of Tfh cells compared with the PBS group and the HDM group, suggesting preferential regulation of the Tfh subset by PM2.5 (Figure 2K,L). It has been reported that cytokine-skewed Tfh cells contribute to class-switch recombination and modify B-cell maturation within GCs.29 Thus, we examined the proportion and number of IL-4–, IL-17–, and IFN-γ–expressing Tfh cells in each group. As expected, the proportion and number of IL-4–expressing Tfh2 cells was significantly higher in the HDM plus PM2.5 group compared with those noted in the PBS group or HDM group (Figure 2M,N).
Figure 2. Mass cytometry analysis of the CD4+ population shift in lung LNs. (A) The CD4+ population, which was plotted by t-SNE, was separated into five clusters according to the expression of the indicated markers. t-SNE plots of an equal number of CD4+ T cells from each treatment group. Each color represents the indicated CD4+ T-cell subset. (B) Heatmap displaying normalized marker expression of each CD4+ T-cell cluster (numerical values are shown in Excel Table S25). (C–L) Relative frequencies and numbers of CD4+ T-cell subsets among the PBS, HDM, and HDM plus PM2.5 groups (numerical values are shown in Excel Tables S26–S35). (M) Percentages of IL-4+, IFN-γ+, and IL-17+ Tfh cells relative to CD4+ T cells in each treatment group (numerical values are shown in Excel Table S36). (N) Numbers of IL-4+, IFN-γ+, and IL-17+ Tfh cells in each treatment group (numerical values are shown in Excel Table S37). n=3–4 per group. Data are shown as the means±SEMs. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 according to ANOVA followed by Bonferroni’s multiple comparisons test. Note: ANOVA, analysis of variance; HDM, house dust mite; IFN, interferon; IL, interleukin; LN, lymph node; PBS, phosphate-buffered saline; PM2.5, ambient particulate matter with an aerodynamic diameter of ≤2.5μm; SEM, standard error of the mean; t-SNE, t-stochastic neighbor embedding; Tfh, T follicular helper; Tfr, T follicular regulatory.
Figure 2A is three two-dimensional T S N E plots, displaying the cell composition of lung lymph nodes from P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript under T S N E 1 and one row, namely, T S N E 2 for Treg, Naive T, Effector T, T follicular helper, and T follicular regulatory. Figure 2B is a heatmap, plotting Treg, Naive T, Effector T, T follicular helper, and T follicular regulatory (y-axis) across C D 44, C D 28, M H C underscore 2, C X C R 5, C T L A 4, C D 80, S T A T 3, C D 25, I C O S, C D 11 c, I L underscore 21 R, K i 67, C C R 7, C D 62 L, I L underscore 4, B C L 6, P D underscore L 1, T C R b, T G F b, C D 40 L, I F N g, I L underscore 17 a, O X 40, F o x p 3, C D 86, P D 1, I L underscore 10, I C O S L, Blimp 1, and C D 4 (x-axis). A color scale depicts the value ranges from 0 to 1 in increments of 0.2. Figures 2C to 2L are graphs titled Naïve T, Naïve T, Effector T, Effector T, Treg, Treg, T follicular regulatory, T follicular regulatory, T follicular helper, T follicular helper, plotting percentage of C D 4 T cells, ranging 0 to 100 in increments of 20; cell numbers (10 caret 7), ranging 0 to 10 in increments of 2; percentage of C D 4 T cells, ranging from 0 to 50 in increments of 10; cell numbers (10 caret 4), ranging from 0 to 10,000 in increments of 2,000; percentage of C D 4 T cells, ranging from 0.0 to 2.5 in increments of 0.5; cell numbers (10 caret 4), ranging from 0 to 250 in increments of 50; percentage of C D 4 T cells, ranging from 0 to 8 in increments of 2; cell numbers (10 caret 4), ranging from 0 to 1,500 in increments of 500; percentage of C D 4 T cells, ranging from 0 to 40 in increments of 10; and cell numbers (10 caret 4), ranging from 0 to 8,000 in increments of 2,000 (y-axis) across P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript (x-axis), respectively. Figures 2M and 2N are graphs titled T follicular helper, plotting percentage of C D 4 T cells, ranging from 0.00 to 0.20 in increments of 0.05 and cell number (10 caret 4), ranging from 0 to 30 in increments of 10 (y-axis) across I L 4, I F N-gamma, I L-17 (x-axis) for P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript.
Next, we used flow cytometry to further confirm Tfh cells and Tfr cells as identified by CD4+CXCR5+ICOS+Foxp3− cells and CD4+CXCR5+ICOS+Foxp3+ cells, respectively. The Tfh cells and Tfr cells were also determined by PD1 expression. As previously reported, mice in the HDM group showed higher percentages of both Tfh and Tfr cell populations compared with mice in the PBS group.30 However, mice exposed to HDM plus PM2.5 had a significantly higher percentage of Tfh cells compared with those exposed to HDM alone, but those mice exposed to HDM plus PM2.5 had only a slightly higher percentage of Tfr cells, which did not reach statistical significance as compared with those exposed to HDM alone (Figure S6A–F).
Measurements of Tfh Differentiation, Allergen-Specific Antibody Production, and Lung Inflammation with Exposure to IP
We next investigated whether IP affects Tfh cell differentiation during the antigen sensitization period. To this end, female C57BL/6 mice were intratracheally sensitized with saline control (PBS), HDM, or HDM plus IP daily from day 0 to day 2. The PBS group was challenged intratracheally with PBS; the HDM group and the HDM plus IP group were challenged intratracheally with HDM from day 9 to day 12 (Figure 3A). As expected, the mice exposed to IP during sensitization had significantly more inflammatory cells in the BALF (Figure 3B,C) and higher serum levels of total IgE, total IgG1, HDM-specific IgE, and HDM-specific IgG1 mAbs (Figure 3D–G). In the mediastinal LNs, the HDM plus IP group had a significantly higher percentage of Tfh cells and PD1hi Tfh cells (Figure 3H,I).
Figure 3. The effects of IP treatment on lung inflammation and Tfh cell differentiation in vivo. (A) Mouse allergic lung inflammation model used in the present study (schematic). Mice in the house dust mite (HDM) group were sensitized by intratracheal inhalation of 10μg of HDM dissolved in 50μL of PBS per mouse on days 0, 1, and 2 and challenged on days 9, 10, 11, and 12 with the same amount of HDM. Mice in the HDM plus IP group were sensitized by intratracheal inhalation of 10μg of HDM mixed with IP (5μM/mouse) in a final volume of 50μL per mouse on days 0, 1, and 2 and challenged on days 9, 10, 11, and 12 with 10μg of HDM only. The PBS group mice received 50μL of PBS during the sensitization and the challenge phases. DMSO was used as the vehicle control in the PBS and the HDM groups to make the comparison with the HDM plus IP group, and the mice were sacrificed and evaluated 48 h after the last allergen challenge. (B) Total cell number in bronchoalveolar lavage fluid (numerical values are shown in Excel Table S44). (C) Cell number of eosinophils, lymphocytes, macrophages, and neutrophils in bronchoalveolar lavage fluid (numerical values are shown in Excel Table S45). The (D) total IgE (numerical values are shown in Excel Table S46), (E) total IgG1 (numerical values are shown in Excel Table S47), (F) HDM-specific IgE (numerical values are shown in Excel Table S48), and (G) HDM-specific IgG1 (numerical values are shown in Excel Table S49) in serum were measured by ELISA. Flow cytometry analysis of the percentage of (H) Tfh (numerical values are shown in Excel Table S50) and (I) PD1hi Tfh (numerical values are shown in Excel Table S51) in CD4+ T cells in lung LNs. All data are representative of two or three independent experiments. n=10 each group. Data are shown as the means±SEMs. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 according to ANOVA followed by Bonferroni’s multiple comparisons test. Note: ANOVA, analysis of variance; BALF, bronchoalveolar lavage fluid; DMSO, dimethyl sulfoxide; ELISA, enzyme-linked immunosorbent assay; Ig, immunoglobulin; IP, indeno[1,2,3-cd]pyrene; LN, lymph node; OD, optical density; PBS, phosphate-buffered saline; SEM, standard error of the mean; Tfh, T follicular helper.
Figure 3A is a schematic illustration, depicting a mouse allergic lung inflammation model used in the present study, where the mice in the house dust mite group were sensitized by intratracheal inhalation of 10 micrograms of the house dust mite dissolved in 50 microliters P B S per mouse on Days 0, 1, and 2 and challenged on Days 9, 10, 11, and 12 with the same amount of the house dust mite. Mice in the house dust mite plus I P group were sensitized by intratracheal inhalation of 10 micrograms house dust mite mixed with I P (5 micromolar per mouse) in a final volume of 50 microliters per mouse on Days 0, 1, and 2 and challenged on Days 9, 10, 11, and 12 with 10 micrograms house dust mite only. P B S group mice received 50 microliters P B S during the sensitization and challenge phases. Dimethyl sulfoxide was used as the vehicle control in P B S and house dust mite groups to make the comparison with the house dust mite plus I P group, and mice were sacrificed and evaluated 48 h after the last allergen challenge. Figures 3B, 3D, 3E, 3F, 3G, 3H, and 3I are graphs titled total cells, total immunoglobulin E, total immunoglobulin G 1, house dust mite specific immunoglobulin E, house dust mite specific immunoglobulin G 1, T follicular helper, and P D 1 begin superscript hi end superscript T follicular helper, plotting cell number in Bronchoalveolar lavage fluid (10 caret 5), ranging from 0 to 20 in increments of 5; nanograms per milliliter, ranging from 0 to 15,000 in increments of 5,000; nanograms per milliliter, ranging from 0 to 40,000 in increments of 10,000; O D, ranging from 0.0 to 0.6 in increments of 0.2; O D, ranging from 0.00 to 0.04 in increments of 0.01; percentage of C D 4 T cells, ranging from 0 to 10 in increments of 2; percentage of C D 4 T cells, ranging from 0 to 10 in increments of 2 (y-axis) across P B S, H D M, and H D M plus particulate matter begin subscript 2.5 end subscript (x-axis), respectively. Figure 3C is a graph, plotting cell number in Bronchoalveolar lavage fluid (10 caret 5), ranging from 0 to 10 in increments of 2 (y-axis) across Eosinophils, Lymphocytes, Macrophages, and Neutrophils (x-axis) for P B S, H D M, and H D M plus I P.
The Effects of AhR Axis on Tfh Cell Differentiation
We hypothesized that the PM2.5–AhR axis may affect the differentiation of Tfh cells. To explore this hypothesis, we tested whether AhR in CD4+ T cells is required for the PM2.5-mediated increase in the percentage of Tfh cells. Mice with a CD4-specific deletion of AhR (CD4ΔAhR) were generated by crossing AhRloxp mice with CD4-Cre mice. Female CD4ΔAhR and AhRloxp (as a WT control) mice were intratracheally sensitized with PBS, HDM, and HDM plus PM2.5 daily from day 0 to day 2 followed by an intratracheal challenge with PBS or HDM from day 9 to day 12. In the HDM group, there was no significant difference in the Tfh or PD1hi Tfh cell population (Figures 4A, S7A, and S8), Tfr, or PD1hi Tfr cell population (Figures 4B, S7B, and S8), PD1hiTfh/PD1hiTfr ratio (Figures 4C and S7C), and lung inflammation (Figure 4D–F) between AhRloxp mice and CD4ΔAhR mice. However, in the HDM plus PM2.5 group, CD4ΔAhR mice showed a significantly lower percentage of Tfh cell population (Figure 4A), Tfh/Tfr ratio (Figure 4C), and lung inflammation (Figures 4D–F and S7D,E) compared with control AhRloxp mice.
Figure 4. The effects of AhR deficiency in CD4+ T cells on PM2.5-induced Tfh cell differentiation. CD4ΔAhR and AhRloxp (WT) mice were sensitized and challenged, as described in Figure 1A, with or without HDM or PM2.5 treatments at the indicated time points, and mice were sacrificed for sampling 48 h after the last allergen challenge. The percentage of (A) Tfh cells (numerical values are shown in Excel Table S52) and (B) Tfr cells (numerical values are shown in Excel Table S54), as well as the (C) Tfh/Tfr ratio (numerical values are shown in Excel Table S56) in CD4+ T cells in lung LNs were examined by flow cytometry analysis. (D) Total cell number in bronchoalveolar lavage fluid (numerical values are shown in Excel Table S58). (E) Cell number of eosinophils, lymphocytes, macrophages, and neutrophils in bronchoalveolar lavage fluid (numerical values are shown in Excel Table S59). (F) Representative H&E and PAS staining of lung tissue sections (scale bar: 500μm in H&E; scale bar: 200μm in PAS). The small boxes show the locations of the big boxes in these panels. All data are representative of two or three independent experiments. n=10 each group. Data are shown as the means±SEMs. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 according to ANOVA followed by Bonferroni’s multiple comparisons test. Note: AhR, aryl hydrocarbon receptor; ANOVA, analysis of variance; BALF, bronchoalveolar lavage fluid; H&E, hematoxylin and eosin; HDM, house dust mite; LN, lymph node; PAS, periodic acid–Schiff; PBS, phosphate-buffered saline; PM2.5, ambient particulate matter with an aerodynamic diameter of ≤2.5μm; SEM, standard error of the mean; Tfh, T follicular helper; Tfr, T follicular regulatory; WT, wild type.
Figures 4A to 4D are graphs, plotting percentage of C D 4 T cells, ranging from 0 to 15 in increments of 5; percentage of C D 4 T cells, ranging from 0.0 to 1.5 in increments of 0.5; fold change, ranging from 0 to 25 in increments of 5; cell number in Bronchoalveolar lavage fluid (10 caret 5), ranging from 0 to 40 in increments of 10 (y-axis) across W T underscore P B S, C D 4 begin superscript uppercase delta Ah R end superscript underscore P B S, W T underscore H D M, C D 4 begin superscript uppercase delta Ah R end superscript underscore H D M, W T underscore H D M plus particulate matter begin subscript 2.5 end subscript, and C D 4 begin superscript uppercase delta Ah R end superscript underscore H D M plus particulate matter begin subscript 2.5 end subscript (x-axis) for T follicular helper, T follicular regulatory, T follicular helper to T follicular regulatory ratio, and Total cells. Figure 4E is a graph, plotting cell number in Bronchoalveolar lavage fluid (10 caret 5), ranging from 0 to 15 in increments of 5 (y-axis) across Eosinophils, Lymphocytes, Macrophages, and Neutrophils (x-axis) for W T underscore P B S, C D 4 begin superscript uppercase delta Ah R end superscript underscore P B S, W T underscore H D M, C D 4 begin superscript uppercase delta Ah R end superscript underscore H D M, W T underscore H D M plus particulate matter begin subscript 2.5 end subscript, and C D 4 begin superscript uppercase delta Ah R end superscript underscore H D M plus particulate matter begin subscript 2.5 end subscript. Figure 4F is a stained tissue displaying six columns, Ah R begin superscript loxP end superscript, C D 4 begin superscript uppercase delta Ah R end superscript for P B S, Ah R begin superscript loxP end superscript, C D 4 begin superscript uppercase delta Ah R end superscript for H D M, and Ah R begin superscript loxP end superscript, C D 4 begin superscript uppercase delta Ah R end superscript for H D M plus particulate matter begin subscript 2.5 end subscript and two rows, namely, H and E, and P A S.
Next, we performed polarization assays using Tfh cells in the presence of IP in vitro. Tfh cells were differentiated from CD4+ naïve T cells under Tfh cell–skewing conditions. IP was added to the cell culture medium for 6 d at a concentration of 30 nM or 300 nM, and the medium and compound were replaced every 3 d. IP treatment significantly promoted Tfh cell differentiation in a dose-dependent manner compared with controls (Figures 5A,B, S9, S10A, S10B, and S11). Consistent with the flow cytometric results, cells exposed to IP exposure also exhibited a higher expression of Tfh-associated genes, including Bcl-6, Batf, c-Maf, interferon regulatory factor 4 (Irf4), thymocyte selection-associated high mobility group box family member 2 (Tox2), and Il21 (Figure 5C–H), as well as the AhR target gene, Cyp1a1 (Figure 1). We also found that cells exposed to IP had higher Il4 expression in Tfh cells, which is required for IgE class switching (Figure 5J). To further characterize the contribution of AhR to IP-induced Tfh cell differentiation, WT Tfh cells and AhR KO Tfh cells were treated with IP or vehicle control, and Bcl-6, Irf4, and Il21 expression levels were analyzed by RT-qPCR. Tfh cells had lower levels of Bcl-6, Irf4, and Il21 in the AhR KO group (Figure 5K–M). Similarly, Tfh cells exposed to IP and the AhR antagonist, CH-223191, had significantly lower expression of Bcl-6, Irf4, and Il21 compared with those exposed to IP alone (Figure 5N–P).
Figure 5. The effects of IP treatment on Tfh cell differentiation in vitro.
IP was added to cell culture medium for 6 d at a concentration of 30 nM or 300 nM, the medium and compound were replaced every 3 d during Tfh cell differentiation. Flow cytometry analysis of the percentages of the (A) CD4+ICOS+CXCR5+ (numerical values are shown in Excel Table S62) and (B) CD4+ICOS+CXCR5+Bcl-6+ (numerical values are shown in Excel Table S63) subsets in CD4+ T cells. (C–J) Tfh-associated genes were evaluated by RT-qPCR (numerical values are shown in Excel Tables S66–S73). (K–M) WT and AhR KO Tfh cells were treated with IP at a concentration of 30 nM or 300 nM, and the mRNA expression levels of Bcl-6, Irf4, and Il21 were examined by RT-qPCR (numerical values are shown in Excel Tables S74–S76). (N–P) The mRNA expression levels of Bcl-6, Irf4, and Il21 were examined in Tfh cells treated with IP or IP plus AhR inhibitor (CH-223191; 10μM) (numerical values are shown in Excel Tables S77–S79). DMSO was the vehicle for IP and CH-223191, and all experiments were done in the presence of the DMSO (0.01%) as vehicle control. All data are representative of two or three independent experiments. n=3–5 each group. Data are shown as the means±SEMs. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 according to ANOVA followed by Bonferroni’s multiple comparisons test. Note: AhR, aryl hydrocarbon receptor; ANOVA, analysis of variance; DMSO, dimethyl sulfoxide; IP, indeno[1,2,3-cd]pyrene; KO, knockout; RT-qPCR, quantitative reverse transcription polymerase chain reaction; SEM, standard error of the mean; Tfh, T follicular helper; WT, wild type.
Figure 5A is a set of three two-dimensional density plots and one graph. The three two-dimensional density plots are titled D M S O, I P 30 nanomolar, and I P 300 nanomolar, plotting C X C R 5, ranging as 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, 10 begin superscript 4 end superscript, 10 begin superscript 5 end superscript (y-axis) across I C O S, ranging as 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, 10 begin superscript 4 end superscript, 10 begin superscript 5 end superscript (x-axis), respectively. The graph, plotting percentage of C D 4 T cell, ranging from 0 to 50 in increments of 10 (y-axis) across T follicular helper, T follicular helper plus 30 nanomolar I P, and T follicular helper plus 300 nanomolar I P (x-axis) for C X C R 5 positive C O S-positive subset. Figures 5B to 5M are graphs, plotting percentage of C D 4 T cell, ranging from 0 to 20 in increments of 5; fold change, ranging from 0 to 3 in unit increments; fold change, ranging from 0 to 4 in unit increments; fold change, ranging from 0 to 3 in unit increments; fold change, ranging from 0.0 to 2.5 in increments of 0.5; fold change, ranging from 0 to 60 in increments of 20; fold change, ranging from 0 to 3 in unit increments; fold change, ranging from 0 to 6 in increments of 2; fold change, ranging from 0 to 5 in unit increments; fold change, ranging from 0 to 4 in unit increments; fold change, ranging from 0.0 to 2.0 in increments of 0.5; fold change, ranging from 0 to 15 in increments of 5 (y-axis) across isotype control, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; Th 0, T follicular helper, T follicular helper plus 30 nanomolar I P, T follicular helper plus 300 nanomolar I P; T follicular helper, T follicular helper plus I P, Ah R K O T follicular helper, Ah R K O T follicular helper plus I P; T follicular helper, T follicular helper plus I P, Ah R K O T follicular helper, Ah R K O T follicular helper plus I P; T follicular helper, T follicular helper plus I P, Ah R K O T follicular helper, Ah R K O T follicular helper plus I P (x-axis) for C X C R 5 positive I C O S-positive Bcl-6 positive subset, Bcl-6, Batf, c-Maf, Irf4, ll21, Tox2, Cyp1a1, ll4, Bcl-6, lrf4, ll21. Figures 5N to 5P are graphs, plotting fold change, ranging from 0 to 5 in unit increments, fold change, ranging from 0 to 15 in increments of 5; fold change, ranging from 0 to 15 in increments of 5 (y-axis) across T follicular helper, T follicular helper plus I P, T follicular helper, T follicular helper plus I P plus C H-223191 (x-axis) for Bcl-6, Irf4, and ll21.
The Effects of IP–AhR–c-Maf Axis on Regulation of Il4 and Il21 Expression in Tfh Cells
We hypothesized that upon ligand binding, AhR would translocate to the nucleus and interact with cellular musculoaponeurotic fibrosarcoma (c-Maf), allowing the AhR–c-Maf complex to bind to the promoter region of Il4 and Il21 to transactivate the expression of Il4 and Il21. To test our hypothesis, c-Maf was knocked down in the Tfh differentiation conditions with or without IP treatment in vitro. We found that IP exposure induced higher percentages of CXCR5+ICOS+ and CXCR5+ICOS+Bcl-6+ Tfh cells in scramble control groups, but this was impaired in the c-Maf knockdown groups (Figure 6A,B). The IP-induced higher expression of Tfh-associated genes, including Bcl-6, Il21, and Il4, was also lower in c-Maf knockdown cells (Figure 6C–F), but c-Maf knockdown did not appear to affect the expression of the AhR downstream gene, Cyp1a1 (Figure 6G). We next sought to identify the transcriptional factor binding motifs of AhR and c-Maf at the Il21 and Il4 promoters. We found that the Il21 promoter has two putative AhR-binding sites [xenobiotic response element 1 (XRE1) and XRE2] and three putative c-Maf-binding sites [Maf recognition element 1 (MARE1), MARE2, and MARE3] (Figure 6H). Moreover, three putative XREs (XRE1, XRE2, and XRE3) and two MAREs (MARE1 and MARE2) were predicted in the Il4 promoter (Figure 6I). To confirm that AhR and c-Maf also interact with their target sequences in the Il21 promoters under IP treatment, we performed ChIP assays with differentiated Tfh cells in vitro. At the Il21 promoter in Tfh cells, c-Maf interacted with MARE1-3, and AhR interacted with XRE1 and XRE2. With IP treatment, the AhR and c-Maf occupancy at the Il21 promoter was greater in Tfh cells than in the DMSO control group. No AhR or c-Maf interaction with the XRE or MARE sequences, respectively, at the Il21 promoter was detected when we used isotype control antibodies (Figure 6J,K). Compared with that seen in Th0 cells, AhR interacted with XRE1 and XRE3 at the Il4 promoter in Tfh cells, and c-Maf interacted with MARE2 at the Il4 promoter in Tfh cells; in addition, IP treatment resulted in higher AhR and c-Maf occupancy of the Il4 promoter in Tfh cells than in the DMSO control group (Figure 6L,M).
Figure 6. The effects of AhR and c-Maf on IP-induced Tfh cell differentiation in vitro. Flow cytometry analysis of the (A) CD4+CXCR5+ICOS+ subset and (B) CD4+CXCR5+ICOS+Bcl-6+ subset in GFP+ cells (c-Maf shRNA-expressing cells) (numerical values are shown in Excel Tables S80 and S81). (C–G) The mRNA expression levels of c-Maf, Bcl-6, Il21, Il4 and Cyp1a1 in c-Maf-knockdown cells were measured by RT-qPCR (numerical values are shown in Excel Tables S82–S86). (H) Schematic diagram of AhR- and c-Maf-binding sites (XRE and MARE) in the Il21 promotor region. (I) Schematic diagram of the AhR- and c-Maf-binding sites (XRE and MARE) in the Il4 promotor region. IP was added to the cell culture medium at a concentration of 30 nM or 300 nM during Tfh cell differentiation. Immunoprecipitation test of c-Maf- and AhR- binding levels in Tfh cells. Chromatin immunoprecipitation (ChIP) assay of the and (J) c-Maf- and (K) AhR-binding levels at the Il21 promotor region in Th0 cells, Tfh cells, IP-treated Tfh cells, and IgG isotype control (numerical values are shown in Excel Tables S87 and S88). ChIP assay of the (L) AhR- and (M) c-Maf-binding levels at the Il4 promotor region in Th0 cells, Tfh cells, IP-treated Tfh cells, and IgG isotype control (numerical values are shown in Excel Tables S89 and S90). (N) The AhR and c-Maf interaction in Tfh cells and IP-treated Tfh cells was examined by immunoprecipitation assay and western blotting. DMSO was the vehicle for IP and all experiments were done in the presence of the DMSO (0.01%) as vehicle control. All data are representative of two or three independent experiments. n=3–4 each group. Data are shown as the means±SEMs. ns, not significant; *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 according to ANOVA followed by Bonferroni’s multiple comparisons test. Note: AhR, aryl hydrocarbon receptor; ANOVA, analysis of variance; ChIP, chromatin immunoprecipitation; c-Maf, cellular musculoaponeurotic fibrosarcoma; DMSO, dimethyl sulfoxide; GFP, green fluorescent protein; Ig, immunoglobulin; IL, interleukin; IP, indeno[1,2,3-cd]pyrene; MARE, Maf recognition element; RT-qPCR, quantitative reverse transcription polymerase chain reaction; SEM, standard error of the mean; shRNA, short hairpin RNA; Tfh, T follicular helper; Th0, immature T helper cell; TSS, transcription start site; XRE, xenobiotic response element.
Figures 6A to 6G are graphs titled C X C R 5 positive I C O S-positive subset, C X C R 5 positive I C O S-positive Bcl-6 positive subset, c-Maf, Bcl-6, ll21, ll4, Cyp1a1, plotting percentage of G F P positive cells, ranging from 55 to 85 in increments of 5; percentage of G F P positive cells, ranging from 40 to 60 in increments of 5; fold change, ranging from 0.0 to 2.0 in increments of 0.5; fold change, ranging from 0.0 to 2.0 in increments of 0.5; fold change, ranging from 0.8 to 1.6 in increments of 0.2; fold change, ranging from 0 to 8 in increments of 2; fold change, ranging from 0 to 10 in increments of 2 (y-axis) across scrambled sh R N A plus D M S O, scrambled control plus I P, c-Maf sh R N A plus D M S O, and c-Maf sh R N A plus I P (x-axis), respectively. Figure 6H is a schematic diagram, depicting the Ah R- and c-Maf-binding sites, including X R E and M A R E from negative 1,200 to T S S in the mouse Il21 promoter region. Figure 6I is a schematic diagram, depicting the Ah R- and c-Maf-binding sites, including X R E and M A R E from negative 1,200 to T S S in the mouse Il4 promoter region. Figures 6J to 6M are graphs titled ll21 promoter (c-maf Ch I P), ll21 promoter (Ah R Ch I P), ll4 promoter (Ah R Ch I P), ll4 promoter (c-maf Ch I P), plotting percentage of input, ranging from 0.0 to 1.5 in increments of 0.5; percentage of input, ranging from 0.0 to 1.0 in increments of 0.2; percentage of input, ranging from 0.0 to 2.0 in increments of 0.5; percentage of input, ranging from 0.0 to 2.0 in increments of 0.5 (y-axis) across Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P, Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P, Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P; Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P; Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P; Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P, Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P, Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P; Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P; Ig G isotype, Th 0, T follicular helper, T follicular helper plus 300 nanomolar I P (x-axis) for M A R E 1, M A R E 2, M A R E 3; X R E 1, X R E 2; X R E 1, X R E 2, X R E 3; and M A R E 1, M A R E 2, respectively. Figure 6N is a western blot displaying five columns, namely, T follicular helper, T follicular helper plus I P under Input and Ig G, T follicular helper, T follicular helper plus I P under Input and I P: Ah R and three rows, namely, Ah R, C-Maf, and lowercase beta actin.
To test whether AhR and c-Maf physically interact with each other in Tfh cells, we performed immunoprecipitation followed by western blotting analysis. Results showed that AhR and c-Maf expression was higher in Tfh cells with IP treatment. Moreover, the c-Maf and AhR complex could be immunoprecipitated in Tfh cells using anti-AhR antibodies and probed with antic-Maf antibodies, suggesting that AhR physically interacted with c-Maf (Figure 6N). In addition, the AhR and c-Maf interaction was enhanced in Tfh cells with IP treatment.
Discussion
Recent studies have implicated the role of Tfh cells in infections and vaccine responses,31 and dysregulation of Tfh and Tfr cells has been suggested as a critical player in various autoimmune diseases,31 inflammatory disorders,32 and cancer.33 However, the molecular mechanism by which Tfh cells are regulated remains to be fully defined. The results from the present study provide evidence supporting a role of environmental PM2.5 and IP in targeting Tfh cell differentiation via AhR, suggesting the potential importance of the PM2.5–AhR–c-Maf axis in the regulation of Tfh cell differentiation.
A previous study has shown that AhR and c-Maf interaction triggers the production of IL-21 in IL-27–induced regulatory type 1 cells.34 c-Maf promotes the differentiation of Tfh cells by increasing IL-21 secretion, and the Tfh function by increasing IL-4 secretion.35 In the present study, using IP as a model of PAH exposure, it was shown that IP activated AhR to bind with c-Maf, which promoted transactivation of the Il21 and Il4 promoters, suggesting that IL-21 production is required for Tfh cell differentiation and IL-4 production in Tfh cells to promote IgE class-switch recombination in B cells. Although a previous study showed the links between PM2.5 exposure, allergic sensitization, asthma, and Th17 activation,36 our data provide another potentially important regulatory mechanism involving the PM2.5–AhR axis in Tfh cells. Collectively, these results add a potentially new dimension in the understanding of Tfh cell differentiation and function, which, together with those potential targets of PM2.5 exposure, including B cells,37 Th2,38 and Th17,36,39 aggravate allergen-induced pulmonary allergic responses.
Tfh cell differentiation is a multistep process, in which DCs are necessary and sufficient to induce the Tfh cell intermediate that requires additional interactions with distinct APCs for promoting the full differentiation program of Tfh cells.40,41 Our data showed that the frequency of the migratory DC-1 subset with high expression of programmed cell death 1 ligand 1 (PDL1) and PDL2 was greater after HDM treatment in vivo. However, there was no difference in the proportion of resident DCs and migratory DCs between the HDM and HDM plus PM2.5 groups, suggesting that PM2.5 exposure did not alter the proportion of resident DCs and migratory DCs in LNs. In addition to DCs, a small population of cells, which were defined as macrophages, was significantly higher in the HDM plus PM2.5 group compared with the PBS and HDM groups. Interestingly, the specific subset of the macrophage population expressed IL-4 and IFN-γ, which may play a role in the overall mechanism of regulation by PM2.5 exposure.
In contrast to our study, in a mouse model of acute influenza infection, administration of AhR ligands, 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), 3,3′,4,4′,5 pentachlorobiphenyl (PCB 126) and an endogenous agonist, 2-(1H-indol-3-ylcarbonyl)-4-thiazolecarboxylic acid methyl ester (ITE) reduced the Tfh cells, whereas administration of another endogenous agonist, 6-formylindolo[3,2-b]carbazole (FICZ) increased their frequency.42 The follow-up study suggested that oral delivery of TCDD reduced Tfh cell differentiation and T-cell–dependent B-cell responses in mice after acute influenza infection,43 although its suppressive mechanism was unclear. These contrasting results suggest that different AhR ligands and their dosage and the route of administration may exert differential outcomes. Similar ligand- and dose-dependent outcomes have been noted in the case of Treg and DC regulation by TCDD vs. PM2.5/IP.24,44–46 Another study reported that TCDD exposure in a rat model before HDM sensitization was shown to suppress the immune response to HDM.47 We also reported that TCDD suppressed, whereas IP increased, total IgE and antigen-specific IgE levels in a mouse allergic lung inflammation model.46,48 Indeed, although the functional impact of the AhR–ligand axis in various pathophysiological contexts has been well documented, the molecular basis and features governing the AhR signaling and its transcriptional activity remain to be fully elucidated. It is now recognized that its functional outcomes are ligand,42 cell type,49,50 and context44,45 dependent, wherein the ligand’s structural feature,51 the dosage used, and the timing of exposure52 are known to influence the activation of AhR and its subsequent nuclear translocation and binding of transcriptional co-activators; also, the ligand’s half-life in vivo may also determine the final outcome. Therefore, differential AhR signaling upon interaction with different ligands can be logically anticipated, particularly in the context of different model systems, such as allergic inflammation vs. respiratory infection, with a distinct molecular and cellular network of regulation.
Because it is challenging to examine human Tfh cells in secondary lymphoid organs, such as tonsils, spleen, and LNs, the alternative approach is to examine circulating CXCR5+CD4+ Tfh (cTfh) cells in human peripheral blood samples. The frequency of cTfh cells is positively correlated with serum IgE levels, and cTfh cells from patients with asthma more effectively promote IgE production than cTfh cells from healthy control study participants.53 Furthermore, the skewing of cTfh cells toward cTfh2 cells has been observed in patients with asthma, and the positive correlation of serum PAH and IL-4 levels has been noted in children with asthma.54 In addition, the patients with asthma treated with inhaled corticosteroids show significantly lower frequencies of cTfh2 and ICOS+ cTfh cells, as well as improved asthma symptoms.55 Our mouse model and single-cell mass cytometry analysis provide a comprehensive analysis and plausible mechanism linking PM2.5 exposure, IgE production, and allergic disease, in which the AhR–c-Maf axis–mediated Tfh cell differentiation is a prominent mechanistic feature.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
K.L., Y.G., W.X., J.F., and S.H. designed and carried out experiments and analyzed the data. S.H., X.X., and X.H. contributed critical reagents and protocols. K.L., S.H., and Y.Z. wrote the manuscript. S.H. and Y.Z. planned, designed, supervised, and coordinated the overall research efforts.
This work was supported by grants from the National Key R&D Program of China (2021YFC2701800 and 2021YFC2701802 to Y.Z.), National Natural Science Foundation of China (82241038 and 81974248 to Y.Z.; 82100033 to S.H.), Program for Outstanding Medical Academic Leader (2019LJ19 to Y.Z.), Shanghai Committee of Science and Technology (21140902400 to Y.Z.; 23ZR1407600, 21ZR1410000 and 19ZR1406400 to JF; 20ZR1408300 to X.H.), the International Joint Laboratory Program of National Children’s Medical Center (EK1125180109 to Y.Z.), Shanghai Municipal Planning Commission of Science and Research Fund (20214Y0440 to S.H.), Shenzhen Third Affiliated Hospital of Shenzhen University Postdoctoral Fellowship, Shenzhen Science and Technology Peacock Team Project (KQTD20170331145453160 to S.H.), and National Health Research Institutes, Taiwan (EM-109-PP-10 to S.H.).
==== Refs
References
1. Fahy JV. 2015. Type 2 inflammation in asthma—present in most, absent in many. Nat Rev Immunol 15 (1 ):57–65, PMID: , 10.1038/nri3786.25534623
2. Gowthaman U, Chen JS, Zhang B, Flynn WF, Lu Y, Song W, et al. 2019. Identification of a T follicular helper cell subset that drives anaphylactic IgE. Science 365 (6456 ):eaaw6433, PMID: , 10.1126/science.aaw6433.31371561
3. Kobayashi T, Iijima K, Dent AL, Kita H. 2017. Follicular helper T cells mediate IgE antibody response to airborne allergens. J Allergy Clin Immunol 139 (1 ):300–313.e7, PMID: , 10.1016/j.jaci.2016.04.021.27325434
4. Xue M, Xu S, Su L, He S, Wu B, Ji C, et al. 2021. Surfactant protein-A inhibits thymic stromal lymphopoietin-mediated T follicular helper cell differentiation and IgE production in asthma. Clin Immunol 231 :108822, PMID: , 10.1016/j.clim.2021.108822.34400320
5. Gowthaman U, Chen JS, Eisenbarth SC. 2020. Regulation of IgE by T follicular helper cells. J Leukoc Biol 107 (3 ):409–418, PMID: , 10.1002/JLB.3RI1219-425R.31965637
6. Ise W, Fujii K, Shiroguchi K, Ito A, Kometani K, Takeda K, et al. 2018. T follicular helper cell-germinal center B cell interaction strength regulates entry into plasma cell or recycling germinal center cell fate. Immunity 48 (4 ):702–715.e4, PMID: , 10.1016/j.immuni.2018.03.027.29669250
7. Qi H, Chen X, Chu C, Liu D, Ma W, Wang Y, et al. 2014. Tfh cell differentiation and their function in promoting B-cell responses. Adv Exp Med Biol 841 :153–180, PMID: , 10.1007/978-94-017-9487-9_6.25261207
8. Olatunde AC, Hale JS, Lamb TJ. 2021. Cytokine-skewed Tfh cells: functional consequences for B cell help. Trends Immunol 42 (6 ):536–550, PMID: , 10.1016/j.it.2021.04.006.33972167
9. Nurieva RI, Chung Y, Martinez GJ, Yang XO, Tanaka S, Matskevitch TD, et al. 2009. Bcl6 mediates the development of T follicular helper cells. Science 325 (5943 ):1001–1005, PMID: , 10.1126/science.1176676.19628815
10. Yu D, Rao S, Tsai LM, Lee SK, He Y, Sutcliffe EL, et al. 2009. The transcriptional repressor Bcl-6 directs T follicular helper cell lineage commitment. Immunity 31 (3 ):457–468, PMID: , 10.1016/j.immuni.2009.07.002.19631565
11. Johnston RJ, Poholek AC, DiToro D, Yusuf I, Eto D, Barnett B, et al. 2009. Bcl6 and Blimp-1 are reciprocal and antagonistic regulators of T follicular helper cell differentiation. Science 325 (5943 ):1006–1010, PMID: , 10.1126/science.1175870.19608860
12. Nurieva RI, Chung Y, Hwang D, Yang XO, Kang HS, Ma L, et al. 2008. Generation of T follicular helper cells is mediated by interleukin-21 but independent of T helper 1, 2, or 17 cell lineages. Immunity 29 (1 ):138–149, PMID: , 10.1016/j.immuni.2008.05.009.18599325
13. Read KA, Powell MD, Oestreich KJ. 2016. T follicular helper cell programming by cytokine-mediated events. Immunology 149 (3 ):253–261, PMID: , 10.1111/imm.12648.27442976
14. Morita R, Schmitt N, Bentebibel SE, Ranganathan R, Bourdery L, Zurawski G, et al. 2011. Human blood CXCR5+CD4+ T cells are counterparts of T follicular cells and contain specific subsets that differentially support antibody secretion. Immunity 34 (1 ):108–121, PMID: , 10.1016/j.immuni.2010.12.012.21215658
15. Akiyama M, Yasuoka H, Yoshimoto K, Takeuchi T. 2018. Interleukin-4 contributes to the shift of balance of IgG subclasses toward IgG4 in IgG4-related disease. Cytokine 110 :416–419, PMID: , 10.1016/j.cyto.2018.05.009.29861381
16. Sahoo A, Alekseev A, Tanaka K, Obertas L, Lerman B, Haymaker C, et al. 2015. Batf is important for IL-4 expression in T follicular helper cells. Nat Commun 6 (1 ):7997, PMID: , 10.1038/ncomms8997.26278622
17. Cheng Q, Yang CY, Guo BY, Wei X, Liu M. 2017. Analysis of mechanism of PM2.5 and house dust mite antigen Der p1 in attack stage of child asthma. Eur Rev Med Pharmacol Sci 21 (10 ):2458–2462, PMID: .28617540
18. Rosenquist NA, Metcalf WJ, Ryu SY, Rutledge A, Coppes MJ, Grzymski JJ, et al. 2020. Acute associations between PM2.5 and ozone concentrations and asthma exacerbations among patients with and without allergic comorbidities. J Expo Sci Environ Epidemiol 30 (5 ):795–804, PMID: , 10.1038/s41370-020-0213-7.32094459
19. Zhao C, Wang Y, Su Z, Pu W, Niu M, Song S, et al. 2020. Respiratory exposure to PM2.5 soluble extract disrupts mucosal barrier function and promotes the development of experimental asthma. Sci Total Environ 730 :139145, PMID: , 10.1016/j.scitotenv.2020.139145.32402975
20. Chang KF, Fang GC, Chen JC, Wu YS. 2006. Atmospheric polycyclic aromatic hydrocarbons (PAHs) in Asia: a review from 1999 to 2004. Environ Pollut 142 (3 ):388–396, PMID: , 10.1016/j.envpol.2005.09.025.16343719
21. Yu SY, Liu W, Xu YS, Yi K, Zhou M, Tao S, et al. 2019. Characteristics and oxidative potential of atmospheric PM2.5 in Beijing: source apportionment and seasonal variation. Sci Total Environ 650 (pt 1 ):277–287, PMID: , 10.1016/j.scitotenv.2018.09.021.30199673
22. Plé C, Fan Y, Ait Yahia S, Vorng H, Everaere L, Chenivesse C, et al. 2015. Polycyclic aromatic hydrocarbons reciprocally regulate IL-22 and IL-17 cytokines in peripheral blood mononuclear cells from both healthy and asthmatic subjects. PLoS One 10 (4 ):e0122372, PMID: , 10.1371/journal.pone.0122372.25860963
23. Tajima H, Tajiki-Nishino R, Watanabe Y, Fukuyama T. 2019. Direct activation of aryl hydrocarbon receptor by benzo[a]pyrene elicits T-helper 2-driven proinflammatory responses in a mouse model of allergic dermatitis. J Appl Toxicol 39 (7 ):936–944, PMID: , 10.1002/jat.3782.30748024
24. Sun L, Fu J, Lin SH, Sun JL, Xia L, Lin CH, et al. 2020. Particulate matter of 2.5 μm or less in diameter disturbs the balance of TH17/regulatory T cells by targeting glutamate oxaloacetate transaminase 1 and hypoxia-inducible factor 1α in an asthma model. J Allergy Clin Immunol 145 (1 ):402–414, PMID: , 10.1016/j.jaci.2019.10.008.31647966
25. Świst K, Pajtasz-Piasecka E. 2011. The influence of transcription factors on CD4+ T cell differentiation [in Polish]. Postepy Hig Med Dosw (Online) 65 :414–426, PMID: , 10.5604/17322693.949499.21734326
26. Qian G, Jiang W, Zou B, Feng J, Cheng X, Gu J, et al. 2018. LPS inactivation by a host lipase allows lung epithelial cell sensitization for allergic asthma. J Exp Med 215 (9 ):2397–2412, PMID: , 10.1084/jem.20172225.30021797
27. Xu W, Zhao X, Wang X, Feng H, Gou M, Jin W, et al. 2019. The transcription factor Tox2 drives T follicular helper cell development via regulating chromatin accessibility. Immunity 51 (5 ):826–839.e5, PMID: , 10.1016/j.immuni.2019.10.006.31732165
28. Zhou Y, Do DC, Ishmael FT, Squadrito ML, Tang HM, Tang HL, et al. 2018. Mannose receptor modulates macrophage polarization and allergic inflammation through miR-511-3p. J Allergy Clin Immunol 141 (1 ):350–364.e8, PMID: , 10.1016/j.jaci.2017.04.049.28629744
29. Mintz MA, Cyster JG. 2020. T follicular helper cells in germinal center B cell selection and lymphomagenesis. Immunol Rev 296 (1 ):48–61, PMID: , 10.1111/imr.12860.32412663
30. Clement RL, Daccache J, Mohammed MT, Diallo A, Blazar BR, Kuchroo VK, et al. 2019. Follicular regulatory T cells control humoral and allergic immunity by restraining early B cell responses. Nat Immunol 20 (10 ):1360–1371, PMID: , 10.1038/s41590-019-0472-4.31477921
31. Crotty S. 2019. T follicular helper cell biology: a decade of discovery and diseases. Immunity 50 (5 ):1132–1148, PMID: , 10.1016/j.immuni.2019.04.011.31117010
32. Uwadiae FI, Pyle CJ, Walker SA, Lloyd CM, Harker JA. 2019. Targeting the ICOS/ICOS-L pathway in a mouse model of established allergic asthma disrupts T follicular helper cell responses and ameliorates disease. Allergy 74 (4 ):650–662, PMID: , 10.1111/all.13602.30220084
33. Cui C, Wang J, Fagerberg E, Chen PM, Connolly KA, Damo M, et al. 2021. Neoantigen-driven B cell and CD4 T follicular helper cell collaboration promotes anti-tumor CD8 T cell responses. Cell 184 (25 ):6101–6118.e13, PMID: , 10.1016/j.cell.2021.11.007.34852236
34. Apetoh L, Quintana FJ, Pot C, Joller N, Xiao S, Kumar D, et al. 2010. The aryl hydrocarbon receptor interacts with c-Maf to promote the differentiation of type 1 regulatory T cells induced by IL-27. Nat Immunol 11 (9 ):854–861, PMID: , 10.1038/ni.1912.20676095
35. Ahearne MJ, Willimott S, Piñon L, Kennedy DB, Miall F, Dyer MJS, et al. 2013. Enhancement of CD154/IL4 proliferation by the T follicular helper (Tfh) cytokine, IL21 and increased numbers of circulating cells resembling Tfh cells in chronic lymphocytic leukaemia. Br J Haematol 162 (3 ):360–370, PMID: , 10.1111/bjh.12401.23710828
36. Zhang J, Fulgar CC, Mar T, Young DE, Zhang Q, Bein KJ, et al. 2018. TH17-induced neutrophils enhance the pulmonary allergic response following BALB/c exposure to house dust mite allergen and fine particulate matter from California and China. Toxicol Sci 164 (2 ):627–643, PMID: , 10.1093/toxsci/kfy127.29846732
37. Villa M, Gialitakis M, Tolaini M, Ahlfors H, Henderson CJ, Wolf CR, et al. 2017. Aryl hydrocarbon receptor is required for optimal B-cell proliferation. EMBO J 36 (1 ):116–128, PMID: , 10.15252/embj.201695027.27875245
38. Yan LI, Gong C, Ying L, Fu W, Liu S, Dai J, et al. 2019. PM2.5 affects establishment of immune tolerance in newborn mice by reducing PD-L1 expression. J Biosci 44 (2 ):41, PMID: , 10.1007/s12038-019-9858-6.31180054
39. Castañeda AR, Vogel CFA, Bein KJ, Hughes HK, Smiley-Jewell S, Pinkerton KE. 2018. Ambient particulate matter enhances the pulmonary allergic immune response to house dust mite in a BALB/c mouse model by augmenting Th2- and Th17-immune responses. Physiol Rep 6 (18 ):e13827, PMID: , 10.14814/phy2.13827.30230272
40. Goenka R, Barnett LG, Silver JS, O’Neill PJ, Hunter CA, Cancro MP, et al. 2011. Cutting edge: dendritic cell-restricted antigen presentation initiates the follicular helper T cell program but cannot complete ultimate effector differentiation. J Immunol 187 (3 ):1091–1095, PMID: , 10.4049/jimmunol.1100853.21715693
41. Deenick EK, Chan A, Ma CS, Gatto D, Schwartzberg PL, Brink R, et al. 2010. Follicular helper T cell differentiation requires continuous antigen presentation that is independent of unique B cell signaling. Immunity 33 (2 ):241–253, PMID: , 10.1016/j.immuni.2010.07.015.20691615
42. Boule LA, Burke CG, Jin GB, Lawrence BP. 2018. Aryl hydrocarbon receptor signaling modulates antiviral immune responses: ligand metabolism rather than chemical source is the stronger predictor of outcome. Sci Rep 8 (1 ):1826, PMID: , 10.1038/s41598-018-20197-4.29379138
43. Houser CL, Lawrence BP. 2022. The aryl hydrocarbon receptor modulates T follicular helper cell responses to influenza virus infection in mice. J Immunol 208 (10 ):2319–2330, PMID: , 10.4049/jimmunol.2100936.35444027
44. Quintana FJ, Basso AS, Iglesias AH, Korn T, Farez MF, Bettelli E, et al. 2008. Control of Treg and TH17 cell differentiation by the aryl hydrocarbon receptor. Nature 453 (7191 ):65–71, PMID: , 10.1038/nature06880.18362915
45. Schulz VJ, van Roest M, Bol-Schoenmakers M, van Duursen MBM, van den Berg M, Pieters RHH, et al. 2013. Aryl hydrocarbon receptor activation affects the dendritic cell phenotype and function during allergic sensitization. Immunobiology 218 (8 ):1055–1062, PMID: , 10.1016/j.imbio.2013.01.004.23433705
46. Wong TH, Lee CL, Su HH, Lee CL, Wu CC, Wang CC, et al. 2018. A prominent air pollutant, indeno[1,2,3-cd]pyrene, enhances allergic lung inflammation via aryl hydrocarbon receptor. Sci Rep 8 (1 ):5198, PMID: , 10.1038/s41598-018-23542-9.29581487
47. Luebke RW, Copeland CB, Daniels M, Lambert AL, Gilmour MI. 2001. Suppression of allergic immune responses to house dust mite (HDM) in rats exposed to 2,3,7,8-TCDD. Toxicol Sci 62 (1 ):71–79, PMID: , 10.1093/toxsci/62.1.71.11399795
48. Xu T, Zhou Y, Qiu L, Do DC, Zhao Y, Cui Z, et al. 2015. Aryl hydrocarbon receptor protects lungs from cockroach allergen-induced inflammation by modulating mesenchymal stem cells. J Immunol 195 (12 ):5539–5550, PMID: , 10.4049/jimmunol.1501198.26561548
49. Hung CH, Yang SN, Wang YF, Liao WT, Kuo PL, Tsai EM, et al. 2013. Environmental alkylphenols modulate cytokine expression in plasmacytoid dendritic cells. PLoS One 8 (9 ):e73534, PMID: , 10.1371/journal.pone.0073534.24039973
50. Su HH, Lin HT, Suen JL, Sheu CC, Yokoyama KK, Huang SK, et al. 2016. Aryl hydrocarbon receptor–ligand axis mediates pulmonary fibroblast migration and differentiation through increased arachidonic acid metabolism. Toxicology 370 :116–126, PMID: , 10.1016/j.tox.2016.09.019.27697457
51. Itkin B, Breen A, Turyanska L, Sandes EO, Bradshaw TD, Loaiza-Perez AI. 2020. New treatments in renal cancer: the AhR ligands. Int J Mol Sci 21 (10 ):3551, PMID: , 10.3390/ijms21103551.32443455
52. Rannug A, Rannug U. 2018. The tryptophan derivative 6-formylindolo[3,2-b]carbazole, FICZ, a dynamic mediator of endogenous aryl hydrocarbon receptor signaling, balances cell growth and differentiation. Crit Rev Toxicol 48 (7 ):555–574, PMID: , 10.1080/10408444.2018.1493086.30226107
53. Gong F, Su Q, Jiang D, Chen J, Pan Y, Huang X. 2014. High frequency of circulating follicular helper T cells in patients with bronchial asthma. Clin Lab 60 (6 ):963–968, PMID: , 10.7754/Clin.Lab.2013.130427.25016701
54. Al-Daghri NM, Abd-Alrahman S, Draz H, Alkharfy K, Mohammed AK, Clerici MS, et al. 2014. Increased IL-4 mRNA expression and poly-aromatic hydrocarbon concentrations from children with asthma. BMC Pediatr 14 (1 ):17, PMID: , 10.1186/1471-2431-14-17.24450480
55. Miyajima S, Shigehara K, Kamekura R, Takaki H, Yabe H, Ikegami I, et al. 2020. Activated circulating T follicular helper cells and skewing of T follicular helper 2 cells are down-regulated by treatment including an inhaled corticosteroid in patients with allergic asthma. Allergol Int 69 (1 ):66–77, PMID: , 10.1016/j.alit.2019.08.008.31648923
|
PMC010xxxxxx/PMC10045040.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36976258
EHP11270
10.1289/EHP11270
Research
The Folic Acid and Creatine Trial: Treatment Effects of Supplementation on Arsenic Methylation Indices and Metabolite Concentrations in Blood in a Bangladeshi Population
https://orcid.org/0000-0002-7241-4574
Abuawad Ahlam K. 1 *
Bozack Anne K. 1 2 *
Navas-Acien Ana 1
Goldsmith Jeff 3
Liu Xinhua 3
Hall Megan N. 4
Ilievski Vesna 1
Lomax-Luu Angela M. 1
Parvez Faruque 1
Shahriar Hasan 5
Uddin Mohammad N. 5
Islam Tariqul 5
Graziano Joseph H. 1
https://orcid.org/0000-0003-4169-1162
Gamble Mary V. 1
1 Department of Environmental Health Sciences, Mailman School of Public Health, Columbia University, New York, New York, USA
2 Division of Environmental Health Sciences, School of Public Health, University of California, Berkeley, Berkeley, California, USA
3 Department of Biostatistics, Mailman School of Public Health, Columbia University, New York, New York, USA
4 Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, New York, USA
5 Columbia University Arsenic Project in Bangladesh, Dhaka, Bangladesh
Address correspondence to Mary V. Gamble, 722 W. 168th St., 1107E, New York, NY 10032 USA. Email: [email protected]
28 3 2023
3 2023
131 3 03701519 3 2022
19 2 2023
24 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Chronic arsenic (As) exposure is a global environmental health issue. Inorganic As (InAs) undergoes methylation to monomethyl (MMAs) and dimethyl-arsenical species (DMAs); full methylation to DMAs facilitates urinary excretion and is associated with reduced risk for As-related health outcomes. Nutritional factors, including folate and creatine, influence one-carbon metabolism, the biochemical pathway that provides methyl groups for As methylation.
Objective:
Our aim was to investigate the effects of supplementation with folic acid (FA), creatine, or the two combined on the concentrations of As metabolites and the primary methylation index (PMI: MMAs/InAs) and secondary methylation index (SMI: DMAs/MMAs) in blood in Bangladeshi adults having a wide range of folate status.
Methods:
In a randomized, double-blinded, placebo (PBO)-controlled trial, 622 participants were recruited independent of folate status and assigned to one of five treatment arms: a) PBO (n=102), b) 400μg FA/d (400FA; n=153), c) 800μg FA/d (800FA; n=151), d) 3g creatine/d (creatine; n=101), or e) 3g creatine+400μg of FA/d (creatine+400FA; n=103) for 12 wk. For the following 12 wk, half of the FA participants were randomly switched to the PBO while the other half continued FA supplementation. All participants received As-removal water filters at baseline. Blood As (bAs) metabolites were measured at weeks 0, 1, 12, and 24.
Results:
At baseline, 80.3% (n=489) of participants were folate sufficient (≥9 nmol/L in plasma). In all groups, bAs metabolite concentrations decreased, likely due to filter use; for example, in the PBO group, blood concentrations of MMAs (bMMAs) (geometric mean±geometric standard deviation) decreased from 3.55±1.89μg/L at baseline to 2.73±1.74 at week 1. After 1 wk, the mean within-person increase in SMI for the creatine+400FA group was greater than that of the PBO group (p=0.05). The mean percentage decrease in bMMAs between baseline and week 12 was greater for all treatment groups compared with the PBO group [400FA: −10.4 (95% CI: −11.9, −8.75), 800FA: −9.54 (95% CI: −11.1, −7.97), creatine: −5.85 (95% CI: −8.59, −3.03), creatine+400FA: −8.44 (95% CI: −9.95, −6.90), PBO: −2.02 (95% CI: −4.03, 0.04)], and the percentage increase in blood DMAs (bDMAs) concentrations for the FA-treated groups significantly exceeded that of PBO [400FA: 12.8 (95% CI: 10.5, 15.2), 800FA: 11.3 (95% CI: 8.95, 13.8), creatine+400FA: 7.45 (95% CI: 5.23, 9.71), PBO: −0.15 (95% CI: −2.85, 2.63)]. The mean decrease in PMI and increase in SMI in all FA groups significantly exceeded PBO (p<0.05). Data from week 24 showed evidence of a reversal of treatment effects on As metabolites from week 12 in those who switched from 800FA to PBO, with significant decreases in SMI [−9.0% (95% CI: −3.5, −14.8)] and bDMAs [−5.9% (95% CI: −1.8, −10.2)], whereas PMI and bMMAs concentrations continued to decline [−7.16% (95% CI: −0.48, −14.3) and −3.1% (95% CI: −0.1, −6.2), respectively] for those who remained on 800FA supplementation.
Conclusions:
FA supplementation lowered bMMAs and increased bDMAs in a sample of primarily folate-replete adults, whereas creatine supplementation lowered bMMAs. Evidence of the reversal of treatment effects on As metabolites following FA cessation suggests short-term benefits of supplementation and underscores the importance of long-term interventions, such as FA fortification. https://doi.org/10.1289/EHP11270
Supplemental Material is available online (https://doi.org/10.1289/EHP11270).
* These authors are joint first authors.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Arsenic (As) contamination of groundwater is common in many regions of the world. At least 94 million people worldwide are exposed to concentrations exceeding the World Health Organization (WHO) guideline of 10μg/L in groundwater.1–3 In Bangladesh, where well water is used as an alternative to microbially contaminated surface water, ∼30 million people remain exposed to drinking water with high As concentrations (>10μg/L).4
Arsenic exposure affects all organ systems, and chronic As exposure is associated with numerous adverse health effects, including cardiovascular disease, diabetes and diabetes-related outcomes, skin lesions (i.e., melanosis, leukomelanosis, and keratosis), impaired intellectual function, and cancers (e.g., bladder, lung, skin cancers).5–7 Arsenic metabolism may modify disease risk. Ingested inorganic As (InAs) undergoes a series of reduction and oxidative methylation reactions. According to the Challenger pathway, arsenite (InAsIII) is methylated to monomethylarsonic acid (MMAsV) by arsenic-3-methyltransferase (AS3MT) using S-adenosylmethionine (SAM) as the methyl donor. MMAsV is then reduced to MMAsIII and methylated to form dimethylarsinic acid (DMAsV) (Figure 1).9 The complete methylation of InAs to DMAs is important because toxicological studies in human lung and bladder cells and hepatocytes have shown that MMAsIII is the most cytotoxic and genotoxic As species.10,11 It is difficult to distinguish between MMAsV and MMAsIII in epidemiological studies owing to oxidation in biological samples; therefore, total MMAs (MMAs(III+V)) is reported. Among As-exposed populations, a higher proportion of MMAs(III+V) (and a lower proportion of DMAs) in urine has been associated with an increased risk for cancers (e.g., bladder, lung, skin cancers), skin lesions, peripheral vascular disease, and atherosclerosis.12,13
Figure 1. Relative distribution of arsenic metabolites in blood and urine. Inorganic arsenic (InAs) in the form of arsenite (InAsIII) is methylated to form monomethylarsonic acid (MMAsV) in a reaction that is catalyzed by arsenic (+3 oxidation state) methyltransferase (AS3MT) using the methyl donor S-adenosylmethionine (SAM). MMAsV is then reduced to form methylarsonous acid (MMAsIII), and a subsequent methylation yields dimethylarsinic acid (DMAsV) and S-adenosylhomocysteine (SAH). Total blood arsenic is composed of ∼25% InAs, 40% MMAs, and 35% DMAs, whereas total urine arsenic is composed of ∼15% InAs, 15% MMAs, and 70% DMAs.8
Figure 1 is an illustration, depicting the relative distribution of arsenic metabolites in blood and urine, where inorganic arsenic in the form of arsenite is methylated to form monomethylarsonic acid in a reaction that is catalyzed by arsenic methyltransferase and utilizes the methyl donor S-adenosylmethionine. The monomethylarsonic acid is then reduced to form methylarsonous acid, and a subsequent methylation yields dimethylarsinic acid and S-adenosylhomocysteine. The scale below titled “Arsenic metabolite distribution” displays the following information: In blood, inorganic arsenic is 25 percent; monomethyl-arsenicals are 40 percent; dimethyl-arsenicals are 35 percent. In urine, inorganic arsenic is 15 percent; monomethyl-arsenicals are 15 percent; dimethyl-arsenicals are 70 percent.
One-carbon metabolism (OCM), the metabolic pathway that generates SAM, is influenced by nutrients, including folate and creatine. Folate, in the form of 5-methyltetrahydrofolate (5-mTHF), donates a methyl group to homocysteine to form methionine, which is activated by methionine adenosyltransferase to form SAM. An estimated 40% of SAM is consumed by the methylation of guanidinoacetate (GAA) to form creatine, an amino acid derivative that is found in tissues with high energy requirements, such as muscle cells.14 Creatine synthesis is down-regulated by dietary intake of creatine, predominantly through the consumption of meat and by creatine supplementation.15 Increased creatine consumption may increase SAM available for the methylation of As.
In a previous randomized controlled trial (RCT), our group investigated the effect of folic acid (FA) supplementation on As methylation capacity among folate-deficient Bangladeshi adults (plasma folate concentrations <9 nmol/L) exposed to As through drinking water. Following 12 wk of treatment, the proportion of DMAs (%DMAs) in urine increased while %InAs and %MMAs in urine, as well as total blood As (bAs) and blood MMAs (bMMAs) concentrations decreased among participants receiving 400μg FA/d compared with the placebo (PBO) group.8,16
More recently, in the Folic Acid and Creatine Trial (FACT), we examined the independent and joint effects of FA and creatine treatment among adults recruited independent of folate status in Bangladesh, a country without FA fortification of food. The a priori primary outcome was change in total bAs concentrations at 12 wk. Among participants receiving 800μg FA/d, there was a greater decrease in total bAs concentrations [percentage change in geometric mean (GM) of −17.8%] compared with PBO (−9.5%) (p<0.05).17 Secondary outcomes included bAs metabolite concentrations (recently measured and reported here) and the proportions of As metabolites in urine (reported by Bozack et al.18). Among participants receiving 400μg FA, 800μg FA, or 400μg FA+creatine daily, there were significantly greater mean decreases in urine %InAs and %MMAs [change in log (%InAs)=−0.09, −0.14, and −0.11 for each treatment group, respectively; change in %MMAs=−1.80, −2.60, and −1.85, respectively] compared with PBO [changes in log (%InAs) and %MMAs=0.05 and 0.15, respectively] (p<0.05). In addition, participants receiving 400μg FA, 800μg FA, or 400μg FA+creatine daily had a greater mean increase in %DMAs (3.25, 4.57, and 3.11, respectively) compared with PBO (−1.17) (p<0.05).18 The decrease in urinary %MMAs among participants receiving 3g creatine/d at week 1 (−0.90) and week 12 (−1.13) also exceeded that of the PBO group (change in %MMAs, week 1=−0.62) (p<0.05).18
Although the relative distribution of As metabolite proportions in urine is an informative indicator of As methylation capacity, the concentrations of As metabolites in blood may better reflect As exposure to tissues. In addition, as a biomarker of exposure, bAs is not complicated by adjustment for dilution, as urinary As is, or—as compared with water As—by difficulties associated with estimating As consumption through water or dietary sources.
The present study uses newly available bAs metabolite data from FACT to investigate the effect of FA and creatine supplementation on As metabolites in blood. We report the effects of supplementation on the bAs primary methylation index (PMI: MMAs/InAs) and the secondary methylation index (SMI: DMAs/MMAs), as well as metabolite concentrations: blood InAs (bInAs), bMMAs, and blood DMAs (bDMAs), in a population of Bangladeshi adults having a wide range of folate nutritional status.
Methods
Study Population and Design
FACT was conducted in Araihazar, Bangladesh, and has previously been described in detail.17 FACT participants were recruited from the Health Effects of Arsenic Longitudinal Study (HEALS), a prospective cohort that originally comprised 11,746 adults living in a 25-km2 area of Araihazar, Bangladesh, recruited in 2000.19 HEALS was subsequently expanded to include >35,000 participants.20 Married adults were eligible for enrollment in HEALS if they had resided in the study area for ≥5 y and had primarily obtained their drinking water from one of the study wells for ≥3 y. Between December 2009 and May 2011, a random subset of 622 participants were recruited from the HEALS cohort to participate in FACT. Five trained teams consisting of one interviewer and one physician recruited participants through house-to-house visits. Male and female participants were eligible for FACT if, upon medical examination, they were 20–65 years of age and had been drinking water from a household well with water As concentrations >50μg/L for ≥1 y. Medical conditions were determined by a study physician who conducted a thorough medical examination at the screening visit. This examination included assessment of current and past medications, physical exam, self-report, and review of medical records. Individuals were excluded if they were pregnant, taking nutritional supplements, had protein in their urine, renal diseases, diabetes, or other health problems, such as gastrointestinal issues.
FACT participants were randomly assigned to one of the following treatment groups: a) PBO (n=104), b) 400μg FA/d (referred to hereafter as 400FA; n=156), c) 800μg FA/d (800FA; n=154), d) 3g creatine/d (creatine; n=104), and e) 3g creatine and 400μg FA/d (creatine+400FA; n=104). The dose of 400μg FA/d was selected because it is the U.S. recommended dietary allowance (RDA)21; the dose of 800μg FA/d was selected to evaluate whether a higher dose would be more effective; and the dose of 3g creatine/d was selected to exceed the average daily creatine loss for a 70-kg male individual (∼1.9g/d) as estimated by Brosnan et al., that is, it was a dose predicted to down-regulate endogenous creatine biosynthesis.14 To reduce bias and ensure a balance of male and female participants in each treatment arm, male and female participants were randomized separately in blocks. By blocks, treatment groups were assigned in the ratio of 1 (PBO):1.5 (400FA):1.5 (800FA):1 (creatine):1 (creatine+400FA). Random permutation was used to assign treatment groups. With the exception of the data management specialists who assigned treatment groups, all participants, field staff, village health workers, laboratory technicians, and investigators were blinded to the treatment for the duration of the study.
The study included two phases (Figure 2). In the first 12-wk phase, participants received supplementation according to treatment group assignments, and in the second 12-wk phase, half of the participants in the 400FA and 800FA groups were randomly switched to PBO (400FA/PBO: n=76 and 800FA/PBO: n=74) and half continued FA supplementation (400FA/FA: n=77; 800FA/FA: n=77) to assess the potential reversal of treatment effects on As metabolites following cessation of FA supplementation. During the second phase, all participants in the PBO, creatine, and creatine+400FA groups received PBO to maintain the study blind.
Figure 2. Folic Acid and Creatine Trial (FACT) study design overview. The PBO, 400μg of FA/d (400FA), 800μg of FA/d (800FA), 3g creatine (creatine), and 3g creatine+400μg of FA/d (creatine+400FA) are shown in brown, green, blue, orange, and purple, respectively. All five treatments were administered daily, with participants switching to PBO at 12 wk as indicated with a lighter color. Modified from Peters et al.17 Note: FA, folic acid; PBO, placebo.
Figure 2 shows a timeline and an illustration. The timeline depicts weeks: 0 to 12 weeks is the first phase, and 13 to 24 weeks is the second phase. The illustration depicts the Folic Acid and Creatine Trial study design and displays the treatment groups with the number of participants per group. 102 participants received placebo for both phases; 77 participants received 400 micrograms of folic acid, of which 76 switched to placebo for the second phase; 151 participants received 800 micrograms of folic acid, of which 74 switched to placebo for the second phase; 101 participants received 3 grams of creatine, all of which were switched to placebo for the second phase; 103 participants received 3 grams creatine plus 400 micrograms folic acid, all of which were switched to placebo for the second phase.
Ethics
The FACT protocol was approved by the institutional review board at Columbia University and the Bangladesh Medical Research Council. Informed consent was obtained by staff physicians in Bangladesh.
Fieldwork and Study Follow-Up
Fieldwork was conducted by the five trained teams, each comprising one interviewer and one physician. Sociodemographic characteristics were obtained at enrollment by field staff–administered questionnaires designed specifically for this study. During home visits, blood samples were collected at baseline and at weeks 1, 12, and 24.
At the beginning of the trial, participants received READ-F As-removal water filters (Brota Services International) and were encouraged to use the filters for all drinking and cooking water. Efficacy of the filters was tested in the field during home visits using the Hach EZ kit (Hach Company); filters were repaired or replaced if As water concentrations exceeded 10μg/L or if participants reported failure of the filter22 (∼50 of the 622 filters failed during the trial). Arsenic concentrations were also measured by laboratory analysis in a random subset of filtered water collected at baseline and at weeks 12 and 24 (described below). Total urinary As concentrations throughout the study were used as a proxy of water filter use. In addition, water filter use 1 y after FACT ended was measured by a follow-up survey, as previously reported by Sanchez et al.22
In the first phase of the trial, all participants received two pill bottles containing a) FA or matched PBO pills and b) creatine or matched PBO pills. In the second phase, all participants received one pill bottle containing FA or matched PBO pills. Village health workers performed daily home visits to ask about compliance or observe participants taking pills. Compliance was assessed using pill counts and circulating folate concentrations as follows: a) the percentage of study pills taken, and b) changes in red blood cell and plasma folate concentrations as reported in changes in GMs over time.17
Sample Handling and Laboratory Measures
Sample handling and laboratory methods have been described in detail.17 Briefly, venous blood was collected in vacutainer tubes with ethylenediaminetetraacetic acid during field visits, stored in IsoRack cool packs (Brinkmann Instruments) at 4°C, and transported to our Araihazar field clinic within 4 h. Plasma was separated by centrifugation at 1,350×g for 10 min at 25°C. Whole blood and plasma samples were stored at −80°C before shipment to Columbia University on dry ice.
Total bAs was measured using a PerkinElmer Elan DRC II inductively coupled plasma mass spectrometer (ICP-MS) with an AS 93+autosampler [intra- and interassay coefficient of variations (CVs): 2.7% and 5.7%, respectively]. AsIII, arsenate (AsV), MMAs, and DMAs were measured in blood and urine using high-performance liquid chromatography (HPLC) coupled to dynamic reaction cell ICP-MS, as previously described.23,24 This method cannot differentiate between reduced and oxidized forms of MMAs and DMAs. Although this method can separate AsV from AsIII, samples can oxidize during sample processing; therefore, AsV and AsIII were added and reported as a single variable reflecting total InAs. The limit of detection (LOD) and interassay CVs were, respectively, 0.11μg/L and 15.9% for arsenobetaine+arsenocholine, 0.22μg/L and 4.4% for AsIII and AsV, 0.22μg/L and 4.4% for MMAs, and 0.22μg/L and 4.7% for DMAs.
Plasma folate and vitamin B12 were analyzed using a radio protein-binding assay according to the manufacturer’s instructions (SimulTRAC-SNB; MP Biomedicals). Intra- and interassay CVs for plasma folate were 5% and 13%, respectively, and for vitamin B12 were 6% and 17%, respectively. Plasma total homocysteine (tHcy) concentrations were measured by HPLC with fluorescence detection according to method described by Pfeiffer et al.25 Separation was achieved with a 150×3.2-mm, 5-μm Prodigy ODS2 analytic column (Phenomenex). Fluorescence was detected with a Waters 474 fluorescence detector (Waters Corporation) with excitation at 385 nm and emission at 515 nm. l-Homocysteine and l-cysteine were used as external calibrators.25 The intra- and interassay CVs for tHcy were 5% and 7%, respectively.
Baseline water As concentrations were measured in nonfiltered water samples collected in polyethylene scintillation vials. Samples were acidified to 1% using high-purity Optima HCl (Fisher Scientific) for 48 h, diluted 1:10, and analyzed with high-resolution ICP-MS.26 No samples were below the LOD of <0.2μg/L, and the intra- and interassay CVs were 2% and 2.6%, respectively.
Study Sample
A total of 622 participants were recruited. Six participants discontinued the trial owing to adverse events [PBO group (1 participant): abdominal cramps; 400FA group (1 participant): hypertension; 800FA group (3 participants): abdominal cramps, severe vertigo, bilateral hydronephrosis; creatine group (1 participant): severe vertigo] and 3 participants discontinued owing to pregnancy (400FA, creatine, and creatine+400FA groups). Two participants dropped out of the trial (PBO and 400FA groups), and 1 participant was excluded from this analysis owing to a missing blood sample. A total of 609 participants completed the study and were used for the present analyses (PBO: n=102, 400FA: n=153, 800FA: n=150, creatine: n=101, creatine+400FA: n=103).
Statistical Methods
Primary and secondary bAs methylation indices were calculated as bMMAs/bInAs concentrations and bDMAs/bMMAs concentrations, respectively. Baseline summary statistics were calculated for participant sociodemographic data, baseline As methylation indices and metabolites, and nutritional parameters using GMs and geometric standard deviations (GSDs) for continuous variables and frequencies for categorical variables. GMs are commonly used in analyses involving environmental chemicals and toxicants, such as metals or metalloids that are measured in blood and urine, because they provide a more accurate measure of the central tendency of lognormally distributed variables, such as bAs metabolites, as compared with arithmetic means.27 Differences between treatment groups at baseline were detected using the Kruskall–Wallis test for continuous variables and chi-square test for categorical variables. Because the distributions of bAs methylation indices and metabolites were right skewed, these variables were natural log-transformed to meet model assumptions, reduce the impact of extreme values, and improve model fitting.
Linear models with repeated measures were used to examine treatment effects on bAs methylation indices and metabolite concentrations in the first 12-wk phase and the potential reversal of treatment effects on As metabolites following FA cessation during phase 2 (weeks 12–24). Model parameters were estimated using generalized estimating equations with an exchangeable correlation structure to account for within-person correlation in repeated measures over time; robust standard errors were used for statistical inference. The models for phase 1 of the RCT included the variables of baseline total bAs, treatment group for the five arms, time (baseline, week 1, and week 12), and group-by-time interactions. Coefficients of interaction terms indicate differences between each treatment group and the PBO group in the mean within-person change from baseline. A Wald test on the coefficients that define group-by-time interactions was used to detect differences in the mean within-person change between specific treatment groups. In addition, to explore potential sex-differences, models examining the treatment effects on bAs methylation indices and metabolite concentrations during phase 1 were further stratified by sex. Effect modification was also assessed using models including group-by-time-by-sex interaction terms. The models for phase 2 of the RCT (effect of FA cessation) included variables for baseline total bAs, time (baseline, week 12, and week 24), treatment groups (two FA doses over 24 wk and two FA doses switched to PBO in the second 12 wk), and interactions of group-by-time. Percentage change in GM over a specific time interval for each treatment group was derived from the estimated model parameters with a 95% CI. All analyses were performed using R (version 3.6.3; R Development Core Team).
Results
By design, approximately half of participants were male (50.2%, n=306). The age (GM±GSD) of participants ranged from 24 to 55 y (37.5±1.2 y). The prevalence of folate deficiency (<9 nmol/L in plasma) was 19.7% (n=120) in this study population. In addition, 38.8% (n=236) of participants had hyperhomocysteinemia (plasma homocysteine ≥13μmol/L) and 24% (n=146) were B12 deficient (<151 pmol/L in plasma). GM baseline well water As concentrations across treatment groups ranged from 119 to 128μg/L, that is, they were ∼12–15 times the WHO-recommended limit of 10μg/L3 but were not statistically different across treatment groups (p=0.95). There were no significant differences in participant characteristics between treatment groups at baseline (p>0.05) (Table 1).
Table 1 Folic Acid and Creatine Trial study participant characteristics [GM±GSD or n (%)] at baseline by treatment group.
Characteristic PBO (n=102) 400FA (n=153) 800FA (n=150) Creatine (n=101) Creatine+400FA (n=103) p-Valuea
Age (y) 37.3±1.2 38.2±1.2 37.3±1.3 37.4±1.3 37.2±1.2 0.85
Sex (n) >0.99
Female 51 (50.0) 76 (49.7) 75 (50.0) 50 (49.5) 51 (49.5)
Male 51 (50.0) 77 (50.3) 75 (50.0) 51 (50.5) 52 (50.5)
Smoking status (n)b 0.69
Never 77 (75.5) 115 (75.2) 106 (70.7) 72 (71.3) 72 (69.9)
Ever 25 (24.5) 36 (23.5) 44 (29.3) 29 (28.7) 31 (30.1)
Betel nut use (n)b 0.76
Never 73 (71.6) 115 (75.2) 113 (75.3) 76 (75.2) 82 (79.6)
Ever 29 (28.4) 36 (23.5) 37 (24.7) 25 (24.8) 21 (20.4)
Land ownership (n)c 0.85
Does not own land 55 (53.9) 76 (49.7) 78 (52.0) 53 (52.5) 58 (56.3)
Owns land 47 (46.1) 77 (50.3) 72 (48.0) 48 (47.5) 44 (42.7)
Body mass index (kg/m2)d 20.2±1.2 19.4±1.1 19.7±1.1 19.8±1.2 19.3±1.1 0.30
Water arsenic (μg/L) 123±2 119±2 119±2 125±2 128±2 0.95
Urinary arsenic (μg/L) 86.2±2.4 98.6±2.5 97.2±2.4 110.0±2.5 117.0±2.3 0.38
Blood arsenice 8.03±1.80 8.26±1.88 8.37±1.72 8.24±1.85 8.76±1.68 0.64
PMIe 1.61±1.27 1.65±1.20 1.63±1.20 1.63±1.20 1.70±1.19 0.56
SMIe 0.62±1.29 0.63±1.29 0.65±1.31 0.63±1.29 0.66±1.30 0.29
InAs (μg/L)e 2.20±1.75 2.22±1.79 2.25±1.65 2.24±1.85 2.27±1.65 0.86
MMAs (μg/L)e 3.55±1.89 3.67±1.97 3.67±1.82 3.64±1.93 3.85±1.73 0.79
DMAs (μg/L)e 2.19±1.83 2.30±1.93 2.38±1.75 2.29±1.83 2.56±1.73 0.37
%InAse 27.5±1.2 26.9±1.2 26.9±1.1 27.1±1.1 25.9±1.1 0.05
%MMAse 44.2±1.1 44.3±1.1 43.8±1.1 44.2±1.1 44.0±1.1 0.75
%DMAse 27.2±1.2 27.8±1.2 28.4±1.2 27.8±1.2 29.2±1.2 0.08
Urinary creatinine (mg/dL) 38.6±2.1 43.7±2.2 40.9±2.2 46.2±2.0 47.5±2.2 0.35
Red blood cell folate (nmol/L)f 451±1 448±2 468±1 NA NA 0.63
Plasma folate (nmol/L) 13.6±1.8 13.5±1.9 14.5±1.8 14.4±1.6 13.3±1.7 0.66
Folate deficient (<9 nmol/L in plasma) (n) 22 (21.6) 36 (23.5) 27 (18.0) 14 (13.9) 21 (20.4) 0.39
Plasma homocysteine (μmol/L) 12.2±1.6 11.9±1.6 11.9±1.6 11.4±1.5 11.8±1.5 0.90
Hyperhomocysteinemia (≥13μmol/L) (n) 43 (42.2) 56 (36.6) 59 (39.3) 39 (38.6) 39 (37.9) 0.93
Plasma vitamin B12 (pmol/L) 204±2 216±2 214±2 224±2 210±2 0.82
Vitamin B12 deficient (<151 pmol/L) (n) 25 (24.5) 37 (24.2) 39 (26.0) 20 (19.8) 25 (24.3) 0.86
Note: %, percentage; 400FA, 400μg FA/d; 800FA, 800μg FA/d; creatine, 3g creatine/d; creatine+400FA, 3g creatine and 400μg FA/d; DMAs, dimethyl-arsenical species; FA, folic acid; GM, geometric mean; GSD, geometric standard deviation; InAs, inorganic arsenic; MMAs, monomethyl-arsenical species; NA, not available/not measured for creatine and creatine+400 FA groups; PBO, placebo; PMI, primary methylation index; SMI, secondary methylation index.
a Kruskal–Wallis test for continuous variables and chi-square test for categorical variables.
b 400FA: n=151.
c Creatine+400FA: n=102.
d PBO: n=101; 400FA: n=150; 800FA: n=147; creatine: n=98; creatine+400FA: n=102.
e 400FA: n=152; creatine: n=100.
f PBO: n=100; 400FA: n=148; 800FA: n=148.
Compliance with supplementation in FACT has been previously reported by Peters et al.17 Pill counts were used to calculate the percentage of study pills taken by each participant throughout the trial. Median compliance was 99.5%, and compliance was similar across treatment groups and study phases (range: 79.1%–100%; IQR: 98.3%–100.0%).17 Compliance, as assessed by plasma folate, was also high. Plasma folate concentrations at week 12 in the 400FA, 800FA, and creatine+400FA groups were significantly greater than baseline concentrations (paired Wilcoxon test p<0.001) and significantly greater than week 12 concentrations in the PBO group (Wilcoxon test p<0.001). The prevalence of folate deficiency did not considerably change in the PBO and creatine arms, whereas all groups receiving FA had a decrease in the prevalence of folate deficiency to <1.5% at week 12 (from 17.9%–23.5% at baseline). Furthermore, participants who remained on FA for the second phase of the trial sustained their plasma folate concentrations, whereas plasma folate returned to near baseline levels among participants who switched to PBO for the second phase.17
Treatment Effects on bAs Methylation Indices
GMs of PMI and SMI at baseline, week 1, and week 12 are provided in Table S1; individual-level data are provided in Excel File 1. Changes in bAs indices between baseline and week 12, ranked by change in PMI and SMI, are shown in Figure S1. The percentage change in the GM of bAs methylation indices between baseline and week 12 are shown in Table 2. As early as week 1, the increase in SMI, an indicator of the extent to which the second methylation reaction is completed, was higher in the creatine+400FA treatment group than the PBO group: within-person change increased by 6.3% (95% CI: 3.4, 9.3) compared with 2.1% (95% CI: −0.8, 5.2) for PBO (p=0.05).
Table 2 GM±GSD at baseline and percentage change in GM (95% CI) at interval (week) of blood arsenic methylation indices and metabolite concentrations from baseline by treatment group.
Arsenic methylation measures by timepoint (week) PBO 400FA 800FA Creatine Creatine+400FA
GM±GSD or % change from baseline (95% CI) GM±GSD or % change from baseline (95% CI) p-Valuea GM±GSD or % change from baseline (95% CI) p-Valuea GM±GSD or % change from baseline (95% CI) p-Valuea GM±GSD or % change from baseline (95% CI) p-Valuea
Arsenic methylation indices
PMI
Baseline 1.61±1.27 1.65±1.20 — 1.63±1.20 — 1.63±1.20 — 1.70±1.19 —
1 −17.0 (−21.6, −12.0) −18.1 (−21.1, −15.0) 0.69 −15.3 (−18.3, −12.1) 0.57 −18.3 (−22.2, −14.2) 0.66 −20.4 (−23.5, −17.2) 0.23
12 −5.5 (−9.8, −1.0) −12.6 (−15.7, −9.4) 0.01 −10.8 (−13.9, −7.7) 0.05 −9.8 (−14.9, −4.5) 0.22 −12.7 (−16.1, −9.2) 0.01
SMI
Baseline 0.62±1.29 0.63±1.29 — 0.65±1.31 — 0.63±1.29 — 0.66±1.30 —
1 2.1 (−0.8, 5.2) 5.4 (2.4, 8.5) 0.12 3.0 (0.2, 6.3) 0.69 3.9 (1.4, 6.5) 0.37 6.3 (3.4, 9.3) 0.05
12 2.2 (−1.5, 5.9) 26.3 (22.1, 30.5) <0.001 23.8 (19.7, 27.9) <0.001 8.9 (4.7, 13.2) 0.02 17.9 (14.3, 21.6) <0.001
Arsenic metabolites (μg/L)
InAs
Baseline 2.20±1.75 2.22±1.79 — 2.25±1.65 — 2.24±1.85 — 2.27±1.65 —
1 13.5 (9.7, 17.5) 13.7 (10.9, 16.5) 0.96 11.6 (8.9, 14.4) 0.42 14.3 (10.5, 18.1) 0.80 15.9 (12.7, 19.1) 0.36
12 4.0 (0.9, 7.2) 2.9 (0.5, 5.5) 0.61 2.1 (0.3, 4.6) 0.36 4.8 (1.5, 8.3) 0.73 5.4 (2.6, 8.3) 0.51
MMAs
Baseline 3.55±1.89 3.67±1.97 — 3.67±1.82 — 3.64±1.93 — 3.85±1.73 —
1 −6.1 (−8.4, −3.7) −7.2 (−8.7, −5.7) 0.42 −5.9 (−7.4, −4.4) 0.92 −7.2 (8.8, 5.5) 0.45 −8.2 (−9.6, −6.8) 0.11
12 −2.0 (−4.0, 0.0) −10.3 (−11.9, −8.8) <0.001 −9.5 (−11.1, −8.0) <0.001 −5.9 (−8.6, −3.0) 0.03 −8.4 (−10.0, −6.9) <0.001
DMAs
Baseline 2.19±1.83 2.30±1.93 — 2.38±1.75 — 2.29±1.83 — 2.56±1.73 —
1 −4.4 (−6.5, −2.2) −2.4 (−4.4, −0.4) 0.20 −3.5 (−5.7, −1.2) 0.57 −3.9 (−5.8, −1.9) 0.74 −2.8 (−4.8, −0.8) 0.30
12 −0.1 (−2.8, 2.6) 12.8 (10.5, 15.2) <0.001 11.3 (8.9, 13.8) <0.001 2.2 (0.5, 5.0) 0.23 7.5 (5.2, 9.7) <0.001
Note: Percentage change was estimated using linear models. GMs for each time point are included in Table S1; individual-level data are included in Excel File 1. —, not applicable; %, percentage; 400FA, 400μg FA/d; 800FA, 800μg FA/d; CI, confidence interval; creatine, 3g creatine/d; creatine+400FA, 3g creatine and 400μg FA/d; DMAs, dimethyl-arsenical species; FA, folic acid; GM, geometric mean; GSD, geometric standard deviation; InAs, inorganic arsenic; MMAs, monomethyl-arsenical species; PBO, placebo; PMI, primary methylation index; SMI, secondary methylation index.
a p-Values were derived from Wald test on coefficients of group-by-time interaction terms in the linear models with repeated measures for differences in changes from baseline between each treatment and PBO group.
At week 1, the increase in SMI from baseline was greater among participants in the 400FA, 800FA, and creatine treatment arms, with respective within-person changes of 5.4% (95% CI: 2.4, 8.5), 3.0% (95% CI: 0.2, 6.3), and 3.9% (95% CI: 1.4, 6.5) compared with PBO, but it did not reach statistical significance (p=0.12, 0.69, and 0.37, respectively). For PMI, in the 400FA, creatine, and creatine+400FA treatment groups, the −18.1% (95% CI: −21.1, −15.0), −18.3% (95% CI: −22.2, −14.2), and −20.4% (95% CI: −23.5, −17.2) decreases at week 1, respectively, were greater than the −17.0% (95% CI: −21.6, −12.0) decrease in PBO but did not reach statistical significance (p=0.69, 0.66, and 0.23, respectively). In the 800FA group, the decrease in PMI of −15.3% (95% CI: −18.3, −12.1) was less than that of PBO, although it did not reach statistical significance (p=0.57).
At week 12, the mean within-person decreases from baseline in PMI among participants in the 400FA, 800FA, and creatine+FA treatment groups were −12.6% (95% CI: −15.7, −9.4), −10.8% (95% CI: −13.9, −7.7), and −12.7% (95% CI: −16.1, −9.2), respectively; these were all significantly greater than the −5.5% (95% CI: −9.8, −1.0) decrease in the PBO group. Among participants in the creatine group, the mean within-person decrease in PMI of −9.8% (95% CI: −14.9, −4.5) was larger than the PBO group, but this did not reach statistical significance (p=0.22). In addition, among all treatment groups, the mean percentage changes in SMI at week 12 were significantly greater (p<0.05) than the 2.2% (95% CI: −1.5, 5.9) increase in the PBO group: 400FA: 26.3% (95% CI: 22.1, 30.5), 800FA: 23.8% (95% CI: 19.7, 27.9), creatine: 8.9% (95% CI: 4.7, 13.2), and creatine+400FA: 17.9% (95% CI: 14.3, 21.6).
Treatment Effects on bAs Metabolite Concentrations
GMs of bInAs, bMMAs, and bDMAs concentrations at baseline, week 1, and week 12 are provided in Table S1; individual-level data are provided in Excel File 1. The changes in participants’ bAs metabolite concentrations between baseline and week 12 by treatment group are shown in Figure 3. Decreases in bAs metabolite concentrations between baseline and week 12 were observed across treatment groups, likely due to water filter use.22 However, Figure 3 illustrates that all treatment groups who received FA showed an overall trend toward a decrease in InAs and MMAs concentrations between baseline and week 12 (i.e., a greater proportion of participants falling below the y-axis, representing a decrease in concentrations).
Figure 3. Change in participants’ blood arsenic metabolite concentrations between baseline and 12 wk by treatment group. The x-axes are ranked by participants with the largest decreases to the largest increases in blood arsenic metabolite concentrations. The PBO, 400μg of FA/d (400FA), 800μg of FA/d (800FA), 3g creatine (creatine), and 3 g creatine+400μg of FA/d (creatine+400FA) are shown in brown, green, blue, orange, and purple, respectively. Individual-level blood arsenic metabolite concentration data are included in Excel File 1. Note: DMAs, dimethyl-arsenical species; FA, folic acid; InAs, inorganic arsenic; MMAs, monomethyl-arsenical species; PBO, placebo.
Figure 3 is a set of graphs plotting the change in the concentrations (micrograms per liter) of inorganic arsenic (panel A), monomethyl-arsenicals (panel B), and dimethyl-arsenicals (panel C) across treatment groups. The treatment groups are shown in the columns labeled: Placebo, 400 folic acid, 800 folic acid, creatine, and creatine plus 400 folic acid. Each plot shows the change in arsenic species concentrations between baseline and 12 weeks, ranging from negative 15 to 5 in increments of 5 (y-axis) across the rank of change in concentration (x-axis).
The percentage change in the GM of bAs metabolite concentrations over time in phase 1 are shown in Table 2. At week 1, compared with baseline, bMMAs decreased across all treatment groups, although these changes were not significantly different than that of the PBO group (p>0.05). At week 12, the decreases in bMMAs were significantly greater (p<0.05) for all treatment groups receiving FA as compared with PBO: the mean within-person decrease was −2.0% (95% CI: −4.0, 0.0) in the PBO group, vs. −10.3% (95% CI: −11.9, −8.8), −9.5% (95% CI: −11.1, −8.0), and −8.4% (95% CI: −10.0, −6.9) for the 400FA, 800FA, and creatine+400FA groups, respectively. The mean within-person decrease in bMMAs in the 800FA group (−9.5%) was not significantly different from the decrease in the 400FA group (−10.3%) (Wald test p>0.05), suggesting that there was not a dose–response effect of FA on bMMAs. The mean within-person change in bDMAs concentrations from baseline to week 12 was 12.8% (95% CI: 10.5, 15.2) for the 400FA group and 11.3% (95% CI: 8.9, 13.8) for the 800FA group vs. −0.1% (95% CI: −2.8, −2.6) for the PBO group (p<0.001). The changes in bInAs concentrations from baseline to weeks 1 and 12 in the FA and creatine treatment groups were not significantly different from changes in the PBO group (p>0.05).
Overall, models stratified by sex were consistent with the unstratified models. Results from sex-stratified linear models are presented in Tables S2 and S3; GMs of bAs methylation indices and metabolite concentrations are provided in Tables S4 and S5. There appear to be sex-specific changes in bAs indices and concentrations that were significantly different (p<0.05) in the creatine group compared with PBO among male participants that were not significant (p>0.05) among female participants. This was reflected in decreases in bMMAs and increases in SMI in the creatine arm that were significantly different from PBO in male participants [creatine vs. PBO: bMMAs: −6.6% (95% CI: −9.5, −3.7) vs. −2.7% (95% CI: −4.8, −0.5), p=0.03; SMI: 12.3% (95% CI: 6.1, 18.8) vs. 3.1% (95% CI: −0.9, 7.3), p=0.02], but not female participants [bMMAs: −5.0% (95% CI: −9.6, −0.2) vs. −1.4% (95% CI: −4.7, 2.1), p=0.31; SMI: 5.4% (95% CI: 0.2, 10.9) vs. 1.3% (95% CI: −4.7, 7.6), p=0.21]. However, effect modification of creatine treatment by sex was not statistically significant as measured by the inclusion of a group-by-time-by-sex interaction term in unstratified models (p>0.05; Table S6).
Treatment Effects during the Second Phase
To evaluate the potential reversal of treatment effects on As metabolites after cessation of FA, we evaluated the percentage changes in bAs indices and metabolites from baseline to week 12, baseline to week 24, and from week 12 to week 24 (Table 3). The percentage increase in SMI between baseline and week 24 was significantly greater among the groups who continued FA treatment compared with groups who discontinued treatment [400FA/FA: 29.2% (95% CI: 23.1, 35.5), 400FA/PBO: 14.1% (95% CI: 9.13, 19.3), 800FA/FA: 23.7% (95% CI: 18.9, 28.6), 800FA/PBO: 13.4% (95% CI: 8.2, 18.9)] (p<0.05). In the 800FA/FA group, percentage changes in bInAs, bMMAs, bDMAs, and PMI over 24 wk were similar to those in the 800FA/PBO group (p>0.05). Data from the second phase, comparing week 12 to week 24, showed evidence of the reversal of treatment effects on As metabolites following FA cessation, with significant decreases in SMI [−9.0% (95% CI: −14.8, −3.5)] and DMAs [−5.9% (95% CI: −10.2, −1.8)] in the 800FA/PBO group and similar trends in the 400FA/PBO group that were not statistically significant (p>0.05). For those who continued 800μg FA supplementation, PMI and MMAs concentrations decreased between weeks 12 and 24 [PMI: −7.2% (95% CI: −14.3, −0.5); MMAs: −3.1% (95% CI: −6.2, −0.1)], whereas changes in the 800FA/PBO group were not significantly different than 0 [PMI: −1.8% (95% CI: −10.0, 5.7), MMAs: 2.8% (95% CI: −0.3, 5.8)]. Similar trends were observed in the 400FA/FA group but were not statistically significant (p>0.05).
Table 3 Percentage change (95% CI) in blood arsenic methylation indices and metabolite concentrations between baseline and week 12, baseline and week 24, and week 12 and week 24 by folic acid supplementation groups.
Arsenic methylation measures by treatment group 400FAa 800FAb
Baseline to week 12 Baseline to week 24 Week 12 to week 24 Baseline to week 12 Baseline to week 24 Week 12 to week 24
Arsenic methylation indices
PMI
FA/FA −11.2 (−15.3, −6.8) −11.1 (−14.8, −7.2) −5.4 (−13.3, 2.0) −8.1 (−12.1, −4.0) −9.5 (−13.3, −5.6) −7.2 (−14.3, −0.5)
FA/PBO −13.3 (−18.0, −8.4) −6.4 (−10.6, −2.0) 2.3 (−4.9, 9.0) −12.4 (−17.1, −7.5) −9.2 (−13.0, −5.3) −1.8 (−10.0, 5.7)
SMI
FA/FA 26.9 (21.1, 33.1) 29.2 (23.1, 35.5) 3.7 (−2.1, 9.1) 20.9 (16.0, 26.0) 23.7 (19.0, 28.6) 4.2 (−1.0, 9.1)
FA/PBO 25.2 (19.6, 31.1) 14.1 (9.1, 19.3) −7.6 (−14.5, −1.1) 26.1 (19.8, 32.7) 13.4 (8.2, 18.9) −9.0 (−14.8, −3.5)
Arsenic metabolites
InAs
FA/FA 1.6 (−1.5, −4.8) 0.9 (−2.2, 4.0) 3.0 (−1.5, 7.2) 0.7 (−2.3, 3.7) 0.9 (−2.0, 3.9) 3.9 (−0.2, 7.2)
FA/PBO 3.6 (−0.1, 7.6) 1.0 (−2.1, 4.2) 1.1 (−3.6, 5.5) 2.4 (−1.4, 6.4) 3.4 (0.2, 6.8) 4.5 (−0.7, 9.5)
MMAs
FA/FA −10.0 (−12.2, −7.8) −10.4 (−12.3, −8.5) −2.4 (−5.9, 1.1) −7.8 (−9.6, −5.9) −8.8 (−10.5, −7.1) −3.1 (−6.2, −0.1)
FA/PBO −10.3 (−12.6, 8.0) −5.5 (−7.6, −3.4) 3.3 (−0.1, 6.6) −10.7 (−13.1, −8.2) −6.3 (−8.3, −4.3) 2.82 (−0.3, 5.8)
DMAs
FA/FA 14.1 (10.8, 17.5) 15.6 (11.8, 19.6) 1.3 (−2.5, 5.0) 11.4 (8.2, 14.6) 6.1 (2.7, 9.6) 1.2 (−2.4, 4.6)
FA/PBO 12.2 (9.0, 15.5) 7.8 (4.7, 10.9) −4.1 (−8.4, 0.1) 12.4 (−8.8, 16.1) 6.1 (2.7, 9.6) −5.9 (−10.2, −1.8)
Note: Percentage change was estimated using linear models. GMs for each time point are included in Table S7; individual-level data are included in Excel File 1. 400FA, 400μg FA/d; 800FA, 800μg FA/d; CI, confidence interval; DMAs, dimethyl-arsenical species; FA, folic acid; FA/FA, continued folic acid treatment; FA/PBO, discontinued folic acid treatment; InAs, inorganic arsenic; MMAs, monomethyl-arsenical species; PBO, placebo; PMI, primary methylation index; SMI, secondary methylation index.
a 400FA/FA: n=77; 400FA/PBO: n=76.
b 800FA/FA: n=76; 800FA/PBO: n=74.
In Figure 4 and Table S7, the GMs for As methylation indices and bAs metabolite concentrations are shown at baseline, week 12, and week 24 for the PBO, 400FA, and 800FA treated groups, with separate groups for those who continued and discontinued FA supplementation; individual-level data are provided in Excel File 1. Across treatment groups, there was a trend toward an increase in bAs metabolite concentrations during the second phase, a period with reduced compliance with water filter use, which has previously been reported by Sanchez et al.22 The plots illustrate similar declines in PMI and increases in SMI for the 400FA and 800FA groups at week 12; these were maintained at week 24 in both FA/FA groups, whereas in the FA/PBO groups, week 24 values tended to revert toward baseline values.
Figure 4. Geometric mean of blood arsenic indices and metabolite concentrations by treatment group from baseline to week 24. The y-axes are the (A) primary methylation index (PMI), (B) secondary methylation index (SMI), (C) inorganic arsenic (InAs), (D) monomethyl-arsenical species (MMAs), and (E) dimethyl-arsenical species (DMAs) levels and the x-axes are weeks. Participants that received PBO, continuous 400μg of FA/d (400FA/FA), continuous 800μg of FA/d (800FA/FA), cessation of 400μg FA/d at 12 wk (400FA/PBO), or cessation of 800μg of FA/d at 12 wk (800FA/PBO) are shown in brown solid lines (triangle points), green solid lines (circle points), blue solid lines (square points), light green dashed lines (circle points), and light blue dotted lines (square points), respectively. The bAs PMI and SMI were calculated as bMMAs/bInAs and bDMAs/bMMAs concentrations, respectively. Percentage change was estimated using linear models. Individual-level data are included in Excel File 1. Note: bDMAs, blood dimethyl-arsenical species; bInAs, blood inorganic arsenic metabolites; bMMAs, blood monomethyl-arsenical species; FA, folic acid; PBO, placebo.
Figures 4A to 4E are line graphs, plotting primary methylation index, ranging from 1.3 to 1.6 in increments of 0.1; secondary methylation index, ranging from 0.6 to 0.9 in increments of 0.1; inorganic arsenic concentrations (micrograms per liter), ranging from 1.0 to 3.0 in increments of 0.5; monomethyl-arsenical species concentrations (micrograms per liter), ranging from 2.0 to 4.0 in increments of 0.5; and dimethyl-arsenical species concentrations (micro grams per liter), ranging from 1.5 to 3.0 in increments of 0.5 (y-axis) across week, ranging from 0 to 24.
Discussion
Reduced As methylation capacity has been linked to increased risk for a number of As-related health outcomes, including cancers of the bladder,28,29 lung,12,30 and skin,31,32 and noncancer outcomes, such as arsenical skin lesions33,34 and atherosclerosis.35 FACT was designed to investigate the efficacy of FA, creatine, and a combination of FA and creatine supplementation on increasing As methylation capacity among a population of mixed folate-deficient and -sufficient individuals in Bangladesh, a country that does not currently have food folate fortification. Our outcomes of interest included changes in bAs methylation indices, PMI and SMI, and bAs metabolite concentrations.
We previously reported findings on total bAs and urinary As metabolites.8,17,18 In short, groups receiving FA supplementation had significantly greater decreases in %MMAs and increases %DMAs in urine compared with PBO (p<0.05).18 In our earlier RCT of 400μg FA supplementation in only folate-deficient individuals,8 the group receiving 400μg FA had a significantly greater decrease in total bAs compared with PBO (p<0.05). In contrast, in the present study of a mixed folate-deficient/replete population, 400μg FA was not effective in lowering total bAs to a significantly greater extent than PBO over the course of the trial, indicating that a higher dose of FA (800μg FA) or a longer duration of supplementation may be required to lower total bAs.17
Here we report new data on As metabolite concentrations in blood, where the distribution of As metabolites differs considerably from that previously reported for urine.18 Although the distribution of As metabolites in urine at baseline ranged from 13.2% to 14.7% InAs, 12.5% to 13.4% MMAs, and 72.0% to 74.4% DMAs across treatment groups,18 the GMs in blood ranged from 25.9% to 27.5% InAs, 44.0% to 44.3% MMAs, and 27.2% to 29.2% DMAs (Table 1). These differences between blood and urine are similar to those observed in our earlier study of folate-deficient participants8 and are related to the more rapid renal excretion of DMAs in urine. The measurement of As, As indices, and As metabolite concentrations in blood has the added advantage that the overall concentration of blood is relatively constant36 as compared with urine, which requires adjustment for hydration status with urinary creatinine or specific gravity37–40 and expression of As metabolites as percentages rather than concentrations. Furthermore, intuitively, As metabolite concentrations in blood should more closely reflect As metabolite exposures to tissues as compared with urine where concentrations of As metabolites are much higher. The bAs methylation indices, PMI and SMI, have the advantage that they are less impacted by As exposure levels and water filter compliance, as observed in the present study, than are bAs metabolite concentrations.
In the first phase of FACT, we observed significant increases in As methylation capacity as reflected by greater decreases in bMMAs and PMI and increases in bDMAs and SMI in all treatment groups receiving FA relative to PBO. In addition, treatment effects on increasing SMI were seen as early as the first week of intervention for the creatine+400FA treatment group. The nonsignificant increase in SMI from baseline to week 1 in the PBO group [2.1% (95% CI: −0.8, 5.2)] was likely due to water filter use,22 given that water As concentrations have been negatively associated with SMI in urine33 and that there is experimental evidence showing that arsenic methylation capacity is saturable.41 Although no significant changes in bInAs concentrations were observed for any treatment groups compared with PBO over time, this was likely due to its methylation to MMAs and continued InAs exposure owing to poor compliance with As-removal filters.22 In exploratory sex-stratified analyses, FA supplementation results were similar among male and female participants, whereas creatine supplementation alone resulted in significantly greater decreases in bMMAs and increases in SMI compared with PBO only among male participants. Although the changes in bMMAs and SMI with creatine treatment compared with PBO were not significant in female participants, they were in the same direction as those in male participants. Thus, given that our study was not powered to detect effect modification, this could be an issue of small sample size, or it could possibly be related to differences in creatine metabolism due to higher muscle mass among male participants. The finding that FA supplementation was effective in increasing As methylation capacity in this population that was predominantly folate sufficient (plasma folate was ≥9.0 nmol/L in 80.3% of the study participants) has important implications. For example, if FA supplementation was only effective in folate-deficient individuals, then targeted approaches would be more appropriate than widespread food FA fortification programs. The latter can influence entire populations and overcome problems associated with FA supplementation, such as poor long-term compliance and difficulties in reaching vulnerable subsets of the population.
In the second phase of FACT, the data showed evidence of the reversal of treatment effects on As methylation following FA cessation from weeks 12 to 24, with significant decreases in SMI and bDMAs in those who switched from 800FA to PBO, whereas for those who remained on 800FA, PMI and bMMAs concentrations continued to decline, showing added benefit to a longer intervention. In addition, as compared with baseline, there was a significantly greater increase in SMI at week 24 among participants who continued supplementation as compared with those who switched to PBO. Data on bDMAs concentrations also suggest the reversal of treatment effects on As methylation after FA cessation: the 7.77% increase in bDMAs concentrations between baseline and week 24 among participants who discontinued 400FA supplementation was roughly half that of those who continued FA supplementation (15.6%). These data indicate that sustainable improvement in As methylation capacity may require long-term interventions, such as food fortification programs, which are generally low-cost, effective approaches that can achieve broad coverage.
The mechanisms of As toxicity are multifactorial; for example, chromosomal instability and altered methylation of DNA and histones are two mechanisms involved in As toxicity, as reviewed in Abuawad et al.42 These mechanisms can be exacerbated by impairments in OCM and may also be improved by folate supplementation. There is some evidence to suggest that folate and/or other OCM-related micronutrients may also be beneficial in reducing As-related toxicity through these mechanisms, in addition to our observed effects of increasing As methylation and lowering bAs, although additional data are needed.43,44
The creatine treatment arms were designed to test our a priori hypothesis that creatine supplementation would reduce the high demand for SAM needed to support endogenous creatine biosynthesis, thereby sparing SAM for the methylation of As. This hypothesis is supported by positive associations between urinary creatinine levels and As methylation capacity in observational studies.17,19,45–47 In FACT, 3g creatine supplementation decreased creatine biosynthesis as indicated by significant decreases in plasma GAA, the precursor to creatine.48 In our previous FACT report on urinary As metabolites, the group receiving creatine supplementation had a significantly greater decrease in %MMAs in urine compared with PBO following treatment, but no significant creatine treatment effects were observed for %InAs or %DMAs.18 In the present analyses, the group receiving creatine supplementation alone had a significantly greater decrease in bMMAs concentrations and increase in SMI from baseline to week 12 compared with PBO. However, the creatine treatment effect was small compared with that of FA supplementation, and there were no significant differences observed in changes of bInAs, bDMAs, or PMI. Potential explanations for limited creatine treatment effects include long-range allosteric regulation of OCM that regulate intracellular SAM concentrations, for example, by changes in activity of glycine-N-methyltransferase (GNMT), a major regulator of SAM concentrations, and in flux through the transsulfuration pathway, which is stimulated by SAM.49–51 The strong cross-sectional associations previously reported between urinary creatinine and As methylation patterns may partially be related to dietary creatine intake, but they could also potentially be influenced by dietary protein/methionine content associated with meat intake or other factors, such as renal function, which could differentially influence excretion/reabsorption of As metabolites in urine.
A limitation of this study includes decreased compliance with water filter use over time, as reported by Sanchez et al. in a study of total urine As and water As levels in FACT,22 which complicates interpretation of changes in bAs metabolite concentrations given that there was a strong decline in bAs for all groups at week 1 owing to initial compliance that reverted afterward. However, owing to the randomized study design, there is no evidence to suggest that compliance differed between treatment groups, which is supported by our data on total urinary As concentrations. Thus, this is not expected to change the conclusions of the study. Furthermore, the use of the As methylation indices, PMI and SMI, circumvents this issue that more directly affects bAs metabolite concentrations. Strengths of this study include the double-blind, randomized, PBO-controlled study design, including evaluation of the potential reversal of treatment effects on As metabolites following cessation of FA treatment and the high participant compliance with treatment. In addition, the use of blood, rather than urinary, measures of As metabolites may better reflect concentrations in tissues and be less affected by hydration levels. That a large proportion of the study population were folate replete increases generalizability to other As-exposed populations.
Conclusion
Our results indicate that FA supplementation increased As methylation capacity, as reflected by larger decreases in PMI and MMAs and increases in SMI and DMAs in blood within FA-treated groups compared with PBO. These findings are consistent with our previous report of the effects of FA supplementation on urinary As methylation profiles in this trial.18 In recent years, the evidence supporting the critical role of nutritional factors in As methylation and corresponding toxicity has grown substantially.42,52 Although As-remediation programs aimed at reducing exposure to As through contaminated drinking water should always remain the top priority, additional approaches to decrease the chronic health effects of As exposure through increasing As methylation capacity and lowering bAs concentrations are needed.
Food fortification programs are one option. Approximately 85 countries mandate grain fortification with FA, but 28 European countries and most Asian countries do not or have only voluntary programs.53 A major goal of FA fortification is to reduce neural tube defects, a lifelong debilitating condition due to failure of the neural folds to fuse during embryonic development. Fortification circumvents issues associated with FA supplementation or other dietary public health interventions, such as unplanned pregnancies, poor supplementation compliance, and difficulties reaching segments of the population.54 However, despite the many known benefits of FA fortification, there are knowledge gaps regarding the potential risks of excessive folate intake and/or high FA levels, particularly in populations where use of folate supplements is common. These are summarized in the proceedings of a National Institutes of Health (NIH) workshop and include the masking of B12 deficiency, metabolic interactions with B12 that may influence neurocognitive outcomes among the elderly, and potential adverse effects on cancer risk, birth outcomes, and other diseases.55 In addition, FA is a synthetic form of folate used in food fortification because it is chemically more stable than 5-mTHF, which is easily degraded by prolonged cooking. Individuals vary in their ability to fully metabolize FA to 5-mTHF. As a result, detectable levels of circulating unmetabolized FA are nearly ubiquitous in populations with FA fortification programs, such as in the United States.56 Evidence of potential unanticipated effects of unmetabolized FA is controversial and is described by the NIH workshop proceedings as inconclusive.55
Many parts of the world affected by high levels of As contamination of drinking water, such as Bangladesh, do not currently have voluntary, let alone mandatory, FA fortification programs, and folate deficiency is common.57 Carefully executed FA fortification programs or other public health interventions are needed to eradicate folate deficiency and its associated adverse health outcomes and could also facilitate a partial reduction in the extensive public health burden of As exposure.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank all of the Folic Acid and Creatine Trial participants and staff who made this work possible. This study was supported by U.S. National Institutes of Health grants, including R01 CA133595, P42 ES010349, R01ES030945, P30 ES0090089, R25 GM062454, and F31 ES032321.
This trial was registered at https://clinicaltrials.gov as NCT01050556.
==== Refs
References
1. Podgorski J, Berg M. 2020. Global threat of arsenic in groundwater. Science 368 (6493 ):845–850, PMID: , 10.1126/science.aba1510.32439786
2. Ravenscroft P, Brammer H, Richards K. 2009. Arsenic Pollution: A Global Synthesis. Chichester, UK: Wiley-Blackwell.
3. WHO (World Health Organization). 2012. Arsenic. WHO fact sheets, http://www.who.int/mediacentre/factsheets/fs372/en/ [accessed 20 March 2023].
4. BBS (Bangladesh Bureau of Statistics). 2019. Progotir Pathey, Bangladesh: Multiple Indicator Cluster Survey 2019, Key Findings. Dhaka, Bangladesh: Government of the Peoples’ Republic of Bangladesh, Bangladeshi Bureau of Statistics, United Nations Children’s Fund. https://bbs.portal.gov.bd/sites/default/files/files/bbs.portal.gov.bd/page/b343a8b4_956b_45ca_872f_4cf9b2f1a6e0/37817b8e25d0d6c1f442e294921ff85e.pdf [accessed 15 February 2022].
5. Naujokas MF, Anderson B, Ahsan H, Aposhian HV, Graziano JH, Thompson C, et al. 2013. The broad scope of health effects from chronic arsenic exposure: update on a worldwide public health problem. Environ Health Perspect 121 (3 ):295–302, PMID: , 10.1289/ehp.1205875.23458756
6. Saba S, Akash MSH, Rehman K, Saleem U, Fiayyaz F, Ahmad T. 2020. Assessment of heavy metals by ICP-OES and their impact on insulin stimulating hormone and carbohydrate metabolizing enzymes. Clin Exp Pharmacol Physiol 47 (10 ):1682–1691, PMID: , 10.1111/1440-1681.13353.32434266
7. Kazemifar AM, Shafikhani AA, Mozhdehipanah H, Khamesi S, Arami M. 2020. Evaluation of different types of arsenic methylation and its relationship with metabolic syndrome in an area chronically exposed to arsenic. Environ Anal Health Toxicol 35 (2 ):e2020006, PMID: , 10.5620/eaht.e2020006.32693558
8. Gamble MV, Liu X, Slavkovich V, Pilsner JR, Ilievski V, Factor-Litvak P, et al. 2007. Folic acid supplementation lowers blood arsenic. Am J Clin Nutr 86 (4 ):1202–1209, PMID: , 10.1093/ajcn/86.4.1202.17921403
9. Challenger F. 1945. Biological methylation. Chem Rev 36 (3 ):315–361, 10.1021/cr60115a003.
10. Moe B, Peng H, Lu X, Chen B, Chen LWL, Gabos S, et al. 2016. Comparative cytotoxicity of fourteen trivalent and pentavalent arsenic species determined using real-time cell sensing. J Environ Sci (China) 49 :113–124, PMID: , 10.1016/j.jes.2016.10.004.28007166
11. Petrick JS, Ayala-Fierro F, Cullen WR, Carter DE, Vasken Aposhian H. 2000. Monomethylarsonous acid (MMAIII) is more toxic than arsenite in Chang human hepatocytes. Toxicol Appl Pharmacol 163 (2 ):203–207, PMID: , 10.1006/taap.1999.8872.10698679
12. Steinmaus C, Yuan Y, Kalman D, Rey OA, Skibola CF, Dauphine D, et al. 2010. Individual differences in arsenic metabolism and lung cancer in a case-control study in Cordoba, Argentina. Toxicol Appl Pharmacol 247 (2 ):138–145, PMID: , 10.1016/j.taap.2010.06.006.20600216
13. Kuo CC, Moon KA, Wang SL, Silbergeld E, Navas-Acien A. 2017. The association of arsenic metabolism with cancer, cardiovascular disease, and diabetes: a systematic review of the epidemiological evidence. Environ Health Perspect 125 (8 ):087001, PMID: , 10.1289/EHP577.28796632
14. Brosnan JT, da Silva RP, Brosnan ME. 2011. The metabolic burden of creatine synthesis. Amino Acids 40 (5 ):1325–1331, PMID: , 10.1007/s00726-011-0853-y.21387089
15. Stead LM, Au KP, Jacobs RL, Brosnan ME, Brosnan JT. 2001. Methylation demand and homocysteine metabolism: effects of dietary provision of creatine and guanidinoacetate. Am J Physiol Endocrinol Metab 281 (5 ):E1095–E1100, PMID: , 10.1152/ajpendo.2001.281.5.E1095.11595668
16. Gamble MV, Liu X, Ahsan H, Pilsner JR, Ilievski V, Slavkovich V, et al. 2006. Folate and arsenic metabolism: a double-blind, placebo-controlled folic acid-supplementation trial in Bangladesh. Am J Clin Nutr 84 (5 ):1093–1101, PMID: , 10.1093/ajcn/84.5.1093.17093162
17. Peters BA, Hall MN, Liu X, Parvez F, Sanchez TR, van Geen A, et al. 2015. Folic acid and creatine as therapeutic approaches to lower blood arsenic: a randomized controlled trial. Environ Health Perspect 123 (12 ):1294–1301, PMID: , 10.1289/ehp.1409396.25978852
18. Bozack AK, Hall MN, Liu X, Ilievski V, Lomax-Luu AM, Parvez F, et al. 2019. Folic acid supplementation enhances arsenic methylation: results from a folic acid and creatine supplementation randomized controlled trial in Bangladesh. Am J Clin Nutr 109 (2 ):380–391, PMID: , 10.1093/ajcn/nqy148.30590411
19. Ahsan H, Chen Y, Parvez F, Argos M, Hussain AI, Momotaj H, et al. 2006. Health Effects of Arsenic Longitudinal Study (HEALS): description of a multidisciplinary epidemiologic investigation. J Expo Sci Environ Epidemiol 16 (2 ):191–205, PMID: , 10.1038/sj.jea.7500449.16160703
20. Shih YH, Scannell Bryan M, Parvez F, Uesugi KH, Shahriar M, Ahmed A, et al. 2020. Gravidity, parity, blood pressure and mortality among women in Bangladesh from the HEALS cohort. BMJ Open 10 (8 ):e037244, PMID: , 10.1136/bmjopen-2020-037244.32847913
21. Institute of Medicine Standing Committee on the Scientific Evaluation of Dietary Reference Intakes and its Panel on Folate, Other B Vitamins, and Choline. 2000. Dietary Reference Intakes for Thiamin, Riboflavin, Niacin, Vitamin B6, Folate, Vitamin B12, Pantothenic Acid, Biotin, and Choline. Washington, DC: National Academies Press.
22. Sanchez TR, Levy D, Shahriar MH, Uddin MN, Siddique AB, Graziano JH, et al. 2016. Provision of well-water treatment units to 600 households in Bangladesh: a longitudinal analysis of urinary arsenic indicates fading utility. Sci Total Environ 563–564 :131–137, PMID: , 10.1016/j.scitotenv.2016.04.112.27135575
23. Csanaky I, Gregus Z. 2003. Effect of selenite on the disposition of arsenate and arsenite in rats. Toxicology 186 (1–2 ):33–50, PMID: , 10.1016/s0300-483x(02)00604-2.12604169
24. Reuter W, Davidowski L, Neubauer K, Di Bussolo J. 2003. Speciation of five arsenic compounds in urine by HPLC/ICP-MS. Shelton, CT: PerkinElmer SciEx. https://www.perkinelmer.com/PDFs/Downloads/app_speciationfivearseniccompounds.pdf [accessed 20 March 2023].
25. Pfeiffer CM, Huff DL, Gunter EW. 1999. Rapid and accurate HPLC assay for plasma total homocysteine and cysteine in a clinical laboratory setting. Clin Chem 45 (2 ):290–292, PMID: , 10.1093/clinchem/45.2.290.9931056
26. Cheng Z, Zheng Y, Mortlock R, Van Geen A. 2004. Rapid multi-element analysis of groundwater by high-resolution inductively coupled plasma mass spectrometry. Anal Bioanal Chem 379 (3 ):512–518, PMID: , 10.1007/s00216-004-2618-x.15098084
27. CDC (Centers for Disease Control and Prevention). 2017. Data sources and data analysis. Blood, serum, and urine samples from NHANES. https://www.cdc.gov/exposurereport/data_sources_analysis.html [accessed 2 February 2022].
28. Steinmaus C, Bates MN, Yuan Y, Kalman D, Atallah R, Rey OA, et al. 2006. Arsenic methylation and bladder cancer risk in case–control studies in Argentina and the United States. J Occup Environ Med 48 (5 ):478–488, PMID: , 10.1097/01.jom.0000200982.28276.70.16688004
29. Pu YS, Yang SM, Huang YK, Chung CJ, Huang SK, Chiu AWH, et al. 2007. Urinary arsenic profile affects the risk of urothelial carcinoma even at low arsenic exposure. Toxicol Appl Pharmacol 218 (2 ):99–106, PMID: , 10.1016/j.taap.2006.09.021.17196235
30. Melak D, Ferreccio C, Kalman D, Parra R, Acevedo J, Pérez L, et al. 2014. Arsenic methylation and lung and bladder cancer in a case-control study in northern Chile. Toxicol Appl Pharmacol 274 (2 ):225–231, PMID: , 10.1016/j.taap.2013.11.014.24296302
31. Yu RC, Hsu KH, Chen CJ, Froines JR. 2000. Arsenic methylation capacity and skin cancer. Cancer Epidemiol Biomarkers Prev 9 :1259–1262, PMID: .11097236
32. Gilbert-Diamond D, Li Z, Perry AE, Spencer SK, Gandolfi AJ, Karagas MR, et al. 2013. A population-based case–control study of urinary arsenic species and squamous cell carcinoma in New Hampshire, USA. Environ Health Perspect 121 (10 ):1154–1160, PMID: , 10.1289/ehp.1206178.23872349
33. Ahsan H, Chen Y, Kibriya MG, Slavkovich V, Parvez F, Jasmine F, et al. 2007. Arsenic metabolism, genetic susceptibility, and risk of premalignant skin lesions in Bangladesh. Cancer Epidemiol Biomarkers Prev 16 (6 ):1270–1278, PMID: , 10.1158/1055-9965.EPI-06-0676.17548696
34. Lindberg AL, Rahman M, Persson LÅ, Vahter M. 2008. The risk of arsenic induced skin lesions in Bangladeshi men and women is affected by arsenic metabolism and the age at first exposure. Toxicol Appl Pharmacol 230 (1 ):9–16, PMID: , 10.1016/j.taap.2008.02.001.18336856
35. Wu MM, Chiou HY, Hsueh YM, Hong CT, Su CL, Chang SF, et al. 2006. Effect of plasma homocysteine level and urinary monomethylarsonic acid on the risk of arsenic-associated carotid atherosclerosis. Toxicol Appl Pharmacol 216 (1 ):168–175, PMID: , 10.1016/j.taap.2006.05.005.16806340
36. Hall M, Chen Y, Ahsan H, Slavkovich V, van Geen A, Parvez F, et al. 2006. Blood arsenic as a biomarker of arsenic exposure: results from a prospective study. Toxicology 225 (2–3 ):225–233, PMID: , 10.1016/j.tox.2006.06.010.16860454
37. Hoet P, Deumer G, Bernard A, Lison D, Haufroid V. 2016. Urinary trace element concentrations in environmental settings: is there a value for systematic creatinine adjustment or do we introduce a bias? J Expo Sci Environ Epidemiol 26 (3 ):296–302, PMID: , 10.1038/jes.2015.23.25827313
38. Nermell B, Lindberg AL, Rahman M, Berglund M, Persson LA, El Arifeen S, et al. 2008. Urinary arsenic concentration adjustment factors and malnutrition. Environ Res 106 (2 ):212–218, PMID: , 10.1016/j.envres.2007.08.005.17900556
39. Kim DK, Song JW, Park JD, Choi BS. 2011. A comparison of the adjustment methods for assessing urinary concentrations of cadmium and arsenic: creatinine vs. specific gravity. Korean J Environ Health Sci 37 (6 ):450–459, 10.5668/JEHS.2011.37.6.450.
40. Abuawad A, Goldsmith J, Herbstman JB, Parvez F, Islam T, LoIacono N, et al. 2022. Urine dilution correction methods utilizing urine creatinine or specific gravity in arsenic analyses: comparisons to blood and water arsenic in the FACT and FOX studies in Bangladesh. Water 14 (9 ):1477, 10.3390/w14091477.
41. Styblo M, Delnomdedieu M, Thomas DJ. 1996. Mono- and dimethylation of arsenic in rat liver cytosol in vitro. Chem Biol Interact 99 (1–3 ):147–164, PMID: , 10.1016/0009-2797(95)03666-0.8620564
42. Abuawad A, Bozack AK, Saxena R, Gamble MV. 2021. Nutrition, one-carbon metabolism and arsenic methylation. Toxicology 457 :152803, PMID: , 10.1016/j.tox.2021.152803.33905762
43. Pilsner JR, Liu X, Ahsan H, Ilievski V, Slavkovich V, Levy D, et al. 2007. Genomic methylation of peripheral blood leukocyte DNA: influences of arsenic and folate in Bangladeshi adults. Am J Clin Nutr 86 (4 ):1179–1186, PMID: , 10.1093/ajcn/86.4.1179.17921400
44. Howe CG, Liu X, Hall MN, Ilievski V, Caudill MA, Malysheva O, et al. 2017. Sex-specific associations between one-carbon metabolism indices and posttranslational histone modifications in arsenic-exposed Bangladeshi adults. Cancer Epidemiol Biomarkers Prev 26 (2 ):261–269, PMID: , 10.1158/1055-9965.EPI-16-0202.27765800
45. Basu A, Mitra S, Chung J, Guha Mazumder DN, Ghosh N, Kalman D, et al. 2011. Creatinine, diet, micronutrients, and arsenic methylation in West Bengal, India. Environ Health Perspect 119 (9 ):1308–1313, PMID: , 10.1289/ehp.1003393.21652291
46. Gamble MV, Liu X, Ahsan H, Pilsner R, Ilievski V, Slavkovich V, et al. 2005. Folate, homocysteine, and arsenic metabolism in arsenic-exposed individuals in Bangladesh. Environ Health Perspect 113 (12 ):1683–1688, PMID: , 10.1289/ehp.8084.16330347
47. Peters BA, Hall MN, Liu X, Neugut YD, Pilsner JR, Levy D, et al. 2014. Creatinine, arsenic metabolism, and renal function in an arsenic-exposed population in Bangladesh. PLoS One 9 (12 ):e113760, PMID: , 10.1371/journal.pone.0113760.25438247
48. Peters BA, Hall MN, Liu X, Parvez F, Siddique AB, Shahriar H, et al. 2015. Low-dose creatine supplementation lowers plasma guanidinoacetate, but not plasma homocysteine, in a double-blind, randomized, placebo-controlled trial. J Nutr 145 (10 ):2245–2252, PMID: , 10.3945/jn.115.216739.26311810
49. Jencks DA, Mathews RG. 1987. Allosteric inhibition of methylenetetrahydrofolate reductase by adenosylmethionine. Effects of adenosylmethionine and NADPH on the equilibrium between active and inactive forms of the enzyme and on the kinetics of approach to equilibrium. J Biol Chem 262 (6 ):2485–2493, PMID: , 10.1016/S0021-9258(18)61530-3.3818603
50. Wagner C, Briggs WT, Cook RJ. 1985. Inhibition of glycine n-methyltransferase activity by folate derivatives: implications for regulation of methyl group metabolism. Biochem Biophys Res Commun 127 (3 ):746–752, PMID: , 10.1016/s0006-291x(85)80006-1.3838667
51. Luka Z, Mudd SH, Wagner C. 2009. Glycine N-methyltransferase and regulation of S-adenosylmethionine levels. J Biol Chem 284 (34 ):22507–22511, PMID: , 10.1074/jbc.R109.019273.19483083
52. Bozack AK, Saxena R, Gamble MV. 2018. Nutritional influences on one-carbon metabolism: effects on arsenic methylation and toxicity. Annu Rev Nutr 38 :401–429, PMID: , 10.1146/annurev-nutr-082117-051757.29799766
53. Food Fortification Initiative. 2017. Global progress. https://www.ffinetwork.org/globalprogress [accessed 2 February 2022].
54. Finer LB, Henshaw SK. 2006. Disparities in rates of unintended pregnancy in the United States, 1994 and 2001. Perspect Sex Reprod Health 38 (2 ):90–96, PMID: , 10.1363/3809006.16772190
55. Maruvada P, Stover PJ, Mason JB, Bailey RL, Davis CD, Field MS, et al. 2020. Knowledge gaps in understanding the metabolic and clinical effects of excess folates/folic acid: a summary, and perspectives, from an NIH workshop. Am J Clin Nutr 112 (5 ):1390–1403, PMID: , 10.1093/ajcn/nqaa259.33022704
56. Pfeiffer CM, Sternberg MR, Fazili Z, Yetley EA, Lacher DA, Bailey RL, et al. 2015. Unmetabolized folic acid is detected in nearly all serum samples from US children, adolescents, and adults. J Nutr 145 (3 ):520–531, PMID: , 10.3945/jn.114.201210.25733468
57. Gamble MV, Ahsan H, Liu X, Factor-Litvak P, Ilievski V, Slavkovich V, et al. 2005. Folate and cobalamin deficiencies and hyperhomocysteinemia in Bangladesh. Am J Clin Nutr 81 (6 ):1372–1377, PMID: , 10.1093/ajcn/81.6.1372.15941889
|
PMC010xxxxxx/PMC10053778.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36988318
EHP11587
10.1289/EHP11587
Research
Suicide and Transportation Noise: A Prospective Cohort Study from Switzerland
https://orcid.org/0000-0003-4906-7719
Wicki Benedikt 1 2
Schäffer Beat 3
Wunderli Jean Marc 3
Müller Thomas J. 4 5
Pervilhac Charlotte 5 6
Röösli Martin 1 2
Vienneau Danielle 1 2
1 Swiss TPH (Swiss Tropical and Public Health Institute), Basel, Switzerland
2 University of Basel, Basel, Switzerland
3 Empa, Swiss Federal Laboratories for Materials Science and Technology, Dübendorf, Switzerland
4 Translational Research Centre, University Hospital of Psychiatry and Psychotherapy, University of Bern, Bern, Switzerland
5 Private Clinic Meiringen, Meiringen, Switzerland
6 Institute of Psychology, Health Psychology and Behavioural Meidicne, University of Bern, Bern, Switzerland
Address correspondence to Danielle Vienneau, Department of Epidemiology and Public Health, Swiss Tropical and Public Health Institute, Kreuzstrasse 2, CH-4123, Allschwil, Switzerland. Telephone: +41 (0)61 284 8398; Fax: +41 (0)61 284 8105. Email: [email protected]
29 3 2023
3 2023
131 3 03701320 5 2022
18 1 2023
10 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Although plausible from a pathophysiological point of view, robust evidence for effects of transportation noise on mental health remains scarce. Meanwhile, psychiatric diseases are among the most prevalent noncommunicable diseases worldwide, and suicide as a mortality outcome highly connected to mental disorders presents a pressing public health issue. The aim of this study was to investigate the association between source-specific transportation noise, particulate matter (PM) air pollution, residential greenness, and suicide by means of a nationwide cohort study.
Methods:
Road traffic, railway and aircraft noise exposure as well as exposure to air pollution [PM with aerodynamic diameter ≤2.5μm (PM2.5)] and greenness [normalized difference vegetation index (NDVI)] were linked to 5.1 million adults (age 15 y and older) in the Swiss National Cohort, accounting for their address history. Mean noise exposure in 5-y periods was calculated. Individuals were followed for up to 15 y (2001–2015). Time-varying Cox regression models were applied to deaths by suicide (excluding assisted suicide). Models included all three noise sources, PM2.5, and NDVI plus individual and spatial covariates, including socioeconomic status. Effect modification by sex, age, socioeconomic indicators, and degree of urbanization was explored.
Results:
During the follow-up, there were 11,265 suicide deaths (10.4% poisoning, 33.3% hanging, 28.7% firearms, 14.7% falls). Road traffic and railway noise were associated with total suicides [hazard ratios: 1.040; 95% confidence interval (CI): 1.015, 1.065; and 1.022 (95% CI: 1.004, 1.041) per 10 dB day-evening-night level (Lden)], whereas for aircraft noise, a risk increase starting from 50 dB was masked by an inverse association in the very low exposure range (30–40 dB). Associations were stronger for females than males. The results were robust to adjustment for residential greenness and air pollution.
Conclusion:
In this longitudinal, nationwide cohort study, we report a robust association between exposure to road traffic and railway noise and risk of death by suicide after adjusting for exposure to air pollution and greenness. These findings add to the growing body of evidence that mental health disorders may be related to chronic transportation noise exposure. https://doi.org/10.1289/EHP11587
Supplemental Material is available online (https://doi.org/10.1289/EHP11587).
The authors declare that they have no competing interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Mental health disorders represent a pressing public health issue. In 2019, the prevalence of mental health disorders globally was estimated to be 13% [95% confidence interval (CI): 12.1, 14.0%], which translates to almost 1 billion people affected. In Switzerland, the estimated prevalence was slightly higher with 17.3% (95% CI: 15.9, 18.8%), translating to about 1.4 million people affected.1 Although most mental health disorders primarily lead to morbidity and decreased quality of life, a mortality outcome closely related to mental illness is suicide.2,3 Suicide is a complex, multicausal phenomenon, involving psychological, social, biological, and environmental factors. A study on suicide in the Swiss National Cohort confirmed that mental and behavioral problems were by far the most prevalent comorbidities in suicide victims across all professions, age groups, and genders.4 Only recently research has started to also explore and identify possible environmental risk factors for suicide, with reported associations of an increased suicide risk with heat,5 air pollution.6 and noise.7 On the other hand, residential greenness and urban green space have been recognized as environmental factors with protective properties.8 Although suicide rates have decreased worldwide and in Switzerland in the last 20 y, the decline is not yet on course to reach the Sustainable Development Goal (SDG) aim of a reduction by one-third by 2030. The World Health Organization (WHO) report, “Suicide Worldwide in 2019,” published in 2021, estimated that 703,000 people died from suicide in 2019 worldwide, which corresponds to 1.3% of all yearly deaths.9 On the grounds of such numbers, reducing the occurrence of mental illnesses is a primary public health interest. Hence, understanding risk factors for the development of mental disorders and therefore for suicide is of utmost importance. In light of the growing urbanization worldwide, studying the role of urban environmental stressors as such potential risk factors can potentially yield insights of consequential importance concerning the promotion of mental health and prevention of psychiatric morbidity and mortality.
In recent years, noise has been recognized as one of the most impactful environmental stressors on health and well-being.10 Among the sources of environmental noise, transportation noise and especially road traffic noise have emerged as the most prevalent and harmful. According to a European Environmental Agency (EEA) report, 20% of the European population (139 million people) in 2017 were estimated to live in areas with transportation noise levels that are considered harmful [>55 dB day-evening-night level (Lden)].10 A more recent study investigating noise exposure in 700 cities in Europe estimated that 42% of the adult urban population are exposed to such harmful levels.11 Due to this widespread occurrence, noise is the second most important driver of the environmental burden of disease in Europe, behind fine particulate matter (PM) air pollution. In numbers, exposure to transportation noise is estimated to be responsible for 400–1,500 disability adjusted life years lost each year per million people in Western Europe.12 However, because more evidence including more outcomes has emerged since 2014, and, for example, effects on mental health were not included in the above-mentioned calculations, these numbers might actually represent an underestimation.
Concerning negative health effects related to noise exposure, there is growing evidence for diverse nonauditory effects, such as arterial hypertension, cardiovascular and metabolic diseases such as type 2 diabetes,13–16 sleep disturbance,17 annoyance,18 as well as reduced quality of life and well-being19—and mental health and neurological disorders.20 Although the exact pathways of the influences of noise on health remain somewhat unclear, there is evidence that noise causes physiological stress reactions involving heightened amygdalar activity,21 allostatic overload,13,15 and disturbed sleep.22 Because these are all established risk factors for multiple mental health disorders, including depression,23–27 an association of noise with poor mental health seems plausible from a pathophysiological perspective. However, the systematic review on noise-related mental health outcomes used for the 2018 WHO guidelines resulted in a judgment of very low-quality evidence for aircraft noise effects on depression and anxiety and a judgment of low-quality evidence for a null effect of road traffic or railway noise.28 The review included 29 predominantly cross-sectional studies. Hence, the poor quality of evidence was attributed to a lack of robust studies investigating the mental health effects of different noise sources. A more recent meta-analysis of five studies found an increased risk of 12% (95% CI: 2%, 23%) for depression per 10 dB Lden of aircraft noise exposure.29 The same study29 also suggested a 2%–3% increased risk for depression per 10 dB Lden for railway noise based on three studies and for road traffic noise based on eleven studies. Another review and meta-analysis including nine studies showed an association between transportation noise and anxiety [odds ratio (OR)=1.09; 95% CI: 0.97, 1.23 per 10 dB], while also rating the quality of evidence as low.30 Since the publication of this meta-analysis, a Swiss prospective cohort study (SAPALDIA) reported a 7% increased risk for the incidence of depression per 10 dB Lden road traffic noise [relative risk (RR)=1.07; 95% CI: 0.93, 1.22] and a 20% increased risk per 10 dB Lden aircraft noise (RR=1.20; 95% CI: 0.92, 1.55). So far, only one study investigating the associations between long-term exposure to environmental noise and suicide has been conducted.7 The authors examined the risk for death by suicide in relation to average nighttime noise exposure (including noise caused by transportation and industrial and recreational activities) in adults in Korea and reported an increased risk per interquartile range (IQR=2.67dB) of nighttime noise of 32% (95% CI: 2%, 70%) in younger adults and 43% (95% CI: 1%, 102%) in older adults. A time-series study from Spain investigating short-term effects of traffic noise exposure on suicides and emergency admission for depression and anxiety reported an increased risk for both outcomes.31
In comparison, the effects of air pollution on mental health have received more attention and have been studied more thoroughly. Air pollutants have been shown to cause oxidative stress and neuroinflammation and to trigger stress responses with stress hormone release, which are the major hypothesized mechanisms linking air pollution and adverse mental health outcomes.32,33 A systematic review and meta-analysis published in 2022 including 39 studies reported significant associations between long-term exposure to various air pollutants (PM2.5, NO2, SO2, CO) and risk of depression. The largest risk increase was observed for PM with aerodynamic diameter ≤10μm (PM10) [RR=1.092 (95% CI: 0.988, 1.206) per 10-μg/m3 increase in exposure]. Smaller effects were also reported per 10-μg/m3 increase of short-term exposure to PM2.5 [RR=1.009 (95% CI: 1.007, 1.011)], PM10 [RR=1.009 (95% CI: 1.006, 1.012)], O3 [RR=1.011 (95% CI: 0.997, 1.026)], NO2 [RR=1.022 (95% CI: 1.012, 1.033)] and SO2 [RR=1.024 (95% CI: 1.010, 1.037)].34 Another systematic review and meta-analysis from 2019 found similar results for depression, and also reported associations between short-term PM10 exposure and suicide risk [RR=1.02 (95% CI: 1.00, 1.03) per 10 μg/m3 at lag 0–2 d, including four studies].6 These results were compiled in a more recent systematic review and meta-analysis from 2021 that included 10 studies, reporting a 2% (95% CI: 1%, 3%) risk increase for suicide per 10 μg/m3 PM2.5 exposure.35 One limitation of these reviews is that most of the included studies have not adjusted for exposure to possible confounders such as transportation noise. A recent large cross-sectional study from the UK Biobank studying PM2.5 and road traffic noise exposure, however, reported an increased risk for depression associated with PM2.5 exposure, but no association with road traffic noise exposure was found.36
Residential greenness or green space is another exposure of interest in environmental epidemiology as a protective factor for health and well-being. For example, higher levels of greenness [normalized difference vegetation index (NDVI)] around people’s place of residence have been associated with a lower risk of natural-cause mortality [HR=0.94 (95% CI: 0.93, 0.95) per IQR (0.14 NDVI in a 500-m buffer)] in a large Swiss cohort study.37 Concerning mental health, a Dutch study found a decreased risk for death by suicide in communities with high green space proportion (>85%) in comparison with communities with little (<25%) green space [RR=0.879 (95% CI: 0.779, 0.991)].38 Beyond suicide, a systematic review from 2020 suggested several beneficial effects of access to green space on adolescents’ mental health, including fewer depressive symptoms and improved general mental health.39 Multiple pathways are posited for this positive association, including that greener living environments or green space availability encourage healthy behavior, and that such factors can aid in stress relief.40
This study investigates the association between exposure to road traffic, railway, and aircraft noise and the risk of death by suicide in a longitudinal, nationwide research cohort in Switzerland. We hypothesized that people exposed to higher levels of transportation noise are more likely to develop mental health disorders such as depression and therefore have a higher risk of death by suicide, independent of coexposure to air pollution and residential greenness as well as socioeconomic position. By using suicide as a surrogate, we aimed to add to the understanding of whether transportation noise exposure affects mental health.
Methods
Study Population
The Swiss National Cohort (SNC) is a longitudinal, population-based research cohort. It links births, mortality, and emigration registries with the former national decennial census and, since 2010, with the annual Registry Based Census.41,42 The linkages in the SNC from 2010 onward are deterministic using a personal identifier, whereas earlier linkages were performed probabilistically based on variables such as date of birth, sex, civil status, nationality, religion, and place of residence. No validation of the probabilistic linkage is available, but comparison with the deterministic linkages from 2010 and onward allows the discovery and exclusion of mismatches. Close to complete representation of the whole population is ensured by compulsory census participation, which is reflected in 98.6% of the population being included in the 4 December 2000 census.43 The SNC was approved by the ethics committees of the Cantons of Zurich and Berne.
For the current study, we used the SNC as a closed cohort that included data from 1 January 2001 to 31 December 2015 for a total of 7.28 million individuals. After excluding individuals below 15 y of age at baseline (17.5% of the full population), data with a mismatch between probabilistic and deterministic SNC linkage (i.e., incorrect probabilistic linkage, 8.2%), missing residential coordinates or individuals living in an institution (5.4%), missing information on covariates (i.e., education or socioeconomic position) (2.5%), or missing exposure data (0.2%), the final sample used for analysis consisted of 5.1 million observations (See Supplement Table S1). No imputations were performed.
Outcome
The outcome of interest was defined as all intentional self-harm [i.e., total suicides; International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10): X60–84, excluding X61.8, X61.9, and X81–82] as cause of death. The SNC contains records from all deaths occurring in Switzerland from 1991 up to 2019 that included cause of death as ICD codes. Regarding exclusions, ICD-Codes X61.8 (right-to-die organization on death certificate) and X61.9 (Poisoning with pentobarbital; the drug used by right-to-die organizations) have been used to indicate assisted suicide since 1998.44 Additionally, we suspected suicides involving vehicles (ICD-10: X81–82) to be spuriously associated with railway noise, due to confounding by proximity and therefore availability of the method. Because preliminary analysis confirmed this suspicion (see Figure S1), these outcomes were also excluded from the main analysis. The specific suicide subclasses Poisoning (ICD10: X60–69, excluding X61.8 and X61.9), Hanging (ICD-10: X70), Firearms (ICD-10: X72–75) and Jumping (ICD-10: X80) were also investigated separately.
Noise Exposure Data
The same noise exposure data used in a previous publication investigating cardiovascular disease and transportation noise in the SNC was used in our study.45 These data were originally developed for the Short and Long Term Effects of Transportation Noise Exposure (SiRENE) project and were available for census years 2001 and 2011.46,47 The database contains modeled noise exposure levels for the three main sources of transportation noise, using the following calculation methods: road traffic (source model sonROAD48 and propagation model StL-8649) railways (source model sonRAIL50 and propagation model SEMIBEL51) and aircraft (FLULA252,53). Concerning aircraft noise, the model included estimates based on air traffic data of the three international civil airports (Zurich, Geneva, and Basel), as well as for the largest military airfield situated in Payerne.
The main noise metric used was the source-specific Lden, which is a weighted logarithmic mean of daily noise exposure with a penalty of 5 dB for evening (1900–2300 hours) and 10 dB for nighttime (2300–0700 hours) noise. The intermittency ratio (IR)54 during the night was also available. This noise metric describes how impactful single noise events are in contrast to background noise. The values of IR range from 0%, meaning single events do not substantially exceed long-term average noise, to 100%, meaning that all noise exposure is produced by individual noise events. Additionally, the number of nighttime noise events (i.e., events 3 dB louder than background noise) was available. Both nighttime IR and number of events were not source-specific but calculated considering all three noise sources. Source-specific Lden at the most exposed façade and corresponding IR as well as the number of events were assigned to participants based on residential location (geocode and floor of residence, using a default middle floor of the building if exact floor was not known55,56). To account for background noise from various sources, Lden values were left censored at 35 dB for road traffic noise and 30 dB for railway and aircraft noise.56
Noise Exposure Assignment
As described in Vienneau et al.,45 the follow-up was divided into three 5-y periods (2001–2005, 2006–2010, 2011–2015) to support time-varying analysis accounting for potential time trends and changes of residence.57 Both residential geocodes and noise estimates were available for 2001 and 2011. Hence, the 2001 noise exposure estimates were assigned for the first period (2001–2005) and the 2011 noise exposure estimates to the third period (2011–2015), based on the residential address at the beginning of the period. Using the 2010 census question “living in the same community 5 years before” and moving dates, the exposure assignment for the middle period was constructed as follows. For people who had not moved or moved after 2006, the 2001 noise data was used for the middle period (2006–2010) because these participants were believed to still be living at the same residence that they lived at in 2001. For people who moved before 2006, the 2011 noise data and updated residential geocodes were used for the middle period (2006–2010).
Covariates
A directed acyclic graph (DAG) was drawn to identify potentially confounding factors (Figure S2). This led to the identification of the following factors: degree of urban, air pollution, green space, civil status, and socioeconomic position. The following individual sociodemographic variables available from the SNC were included to represent socioeconomic position: education level (compulsory education or less, upper secondary level education, tertiary level education), mother tongue (German and Rhaeto-Romansh, French, Italian, other language), nationality (Swiss, non-Swiss), and local index of socioeconomic position (local SEP in quartiles). The local-SEP index used is calculated for a small local area of 50 nearest neighbors and considered median rent per living space, education level and type of occupation of the household head and number of inhabitants per room.58 Additionally, civil status (single, married, widowed, divorced) and degree of urban (urban, peri-urban, rural) were included as potential confounders and sex (female, male) as a covariate. The definition of degree of urban was performed for every community by the Swiss Federal Office for statistics based on morphological criteria such as population number and density as well as functional criteria such as commuter flows59 and is part of the SNC data set. Because socioeconomic status was identified as one of the most important possible confounders in this study, area-level SEP and unemployment rate were also calculated at community (n=2,896 in 2001, n=2,585 in 2011) and cantonal (n=26) level to reflect different aspects of SEP on different levels. The community- and cantonal-level SEP variables were derived by averaging the local-SEP index of all individuals within the respective area. Unemployment rates were available from the SNC and defined as percentage of the working-age population (20–65 y) unemployed. All covariates were available at baseline in 2001, whereas some covariates were also available at the start of the third period coinciding with the 2011 census. Those updated included civil status, nationality, local SEP, area SEP, and unemployment rate, whereas for other variables the baseline values were retained.
Concerning potentially confounding environmental exposures, PM2.5 concentration (micrograms per cubic meter) was selected as marker for air pollution in the main model, whereas NO2 concentrations (micrograms per cubic meter) were used for sensitivity analyses. Maps for both pollutants for the year 2010 were available from validated European 100m×100m hybrid land use regression models developed based on AirBase routine monitoring data, satellite observations, dispersion model estimates, and land use and traffic data. The model predictions for 2010 correlated highly with predictions in other years60; thus the 2010 estimates were considered relevant for all three 5-y periods. Air pollution exposure was updated according to residence history at the beginning of each of the three time periods.
As a possible confounder, greenness measured as mean NDVI with a 500-m buffer around participants’ addresses was included as a covariate in the main model. NDVI exposure derived from a data set for 2014 previously constructed for and used in the SNC (using 30×30m resolution, cloud-free Landsat scenes from summer months).37 NDVI exposure was applied to the geocodes at the beginning of each of the three time periods (2001, 2006, and 2011), thus updating residential greenness for individuals who moved during our study period.
Statistical Analysis
The Cox proportional hazards model was applied to assess associations between death by all intentional self-harm (ICD-10: X60–84, excluding X61.8, X61.9, and X81–82) as well as the specific outcome subclasses [Poisoning (ICD10: X60–69, excluding X61.8 and X61.9), Hanging (ICD-10: X70), Firearms (ICD-10: X72–75), and Jumping (ICD-10: X80)] and exposure to each transportation noise source, air pollution, and NDVI, with age as timescale and stratified by sex. Adherence to the proportional hazards assumption was tested by calculating covariate-specific Schoenfeld residuals over time. Sex, time period, local SEP, civil status, and education level were included as strata because these covariates violated the proportional hazards assumption. To consider residential history and adjust for time trends in noise exposure and mortality, calendar time was adjusted for by dividing follow-up into three periods of 5 y each (2001–2005, 2006–2010, 2011–2015). Follow-up was continued until failure (i.e., death by suicide) or censoring (i.e., death by any other cause; emigration) or end of the follow-up period on 31 December 2015. Because some participants have exposure to more than one transportation noise source, we included road traffic, railway, and aircraft noise as well as air pollution and NDVI in a single model. As done previously,55 this approach allows identification of mutually independent associations of any single exposure with the outcome. Results were calculated and reported as hazard ratio (HR) and 95% CIs per 10-dB increase in Lden for each transportation noise source, per 10-μg/m3 increase in PM2.5 concentration and per 0.1 increase in NDVI.
E-values were calculated for the main findings. The e-value is a measure for the potential effect of residual confounding, which is interpreted as the strength of association that an unmeasured confounder would need to have with both the exposure and the outcome, conditional on the measured covariates in the model, to be fully responsible for the observed exposure–outcome association.61 The absolute excess risk for the main findings was calculated by multiplying the suicides/100,000 person-years (PY) by the excess risk (HR-1). Natural splines with 3 degrees of freedom (df) were used to assess the exposure–response relationship.
Incremental model adjustments were applied. Model 0, or the base model, included the Lden variables for the three noise sources, age as time scale, strata sex, and 5-y period. In model 1, the individual sociodemographic covariates (civil status, education, mother tongue, nationality, local-SEP index) were added. Model 2 added to model 1 the area-SEP and unemployment variables. Model 3 additionally adjusted for air pollution measured as continuous PM2.5 exposure and NDVI. As a sensitivity analysis, Model 3b included continuous NO2 exposure instead of PM2.5. Two additional models also included noise eventfulness at night, parameterized in model 4a as quartiles of IR and in model 4b as quartiles of number of events. Variance inflation factor (VIF<5) was used a posteriori to evaluate multicollinearity between the three Lden variables, IR, number of events, and the air pollution variables.62,63 Pearson correlation coefficient was calculated to describe correlation between the different exposures.
The main analysis was conducted for the full cohort, combining both sexes and all ages. Separate HRs were also calculated for males and females, and for three separate age groups (15–29, 30–65, over 65 y). Effect modification by SEP was explored by stratifying the analysis by quartiles of the local-SEP index. Additional analyses included stratified analysis by degree of urban and civil status (married vs. single/divorced/widowed). Interaction between air pollution and road traffic noise was investigated using a model with categorical exposures corresponding to quartiles. Likelihood-ratio testing was applied to test whether the interaction term improved model fit. A separate analysis was conducted that investigated the risk in groups exposed to one, two, or three noise sources above 50 dB Lden in comparison with that of participants with exposure to all sources below 50 dB (=reference group). This cutoff was determined based on the shape of the exposure–response functions we derived in this study, as well as the distribution of noise in our sample. This last analysis was adjusted for PM2.5, NDVI, and the same individual sociodemographic and regional covariates as in the main analysis.
Analyses were conducted in Stata 16 (StataCorp LLC), and plots and splines were developed in R (version 4.0; R Development Core Team).
Results
Study Population
A total of 5,084,838 individuals living in Switzerland and age 15 y or older at baseline (1 January 2001) were included (Table 1). Follow-up lasted until 31 December 2015, resulting in 69,440,133 PY. Our sample consisted of slightly more females (51.6%), individuals with mostly Swiss nationality (81.4%), and predominantly speaking German (or Rhaeto-Romansh) as native language (65.1%). A majority were married (60.3%) and had more than compulsory education (71.7%). Almost half of the study population lived in peri-urban settings (45%), with more similar proportions living in urban (29.1%) and rural (25.9%) areas.
Table 1 Population characteristics of the eligible participants from the Swiss National Cohort at baseline (2001).
Characteristic 2001 (Baseline)
Number of participants 5,084,838
Sex [% (n)]
Female 51.6% (2,624,262)
Male 48.4% (2.460,576)
Age [% (n)]
15–29 y 18.7% (948,618)
30–64 y 62.2% (3,163,489)
≥65 19.1% (972,731)
Mother tongue [% (n)]
German and Rhaeto-Romansh 65.1% (3,312,465)
French 19.7% (999,495)
Italian 7.1% (360,538)
Other 8.1% (412,340)
Education [% (n)]
Compulsory education or less 27.5% (1,398,715)
Upper secondary level 51.8% (2,633,811)
Tertiary level education 19.9% (1,011,479)
Child/unknown 0.8% (40,833)
Urbanization [% (n)]
Urban 29.1% (1,478,470)
Peri-urban 45% (2,289,923)
Rural 25.9% (1,316,445)
Marital status [% (n)]
Single 26% (1,321,024)
Married 60.3% (3,066,705)
Divorced 7% (355,994)
Widowed 6.7% (341,115)
Nationality [% (n)]
Swiss 81.4% (4,137,934)
Non-Swiss 18.6% (946,904)
Local-SEP [mean (SD)] 63.0 (10.6)
Area SEP region [mean (SD)] 62.8 (4.2)
Area SEP community-region [mean (SD)] 0.04 (5.2)
Area unemployment community [%, mean (SD)] 3.5 (0.7)
Area unemployment community-region [%, mean (SD)] 0 (1.2)
Note: SD, standard deviation; SEP, socioeconomic position.
During the 15-y follow-up period, 11,265 deaths from intentional self-harm (excluding assisted suicide and suicide involving vehicles) occurred. Of these, 14.0% concerned people between 15 and 30 y of age, 64.8% people between 31 and 65 y, and 21.2% people older than 65 y. Roughly three-quarters (74.1%) of the deceased by suicide were males.
The mean exposure for road traffic noise was highest (54.4 dB Lden), followed by railway noise (38.6 dB) and aircraft noise (34.5 dB) (Table 2). Correlations between the different noise sources were low (Pearson r=0.04–0.13). Both PM2.5 and NO2 concentrations were somewhat correlated with aircraft noise (r=0.41 and 0.40, respectively), although only NO2 showed some correlation with road traffic noise (r=0.42; r=0.24 for PM2.5). Correlation between the two air pollutants was high (r=0.70). See Figure S3 for the full correlation matrix.
Table 2 Levels of the different environmental exposures of the eligible participants from the Swiss National Cohort at baseline (2001).
Exposure Mean (SD)
Road traffic noise [Lden (dB)] 54.4 (8.2)
Railway noise [Lden (dB)] 38.6 (11.1)
Aircraft noise [Lden (dB)] 34.5 (7.8)
PM2.5 concentration (μg/m3) [mean (SD)] 15.9 (2.4)
NO2 concentration (μg/m3) [mean (SD)] 23.7 (7.5)
NDVI exposure (no unit) [mean (SD)] 0.57 (0.11)
Note: dB, decibel; ICD, International Statistical Classification of Diseases and Related Health Problems, 10th Revision; Lden, day-evening-night level; NDVI, normalized difference vegetation index; SD, standard deviation.
Main Findings
Road traffic noise was associated with an increased risk of death by suicide in all of the models, with an HR of 1.040 (95% CI: 1.015, 1.065) per 10-dB increase in noise exposure in the full model (Model 3) adjusting for SEP, PM2.5 exposure, and NDVI at place of residence (Figure 1; Table 3). Railway noise exposure was also associated with an increased risk of death by intentional self-harm, but it was of a smaller magnitude (HR=1.022; 95% CI: 1.004, 1.041). For aircraft noise, no linear association was observed (HR=0.997; 95% CI: 0.965, 1.029). These results were robust across models, with smaller effect estimates mainly for road traffic noise after adding individual sociodemographic covariates (Model 0 to Model 1), but otherwise no major changes in the tendencies of the observed associations (Table S2). The observed increased risks translate to an absolute excess risk of 0.63 additional suicide deaths/100,000 PY for each 10-dB increase in road traffic noise and 0.36 additional suicide deaths/100,000 PY for each 10-dB increase in railway noise.
Figure 1. Association (HR and 95% CI) between transportation noise source (Lden) and mortality from all intentional self-harm (Main Model 3). Results from main model (M3) including noise exposures (road traffic noise, railway noise, and aircraft noise), PM2.5 exposure, NDVI, age as timescale, sex as strata, individual sociodemographic covariates (civil status, education, mother tongue, nationality, urbanization, local SEP) and area covariates (area SEP-Index and unemployment rate). Outcome is mortality from intentional self-harm (ICD-10: X60–84, excl. ICD-10 X61.8, X61.9, X81–82). The numerical values of the results displayed in this figure can be found in Table 3. Note: CI, confidence interval; HR, hazard ratio; ICD-10, International Statistical Classification of Diseases and Related Health Problems, 10th Revision; NDVI, normalized difference vegetation index; SEP, socioeconomic position.
Figure 1 is a line graph plotting hazard ratio per 10 decibels day evening night level ranging from 0.96 to 1.06 in increments of 0.02 (y-axis) across noise exposure of road, rail, and air traffic (x-axis).
Table 3 HR (95% CI) per 10-dB increase in Lden, 10-μg/m3 increase in PM2.5, and 0.1 increase per 0.1 NDVI for death by intentional self-harm, in mutually adjusted models.
All intentional self-harm (ICD-10: X60–84, excluding X61.8, X61.9 and X81–82; N cases=11,265) Poisoning (X60–69) Hanging (X70) Firearms (X72–75) Jumping (X80)
All (n=11,265) Male (n=8,476) Female (n=2,789) Age 15–30 y (n=1,508) Age >30–65 y (n=7,240) Age >65 (n=2,517) All (n=1,178) All (n=3,755) All (n=3,236) All (n=1,651)
Road traffic noise 1.040 (1.015, 1.065)a 1.034 (1.006, 1.063) 1.058 (1.007, 1.112) 1.079 (1.011, 1.152) 1.05 (1.019, 1.082) 0.994 (0.944, 1.046) 1.106 (1.025, 1.193) 1.060 (1.017, 1.105) 1.007 (0.963, 1.053) 1.009 (0.964, 1.055)
Railway noise 1.022 (1.004, 1.041)b 1.021 (1.000, 1.043) 1.028 (0.992, 1.066) 0.982 (0.934, 1.032) 1.027 (1.004, 1.051) 1.037 (0.998, 1.077) 1.053 (0.997, 1.111) 1.006 (0.974, 1.039) 1.014 (0.980, 1.050) 1.015 (0.98, 1.051)
Aircraft noise 0.997 (0.965, 1.029) 0.995 (0.959, 1.032) 1.005 (0.943, 1.071) 0.983 (0.901, 1.072) 1.015 (0.976, 1.056) 0.950 (0.887, 1.017) 0.994 (0.902, 1.094) 0.970 (0.917, 1.027) 0.989 (0.933, 1.049) 0.991 (0.935, 1.051)
PM2.5 0.900 (0.811, 0.998) 0.899 (0.798, 1.012) 0.918 (0.74, 1.137) 0.978 (0.734, 1.303) 0.882 (0.776, 1.002) 0.915 (0.729, 1.149) 0.943 (0.671, 1.324) 0.913 (0.767, 1.086) 0.961 (0.792, 1.167) 0.906 (0.745, 1.101)
NDVI 0.999 (0.978, 1.020) 1.016 (0.992, 1.041) 0.946 (0.908, 0.986) 1.002 (0.948, 1.058) 0.986 (0.961, 1.012) 1.036 (0.991, 1.084) 0.923 (0.868, 0.982) 1.042 (1.005, 1.081) 1.051 (1.009, 1.094) 1.048 (1.007, 1.091)
Note: Results from main model (M3) including noise exposures (road traffic noise, railway noise, and aircraft noise), PM2.5 exposure, NDVI within 500 m around the residence, age as timescale, sex as strata, individual sociodemographic covariates (civil status, education, mother tongue, nationality, urbanization, local SEP) and area covariates (area SEP-Index and unemployment rate). CI, confidence interval; HR, hazard ratio; Lden, day-evening-night level; NDVI, normalized difference vegetation index.
The E-value is a measure for the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome, conditional on the measured covariates, to fully explain the observed exposure–outcome association.
a E-Value for point estimate = 1.24.
b E-Value for point estimate = 1.17.
The observed tendencies were consistent across all outcome subgroups, with the exception of intentional self-harm involving guns, where no associations with transportation noise from any source were observed (Figure S4). The strongest associations were observed for poisoning, which is also referred to as a nonviolent suicide method [road traffic: HR=1.106 (95% CI: 1.025, 1.193), railway: HR=1.053 (95% CI: 0.997, 1.111)]. See Table S3 for all HRs and CIs. No association was found between measures for eventfulness of noise at night (number of events or IR) and risk for death by suicide (Table S4). In the analysis considering the number of noise sources above 50 dB Lden as exposure, a notable upward trend in risk was observed [One: HR=1.053 (95% CI: 1.006, 1.102), Two: HR=1.118 (95% CI: 1.049, 1.192), Three: HR=1.252 (95% CI: 0.969, 1.619); Figure S5].
For air pollution, the main results (from Model 3) indicated a negative association of PM2.5 exposure with death by intentional self-harm with large CIs [All individuals: HR=0.900 (95% CI: 0.811, 0.998), Males: HR=0.899 (95% CI: 0.798, 1.012), Females: HR=0.918 (95% CI: 0.740, 1.137)] after adjustment for the sources of transportation noise and NDVI (Table 3).
Investigation of an interaction between road traffic noise and PM2.5 exposure using a categorical model (quartiles as exposure categories) did not indicate an interaction between these two exposures (see Table S5). Likelihood-ratio testing revealed that adding the interaction term did not significantly improve model fit (p=0.160).
After adjusting for the three transportation noise sources and PM2.5, residential greenness at the place of residence, measured as NDVI with a 500-m buffer, showed a negative association with risk of death by suicide in females, whereas no clear associations where observed in males or the total sample [All individuals: HR=0.999 (95% CI: 0.978, 1.020), Males: HR=1.016 (95% CI: 0.992, 1.041), Females: HR=0.946 (95% CI: 0.908, 0.986)] (Table 3).
Exposure–Response Relationship
Based on the main Model 3, natural splines with 3 df showed a near linear association between intentional self-harm mortality and exposure to road traffic noise starting at around 50 dB Lden (Figure 2). For railway noise, a linear risk increase was observed beginning below 35 dB. Similarly, the risk started to increase linearly from just below 45 dB Lden for aircraft noise; however, below this value, where most of the observations were located, the exposure–response association was inverse.
Figure 2. Exposure–response relationships for the association between transportation noise source [Lden (dB)] and mortality from intentional self-harm (ICD-10: X60–84, excl. ICD-10 X61.8, X61.9, X81–82). Natural splines (3 df, knots placed at tertiles of noise distribution) for the association between road traffic, railway, or aircraft noise (Lden, dB) and mortality from all intentional self-harm (ICD-10: X60–84, excluding ICD-10 X61.8, X61.9, X81–82). Same adjustments as in main model (M3), including noise exposures (road traffic noise, railway noise, and aircraft noise), PM2.5 exposure, NDVI within 500 m around the residence, age as timescale, sex as strata, individual sociodemographic covariates (civil status, education, mother tongue, nationality, urbanization, local SEP, area SEP, and unemployment rate) were used. Vertical dashed red lines show source-specific WHO guideline levels: road traffic=53 dB, railway=54 dB, aircraft=45 dB. (For interpretation of the references to color in this figure legend, see the web version of this article.) Internal knots placed at the following values (tertiles of respective noise distribution): road traffic noise: 50.64 dB, 57.84 dB; railway noise: 30 dB (lower bound and first tertile), 39.66 dB; aircraft noise: 30 dB (lower bound and first tertile), 32.62 dB. Mean and SD of the noise distribution can be found in Table 2. HR and 95% CI at Lden values indicated on the x-axis can be found in Table S9A–C. Note: CI, confidence interval; dB, decibel; df, degrees of freedom; HR, hazard ratio; ICD, International Statistical Classification of Diseases and Related Health Problems, 10th Revision; Lden, day-evening-night level; NDVI, normalized difference vegetation index; SD standard deviation; SEP, socioeconomic position; WHO, World Health Organization.
Figure 2 is a set of three line graphs, where the first two graphs plot hazard ratio ranging from 1.00 to 1.30 in increments of 0.05 (y-axis) across road traffic noise open parenthesis day evening night level open brackets decibel closed bracket closed parenthesis ranging from 40 to 70 in increments of 10 (x-axis) and railway noise open parenthesis day evening night level open bracket decibel closed bracket closed parenthesis ranging from 30 to 80 in increments of 10 (x-axis) respectively. The third graph plots hazard ratio ranging from 1.00 to 1.15 in increments of 0.05 (y-axis) across aircraft noise open parenthesis day evening night level open bracket decibel closed bracket closed parenthesis ranging from 30 to 55 in increments of 5 (x-axis).
Sensitivity Analysis
For all outcomes, adjusting for NO2 instead of PM2.5 did not change the associations with exposure to the different noise sources (see Figure S6). Not adjusting for transportation noise in the sensitivity analysis did not influence the null air pollution associations found in the main model 3, which included copollutant adjustment (Figure S7).
Effect Modification
The observed increased risk of death by suicide in the main model (Model 3) for road traffic and railway noise was higher in females [road traffic: HR=1.058 (95% CI: 1.007, 1.112), railway: HR=1.028 (95% CI: 0.992, 1.066)] than in males [road traffic: HR=1.034 (95% CI: 1.006, 1.063), railway: HR=1.021 (95% CI: 1.000, 1.043)] (Table 3). These tendencies were robust across all outcome subcategories, with an association also seen in females among suicides using firearms (see Table S3; Figure S8).
Concerning age groups, the effect of road traffic noise was comparable in the two younger age groups [15–30 y: HR=1.079 (95% CI: 1.011, 1.152), 30–65 y: HR=1.050 (95% CI: 1.019, 1.082)], whereas no effect was observed for individuals older than 65 y [HR=0.994 (95% CI: 0.944, 1.047)]. For railway noise, associations were observed only in the two older age groups [30–65 y: HR=1.027 (95% CI: 1.004, 1.051), >65y: HR=1.037 (95% CI: 0.998, 1.077)], whereas again no association was observed for aircraft noise in any age group (see Table 3; Figure S9).
Looking at effect modification by local-SEP index, an increased risk for both road traffic and railway noise was observed across all local-SEP quartiles, with the largest association with road traffic noise in the second quartile (HR=1.085; 95% CI: 1.035, 1.138) and with railway noise in the third quartile (HR=1.033; 95% CI: 0.996, 1.071). There were no indications of a trend toward lower or higher SEP categories (Figure S10; Table S6). No relevant differences were observed according to civil status (Table S7).
Stratified analysis according to urbanization revealed larger risk increases for suicide deaths associated with road traffic noise in urban (HR=1.050; 95% CI: 1.004, 1.098) and peri-urban (HR=1.045; 95% CI: 1.005, 1.087) areas than in rural settings (HR=1.022; 95% CI: 0.979, 1.066). For railway noise, the largest risk increase was seen in the peri-urban setting (HR=1.043; 95% CI: 1.014, 1.072). NDVI showed a negative association with risk of death by suicide in the urban setting (HR=0.942; 95% CI: 0.912, 0.973), whereas no association was observed in the peri-urban setting and a positive association in the rural areas (HR=1.072; 95% CI: 1.027, 1.119) (Table S8; Figure S11).
Discussion
Our findings suggest an association between exposure to transportation noise at the place of residence and the risk of death by intentional self-harm. Risk started to increase at levels of 50 dB or even lower, i.e., below the current WHO guideline levels for all noise sources. The observed associations were stronger in females than in males. We did not find any evidence for an increased risk of death by intentional self-harm due to air pollution. An inverse association with residential greenness was observed in females and in the urban setting.
There is hardly any research investigating long-term exposure to transportation noise and risk of suicide. The only previous study that investigated long-term noise exposure and suicide was conducted in Korea using environmental noise measurement data from a nationwide noise monitoring system, and not differentiating the exposure by noise source (e.g., noise caused by transportation and industrial and recreational activities). Mean monthly nighttime noise levels from the closest measurement stations were used as exposure. The authors reported a significantly increased risk for death by suicide per IQR increase of nighttime noise of 32% (95% CI: 2%, 70%) in younger adults (20–54 y, n=124,994) and 43% (95% CI: 1%, 102%) in older adults (≥55y, n=30,498).7 Although these results are difficult to directly compare to ours, it is notable that both suggest a risk increase for death by suicide in relation to noise exposure.
Overall, however, we believe our results should not be interpreted as suggesting that transportation noise has a direct influence on suicide or suicidal behavior, but rather that suicide as a surrogate for underlying mental health disorders is associated with transportation noise exposure. There is conclusive evidence that mental and behavioral disorders are the predominant comorbidities in suicide victims.2,64 Hence, we reasoned to use suicide as a surrogate for underlying mental health disorders. The advantage of this approach is that it enables the use of the extensive mortality data in the SNC to study mental disorders. However, it is clear that there are also some limitations. For example, deaths by suicide represent only the “tip of the iceberg.” Estimations suggest that worldwide, there are ∼20 suicide attempts for every death by suicide,9 and a Swiss study even noted 32 attempts for each death.65 Additionally, suicide is a highly complex issue with many influencing factors, however, with psychiatric diseases and especially depressive disorders representing an important factor.64
When comparing our results on associations of transportation noise with suicide to existing literature on transportation noise and mental health, there is mixed agreement. For example, a systematic review and meta-analysis from 2020 reported an association between exposure to road traffic noise and anxiety [odds ratio (OR)=1.08; 95% CI: 1.01, 1.15 per 10 dB Lden], whereas no effect was found for railway and aircraft noise.30 Another systematic review and meta-analysis from the same year, in contrast, reported an association of aircraft noise exposure with risk for depression [12% (95% CI: 2%, 23%) increased risk per 10 dB Lden], whereas smaller risk increases were found for road traffic [3% (95% CI: −1%, 6%) per 10 dB Lden] and railway noise [2% (95% CI: −5%, 8%) per 10 dB Lden].29 A more recent longitudinal study reported an association of road traffic noise and psychological ill health.66 A study from Switzerland, also published after the above-mentioned reviews, found an association between incidence of depression and noise annoyance, whereas no significant association was found with noise exposure of any source.67 An interesting observation is that results from a 2022 UK Biobank study suggested the opposite of the results in our study. In their large cross-sectional study, the authors reported an increased risk for major depression associated with PM2.5 exposure but not an association with transportation noise exposure.36 Although numerous studies report associations of transportation noise and mental health, the quality of evidence is considered low due to study design (mostly cross-sectional) and small sample sizes.28,68,69 Additionally, heterogeneity in exposure assessments, outcome definitions and effect measures complicate a conclusive comparison of results. The consensus from the existing literature, however, is that an impact of transportation noise on mental health is highly probable,70 which our findings further support.
One difference in our results from existing evidence is that we did not find a clear association between aircraft noise and death by intentional self-harm. This finding in particular contradicts the results from the previously mentioned meta-analysis by Hegewald et al.,29 which reported a rather large increased risk of depression per 10 dB aircraft noise. These results are mainly driven by one study by Seidler et al.71 (Weight 98%), which was set around the airport of Frankfurt before the night flying ban was established in 2011. In contrast, in Zurich, the largest airport in Switzerland, a ban had already been established in 1972.72 Hence, an interpretation of this contrasting result could be that noise (and specifically aircraft noise) during the night is the main contributor of negative effects on mental health. This finding is in agreement with the study by Min and Min,7 which reported an increased risk of suicide with increasing nighttime environmental noise exposure. Additionally, this conclusion is supported by psychiatric literature judging sleep disturbances as an independent risk factor for most psychiatric disorders.73 Alternatively, the absence of an association between aircraft noise and risk for death by suicide might also be due to residual confounding and exposure distribution, because aircraft noise in our study area (Switzerland) is mostly concentrated around larger cities (Basel, Geneva, Zurich), areas for which a 2016 mortality atlas of Switzerland showed average to lower suicide rates.74 Additionally, the exposure–response curve for aircraft noise showed an inverse association in the very low exposure. Areas with low aircraft noise exposure are also more likely very rural, which might contribute to confounding. This theory is supported by results from another SNC study by Guseva Canu et al.75 that identified men working in agriculture, hunting, and forestry, who tend to live in more rural settings, to be at an increased suicide risk in comparison with the risk found for men working in other professions.
Concerning effect modification, we consistently saw stronger associations of transportation noise and suicide risk in females than in males and also a protective effect of greenness exclusively in females in the main analysis. Because gender differences in suicide are well known, these associations are not surprising. In general, women make more suicide attempts, but suicide mortality is generally higher in men.76 Accordingly, in our nationwide sample, 74.1% of the suicide victims were male. This discrepancy is often referred to as the “gender paradox of suicide.” Among other factors, differences in psychopathology have been proposed as possible reasons for these gender differences.77 Already in 2004, differences in underlying psychiatric disorders in male and female suicides were reported: although diagnoses with personality, childhood, and alcohol or substance disorders were more common in males, females had more often been diagnosed with depressive or affective disorders.2 This underlying psychopathology may partly explain the observed effect modification in our results. The suspected mechanisms linking transportation noise exposure with mental health (prolonged stress reactions,21 allostatic overload,15 and sleep disturbances22) are thought to mostly increase the risk for affective disorders such as depression and anxiety disorders,24,25,27 whereas effects on personality disorders and other psychiatric disorders such as schizophrenia seem less plausible. This evidence is also consistent with the existing, albeit limited, evidence investigating transportation noise exposure and mental health outcomes.28 Hence, we interpret our results indicating a stronger effect of road traffic noise exposure on suicide risk in females as suggesting that such associations are mostly mediated through an increased risk for affective and anxiety disorders. Because reliable data on psychiatric diagnosis or medication prescription were not available, this hypothesis could not be tested with mediation analysis. Further studies are needed to elucidate the pathways through which transportation noise influences mental health.
Another interesting finding from our study concerns those on greenness. We observed a substantial risk reduction [HR=0.942 (95% CI: 0.912, 0.973) per 0.1 NDVI] in the urban setting. This finding is in line with the findings of a Dutch longitudinal case–control study,8 which also reported a reduced risk for suicide associated with more residential green space in urban but not rural regions. Conversely, however, we saw a strong association of higher NDVI with increased risk of suicide in the rural setting. We argue that this association might be spurious, because those areas with the highest NDVI in the rural setting are probably very rural areas, where a higher percentage of the population are agricultural workers. As mentioned above, this population is among the most at-risk professional groups in Switzerland.75 Also, remote regions showed above-average suicide mortality in a 2016 mortality atlas.74
Regarding air pollution, we found no association between either PM2.5 or NO2 and suicide mortality. The lack of association was found in models both with and without adjusting for transportation noise (Figure S6). This lack of association is in contrast to many studies reporting an association between long-term and short-term exposure to air pollution and an increased risk for depression, as well as effects on suicides.6,34–36 Most of these studies, except for the UK Biobank study,36 however, did not adjust for exposure to transportation noise. It may thus also be that some of these previous results on air pollution have been confounded by transportation noise. Future studies exploring either the effects of air pollution or transportation noise exposure on mental health should consider that both exposures may play a role.
Strengths and Limitations
To our knowledge, this is the first study investigating long-term exposure to source-specific transportation noise and suicide mortality. The use of the Swiss National Cohort enabled following more than 5 million Swiss residents over 15 y of age in combination with high-quality noise models providing energy-based metrics is a strength of this study. This, in combination with adjustment for air pollution exposure using data from a validated land use regression model as well as NDVI, is a further asset.
Even though this study is based on very comprehensive data and noise exposure assessment, some exposure misclassification is unavoidable, for instance, for individuals who have moved during the study period. To minimize this, we implemented an approach to account for this spatial change using census data and address history to update the estimated noise exposure at the beginning of each 5-y period. Additionally, as is always the case when using noise exposure estimates at participants’ home addresses, our estimated exposures do not reflect the exposure the participants experienced when away from home.78
Some residual confounding, mostly by SEP, can also not be dismissed. To diminish this, we adjusted for SEP on different levels. The rationale behind this approach was that some levels might better correspond to different types of possible confounders. For example, regional markers might better reflect the quality of health services, community markers contribute information about the population mix, and the near-individual local-SEP index plus actual individual covariates (e.g., civil status and education) would correspond best with health behavior. However, this probably still did not result in a perfect reflection of individual participants’ SEP. Another possible source of residual confounding is urbanization. However, considering the spatial pattern of suicide in Switzerland, urban/rural differences mostly occur in specific age groups, vary across language regions,74 and might also partially be related to religion.79 We do not assume that such patterns are systematically correlated with transportation noise. One exception could be in the very rural setting, where noise (and air pollution) exposure is typically very low, but suicide risk might be higher. Such confounding would, however, have led to an underestimation of our effect estimates. Looking at the e-value of 1.24 for our main findings, we conclude that it is unlikely that uncaptured features of urbanization or SEP could be that strongly correlated with both transportation noise and suicide across our sample of more than 5 million individuals.
Additional limitations are the lack of data on medical records regarding psychiatric diagnoses and medication intake as well as lifestyle factors such as smoking or alcohol consumption in the SNC. Knowledge about suicide attempts as an additional outcome reflective of underlying severe mental health disorders would have further improved our study.
Conclusion
In this nationwide administrative cohort study, we found a robust association of exposure to transportation noise and the risk for death by intentional self-harm. Though information on mental health status was not available, these findings suggest that suicide as a surrogate for mental health disorders may be related to transportation noise, adding to the growing body of evidence for such effects. Further research is needed to solidify the understanding of the complex relationship between noise exposure, other environmental stressors such as air pollution, socioeconomic factors, and mental health. However, our results emphasize the public health importance of efforts to reduce the population exposed to high levels of transportation noise.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank the Federal Statistical Office (FSO) for providing mortality and census data and for the support that made the Swiss National Cohort and this study possible. This work was supported by the Swiss National Science Foundation (grant nos. 3347CO-108806, 33CS30_134273 and 33CS30_148415). The members of the Swiss National Cohort Scientific Board are M. Zwahlen (University of Berne), M. Egger (University of Berne), V. von Wyl (University of Zurich), O. Hämmig (University of Zurich), M. Bochud (University of Lausanne), M. Röösli (University of Basel), and M. Schwyn (Federal Statistical Office).
The SNC was approved by the Ethics Committees of the Canton Bern (No KeK 153/2014, PB_2020-00050).
B.W., D.V., and M.R. worked on study concept and study design; D.V. conducted data preparation; B.W. and D.V. performed statistical modeling; and B.W. wrote and revised manuscript and performed all data interpretation, review, and commentary on the manuscript.
This work was supported by the Swiss National Science Foundation (grant nos. 3347CO-108806, 33CS30_134273 and 33CS30_148415, 324730B_201272).
==== Refs
References
1. Institute for Health Metrics and Evaluation. 2019. GBD Results. https://vizhub.healthdata.org/gbd-results/ [accessed 21 December 2022].
2. Arsenault-Lapierre G, Kim C, Turecki G. 2004. Psychiatric diagnoses in 3275 suicides: a meta-analysis. BMC Psychiatry 4 :37, PMID: , 10.1186/1471-244X-4-37.15527502
3. Henriksson MM, Aro HM, Marttunen MJ, Heikkinen ME, Isometsä ET, Kuoppasalmi KI, et al. 1993. Mental disorders and comorbidity in suicide. Am J Psychiatry 150 (6 ):935–940, PMID: , 10.1176/ajp.150.6.935.8494072
4. Schmid M, Michaud L, Bovio N, Guseva Canu I, Swiss National Cohort (SNC). 2020. Prevalence of somatic and psychiatric morbidity across occupations in Switzerland and its correlation with suicide mortality: results from the Swiss National Cohort (1990–2014). BMC Psychiatry 20 (1 ):324, PMID: , 10.1186/s12888-020-02733-7.32571249
5. Bär S, Bundo M, de Schrijver E, Müller TJ, Vicedo-Cabrera AM. 2022. Suicides and ambient temperature in Switzerland: a nationwide time-series analysis. Swiss Med Wkly 152 :w30115, PMID: , 10.4414/smw.2022.w30115.35262317
6. Braithwaite I, Zhang S, Kirkbride JB, Osborn DPJ, Hayes JF. 2019. Air pollution (particulate matter) exposure and associations with depression, anxiety, bipolar, psychosis and suicide risk: a systematic review and meta-analysis. Environ Health Perspect 127 (12 ):126002, PMID: , 10.1289/EHP4595.31850801
7. Min JY, Min KB. 2018. Night noise exposure and risk of death by suicide in adults living in metropolitan areas. Depress Anxiety 35 (9 ):876–883, PMID: , 10.1002/da.22789.29953702
8. Helbich M, O’Connor RC, Nieuwenhuijsen M, Hagedoorn P. 2020. Greenery exposure and suicide mortality later in life: a longitudinal register-based case-control study. Environ Int 143 :105982, PMID: , 10.1016/j.envint.2020.105982.32712421
9. WHO (World Health Organization). 2021. Suicide Worldwide in 2019: Global Health Estimates. Geneva, Switzerland: World Health Organziation.
10. EEA (European Environment Agency). 2020. Healthy Environment, Healthy Lives: How the Environment Influences Health and Well-Being in Europe. European Environment Agency, Denmark, Copenhagen. https://www.eea.europa.eu/publications/healthy-environment-healthy-lives [accessed 16 March 2023].
11. Khomenko S, Cirach M, Barrera-Gómez J, Pereira-Barboza E, Iungman T, Mueller N, et al. 2022. Impact of road traffic noise on annoyance and preventable mortality in european cities: a health impact assessment. Environ Int 162 :107160, PMID: , 10.1016/j.envint.2022.107160.35231841
12. Hänninen O, Knol AB, Jantunen M, Lim T-A, Conrad A, Rappolder M, et al. 2014. Environmental burden of disease in Europe: assessing nine risk factors in six countries. Environ Health Perspect 122 (5 ):439–446, PMID: , 10.1289/ehp.1206154.24584099
13. Eriksson C, Pershagen G, Nilsson M. 2018. Biological mechanisms related to cardiovascular and metabolic effects by environmental noise. World Health Organization. Regional Office for Europe. https://apps.who.int/iris/handle/10665/346548 [accessed 16 March 2023].
14. Recio A, Linares C, Banegas JR, Díaz J. 2016. Road traffic noise effects on cardiovascular, respiratory, and metabolic health: an integrative model of biological mechanisms. Environ Res 146 :359–370, PMID: , 10.1016/j.envres.2015.12.036.26803214
15. Basner M, Babisch W, Davis A, Brink M, Clark C, Janssen S, et al. 2014. Auditory and non-auditory effects of noise on health. Lancet 383 (9925 ):1325–1332, PMID: , 10.1016/S0140-6736(13)61613-X.24183105
16. Münzel T, Hahad O, Sørensen M, Lelieveld J, Duerr GD, Nieuwenhuijsen M, et al. 2022. Environmental risk factors and cardiovascular diseases: a comprehensive expert review. Cardiovasc Res 118 (14 ):2880–2902, PMID: ,10.1093/cvr/cvab316.34609502
17. Brink M, Schäffer B, Vienneau D, Pieren R, Foraster M, Eze IC, et al. 2019. Self-reported sleep disturbance from road, rail and aircraft noise: exposure-response relationships and effect modifiers in the SiRENE study. Int J Environ Res Public Health 16 (21 ):4186, PMID: , 10.3390/ijerph16214186.31671890
18. van Kamp I, Simon S, Notley H, Baliatsas C, van Kempen E. 2020. Evidence relating to environmental noise exposure and annoyance, sleep disturbance, cardio-vascular and metabolic health outcomes in the context of IGCB (N): a scoping review of new evidence. Int J Environ Res Public Health 17 (9 ):3016, PMID: , 10.3390/ijerph17093016.32357581
19. Cerletti P, Eze IC, Schaffner E, Foraster M, Viennau D, Cajochen C, et al. 2020. The independent association of source-specific transportation noise exposure, noise annoyance and noise sensitivity with health-related quality of life. Environ Int 143 :105960, PMID: , 10.1016/j.envint.2020.105960.32682053
20. Hahad O, Bayo Jimenez MT, Kuntic M, Frenis K, Steven S, Daiber A, et al. 2022. Cerebral consequences of environmental noise exposure. Environ Int 165 :107306, PMID: , 10.1016/j.envint.2022.107306.35635962
21. Hahad O, Daiber A, Münzel T. 2021. Heightened amygdalar activity mediates the cardiometabolic effects of transportation noise stress. Psychoneuroendocrinology 131 :105347, PMID: , 10.1016/j.psyneuen.2021.105347.34304945
22. Smith MG, Cordoza M, Basner M. 2022. Environmental noise and effects on sleep: an update to the WHO systematic review and meta-analysis. Environ Health Perspect 130 (7 ):76001, PMID: , 10.1289/EHP10197.35857401
23. Harvey AG, Murray G, Chandler RA, Soehner A. 2011. Sleep disturbance as transdiagnostic: consideration of neurobiological mechanisms. Clin Psychol Rev 31 (2 ):225–235, PMID: , 10.1016/j.cpr.2010.04.003.20471738
24. Guidi J, Lucente M, Sonino N, Fava GA. 2021. Allostatic load and its impact on health: a systematic review. Psychother Psychosom 90 (1 ):11–27, PMID: , 10.1159/000510696.32799204
25. Beck MDAT, Alford PDBA. Depression: Causes and Treatment. Philadelphia, PA: University of Pennsylvania Press.
26. Bebbington P. 1987. Misery and the of disease theories of depression. Int J Soc Psychiatry 33 (1 ):13–20, PMID: , 10.1177/002076408703300102.3583600
27. Baglioni C, Nanovska S, Regen W, Spiegelhalder K, Feige B, Nissen C, et al. 2016. Sleep and mental disorders: a meta-analysis of polysomnographic research. Psychol Bull 142 (9 ):969–990, PMID: , 10.1037/bul0000053.27416139
28. Clark C, Paunovic K. 2018. WHO environmental noise guidelines for the european region: a systematic review on environmental noise and quality of life, wellbeing and mental health. Int J Environ Res Public Health 15 (11 ):2400, PMID: , 10.3390/ijerph15112400.30380665
29. Hegewald J, Schubert M, Freiberg A, Romero Starke K, Augustin F, Riedel-Heller SG, et al. 2020. Traffic noise and mental health: a systematic review and meta-analysis. Int J Environ Res Public Health 17 (17 ):6175, PMID: , 10.3390/ijerph17176175.32854453
30. Lan Y, Roberts H, Kwan M-P, Helbich M. 2020. Transportation noise exposure and anxiety: a systematic review and meta-analysis. Environ Res 191 :110118, PMID: , 10.1016/j.envres.2020.110118.32835678
31. Díaz J, López-Bueno JA, López-Ossorio JJ, Gónzález JL, Sánchez F, Linares C. 2020. Short-term effects of traffic noise on suicides and emergency hospital admissions due to anxiety and depression in Madrid (Spain). Sci Total Environ 710 :136315, PMID: , 10.1016/j.scitotenv.2019.136315.31923678
32. Babadjouni RM, Hodis DM, Radwanski R, Durazo R, Patel A, Liu Q, et al. 2017. Clinical effects of air pollution on the central nervous system; a review. J Clin Neurosci 43 :16–24, PMID: , 10.1016/j.jocn.2017.04.028.28528896
33. Li H, Cai J, Chen R, Zhao Z, Ying Z, Wang L, et al. 2017. Particulate matter exposure and stress hormone levels. Circulation 136 (7 ):618–627, PMID: , 10.1161/CIRCULATIONAHA.116.026796.28808144
34. Borroni E, Pesatori AC, Bollati V, Buoli M, Carugno M. 2022. Air pollution exposure and depression: a comprehensive updated systematic review and meta-analysis. Environ Pollut 292 (pt A ):118245, PMID: , 10.1016/j.envpol.2021.118245.34600062
35. Liu Q, Wang W, Gu X, Deng F, Wang X, Lin H, et al. 2021. Association between particulate matter air pollution and risk of depression and suicide: a systematic review and meta-analysis. Environ Sci Pollut Res Int 28 (8 ):9029–9049, PMID: , 10.1007/s11356-021-12357-3.33481201
36. Hao G, Zuo L, Xiong P, Chen L, Liang X, Jing C. 2022. Associations of PM2.5 and road traffic noise with mental health: evidence from UK Biobank. Environ Res 207 :112221, PMID: , 10.1016/j.envres.2021.112221.34656633
37. Vienneau D, de Hoogh K, Faeh D, Kaufmann M, Wunderli JM, Röösli M. 2017. More than clean air and tranquillity: residential green is independently associated with decreasing mortality. Environ Int 108 :176–184, PMID: , 10.1016/j.envint.2017.08.012.28863390
38. Helbich M, de Beurs D, Kwan M-P, O’Connor RC, Groenewegen PP. 2018. Natural environments and suicide mortality in The Netherlands: a cross-sectional, ecological study. Lancet Planet Health 2 (3 ):e134–e139, PMID: , 10.1016/S2542-5196(18)30033-0.29546252
39. Zhang Y, Mavoa S, Zhao J, Raphael D, Smith M. 2020. The association between green space and adolescents’ mental well-being: a systematic review. Int J Environ Res Public Health 17 (18 ):6640, PMID: , 10.3390/ijerph17186640.32932996
40. Zhang R, Zhang CQ, Rhodes RE. 2021. The pathways linking objectively-measured greenspace exposure and mental health: a systematic review of observational studies. Environ Res 198 :111233, PMID: , 10.1016/j.envres.2021.111233.33933490
41. Bopp M, Spoerri A, Zwahlen M, Gutzwiller F, Paccaud F, Braun-Fahrländer C, et al. 2009. Cohort profile: the Swiss National Cohort—a longitudinal study of 6.8 million people. Int J Epidemiol 38 (2 ):379–384, PMID: , 10.1093/ije/dyn042.18326512
42. Spoerri A, Zwahlen M, Egger M, Bopp M. 2010. The Swiss National Cohort: a unique database for national and international researchers. Int J Public Health 55 (4 ):239–242, PMID: , 10.1007/s00038-010-0160-5.20556473
43. Renaud AS. 2004. Coverage Estimation for the Swiss Population Census 2000: Estimation Methodology and Results, Neuchatel, Switzerland. https://www.bfs.admin.ch/bfsstatic/dam/assets/341896/master [accessed 16 March 2023].
44. Steck N, Zwahlen M, Egger M, Swiss National Cohort. 2015. Time-trends in assisted and unassisted suicides completed with different methods: Swiss National Cohort. Swiss Med Wkly 145 :w14153, PMID: , 10.4414/smw.2015.14153.26099005
45. Vienneau D, Saucy A, Schäffer B, Flückiger B, Tangermann L, Stafoggia M, et al. 2022. Transportation noise exposure and cardiovascular mortality: 15-years of follow-up in a nationwide prospective cohort in Switzerland. Environ Int 158 :106974, PMID: , 10.1016/j.envint.2021.106974.34775186
46. Karipidis I, Vienneau D, Habermacher M, Köpflii M, Brink M, Probst-Hensch N, et al. 2014. Reconstruction of historical noise exposure data for environmental epidemiology in Switzerland within the SiRENE project. Noise Mapping 1 (1 ):000010247820140002, 10.2478/noise-2014-0002.
47. Röösli M, Wunderli J, Brink M, Cajochen C, Probst-Hensch N. 2019. Die SiRENE-Studie. Swiss Med Forum 19 (0506 ):77–82, 10.4414/smf.2019.03433.
48. Heutschi K. 2004. SonRoad: new Swiss road traffic noise model. Acta Acustica United with Acustica 90 (3 ):548–554.
49. FOEN (Federal Office for the Environment). 1987. Computermodell zur Berechnung von Strassenlärm. Bern, Switzerland. https://www.bafu.admin.ch/bafu/en/home/topics/noise/publications-studies/publications/computermodell-berechnung-strassenlaerm.html [accessed 16 March 2023].
50. Thron T, Hecht M. 2010. The sonRAIL emission model for railway noise in Switzerland. Acta Acustica United with Acustica 96 (5 ):873–883, 10.3813/AAA.918346.
51. BUWAL (Bundesamt für Umwelt, Wald und Landschaft). 1990. Schweizerisches Emissions- und Immisionsmodell für die Berechnung von Eisenbahnlärm (SEMIBEL). Schriftenreihe Umweltschutz Nr. 116, Bern, Switzerland.
52. Krebs W, Bütikofer R, Plüss S, Thomann G. 2004. Sound source data for aircraft noise simulation. Acta Acustica United with Acustica 90 (1 ):91–100.
53. Krebs W, Thomann G, Bütikofer R. 2010. FLULA2, Ein Verfahren zur Berechnung und Darstellung der Fluglärmbelastung, Version 4. Dübendorf, Switzerland: Empa. https://www.empa.ch/documents/56129/103151/SaT_FLULA2_Dokument/62e3c7e1-e395-4975-9eba-fda3adf17962 [accessed 16 March 2023].
54. Wunderli JM, Pieren R, Habermacher M, Vienneau D, Cajochen C, Probst-Hensch N, et al. 2016. Intermittency ratio: a metric reflecting short-term temporal variations of transportation noise exposure. J Expo Sci Environ Epidemiol 26 (6 ):575–585, PMID: , 10.1038/jes.2015.56.26350982
55. Héritier H, Vienneau D, Foraster M, Eze IC, Schaffner E, Thiesse L, et al. 2017. Transportation noise exposure and cardiovascular mortality: a nationwide cohort study from Switzerland. Eur J Epidemiol 32 (4 ):307–315, PMID: , 10.1007/s10654-017-0234-2.28280950
56. Vienneau D, Héritier H, Foraster M, Eze IC, Schaffner E, Thiesse L, et al. 2019. Façades, floors and maps—influence of exposure measurement error on the association between transportation noise and myocardial infarction. Environ Int 123 :399–406, PMID: , 10.1016/j.envint.2018.12.015.30622064
57. Canchola AJ, Stewart SL, Bernstein L, West DW, Ross RK, Deapen D, et al. 2003. Cox regression using different time-scales. Western users of SAS software annual meeting. SAS Institute, San Francisco, CA, USA, 2003.
58. Panczak R, Galobardes B, Voorpostel M, Spoerri A, Zwahlen M, Egger M. 2012. A Swiss neighbourhood index of socioeconomic position: development and association with mortality. J Epidemiol Community Health 66 (12 ):1129–1136, PMID: , 10.1136/jech-2011-200699.22717282
59. Raum mit städtischem Charakter der Schweiz 2012. Neuchâtel, Switzerland: Bundesamt für Statistik (BFS).
60. de Hoogh K, Chen J, Gulliver J, Hoffmann B, Hertel O, Ketzel M, et al. 2018. Spatial PM2.5, NO2, O3 and BC models for Western Europe—evaluation of spatiotemporal stability. Environ Int 120 :81–92, PMID: , 10.1016/j.envint.2018.07.036.30075373
61. VanderWeele TJ, Ding P. 2017. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med 167 (4 ):268–274, PMID: , 10.7326/M16-2607.28693043
62. Héritier H, Vienneau D, Foraster M, Eze IC, Schaffner E, de Hoogh K, et al. 2019. A systematic analysis of mutual effects of transportation noise and air pollution exposure on myocardial infarction mortality: a nationwide cohort study in Switzerland. Eur Heart J 40 (7 ):598–603, PMID: , 10.1093/eurheartj/ehy650.30357335
63. Vienneau D, de Hoogh K, Bechle MJ, Beelen R, van Donkelaar A, Martin RV, et al. 2013. Western European land use regression incorporating satellite- and ground-based measurements of NO2 and PM10. Environ Sci Technol 47 (23 ):13555–13564, PMID: , 10.1021/es403089q.24156783
64. Franklin JC, Ribeiro JD, Fox KR, Bentley KH, Kleiman EM, Huang X, et al. 2017. Risk factors for suicidal thoughts and behaviors: a meta-analysis of 50 years of research. Psychol Bull 143 (2 ):187–232, PMID: , 10.1037/bul0000084.27841450
65. Peter C, Tuch A. 2019. Suizidgedanken und Suizidversuche in der Schweizer Bevölkerung Obsan Bulletin 7/2019. Schweizerisches Gesund-heitsobservatorium Obsan, Neuchâtel, Switzerland.
66. Stansfeld S, Clark C, Smuk M, Gallacher J, Babisch W. 2021. Road traffic noise, noise sensitivity, noise annoyance, psychological and physical health and mortality. Environ Health 20 (1 ):32, PMID: , 10.1186/s12940-021-00720-3.33766056
67. Eze IC, Foraster M, Schaffner E, Vienneau D, Pieren R, Imboden M, et al. 2020. Incidence of depression in relation to transportation noise exposure and noise annoyance in the SAPALDIA study. Environ Int 144 :106014, PMID: , 10.1016/j.envint.2020.106014.32763645
68. Dickerson AS, Wu AC, Liew Z, Weisskopf M. 2020. A scoping review of non-occupational exposures to environmental pollutants and adult depression, anxiety, and suicide. Curr Environ Health Rep 7 (3 ):256–271, PMID: , 10.1007/s40572-020-00280-7.32533365
69. Dzhambov AM, Lercher P. 2019. Road traffic noise exposure and depression/anxiety: an updated systematic review and meta-analysis. Int J Environ Res Public Health 16 (21 ):4134, PMID: , 10.3390/ijerph16214134.31717834
70. Hahad O, Beutel ME, Gilan DA, Michal M, Daiber A, Münzel T. 2020. Auswirkungen von umweltrisikofaktoren wie lärm und luftverschmutzung auf die psychische gesundheit: was wissen wir? Dtsch Med Wochenschr 145 (23 ):1701–1707, PMID: , 10.1055/a-1201-2155.32757179
71. Seidler A, Hegewald J, Seidler AL, Schubert M, Wagner M, Dröge P, et al. 2017. Association between aircraft, road and railway traffic noise and depression in a large case-control study based on secondary data. Environ Res 152 :263–271, PMID: , 10.1016/j.envres.2016.10.017.27816007
72. Bissegger M. 2013. Noise management in the light of airport development. In: Proceedings of the Inter-Noise 2013, 42nd International Congress and Exposition on Noise Control Engineering, Noise Control for Quality of Life, 15–18 September 2013. ÖAL Österreichischer Arbeitsring für Lärmbekämpfung, eds. Innsbruck, Austria: Institute of Noise Control Engineering, 247 (3 ):5070 – 5076.
73. Freeman D, Sheaves B, Waite F, Harvey AG, Harrison PJ. 2020. Sleep disturbance and psychiatric disorders. Lancet Psychiatry 7 (7 ):628–637, PMID: , 10.1016/S2215-0366(20)30136-X.32563308
74. Chammartin F, Probst-Hensch N, Utzinger J, Vounatsou P. 2016. Mortality atlas of the main causes of death in Switzerland, 2008–2012. Swiss Med Wkly 146 :w14280, PMID: , 10.4414/smw.2016.14280.26901250
75. Guseva Canu I, Bovio N, Mediouni Z, Bochud M, Wild P, Swiss National Cohort (SNC). 2019. Suicide mortality follow-up of the Swiss National Cohort (1990–2014): sex-specific risk estimates by occupational socio-economic group in working-age population. Soc Psychiatry Psychiatr Epidemiol 54 (12 ):1483–1495, PMID: , 10.1007/s00127-019-01728-4.31127347
76. Värnik A, Kõlves K, van der Feltz-Cornelis CM, Marusic A, Oskarsson H, Palmer A, et al. 2008. Suicide methods in Europe: a gender-specific analysis of countries participating in the “European Alliance Against Depression”. J Epidemiol Community Health 62 (6 ):545–551, PMID: , 10.1136/jech.2007.065391.18477754
77. Schrijvers DL, Bollen J, Sabbe BG. 2012. The gender paradox in suicidal behavior and its impact on the suicidal process. J Affect Disord 138 (1–2 ):19–26, PMID: , 10.1016/j.jad.2011.03.050.21529962
78. Helbich M. 2018. Toward dynamic urban environmental exposure assessments in mental health research. Environ Res 161 :129–135, PMID: , 10.1016/j.envres.2017.11.006.29136521
79. Steck N, Egger M, Zwahlen M, Swiss National Cohort. 2016. Assisted and unassisted suicide in men and women: longitudinal study of the Swiss population. Br J Psychiatry 208 (5 ):484–490, PMID: , 10.1192/bjp.bp.114.160416.26743810
|
PMC010xxxxxx/PMC10056221.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36989077
EHP11524
10.1289/EHP11524
Research
Probabilistic Points of Departure and Reference Doses for Characterizing Human Noncancer and Developmental/Reproductive Effects for 10,145 Chemicals
https://orcid.org/0000-0003-3651-1307
Aurisano Nicolò 1
Jolliet Olivier 1 2
Chiu Weihsueh A. 3
Judson Richard 4
Jang Suji 3
Unnikrishnan Aswani 4
Kosnik Marissa B. 1
https://orcid.org/0000-0001-7148-6982
Fantke Peter 1
1 Quantitative Sustainability Assessment, Department of Environmental and Resource Engineering, Technical University of Denmark, Lyngby, Denmark
2 Department of Environmental Health Sciences, School of Public Health, University of Michigan, Ann Arbor, Michigan, USA
3 Department of Veterinary Integrative Biosciences, College of Veterinary Medicine and Biomedical Sciences, Texas A&M University, College Station, Texas, USA
4 National Center for Computational Toxicology, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA
Address correspondence to Peter Fantke. Telephone: 45 45254452. Email: [email protected]
29 3 2023
3 2023
131 3 03701606 5 2022
06 2 2023
03 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Regulatory toxicity values used to assess and manage chemical risks rely on the determination of the point of departure (POD) for a critical effect, which results from a comprehensive and systematic assessment of available toxicity studies. However, regulatory assessments are only available for a small fraction of chemicals.
Objectives:
Using in vivo experimental animal data from the U.S. Environmental Protection Agency’s Toxicity Value Database, we developed a semiautomated approach to determine surrogate oral route PODs, and corresponding toxicity values where regulatory assessments are unavailable.
Methods:
We developed a curated data set restricted to effect levels, exposure routes, study designs, and species relevant for deriving toxicity values. Effect levels were adjusted to chronic human equivalent benchmark doses (BMDh). We hypothesized that a quantile of the BMDh distribution could serve as a surrogate POD and determined the appropriate quantile by calibration to regulatory PODs. Finally, we characterized uncertainties around the surrogate PODs from intra- and interstudy variability and derived probabilistic toxicity values using a standardized workflow.
Results:
The BMDh distribution for each chemical was adequately fit by a lognormal distribution, and the 25th percentile best predicted the available regulatory PODs [R2≥0.78, residual standard error (RSE)≤0.53 log10 units]. We derived surrogate PODs for 10,145 chemicals from the curated data set, differentiating between general noncancer and reproductive/developmental effects, with typical uncertainties (at 95% confidence) of a factor of 10 and 12, respectively. From these PODs, probabilistic reference doses (1% incidence at 95% confidence), as well as human population effect doses (10% incidence), were derived.
Discussion:
In providing surrogate PODs calibrated to regulatory values and deriving corresponding toxicity values, we have substantially expanded the coverage of chemicals from 744 to 8,023 for general noncancer effects, and from 41 to 6,697 for reproductive/developmental effects. These results can be used across various risk assessment and risk management contexts, from hazardous site and life cycle impact assessments to chemical prioritization and substitution. https://doi.org/10.1289/EHP11524
Supplemental Material is available online (https://doi.org/10.1289/EHP11524).
While not affecting the method and results presented in this paper, O.J. discloses his role as a member of the USEtox Center scientific advisory board and chair of the project on Global Guidance for Life Cycle Impact Assessment a project supported by the Life Cycle Initiative, hosted at UN-Environment Programme. The views expressed in this paper are those of the authors and do not necessarily reflect the views or policies of the U.S. Environmental Protection Agency. All other authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Chemical management and assessment frameworks, whether for site cleanup, life cycle impact assessment (LCIA), chemical alternatives assessment (CAA), or comparative risk screening, all aim to evaluate toxicological impacts on human health from chemical exposures.1,2 These frameworks rely on chemical-specific points of departure (PODs) for deriving the quantitative toxicity values necessary for such evaluations. The POD represents the point on the dose–response curve marking the beginning of a low-dose extrapolation for risk assessment3 and is derived from effect levels from in vivo studies, such as the lowest observed adverse effect level (LOAEL), the no observed adverse effect level (NOAEL), and the statistically derived benchmark dose lower confidence limit (BMDL).4 Moreover, these PODs are typically required to be based on regulatory assessments that review and synthesize the available toxicity data, such as the U.S. Environmental Protection Agency’s (EPA’s) Integrated Science Assessments and Integrated Risk Information System (IRIS) toxicological reviews and Provisional Peer Reviewed Toxicity Values (PPRTV), among others. Yet, regulatory data sources are only available for a very limited share of the several tens of thousands of chemical substances commonly used worldwide,5–7 mainly because developing such regulatory assessments is highly data-, time-, and resource-intensive.8 Regulatory assessment being generally based on the most sensitive end points, the number of chemicals with developmental/reproductive regulatory PODs is even more restricted.
For chemical risk assessment purposes, the World Health Organization International Programme on Chemical Safety (WHO/IPCS) developed a unified framework for dose–response assessment able to derive probabilistic reference doses (RfDs) from PODs.9–12 This framework provides a consistent and transparent approach for both health-based risk assessment as well as comparative risk. Moreover, in the LCIA context its implementation was recommended for deriving human dose–response factors for noncancer end points,1 using human population effect doses with an incidence response level I=10%. However, the WHO/IPCS framework has only been applied to n=608 substances with regulatory data to calculate probabilistic RfDs12 and to n=115 organic chemicals to calculate human population effect doses (I=10%).1
With the increasing availability of online experimental animal databases, it is possible to obtain in vivo toxicity data for tens of thousands of chemical substances. Examples of such large toxicity data sources include the U.S. EPA’s Toxicity Value Database (ToxValDB)13 and the International Uniform Chemical Information Database (IUCLID; https://iuclid6.echa.europa.eu/) developed under the European Registration, Evaluation, Authorisation, and Restriction of Chemicals (REACH) regulation (EC 1907/2006). We propose that through the application of rigorous curation and statistical approaches, these data sources can be used to derive “surrogate” animal-based PODs in a quantitative high-throughput approach, systematically evaluating separate PODs for both reproductive/developmental effects and nonreproductive/developmental effects. Specifically, for substances for which regulatory PODs are not available, such experimental animal data could be alternatively used to estimate a POD that closely mimics one that would be selected in a regulatory assessment context (Figure S1).1
However, such an approach needs to address numerous challenges presented by these databases.14 For example, a chemical with multiple studies reported can have multiple effect-level values (i.e., experimental values of toxicity from individual studies) associated with it. A repeat dose toxicity data set for a single chemical may include several effect-level types (e.g., NOAELs, LOAELs) covering different observed critical effects (e.g., body weight, reproduction) for various tested species (e.g., rats, dogs), with orders of magnitude in the variability of the reported effect-level values.3,12,15 Thus, systematic methods for data selection and harmonization for human toxicity information, similar to those proposed for physico-chemical properties16 and freshwater ecotoxicity information,17 need to be developed.18,19 Given that regulatory PODs are intended to be protective of all potential adverse effects, the estimated POD should be at the lower end of the distribution of available toxicity values,20 following careful data curation where needed.21
Therefore, our approach to expand the coverage of chemicals for which toxicity values could be derived consisted of four specific objectives, namely: To create a consistent and curated data set of chronic dose–response toxicity data for multiple noncancer end points for oral exposure
To develop a statistical approach to determine oral PODs by comparing curated toxicity data against available regulatory values
To provide an extended set of oral PODs with quantified uncertainties for a wide range of chemicals, differentiating between reproductive/developmental and nonreproductive/developmental effects
To determine probabilistic RfDs for health-based or comparative risk assessments and human population effect doses (I=10%) for LCIA, both calculated from the extended set of oral PODs using the WHO/IPCS framework
Throughout this paper, we separately consider reproductive/developmental effects and nonreproductive/developmental effects (the latter hereafter referred to as “general noncancer effects”) owing to an average factor of roughly 20 difference in severity to affect human lifetime loss,1,22 as well as the differences in applicable life stages and exposure durations. The surrogate PODs we develop along with their corresponding probabilistic RfDs and human population effect doses are suitable for implementation into various chemical management and exposure and impact assessment frameworks for application in high-throughput risk screening, LCIA, CAA for chemical substitution, and exposure and risk prioritization.1,23,24
Methods
Figure 1 provides an overview of the overall workflow followed in this paper. First, we curated and selected experimental animal toxicity data and split them into two distinct data sets covering general noncancer effects and reproductive/developmental effects (Figure 1A). Second, we collected POD values from regulatory data sources (PODreg) (Figure 1B) and compared these PODreg with the curated dose–response toxicity data to identify a statistical approach for deriving surrogate oral PODs (Figure 1C). Third, we systematically applied this approach to determine a surrogate POD for each substance in the two curated data sets (Figure 1D). We then characterized the uncertainty around each of the surrogate PODs that was due to intrastudy and interstudy variability through a bootstrapping approach (Figure 1E). Finally, using the surrogate PODs and their uncertainty, we derived both probabilistic RfDs and human population effect doses (I=10%) for use in health-based or comparative risk assessments and LCIA, respectively (Figure 1F). The following sections detail each of these main steps.
Figure 1. Overview of the workflow: (A) semiautomated data curation and selection process applied to the collected in vivo data from ToxValDB; (B) collection and extrapolation of regulatory PODs; (C) analysis of the correlation between ToxValDB and regulatory POD data; (D) systematic derivation of oral PODs from the curated data sets, differentiating between general noncancer (nonreproductive/developmental) and reproductive/developmental effects; (E) quantification of the substance-specific uncertainty of the derived PODs from intra- and interstudy variability; and (F) derivation of probabilistic RfD and human population effect doses (I=10%). Note: BMDh, human equivalent benchmark dose; nchem, number of chemicals; ndata, number of data points (records); NOAEL, no observed adverse effect level; non-rep/dev, nonreproductive or developmental; POD, point of departure; rep/dev, reproductive or developmental; RfD, probabilistic reference dose; ToxValDB, Toxicity Value Database.
Figure 1A is an illustration depicting the workflow of a semiautomated data curation and selection process applied to the collected in vivo data from the Toxicity Value Database. It has two steps. Step 1: A toxicity value database with 30,654 chemicals divides 427,508 data points. It includes effect-level type, for example, no observed adverse effect level; exposure route, for example, oral; effect value and unit; study type, for example, reproductive; tested species; qualifier, for example, greater than, less than, approximately; critical effect, for example, body weight; conceptual model, for example, quantal deterministic; and extrapolation to human equivalent benchmark dose. Step 2: The curated data set includes nonreproductive or developmental effects with 8,023 chemicals and 43,528 data points, and reproductive or developmental effects with 6,697 chemicals and 46,565 data points. Figure 1B is a flowchart titled Preparing regulatory data set and has two steps. Step 1: Regulatory points of departure leads to extrapolation to human equivalent benchmark dose. Step 2: The human equivalent benchmark dose leads to nonreproductive or developmental effects with 744 chemicals and reproductive or developmental effects with 41 chemicals. Figure 1C is a set of two scatter plots titled Correlation between toxicity value database and regulatory data set, plotting log to the base 10 points of departure begin subscript regulatory, human equivalent benchmark dose end subscript (milligrams per kilogram per day), ranging from negative 4 to 4 in unit increments (y-axis) across log to the base 10 human equivalent benchmark dose (milligrams per kilogram per day), ranging from negative 4 to 4 in unit increments (x-axis). Figure 1D is an image displaying the following information: Deriving points of departure per substance including 8,023 chemicals under nonreproductive or developmental effects and 6,697 chemicals under reproductive or developmental effects. Figure 1E is a line graph titled Quantifying uncertainty, plotting percentile, ranging from 0.00 to 1.00 in increments of 0.25 (y-axis) across log to the base 10 human equivalent benchmark dose (milligrams per kilogram per day), ranging from negative 1 to 4 in unit increments (x-axis) for interstudy and intrastudy variability for estimating 95 percent confidence interval around derived points of departures. Figure 1F is an image titled Probabilistic reference doses and human effect doses and displays the following information: 10,145 chemicals under probabilistic reference doses and 10,145 chemicals under human effect doses by 10 percent.
Description of the in Vivo Input Data Set
The in vivo data were collected in March 2021 from the U.S. EPA’s ToxValDB (version 9.1), an experimental toxicity database compiled from >40 publicly available sources.13 These include—among others—the Toxicity Reference Database (ToxRefDB; version 2.0),25,26 IRIS (https://www.epa.gov/iris), Office of Pesticide Programs (OPP; https://www.epa.gov/pesticides), PPRTV (https://www.epa.gov/pprtv), European Chemicals Agency’s eChem Portal (https://www.echemportal.org/echemportal), and European Food Safety Authority’s Chemical Hazards Database (https://www.efsa.europa.eu/en/data/chemical-hazards-data). The current version of ToxValDB is accessible through the EPA’s CompTox Chemicals Dashboard (https://comptox.epa.gov/dashboard).27 The accessed database contained 427,506 records providing toxicity information on >30,000 chemicals.
Input Data Curation and Selection
We curated and selected the toxicity data from the ToxValDB with a semiautomated process based on a set of specific criteria derived from the WHO/IPCS recommendations in dose–response modeling (Figure 1A).10–12 The curation aimed first to harmonize the reported information to facilitate the data processing in our study; second, to filter out all records not relevant for our analysis (e.g., exposure route different from oral); and third, to make reported toxicity animal data directly comparable across different tested species and study types. We summarize below the steps of the curation and selection process with a few examples and actions taken (e.g., filtering, extrapolation), and Tables S1–S3 detail the process, including additional examples and further explanations of the choices made. Effect-level types: we focused on the three effect-level types used for deriving PODs (i.e., NOAELs, LOAELs, and BMDLs) and excluded all the records referring to other effect-level types. The curation included, for example, grouping effect levels reported as no effect level (NEL) and no observed effect level (NOEL) to NOAEL, or lowest effect level (LEL) and lowest observed effect level (LOEL) to LOAEL. In addition, we disregarded all records with NELs (or LELs) as effect-level types in all cases in which another record from the same study and with effect-level types equal to NOAEL (or LOAEL) was already available.
Exposure route: we focused on oral exposure as the route of interest in the present study and thus excluded all records referring to other routes. During the curation, we grouped exposure routes reported as “food,” “gavage,” “diet, unspecified,” “oral via capsule,” “drinking water,” “stomach intubation,” “oral, intragastric,” “oral, gavage,” “feed,” “diet,” “drinking water,” and “liquid diet” to oral. In cases of missing information, we assigned an exposure route as oral for those with reported units equal to milligrams per kilogram per day or equivalent.
Effect values and units: we converted reported effect values into a consistent unit of milligrams per kilogram per day. We excluded all the records with missing effect values or unconvertible and unclear units (e.g., “mg/mg3,” “ppm urine,” or “mg/kg ash femur”). Specifically for records with REACH as source and “mg/kg diet” as the reported unit of effect value, we converted the reported effect values to milligrams per kilogram per day by dividing the reported effect value by 16 if the tested species was rat and by 4.5 if the tested species was mouse. Single-dose data (acute tests, unit typically in “mg/kg”) were also excluded.
Study type: we focused on five study types: chronic, subchronic, subacute, reproductive, and developmental. Harmonization of reported study types included, for example, “fertility” being assigned to reproductive. In addition, for records with subacute or subchronic as the study type, we extrapolated their effect values to chronic by applying a subchronic-to-chronic factor of 2 and a subacute-to-chronic factor of 5.12,28 For records with reproductive or developmental as the study type, we did not apply any extrapolation because we assumed that the study covered the relevant window of susceptibility. The records with reported study type being different (and unconvertible) to one of the five considered were disregarded.
Tested species: we focused only on records providing toxicity information on mammals, excluding other species. We harmonized reported species names and grouped them into commonly tested species. For example, we grouped records with tested species reported as mice or hamster into mouse. If no tested species were reported, we flagged the record and assumed the tested species to be rat (the predominant tested species across the retrieved data). In addition, we extrapolated the effect values of all records to humans. The interspecies body weight scaling was performed by dividing reported effect values by conversion factors (CFs) to humans estimated as follows: CF=BWh0.25BWa0.25,
where BWh is the average body weight of humans of 70kg, and BWa is the body weight of the tested species. As an example, by assuming an average weight for a mouse BWa=0.025 kg, a CF=7.3 is estimated, and in case of an effect value of 10mg/kg per day, the dose tested with a mouse is converted to an effect value for humans of 1.4mg/kg per day.
Qualifiers: in cases of reported effect values accompanied by numeric qualifiers (e.g., “<,” “>,” “≥”), after analyzing the original sources for these records and based on expert judgment, we decided to disregard the presence of the numeric qualifiers except for NOAELs accompanied by “<.” The effect-level types of these “NOAEL <x” records were converted to LOAEL given that actual effects were observed at the tested dose in the original studies.
Critical effects: the reported effects studied were standardized to one of the following categories: body weight, clinical chemistry, clinical signs, development, enzyme activity, food or water consumption, gross pathology, hematology, mortality/survival, multiple, neurobehavior, none, nonneoplastic histopathology, organ weight, other, reproduction, or urinalysis. In addition, for all records for which we allocated development or reproduction as critical effect category, we cross-checked this information with the previously harmonized study type category and overwrote the study type category in case of mismatch.
Conceptual model: based on the previously assigned standardized effect categories and study types, we assigned to each data record one of the following conceptual models: continuous, quantal-deterministic, quantal-stochastic, or multiple, following the WHO/IPCS recommendations in dose–response modeling (Table S2).10–12 For example, chronic records with body weight as a standardized effect category were assigned the conceptual model continuous.
Extrapolation to benchmark dose: based on the curated effect-level type, study type, and assigned conceptual model, we extrapolated the effect value of each record to a chronic human equivalent benchmark dose (BMDh) based on the WHO/IPCS framework (Table S3).10,12 In the case of multiple possible conceptual models assigned to the same record, we calculated the BMDh value based on the averaged results of the two assigned conceptual models. At each extrapolation step to convert the in vivo data to BMDh (e.g., interspecies body weight scaling), uncertainty distributions were assigned to BMDh. Assuming lognormal distribution for each factor, the uncertainties were combined probabilistically by applying the approximation described in the latest work in dose–response modeling.10–12 The probabilistically combined uncertainties quantified a total uncertainty around each extrapolated effect value (i.e., BMDh). In Table S3, uncertainty factors are provided as the ratio between the 95th percentile and the median of the lognormal distribution (P95/P50).
After curating and applying the described semiautomated process to select the retained toxicity data from ToxValDB, we split the curated data into two distinct data sets covering general noncancer effects and reproductive/developmental effects, respectively. This repartition was performed based on each record’s derived study type and critical effects (Figure 1A).
Regulatory Data
We gathered regulatory data from a previously published database of publicly available, peer-reviewed human health toxicity values reported in specific public sources, including—among others—U.S. EPA (e.g., IRIS, OPP) and California EPA (Office of Environmental Health Hazard Assessment).8,29 We then cross-checked these values with the November 2019 release of the U.S. EPA Regional Screening Levels (RSLs), adding additional chemicals not previously identified for which PODs could be identified.30 In our study, a PODreg is defined as the NOAEL, LOAEL, or BMDL associated with a reported reference dose. To ensure a consistent comparison, we extrapolated the gathered PODreg to chronic human equivalent benchmark dose (PODreg,BMDh), applying the same procedure as described previously for the curated and selected ToxValDB records while also differentiating between general noncancer and reproductive/developmental effects (Figure 1B).
Comparison and Approach for Deriving Oral PODs
To systematically determine oral PODs for substances for which regulatory values were not available, we started by comparing the curated toxicity data from ToxValDB against the available PODreg,BMDh (Figure 1C). The comparison was carried out separately for general noncancer and reproductive/developmental effects for chemicals for which both PODreg,BMDh and in vivo data were available. For each of these substances, we assumed a lognormal distribution across BMDh and derived a POD from the x-percentile of the fitted lognormal distribution (PODpxBMDh). The resulting PODpxBMDh values were then compared against the respective PODreg,BMDh. Although the curated data from ToxValDB did not necessarily cover the exact data sets used by health risk assessors to select PODreg, the comparison between the resulting values informed us about the importance of potential differences.
We hypothesized that PODpxBMDh on the lower end of the effect values distribution was a suitable proxy for PODreg,BMDh across different chemicals.20 To evaluate this hypothesis and to identify the most suitable x-percentile, we analyzed the correlation of PODreg,BMDh values against four different PODpxBMDh values, from the 5th to the 35th percentile (i.e., PODp05BMDh, PODp15BMDh, PODp25BMDh, and PODp35BMDh). In addition, to put our approach into perspective, we investigated via the Shapiro–Wilk normality test31 whether the BMDh distribution for each chemical could be adequately fit by a lognormal distribution.
The two function moments used for fitting the lognormal distribution were mu (μ) and sigma (σ), which respectively denoted the log-scale population median and standard deviation of the available effect values for a substance.32 For all substances, μ was calculated from the available BMDh. In contrast, σ was calculated from the available BMDh only for data-rich chemicals (≥10 records available), whereas for data-poor chemicals (<10 records available), we applied a fixed standard deviation (σfixed) derived from the average across σ of data-rich chemicals. We derived two distinct σfixed, one to be applied for general noncancer effects (σfixednon-rep/dev) and one for reproductive/developmental effects (σfixedrep/dev). Given that the estimates of σ from <10 available records were highly unstable, we used an average shaped distribution instead of relying on the few available effect values. The derived x-percentile from the fitted lognormal distribution (PODpxBMDh) were expected to be more representative for the considered data-poor chemical.
In addition, to investigate the potential influence of remaining double entries (i.e., duplicate records) in the curated ToxValDB, we studied how much the surrogate PODs were affected in the case of keeping only records with unique derived BMDh values, effect-level types, and tested species.
Deriving PODs per Substance
After identifying the most suitable x-percentile to be used as a surrogate of PODreg,BMDh, we systematically derived PODpxBMDh for each substance from the available records in the two curated in vivo data sets (Figure 1D). For a substance for which records were available in both data sets, two distinct PODpxBMDh values were derived separately, one for general noncancer effects (PODpxBMDhnon-rep/dev) and one for reproductive/developmental effects (PODpxBMDhrep/dev).
Quantifying Uncertainty around the Derived PODs
To characterize the uncertainty around the derived PODpxBMDh, we took into account both interstudy variability and intrastudy variability (Figure 1E). These two aspects were quantified separately and expressed as the squared geometric standard deviation (GSDinter2 and GSDintra2) and then combined (GSDtotal2)33 to provide a 95% confidence interval (CI) for each PODpxBMDh in the two data sets: GSDtotal2=10(log10GSDinter2)2+(log10GSDintra2)2,
GSDtotal2 being a unitless factor equal to P97.5/P50 or to (P95/P50)2/1.65 and denoting that the distribution of 95% of all values fall within PODpxBMDh divided by GSDtotal2 and PODpxBMDh multiplied by GSDtotal2.
We calculated GSDinter2, which reflects the variability across available effect values, for each PODpxBMDh in the two distinct data sets. To estimate GSDinter2, we started from the lognormal distribution fitted through the available effect values (extrapolated to BMDh) for deriving PODpxBMDh. When fitting the lognormal distribution, one of the two moments used was σ (standard deviation of the available BMDh for a substance). We thus estimated the 95% CI of σ via the function fitdistr in the R package MASS,34 and from this 95% CI we derived an upper and lower bound for PODpxBMD (PODpxBMDhinter,upper and PODpxBMDhinter,lower)35 by fitting two new lognormal distributions using instead of σ its 95% CI. GSDinter2 was then calculated as: GSDinter2=PODpxBMDhinter,upper/PODpxBMDhinter,lower.
In contrast, GSDintra2 reflects the variability specific to the effect values. To estimate GSDintra2, we started from the record-specific distribution around the extrapolated effect value. This record-specific distribution was based on the uncertainty distributions assigned when converting the in vivo data to human BMDs at the following extrapolation steps: LOAEL to NOAEL, NOAEL or BMDL to BMD, subchronic/subacute to chronic, interspecies body weight scaling and, interspecies toxicokinetics (TKs) and toxicodynamics (TDs) (Table S3).10–12 The record-specific uncertainty was propagated from the available records (BMDh) to the derived PODpxBMDh via a bootstrap method. First, for each substance, 1,000 bootstrap samples were sampled from the estimated distributions around BMDh of the available records. Second, 1,000 lognormal distributions were fitted to the bootstrap samples using μ as the median of the resampled effect values, and σ as the same σ used to derive BMDh, based on the originally available effect values (in practice only μ varied and the same shaped distribution was always fitted to the resamples). Third, from the 1,000 fits, we derived an upper and lower bound for PODpxBMDh (PODpxBMDhintra,upper and PODpxBMDhintra,lower). GSDintra2 was then calculated as follows: GSDintra2=PODpxBMDhintra,upper/PODpxBMDhintra,lower.
When using the derived PODpxBMDh as a surrogate of regulatory value, it was necessary to consider the additional uncertainty associated with the prediction of this regulatory value, which was obtained from the residual standard error between PODreg,BMDh and PODpxBMDh. Because this residual standard error already accounted for the uncertainty on the PODpxBMDh for regulated chemicals, we took as GSDfinal2 the maximum between the uncertainty related to the residual standard error (GSDpx→reg2) and the substance-specific GSDtotal2 on the derived POD.
Deriving Probabilistic RfDs and Human Effect Doses
Following the automated workflow developed by Chiu et al.,12 probabilistic RfDs were derived for risk assessment purposes as the lower 95% confidence bound of HDM1%, that is, the daily human dose at which 1% of the population shows a level of effect M corresponding to the effect-level type (e.g., LOAEL, NOAEL, or BMDL) reported in the database as well as the type of end point (e.g., continuous, quantal-deterministic, or quantal-stochastic) (Figure 1F). HDM1% values were calculated from the provided PODpxBMDh by dividing it by an extrapolation factor of 9.7 (P50) to account for variability in sensitivity between the median human and the first percentile human.10 The 90% CI of HDM1% was calculated combining probabilistically GSDfinal2 and the uncertainty factor (i.e., P95/P50=4.3) assigned to the human variability at the first percentile to yield the 90% CI of GSDfinal1.65=(GSDfinal2)1.652.10 We directly implemented the approximate approach by Chiu et al.12 given that in their study it yielded results within 20–30% of the Monte Carlo simulation. We then compared the derived lower 95% confidence bound of HCM1% against the related regulatory RfD (if available) to investigate the potential influence of the database uncertainty factor (UFd). This factor accounts for data gaps and is typically equal to 1, 3, and 10 as a function of the data coverage for different end points.36 UFd was applied when deriving regulatory RfDs but it was not directly included in the WHO/IPCS framework.12 This helped us understand whether the derived toxicity values were consistent with regulatory RfDs and identify potential biases. To put the obtained results into perspective, the derived probabilistic RfDs were finally compared against the population median chemical intake rates provided by the Systematic Empirical Evaluation of Models (SEEM) meta-model.37
For LCIA purposes, we derived effect doses at which 10% of the population shows a level of effect M (HDM10%). HDM10% was derived from the provided PODpxBMDh by dividing it by 3.49 (P50) as an extrapolation factor to account for the human variability between the 50% and the 10% incidence level.10 HDM10%-related uncertainty was calculated by combining probabilistically GSDfinal2 of PODpxBMDh and the uncertainty factor assigned to the human variability at the 10th percentile, that is, P97.5/P50=2.67,10 HDM10% being also defined as ED10 by Fantke et al.1
Data Analysis
The curation and selection process of the toxicity data and all the analyses were carried out using the open source statistical software R (version 3.6.1; R Development Core Team). All figures were generated by ggplot2 package38 in R. The R code for deriving PODs from the curated data sets is available in the Supplemental Material, “R code for deriving points of departure from the curated datasets.”
Results
Curated Toxicity Test Data Sets
After the application of the semiautomated data curation and selection process, we obtained two distinct data sets, the first covering general noncancer effects composed of n=43,528 records and providing toxicity information for n=8,023 substances, the second covering reproductive/developmental effects composed of n=46,565 records for n=6,697 substances. The fraction of records excluded at each step of the curation and selection process is provided in Table S1. Table S4 presents the summary statistics of the two curated data sets, and Figure 2A,B visualizes the effect values (all extrapolated to BMDh) distribution across curated records and the underlying effect-level types and study types information. Most of the records had NOAEL as effect-level type in both data sets, 71% (n=31,082) for the general noncancer effects data set (Figure 2A) and 78% (n=36,381) for the reproductive/developmental effects data set (Figure 2B). Only a small share of the available records reported BMDL as an effect-level type (≤1%, n=581), the rest of the data being reported as LOAEL. In the general noncancer data set, 33% (n=14,605) of the records were reported as chronic, a majority of 58% (n=25,342) as subchronic, and only 8% (n=3,581) as subacute.
Figure 2. Distribution across curated records of (A) the effect values (BMDh) and the underlying effect-level and study types for the general noncancer effects data set (n=43,528) and for (B) the reproductive/developmental effects data set (n=46,565), and number of available records for each chemical in (C) the general noncancer effects data set (n=43,528) and in (D) the reproductive/developmental effects data set (n=46,565). The red dashed lines in (C) and (D) divide data-poor chemicals (<10 records available) and data-rich chemicals (≥10 records available). Corresponding numeric data for (A) and (C) are available in Excel Table S1, and for (B) and (D) in Excel Table S2. Note: BMDh, chronic human equivalent benchmark dose; BMDL, benchmark dose lower confidence limit; LOAEL, lowest observed adverse effect level; NOAEL, no observed adverse effect level.
Figures 2A and 2B are histograms, plotting number of data points, ranging from 0 to 6,000 in increments of 2,000 and 0 to 10,000 in increments of 2,500 (y-axis) across log to the base 10 chronic human equivalent benchmark dose (milligrams per kilogram per day), ranging from negative 5.0 to 5.0 in increments of 2.5 (x-axis) for Effect-level type-study type, including no observed adverse effect level–chronic, no observed adverse effect level–subchronic, no observed adverse effect level–subacute, lowest observed adverse effect level–chronic, lowest observed adverse effect level–subchronic, lowest observed adverse effect level–subacute, benchmark dose lower confidence limit–chronic, benchmark dose lower confidence limit–subchronic, and benchmark dose lower confidence limit–subacute; and Effect-level type, including no observed adverse effect level, lowest observed adverse effect level, and benchmark dose lower confidence limit. Figures 2C and 2D are histograms, plotting number of chemicals, ranging from 0 to 3,000 in increments of 1,000 and 0 to 1,200 in increments of 200 (y-axis) across number of data points per chemical, ranging from 1 to 10 in unit increments and 11 to 15, 16 to 20, 21 to 30, 31 to 40, 41 to 50, 51 to 75, and greater than 75 (x-axis).
In both data sets, BMDh ranged substantially across records by >10 orders of magnitude. More specifically, for general noncancer effects, BMDh ranged from 6×10−9 to 2.2×105 mg/kg per day, with a median value of 31mg/kg per day, and for reproductive/developmental effects from 7.3×10−10 to 2.6×105 mg/kg per day, with a median value of 100mg/kg per day. Figure 2C,D shows the number of chemicals falling within different bins of reported data points per chemical, differentiating between the two curated data sets. We observed that only a limited number of records were available for the majority of the substances. For example, 47% (n=3,169) of the chemicals in the reproductive/developmental effects data set had less than four available records (Figure 2D).
The rat was the most commonly reported tested species in both data sets, followed by the mouse. Together, these two tested species represented >80% (n=76,548) of the reported tested species across the two data sets. The third most common tested species were dog in the general noncancer effects data set and rabbit in the reproductive/developmental effects data set. The tested species was not reported for 5% (n=3,827) of the records across both data sets; we flagged these records and assumed rats as the tested species (Figure S2).
Figure 3 presents the effect values (BMDh), related effect-level types, and PODreg,BMDh (when available) for all the chemicals covered in the general noncancer effects data set (Figure 3A) and the reproductive/developmental effects data set (Figure 3B). For a given chemical, the observed variability in BMDh spanned up to 7 orders of magnitude, and across chemicals, we observed an average standard deviation of half an order of magnitude. As a general trend, we observed that PODreg,BMDh fell on the lower half of the effect values distribution across different chemicals. In addition, based on the Shapiro–Wilk tests carried out, we found that the BMDh distribution for each chemical could be adequately fit by a lognormal distribution, with p>0.05 for the majority of the chemicals in the two data sets.
Figure 3. Curated effect values (extrapolated to BMDh), their underlying effect-level types, the corresponding regulatory PODs (PODreg,BMDh) and derived PODp25BMDh (gray data points) for each of the chemicals covered by the in vivo data, differentiating between (A) general noncancer effects and (B) reproductive/developmental effects. Chemicals are ranked by derived PODs, the gray curve representing the percentage of chemicals above a certain POD. Corresponding numeric data for NOAELs, LOAELs, and BMDLs in (A) and (B) are available in Excel Table S1 and Excel Table S2, respectively; corresponding numeric data for regulatory PODs in (A) and (B) are available in Excel Table S3 and Excel Table S4, respectively; corresponding numeric data for derived PODs in (A) and (B) are available in Excel Table S5. Note: BMDh, chronic human equivalent benchmark dose; BMDL, benchmark dose lower confidence limit; LOAEL, lowest observed adverse effect level; NOAEL, no observed adverse effect level; POD, point of departure; PODp25BMDh, point of departure derived from the 25th percentile of the fitted lognormal distribution to the curated effect values extrapolated to chronic human equivalent benchmark dose; PODreg,BMDh, point of departure associated with a reported reference dose extrapolated to chronic human equivalent benchmark dose.
Figures 3A and 3B are graphs, plotting percentage of ranked chemicals, ranging from 0 to 100 percent in increments of 25 (left y-axis) and 8,023 chemicals and 6,697 chemicals (right y-axis) across log to the base 10 chronic human equivalent benchmark dose (milligrams per kilogram per day), ranging from negative 5 to 5 in unit increments (x-axis) for benchmark dose lower confidence limit, lowest observed adverse effect level, no observed adverse effect level, point of departure, and point of departure associated with a reported reference dose.
The n=90,093 curated and selected toxicity records from the ToxValDB are provided in the Supplemental Material, differentiating between the general noncancer effects data set (Excel Table S1) and the reproductive/developmental effects data set (Excel Table S2). PODreg,BMDh extrapolated to chronic human equivalent benchmark doses are available for n=744 chemicals for general noncancer effects (Excel Table S3) and for n=41 chemicals for reproductive/developmental effects (Excel Table S4).
Comparison with Regulatory Toxicity Values
To characterize the distribution of the toxicity values for data-rich chemicals with at least 10 records available, we directly used the available effect values (BMDh) to derive a chemical-specific standard deviation given that the available records were sufficient to represent and cover different potential effects. We also derived average standard deviations across data-rich chemicals of log10σfixednon-rep/dev=0.55 for general noncancer effects and log10σfixedrep/dev=0.45 for reproductive/developmental effects (Figure S3). We then applied these averages to all data-poor chemicals with <10 records, for which chemical-specific σ would not be reliable.
Using these standard deviations, we constructed lognormal distributions of BMDh and compared four different PODpxBMDh (i.e., 5th, 15th, 25th, and 35th percentiles) to the curated PODreg,BMDh values. This analysis identified the 25th percentile (i.e., PODp25BMDh) as the best approximation of PODreg,BMDh for both effect data sets (Figure S4). Figure 4 compares the estimated PODp25BMDh and the respective PODreg,BMDh for general noncancer effects (Figure 4A) and for reproductive/developmental effects (Figure 4B). For both data sets, the estimated PODp25BMDh correlated well with the available PODreg,BMDh, with a coefficient of determination R2≥0.78 and a residual standard error (RSE)≤0.53 of the log-transformed values. In addition, we investigated the few outliers present in Figure 4 and did not identify specific trends or clusters. These outliers covered a large chemical space and included metals, insecticides, and phthalate plasticizers. Hence, no chemical categories appeared to be more problematic than others.
Figure 4. Comparison between estimated PODp25BMDh and available regulatory POD values (PODreg,BMDh) for data-rich (dark green rectangle, ≥10 records available) and data-poor chemicals (light green triangle, <10 records available), differentiating between (A) general noncancer effects and (B) reproductive/developmental effects. The dashed line represents the 1:1 line, and the solid line represents the best fit. Corresponding numeric data for PODreg,BMDh in (A) and (B) are available in Excel Table S3 and Excel Table S4, respectively; corresponding numeric data for PODp25BMDh is available in Excel Table S5. Note: POD, point of departure; PODp25BMDh, point of departure derived from the 25th percentile of the fitted lognormal distribution to the curated effect values extrapolated to chronic human equivalent benchmark dose; PODreg,BMDh, point of departure associated with a reported reference dose extrapolated to chronic human equivalent benchmark dose; RSE, residual standard error.
Figures 4A and 4B are scatter plots, plotting log to the base 10 point of departure associated with a reported reference dose extrapolated to chronic human equivalent benchmark dose (milligrams per kilogram per day), ranging from negative 5 to 5 in unit increments (y-axis) across log to the base 10 point of departure derived from the 25th percentile of the fitted lognormal distribution to the curated effect values extrapolated to chronic human equivalent benchmark dose, ranging from negative 5 to 5 in unit increments (x-axis) for data records count, including less than 10 and greater than or equal to 10.
Recommended PODs
After identifying the 25th percentile of the distribution as the best approximation of PODreg,BMDh, we derived surrogate PODs (i.e., PODp25BMDh) for n=10,145 substances. More specifically, from the general noncancer effects data set, we derived surrogate PODs for n=8,023 substances, and from the reproductive/developmental effects data set, we derived surrogate PODs for n=6,697 substances. For each of the n=4,575 substances common to both data sets, two distinct surrogate PODs were thus derived, one from each data set.
Figure 3 presents the derived surrogate PODs ranked in increasing order (dark gray curve), together with the underlying BMDh, as well as the related PODreg,BMDh where available. The derived surrogate POD values range across chemical substances >8–10 orders of magnitude, from 3.1×10−6 to 1.1×104 mg/kg per day, with a median value of 22mg/kg per day for general noncancer effects, and from 2.8×10−4 to 1.3×104 mg/kg per day, with a median value of 76 mg/kg per day for reproductive/developmental effects. Examples of substances with the lowest POD estimates (i.e., highest potential toxicity) in both data sets include dioxins [e.g., 2,3,7,8-tetrachlorodibenzo-p-dioxin, Chemical Abstract Service (CAS): 1746-01-6], polychlorinated dibenzofurans (e.g., 2,3,4,7,8-pentachloro-dibenzofuran, CAS: 57117-31-4) and, heavy metals (e.g., lead, CAS: 7439-92-1). The gaps observed in Figure 3 were linked to those substances for which only one record was available with original reported effect values corresponding to standard tested dosimetry doses (e.g., 10, 100, 1,000mg/kg per day). Excel Table S5 provides all derived PODs and the number of underlying effect values.
We compared the derived POD values for the n=4,575 substances with two distinct surrogate PODs derived from each data set. As a general trend, we observed that the higher the toxicity for general noncancer, the higher for reproductive/developmental effects. However, we also observed outliers. For the same chemical, PODs covering general noncancer effects were lower (i.e., higher toxicity) than the ones covering reproductive/developmental effects by a median factor of 2. We also observed a high variability of the ratio of the two POD values across chemicals, going from a factor of 0.0003 to 4,000 (Figure S5). In addition, we investigated the potential influence of duplicate records in the curated data set when deriving POD values. With this analysis, we found that the difference of POD values was limited for the majority of the substances, with an average increase of only 7%, with no greater than a factor 2 increase for 95% (n=4,806) of substances for both general noncancer and reproductive/developmental effects (Excel Table S6).
Concerning regulatory values, PODreg,BMDh were available for both general noncancer effects and reproductive/developmental effects for 23 chemicals, of which 15 were pesticides with extensive testing requirements. The range of derived POD values for these 23 chemicals spanned almost 5 orders of magnitude (from 2.4 to 1,024mg/kg per day), but the ratios between the two POD values across chemicals were all less than 1 order of magnitude, which corresponded to the uncertainty in the POD from any one study. The results of this comparison are given in Excel Table S7.
Uncertainty Estimates for PODs
To derive a 95% CI around the derived PODs (PODp25BMDh), we first estimated the two types of uncertainty for each POD, namely GSDinter2 and GSDintra2, reflecting interstudy and intrastudy variability (Figure S6).
In the two data sets, GSDinter2 of data-rich chemicals increased with the corresponding σ of the available BMDh while decreasing with the number of data points, from a maximum value of a factor GSDinter2=3.1, down to a factor <1.2 with >100 data points (Figure S7A,B). For data-poor chemicals (<10 records available), we assigned a fixed GSDinter2=2.4, calculated as the 97.5th percentile of the estimated GSDinter2 across substances with n=10 records available, given that GSDinter2 might be unreliable and highly biased by the limited number of effect values available.
For intrastudy variability, GSDintra2 values were estimated via the 1,000-bootstrap-samples approach across PODs in the two data sets and ranged from a factor of 1.1 to a factor of 14.4 (Figure S7C,D). For chemicals with a single record available, we defined GSDintra_single2 as the upper-bound of the estimated GSDintra2 across substances with two records available, differentiating between general noncancer effects (GSDintra_single2=15) and reproductive/developmental effects (GSDintra_single2=12) (Figure S7C,D).
Finally, we combined these two uncertainties to characterize an overall substance-specific GSDtotal2 for each derived POD. Even though there was variability in GSDtotal2 across substances with the same number of records due to differences in the variability of the underlying data, this variability systematically decreased with the increase in the number of records available (Figure S7). When comparing with regulatory values, the uncertainty factor of GSDp25→reg2=101.96×0.46=8 for general noncancer and GSDp25→reg2=102.02×0.53=12 for reproductive/developmental effects were also considered to reflect the use of PODp25BMD as a suitable approximation of PODreg. We took as the final substance-specific GSDfinal2 the maximum between GSDtotal2 and GSDp25→reg2. Estimated GSDfinal2 ranged from GSDfinal2=8 up to GSDtotal2=17.2 for general noncancer effects, and up to GSDfinal2=13.9 for reproductive/developmental effects. The distributions of the resulting surrogate PODs (PODp25BMDh) with their characterized 95% CIs are displayed in Figure S8.
In addition, we investigated for which fraction of substances the available regulatory PODs (PODreg,BMDh) were falling within the 95% CI of the derived surrogate PODs to put the provided results in perspective. From this analysis, we observed that for the majority of the considered chemicals, PODreg,BMDh were well within the estimated 95% CI, which corresponded to 707 of 744 chemicals for general noncancer effects and 3 of 41 for reproductive/developmental effects (Figure S9).
Probabilistic RfDs and Human Effect Doses
Starting from the recommended PODs, we first derived probabilistic RfDs as the lower 95% confidence bound of HDM1%, using the WHO/IPCS framework. Because this framework focuses on end point–specific uncertainties and RfDs, an additional database uncertainty factor (UFd) needed to be included when deriving probabilistic RfDs comparable to and consistent with regulatory RfDs.
To derive probabilistic RfDs, the following additional UFd were thus applied: The lower 95% confidence bound of HCM1% was divided by UFd=10 for substances with very poor data availability (n≤3 records), by UFd=3 for substances with intermediary data availability (3<n<10 records), and by UFd=1 for data-rich substances (n≥10 records). For data-rich chemicals, the probabilistic RfD value was thus equal to the lower 95% confidence bound of HDM1%. The derived probabilistic RfDs showed a good correlation with the regulatory RfDs, with a R2=0.58 and RSE=0.79 evaluated on log-scale for the 1:1 line (Figure S10B). In contrast, neglecting UFd would lead to a systematic overestimation of the RfDs (Figure S10A; R2=0.54, RSE=0.82).
Derived probabilistic RfDs were on average lower than surrogate PODs by a factor of 800 and ranged across chemicals by 8–10 orders of magnitude, with a median value of 0.04mg/kg per day for general noncancer effects (Figure 5A) and 0.1mg/kg per day for reproductive/developmental effects (Figure 5B). The derived probabilistic RfDs could then be used to put exposures into perspective by comparing them with the population median chemical intake rates and their upper 95% confidence bound estimated via the SEEM meta-model. This analysis highlighted that only for n=14 chemicals, the best estimate of the median intake rates were higher than derived probabilistic RfDs. In contrast, when considering the upper 95% confidence bound, median intake rates were higher than derived probabilistic RfDs for ∼23% (n=1,127) of the substances for which SEEM intake rates were available (Figure 5), substances that might deserve further scrutiny in priority.
Figure 5. Derived probabilistic reference doses (RfD=lower 95% confidence bound of HDM1%) and population median chemical intake rates, differentiating between (A) general noncancer effects and (B) reproductive/developmental effects. Substances are ranked in increasing order based on the derived probabilistic RfDs. The upper 95% confidence bound of the SEEM Intake rates (error bars) reflects uncertainty around the population median intake rate and does not reflect population variability. Corresponding numeric data for probabilistic RfDs are available in Excel Table S5. Note: HDM1%, the daily human dose at which 1% of the population shows a level of effect M corresponding to the effect-level type reported in the database and the end point type; SEEM, Systematic Empirical Evaluation of Models.
Figures 5A and 5B are line graphs, plotting percentage of ranked chemicals, ranging from 0 to 100 percent in increments of 25 (left y-axis) and 8,023 chemicals and 6,697 chemicals (right y-axis) across log to the 10 milligrams per kilogram per day, ranging from negative 8 to 3 in unit increments (x-axis) for probabilistic reference doses and S E E M intake rate.
Second, from the surrogate PODs we derived best estimates of human population effect doses 10% (HDM10%) following the WHO/IPCS framework and the latest recommendations for deriving human dose–response factors for noncancer end points for LCIA. The derived HDM10% ranged across chemicals by 8–10 orders of magnitude, with a median value of 6.3mg/kg per day for general noncancer effects and 21.8mg/kg per day for reproductive/developmental effects. In both data sets, the characterized uncertainties of HDM10% (i.e., 95% CI) were on average equal to a factor of 14 and spanned up to a factor of 20.3 (Figure S11). Excel Table S5 provides the derived probabilistic RfDs and HDM10% with related uncertainties, and Excel Table S8 provides an example of their calculation from two records for an arbitrary substance.
Discussion
Applicability of the Derived Toxicity Values
By applying the presented semiautomated curation and extrapolation approach, we provided PODs consistent with regulatory values for >10,000 substances, substantially expanding the chemicals coverage for which toxicity values could be derived, from 744 to 8,023 chemicals for general noncancer effects, and in an even higher proportion, from 41 to 6,697 chemicals for reproductive/developmental effects. The derived probabilistic RfDs and HDM10% can be used in a variety of chemical management and exposure and impact assessment frameworks, including the evaluation of human toxicity impacts in LCIA,1,23,24 ranking and prioritization of chemicals for additional study and evaluation, chemical safety and risk management, as well as alternatives assessment for chemical substitution. By increasing the coverage of chemical substances for human toxicity effect modeling, our results also fill a critical gap in toxicity information availability highlighted, for example, in recent high-throughput exposure and risk screening studies.39–41 Indeed, even though exposure estimates were quantified for hundreds of different substances in such studies, the lack of toxicity data prevented a comprehensive risk evaluation for all the studied chemicals.
In addition, the derived PODs and related HDM10% are following the globally recommended approach for deriving health effect factors for noncancer end points by differentiating between general noncancer effects and reproductive/developmental effects, enabling us to then account for the average ∼20-fold highest severity of reproductive/development effect when evaluating disability adjusted life years.1,22 At the same time, the proposed approach is analogous to current practices of environmental risk assessment and LCIA for ecotoxicity characterization,42 where species sensitivity distributions are derived by fitting a lognormal distribution to effect values to quantify critical effect levels and related impacts in ecosystems.32
Finally, by estimating oral doses as surrogates of regulatory values also for data-poor chemicals (<10 records), our proposed approach potentially helps in reducing costs and time (as well as ethical concerns) related to using large numbers of animals to derive a complete set of toxicity studies covering different effects. Although our approach cannot substitute for the rigorous health assessments of chemicals potentially of concern, nevertheless, it might support the work of health risk assessors at multiple levels for screening purposes when a chemical of concern has not yet been thoroughly tested or reviewed.8 Most importantly, our approach provides a reliable alternative wherever regulatory toxicity values are absent but where other subacute, subchronic, or chronic toxicity data are available.
Limitations of the Proposed Approach
Our proposed approach also comes with limitations. First, we also derived PODs for many data-poor chemicals (<10 records) and so potentially missed critical effects not covered by the considered studies, thus underestimating the actual chemical toxicity. We addressed this limitation by assigning a fixed value of the standard deviation derived from the set of chemicals with sufficient reported data records to substances with <10 available effect values, thus fitting a lognormal distribution with a predefined average shape. This higher uncertainty for data-poor chemicals is reflected via the high GSDtotal2 values and related 95% CIs on the reported PODs, which are highly dependent on the number of effect values available for fitting the distribution, that is, the lower the number of records, the higher the GSDtotal2 (Figure S7). For a given chemical, the observed average standard deviation of half an order of magnitude for BMDh values could arise from different sources, including different critical effects studied, different species tested in different environmental conditions (i.e., biological variability), as well as systematic errors (e.g., measurement errors, different experimental protocols, measurement tools).3,43,44 However, given that the rat was the most commonly tested species, the observed variability was rather associated with the different effects studied in addition to the intrinsic variability in measuring toxicity effects.
Second, specifically for reproductive/developmental effects, the comparison against PODreg,BMDh was carried out for only n=41 substances. Thus, the choice of selecting the 25th percentile (PODp25BMDh) of the fitted lognormal distribution to the available effect values is less reliable than for general noncancer effects, for which regulatory data were available for n=744 chemicals. We made this choice for consistency and, considering that, we still observed a good correlation compared with the other percentiles tested. Nevertheless, increasing the pool of regulatory data covering reproductive/developmental effects is urgently needed to verify that PODp25BMDh would still be the best approximation of PODreg,BMDh also for these effects.
Third, there are possible remaining double entries (i.e., duplicate records) in the retrieved ToxValDB. More specifically, these potential double entries are due to the fact that ToxValDB is collecting experimental toxicity data from >40 publicly available sources, which in turn are also gathering experimental toxicity data from different sources, as well as running actual experimental tests. Therefore, there is the risk that in ToxValDB for the same substance, various records from different sources might be available but reporting the same results from a given experimental test. Based on testing the potential influence of keeping only records with unique derived BMDh values, effect-level types, and tested species on our results, we found that the difference in POD values was substantial only for a small fraction of substances for both general noncancer and reproductive/developmental effects. Furthermore, the presence of duplicates is already accounted for statistically through the choice of the 25th percentile to represent the surrogate POD given that this percentile was derived using data sets that may have included duplicates.
Finally, there is a limit on how accurately a toxicity value can be predicted, and this is an intrinsic limitation of our approach and of any other approach that uses reported toxicity test data as a starting point. This is because risk estimates can vary widely across regulatory settings, even for the same chemical, despite the same underlying toxicity data set and the rigorous scientific judgment involved in developing toxicity values.8,45 Nevertheless, our results suggest that using the 25th percentile (PODp25BMDh) of the fitted lognormal distribution to the available effect values for a substance is an efficient method for estimating a POD that would be selected in a regulatory context.
Future Research Needs
To further advance this effort toward using experimental animal data to derive PODs for human toxicity effects, future research needs include extending the current approach to cover additional exposure routes, such as inhalation and dermal exposure, given that we focused our work on oral toxicity owing to the higher data availability than for other exposure routes. Covering additional exposure routes is crucial, especially for exposure and impact assessment frameworks aiming at comparing chemicals impacts across exposure routes.39,46 Similarly, in our study, we differentiate between PODs for reproductive/developmental effects and general noncancer effects owing to the difference in severity of these two disease categories to affect human lifetime loss.1,22 Nevertheless, future work should focus on increasing this differentiation and providing more critical effect-specific PODs, such as endocrine disruption effects.47
Conclusions
Given the large number of new and existing substances requiring assessment, there is a pressing need for cost-effective and rapid nonanimal alternatives,48 which is in line with the need for a transition toward more sustainable chemistries49,50 through the use of novel and innovative digitalization methods.51 Such methods will facilitate a broader coverage of chemicals that can be considered in a rapid screening, quantitative assessments of chemical emissions, along with product life cycles, chemical substitution, and risk prioritization. Our proposed surrogate PODs, probabilistic RfDs, and HDM10% constitute a valuable starting point for addressing these needs for substances lacking regulatory assessments.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank Y. Emara (Technical University of Denmark) for the discussion of the method for quantifying the uncertainty around points of departure. This research was funded in part, by grants P42 ES027704 and P30 ES029067 from the National Institute of Environmental Health Sciences. This work was supported by the Global Best Practices on Emerging Chemical Policy Issues of Concern under the UN Environment’s Strategic Approach to International Chemicals Management (SAICM; GEF project 9771, grant S1-32GFL-000632), by the Safe and Efficient Chemistry by Design (SafeChem) project funded by the Swedish Foundation for Strategic Environmental Research (grant DIA 2018/11), and by the Partnership for the Assessment of Risks from Chemicals (PARC) project (grant 101057014) funded under the European Union’s Horizon Europe Research and Innovation program.
==== Refs
References
1. Fantke P, Chiu WA, Aylward L, Judson R, Huang L, Jang S, et al. 2021. Exposure and toxicity characterization of chemical emissions and chemicals in products: global recommendations and implementation in USEtox. Int J Life Cycle Assess 26 (5 ):899–915, PMID: , 10.1007/s11367-021-01889-y.34140756
2. Fantke P, Huang L, Overcash M, Griffing E, Jolliet O. 2020. Life cycle based alternatives assessment (LCAA) for chemical substitution. Green Chem 22 (18 ):6008–6024, 10.1039/D0GC01544J.
3. Pradeep P, Friedman KP, Judson R. 2020. Structure-based QSAR models to predict repeat dose toxicity points of departure. Comput Toxicol 16 :100139, PMID: , 10.1016/j.comtox.2020.100139.34017928
4. Pham LL, Watford SM, Pradeep P, Martin MT, Thomas RS, Judson RS, et al. 2020. Variability in in vivo studies: defining the upper limit of performance for predictions of systemic effect levels. Comput Toxicol 15 :100126, PMID: , 10.1016/j.comtox.2020.100126.33426408
5. Jolliet O, Huang L, Hou P, Fantke P. 2021. High throughput risk and impact screening of chemicals in consumer products. Risk Anal 41 (4 ):627–644, PMID: , 10.1111/risa.13604.33073419
6. Judson R, Richard A, Dix DJ, Houck K, Martin M, Kavlock R, et al. 2009. The toxicity data landscape for environmental chemicals. Environ Health Perspect 117 (5 ):685–695, PMID: , 10.1289/ehp.0800168.19479008
7. Wang Z, Walker GW, Muir DCG, Nagatani-Yoshida K. 2020. Toward a global understanding of chemical pollution: a first comprehensive analysis of national and regional chemical inventories. Environ Sci Technol 54 (5 ):2575–2584, PMID: , 10.1021/acs.est.9b06379.31968937
8. Wignall JA, Muratov E, Sedykh A, Guyton KZ, Tropsha A, Rusyn I, et al. 2018. Conditional toxicity value (CTV) predictor: an in silico approach for generating quantitative risk estimates for chemicals. Environ Health Perspect 126 (5 ):057008, PMID: , 10.1289/EHP2998.29847084
9. Chiu WA, Paoli GM. 2021. Recent advances in probabilistic dose–response assessment to inform risk-based decision making. Risk Anal 41 (4 ):596–609, PMID: , 10.1111/risa.13595.32966629
10. WHO/IPCS (World Health Organization and International Programme on Chemical Safety). 2018. Guidance Document on Evaluating and Expressing Uncertainty in Hazard Characterization, 2nd ed. https://apps.who.int/iris/handle/10665/259858 [accessed 17 March 2020].
11. Chiu WA, Slob W. 2015. A unified probabilistic framework for dose–response assessment of human health effects. Environ Health Perspect 123 (12 ):1241–1254, PMID: , 10.1289/ehp.1409385.26006063
12. Chiu WA, Axelrad DA, Dalaijamts C, Dockins C, Shao K, Shapiro AJ, et al. 2018. Beyond the RfD: broad application of a probabilistic approach to improve chemical dose–response assessments for noncancer effects. Environ Health Perspect 126 (6 ):067009, PMID: , 10.1289/EHP3368.29968566
13. Judson R. 2019. ToxValDB: Compiling Publicly Available In Vivo Toxicity Data. The United States Environmental Protection Agency’s Center for Computational Toxicology and Exposure. Presentation. 10.23645/epacomptox.7800653.v1.
14. Li L, Zhang Z, Men Y, Baskaran S, Sangion A, Wang S, et al. 2022. Retrieval, selection, and evaluation of chemical property data for assessments of chemical emissions, fate, hazard, exposure, and risks. ACS Environ Au 2 (5 ):376–395, 10.1021/acsenvironau.2c00010.37101455
15. Zeise L, Bois FY, Chiu WA, Hattis D, Rusyn I, Guyton KZ. 2013. Addressing human variability in next-generation human health risk assessments of environmental chemicals. Environ Health Perspect 121 (1 ):23–31, PMID: , 10.1289/ehp.1205687.23086705
16. Aurisano N, Fantke P. 2022. Semi-automated harmonization and selection of chemical data for risk and impact assessment. Chemosphere 302 :134886, PMID: , 10.1016/j.chemosphere.2022.134886.35537623
17. Aurisano N, Albizzati PF, Hauschild M, Fantke P. 2019. Extrapolation factors for characterizing freshwater ecotoxicity effects. Environ Toxicol Chem 38 (11 ):2568–2582, PMID: , 10.1002/etc.4564.31393623
18. Smith MN, Cohen Hubal EA, Faustman EM. 2020. A case study on the utility of predictive toxicology tools in alternatives assessments for hazardous chemicals in children’s consumer products. J Expo Sci Environ Epidemiol 30 (1 ):160–170, PMID: , 10.1038/s41370-019-0165-y.31501452
19. Aurisano N, Weber R, Fantke P. 2021. Enabling a circular economy for chemicals in plastics. Curr Opin Green Sustain Chem 31 :100513, 10.1016/j.cogsc.2021.100513.
20. Friedman KP, Gagne M, Loo LH, Karamertzanis P, Netzeva T, Sobanski T, et al. 2020. Utility of in vitro bioactivity as a lower bound estimate of in vivo adverse effect levels and in risk-based prioritization. Toxicol Sci 173 (1 ):202–225, PMID: , 10.1093/toxsci/kfz201.31532525
21. Fantke P, Aurisano N, Provoost J, Karamertzanis PG, Hauschild M. 2020. Toward effective use of REACH data for science and policy. Environ Int 135 :105336, PMID: , 10.1016/j.envint.2019.105336.31884133
22. Huijbregts MAJ, Rombouts LJA, Ragas AMJ, van de Meent D. 2005. Human-toxicological effect and damage factors of carcinogenic and noncarcinogenic chemicals for life cycle impact assessment. Integr Environ Assess Manag 1 (3 ):181–244, PMID: , 10.1897/2004-007R.1.16639884
23. Jolliet O, Ernstoff AS, Csiszar SA, Fantke P. 2015. Defining product intake fraction to quantify and compare exposure to consumer products. Environ Sci Technol 49 (15 ):8924–8931, PMID: , 10.1021/acs.est.5b01083.26102159
24. Fantke P, Ernstoff AS, Huang L, Csiszar SA, Jolliet O. 2016. Coupled near-field and far-field exposure assessment framework for chemicals in consumer products. Environ Int 94 :508–518, PMID: , 10.1016/j.envint.2016.06.010.27318619
25. Martin MT, Judson RS, Reif DM, Kavlock RJ, Dix DJ. 2009. Profiling chemicals based on chronic toxicity results from the U.S. EPA ToxRef Database. Environ Health Perspect 117 (3 ):392–399, PMID: , 10.1289/ehp.0800074.19337514
26. Watford S, Pham LL, Wignall J, Shin R, Martin MT, Friedman KP. 2019. ToxRefDB version 2.0: improved utility for predictive and retrospective toxicology analyses. Reprod Toxicol 89 :145–158, PMID: , 10.1016/j.reprotox.2019.07.012.31340180
27. Williams AJ, Grulke CM, Edwards J, McEachran AD, Mansouri K, Baker NC, et al. 2017. The CompTox Chemistry Dashboard: a community data resource for environmental chemistry. J Cheminform 9 (1 ):61, PMID: , 10.1186/s13321-017-0247-6.29185060
28. Guth S, Roth A, Engeli B, Lachenmeier DW, Cartus AT, Hüser S, et al. 2020. Comparison of points of departure between subchronic and chronic toxicity studies on food additives, food contaminants and natural food constituents. Food Chem Toxicol 146 :111784, PMID: , 10.1016/j.fct.2020.111784.32998026
29. Wignall JA, Shapiro AJ, Wright FA, Woodruff TJ, Chiu WA, Guyton KZ, et al. 2014. Standardizing benchmark dose calculations to improve science-based decisions in human health assessments. Environ Health Perspect 122 (5 ):499–505, PMID: , 10.1289/ehp.1307539.24569956
30. U.S. EPA (U.S. Environmental Protection Agency). 2019. Regional Screening Levels (RSLs) - Generic Tables. https://www.epa.gov/risk/regional-screening-levels-rsls-generic-tables [accessed 17 March 2020].
31. Shapiro SS, Wilk MB. 1965. An analysis of variance test for normality (complete samples). Biometrika 52 (3–4 ):591–611, 10.1093/biomet/52.3-4.591.
32. Posthuma L, van Gils J, Zijp MC, van de Meent D, de Zwartd D. 2019. Species sensitivity distributions for use in environmental protection, assessment, and management of aquatic ecosystems for 12 386 chemicals. Environ Toxicol Chem 38 (4 ):905–917, PMID: , 10.1002/etc.4373.30675920
33. Hong J, Shaked S, Rosenbaum RK, Jolliet O. 2010. Analytical uncertainty propagation in life cycle inventory and impact assessment: application to an automobile front panel. Int J Life Cycle Assess 15 (5 ):499–510, 10.1007/s11367-010-0175-4.
34. Venables WN, Ripley BD. 2002. Modern Applied Statistics with S. New York, NY: Springer.
35. Rosenbaum RK, Georgiadis S, Fantke P. 2018. Uncertainty management and sensitivity analysis. In: Life Cycle Assessment. Hauschild M, Rosenbaum R, Olsen S, eds. Cham, Switzerland: Springer.
36. California Environmental Protection Agency, Office of Environmental Health Hazard Assessment. 2008. Technical Support Document for the Derivation of Noncancer Reference Exposure Levels. Air Toxic Hot Spots, Risk Assessment Guidelines. https://oehha.ca.gov/media/downloads/crnr/noncancertsdfinal.pdf [accessed 17 March 2020].
37. Ring CL, Arnot JA, Bennett DH, Egeghy PP, Fantke P, Huang L, et al. 2019. Consensus modeling of median chemical intake for the U.S. population based on predictions of exposure pathways. Environ Sci Technol 53 (2 ):719–732, PMID: , 10.1021/acs.est.8b04056.30516957
38. Wickham H. 2016. ggplot2: Elegant Graphics for Data Analysis. New York, NY: Springer.
39. Aurisano N, Huang L, Milà I Canals L, Jolliet O, Fantke P. 2021. Chemicals of concern in plastic toys. Environ Int 146 :106194, PMID: , 10.1016/j.envint.2020.106194.33115697
40. Aurisano N, Fantke P, Huang L, Jolliet O. 2022. Estimating mouthing exposure to chemicals in children’s products. J Expo Sci Environ Epidemiol 32 (1 ):94–102, PMID: , 10.1038/s41370-021-00354-0.34188178
41. Huang L, Fantke P, Ritscher A, Jolliet O. 2022. Chemicals of concern in building materials: a high-throughput screening. J Hazard Mater 424 (pt C ):127574, PMID: , 10.1016/j.jhazmat.2021.127574.34799153
42. Fantke P, Aurisano N, Bare J, Backhaus T, Bulle C, Chapman PM, et al. 2018. Toward harmonizing ecotoxicity characterization in life cycle impact assessment. Environ Toxicol Chem 37 (12 ):2955–2971, PMID: , 10.1002/etc.4261.30178491
43. Browne P, Judson RS, Casey WM, Kleinstreuer NC, Thomas RS. 2015. Screening chemicals for estrogen receptor bioactivity using a computational model. Environ Sci Technol 49 (14 ):8804–8814, PMID: , 10.1021/acs.est.5b02641.26066997
44. Kleinstreuer NC, Ceger PC, Allen DG, Strickland J, Chang X, Hamm JT, et al. 2016. A curated database of rodent uterotrophic bioactivity. Environ Health Perspect 124 (5 ):556–562, PMID: , 10.1289/ehp.1510183.26431337
45. National Research Council. 2009. Science and Decisions: Advancing Risk Assessment. Washington, DC: National Academies Press.
46. Fantke P, Bruinen de Bruin Y, Schlüter U, Connolly A, Bessems J, Kephalopoulos S, et al. 2022. The European exposure science strategy 2020–2030. Environ Int 170 :107555, PMID: , 10.1016/j.envint.2022.107555.36244229
47. Emara Y, Fantke P, Judson R, Chang X, Pradeep P, Lehmann A, et al. 2021. Integrating endocrine-related health effects into comparative human toxicity characterization. Sci Total Environ 762 :143874, PMID: , 10.1016/j.scitotenv.2020.143874.33401053
48. Mansouri K, Karmaus AL, Fitzpatrick J, Patlewicz G, Pradeep P, Alberga D, et al. 2021. CATMoS: Collaborative Acute Toxicity Modeling Suite. Environ Health Perspect 129 (4 ):47013, PMID: , 10.1289/EHP10369.33929906
49. Fantke P, Illner N. 2019. Goods that are good enough: introducing an absolute sustainability perspective for managing chemicals in consumer products. Curr Opin Green Sustain Chem 15 :91–97, 10.1016/j.cogsc.2018.12.001.
50. Kosnik MB, Hauschild M, Fantke P. 2022. Toward assessing absolute environmental sustainability of chemical pollution. Environ Sci Technol 56 (8 ):4776–4787, PMID: , 10.1021/acs.est.1c06098.35349278
51. Fantke P, Cinquemani C, Yaseneva PD, De Mello J, Schwabe H, Ebeling B, et al. 2021. Transition to sustainable chemistry through digitalization. Chem 7 (11 ):P2866–P2882, 10.1016/j.chempr.2021.09.012.
|
PMC010xxxxxx/PMC10056314.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
36989076
EHP11016
10.1289/EHP11016
Research
Widespread Clean Cooking Fuel Scale-Up and under-5 Lower Respiratory Infection Mortality: An Ecological Analysis in Ecuador, 1990–2019
https://orcid.org/0000-0002-7736-3905
Gould Carlos F. 1 2
Bejarano M. Lorena 3
Kioumourtzoglou Marianthi-Anna 1
Lee Alison G. 4
Pillarisetti Ajay 5 6
Schlesinger Samuel B. 7
https://orcid.org/0000-0001-6979-5655
Terán Enrique 8
Valarezo Alfredo 3
Jack Darby W. 1
1 Department of Environmental Health Sciences, Columbia University Mailman School of Public Health, New York, New York, USA
2 Department of Earth System Science, Stanford University, Stanford, California, USA
3 Institute for Energy and Materials, Department of Mechanical Engineering, Universidad San Francisco de Quito, Quito, Ecuador
4 Division of Pulmonary, Critical Care and Sleep Medicine, Icahn School of Medicine at Mount Sinai, New York, New York, USA
5 Gangarosa Department of Environmental Health Science, Emory University Rollins School of Public Health, Atlanta, Georgia, USA
6 Environmental Health Sciences, University of California, Berkeley, California, USA
7 Independent Consultant, Quito, Ecuador
8 Colegio de Ciencias de la Salud, Universidad San Francisco de Quito, Quito, Ecuador
Address correspondence to Carlos F. Gould, Stanford University, Department of Earth System Science, 473 Via Ortega, Stanford, CA 94305 USA. Telephone: (812) 322-4875. Email: [email protected]
29 3 2023
3 2023
131 3 03701726 1 2022
09 1 2023
10 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Nationwide household transitions to the use of clean-burning cooking fuels are a promising pathway to reducing under-5 lower respiratory infection (LRI) mortality, the leading cause of child mortality globally, but such transitions are rare and evidence supporting an association between increased clean fuel use and improved health is limited.
Objectives:
This study aimed to investigate the association between increased primary clean cooking fuel use and under-5 LRI mortality in Ecuador between 1990 and 2019.
Methods:
We documented cooking fuel use and cause-coded child mortalities at the canton (county) level in Ecuador from 1990 to 2019 (in four periods, 1988–1992, 1999–2003, 2008–2012, and 2015–2019). We characterized the association between clean fuel use and the rate of under-5 LRI mortalities at the canton level using quasi-Poisson generalized linear and generalized additive models, accounting for potential confounding variables that characterize wealth, urbanization, and child health care and vaccination rates, as well as canton and period fixed effects. We estimated averted under-5 LRI mortalities accrued over 30 y by predicting a counterfactual count of canton-period under-5 LRI mortalities were clean fuel use to not have increased and comparing with predicted canton-period under-5 LRI mortalities from our model and observed data.
Results:
From 1990 to 2019, the proportion of households primarily using a clean cooking fuel increased from 59% to 95%, and under-5 LRI mortality fell from 28 to 7 per 100,000 under-5 population. Canton-level clean fuel use was negatively associated with under-5 LRI mortalities in linear and nonlinear models. The nonlinear association suggested a threshold at approximately 60% clean fuel use, above which there was a negative association. Increases in clean fuel use between 1990 and 2019 were associated with an estimated 7,300 averted under-5 LRI mortalities (95% confidence interval: 2,600, 12,100), accounting for nearly 20% of the declines in under-5 LRI mortality observed in Ecuador over the study period.
Discussion:
Our findings suggest that the widespread household transition from using biomass to clean-burning fuels for cooking reduced under-5 LRI mortalities in Ecuador over the last 30 y. https://doi.org/10.1289/EHP11016
Supplemental Material is available online (https://doi.org/10.1289/EHP11016).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Lower respiratory infections (LRIs) are the leading cause of death for children under 5 y old (hereafter, “under-5”) globally, with the largest burden of morbidity and mortality occurring in low- and middle-income countries (LMICs).1 The factors that contribute to LRI mortality are primarily related to poverty and include malnutrition and micronutrient deficiency; poor health care access; inadequate use of health care resources; low vaccine coverage; inadequate water, sanitation, and hygiene; and elevated air pollution exposure.2,3 Under-5 LRI incidence and mortality have declined globally over the last 30 y, with modeling studies suggesting these improvements are likely due to advances in economic welfare and changes in modifiable risk factors like air pollution, hygiene, and vaccine coverage.2 Improving the evidence base for changes in modifiable risk factors to reduce under-5 LRI incidence or mortality can help to guide investments in addressing the persistently large LRI burden of disease among children.
Exposure to household air pollution (HAP) from burning biomass inefficiently for daily cooking and heating needs is a leading environmental risk factor for under-5 mortality around the world.1,4 Although results from cookstove intervention trials have not yielded improved health in intention-to-treat analyses in large part5 (one exception being study in Guatemala where a biomass stove with a chimney reduced physician-diagnosed severe pneumonia incidence),6 they have established a robust exposure–response relationship showing that higher HAP exposure is associated with increased risk of children’s respiratory infections.7–9 One main hypothesis for the nonexistent or smaller-than-expected observed health benefits is that air pollution exposure reductions were insufficient due to a) households continued use of polluting fuels and/or b) elevated ambient concentrations from continued polluting fuel use among nonstudy households in the community or other noncooking emissions sources (e.g., trash burning, road sources, dust).5,10–12 Still, evidence from randomized controlled trials and observational studies shows that, when clean-burning fuels like gas and electricity largely displace the use of polluting fuels like firewood, dung, and charcoal, personal air pollution exposures can be dramatically reduced—and even be close—to exposures designated in health-based exposure guidelines.6,11,13–20 These studies are supported by laboratory- and field-based emissions estimates that indicate that the magnitude of health-damaging pollutants released during cooking is dramatically reduced when using gas stoves vs. biomass stoves.21,22
Existing studies also suggest that residential biomass burning contributes significantly to ambient air pollution.23–27 A recent modeling study exemplified the implications of this source apportionment, showing that completely mitigating household biomass burning in India can effectively allow for the population-weighted ambient exposure to particulate matter (PM) with aerodynamic diameter ≤2.5μm (PM2.5) to fall to the Indian annual ambient air quality standard (40 μg/m3), in the absence of other control measures.28 Therefore, widespread community transitions to clean-burning cooking fuels (CFs) may be a strategy to address elevated ambient air pollution. Nevertheless, a household that reduces its own cooking-related air pollution may still face high air pollution–related health risks if ambient air pollution concentrations remain elevated in a community. Given that the risk of key health outcomes like LRIs decline nonlinearly with lower air pollution exposure,7 persistently elevated ambient air pollution concentrations could curb the potential health benefits of household-level displacement of polluting CFs with clean-burning fuels.
Given this evidence, a central policy challenge has been facilitating situations where clean fuel adopters can drastically reduce their polluting fuel use. However, for many in LMICs, clean fuels are too costly and difficult to access to use consistently, and an estimated 2 billion people will continue to rely on biomass fuels for their household energy needs in 2030.29–32 Around the world, there are very few examples of nationwide transitions to clean CFs in recent history when reliable census and mortality data have been recorded.33 As a result, there is a paucity of evidence related to the long-term population health impacts of such clean energy transitions.
The results of policy changes in Ecuador provide a unique research opportunity to address this evidence gap. Universal direct-to-consumer government subsidies for liquified petroleum gas (LPG), first introduced in the 1970s and later increased and stabilized in 2000, have reduced the cost of LPG for household uses to approximately 10% of market price, with most households paying between USD $2.50 and USD $3.50 for a 15-kg cylinder refill, representing <1% of total monthly expenditures for the majority of households.34,35 Although in the 1970s 80% of households cooked primarily with biomass (generally firewood), by 2010 just 9% of all households and 19% of rural households reported cooking primarily with biomass fuels.36 High LPG use in Ecuador contrasts with that of neighboring countries like Peru, where 80% of rural households cooked primarily with biomass fuels in 2012.37
Whether Ecuador’s nationwide household transition from firewood to LPG for cooking has yielded health benefits remains a crucial open question. We studied the association between historical increases in clean fuel use on under-5 LRI mortality in Ecuador between 1990 and 2019 and quantified health benefits. These results have important implications for the potential health benefits of ongoing clean fuel promotion programs globally, such as those in India, Ghana, and Peru, among others.
Materials and Methods
Our analysis aimed to model the association between increased clean fuel use and under-5 LRI mortality in Ecuador at the canton-level over the past 30 y. To do this, we aggregated public use data on mortality, clean fuel use, and household and individual characteristics into 5-y periods (1988–1992, 1999–2003, 2008–2012, and 2015–2019), using national census data and regionally representative surveys that combine to provide national coverage of all cantons in the country (data sources described in Table S1 and below). These time periods center around the three most recent full national censuses (1990, 2001, 2010) and the 5 most recent years during which there have been regular surveys that can be combined to provide national coverage of all cantons in the country.
Since the first period, several cantons split into two or more cantons, largely due to population growth and political motivations. In each of these cases, no external borders were altered, so we rejoined the split cantons into the original cantons to maintain a consistent population for analysis. Therefore, although there are now 224 cantons in Ecuador, in our analysis we used 173 cantons, as there were in 1990, because we cannot allocate cantonal data from 1990 to present-day divisions. In addition, we considered four cantons to be missing data: three areas that are considered “Nondelimited zones” (those that do not belong to a province) and that do not have corresponding administrative data (equivalent to three cantons; the combined population in 2010 was approximately 4,580 children under 5 y old and 33,000 total), and one that was a part of Peru in 1990 and did not have data from that time period. Additionally, two cantons had no observed under-5 LRI mortalities throughout the entire study period and were thus dropped from the analysis. According to our socioeconomic and demographic variables (described in the “Results” section titled “Exposure–Response Relationship”), these two cantons were somewhat less populated, more rural, and less economically developed than the cantons with observed under-5 LRI mortality (Table S2). The final sample size was 167 cantons across four time periods (n=668 canton-period observations).
Mortality Data and Outcome Definition
We accessed publicly available mortality data that aim to record every individual death in Ecuador since 1990 (50,000–68,000 deaths/year), including date of birth, date of death, location of death, and sex. These data are collected from physical or digital reports of all individual deaths in Ecuador each year and are managed by the National Statistical and Census Institute (INEC). Deaths were coded according to the International Classification of Diseases (ICD). Following the Global Burden of Disease classifications,2 we defined deaths caused by LRIs in children under 5 y old using ICD-9 codes 73, 79, 466–470, 480–489, 513, and 770 (1990–1996) and ICD-10 codes A48, A70, B97, J09–J22, J85, P23, and U04 (1997–2019) (causes and distribution shown in Table S3). Although the sensitivity and specificity of ICD codes to assess LRIs in children have not been widely determined, existing evidence suggests that this combination of ICD codes has high sensitivity for detecting LRIs.38–40 Previously identified limitations of detecting LRIs using ICD codes are often due to chronic comorbidities unlikely to exist in children, suggesting higher accuracy for this study’s outcome (under-5 LRI mortality).41 Ecuador’s mortality registry, which was used in this study, is classified as “Medium-High Quality” in an evaluation of vital statistics based on completeness of death reporting, quality of death reporting, level of cause-specific detail, internal consistency, quality of age and sex reporting, and data availability and timeliness.42
Clean Cooking Fuel Data
We estimated the fraction of households in a canton using a clean CF as their primary CF (%CF) in each of our time periods using a) the national censuses collected in 1990, 2001, and 2010 covering the full Ecuadorian population and b) the “Survey of Employment, Unemployment, and Subemployment” [Encuesta de empleo, desempleo, y subempleo (ENEMDU)], a survey administered to a rotating panel of households three times per year regularly between 2015 and 2019. In both surveys, respondents were asked, “What is the primary fuel or energy source that this household uses for cooking?” (“Cuál es el principal combustible o energia que utiliza este hogar para cocinar?”). Responses to this question have no bearing on obtaining subsidies or other government benefits. Based on existing literature on HAP concentrations or personal exposures when a household relies primarily on a given CF, we coded gas, gas (tank or cylinder), centralized/piped gas, and electricity as clean fuels and all other fuels [firewood, kerosene (locally known as kerex), diesel, gasoline, agricultural residues, and animal dung] as not clean fuels.11,12,43 In the census years, we divided the number of households using a clean fuel by the total number of households responding to the question to estimate %CF at the canton level. When using the ENEMDU, we pooled all available data collected from 2015 to 2019 (14 surveys) and used the “srvyr” package in R (version 4.2.2; R Development Core Team) to leverage expansion factors provided by the Ecuadorian INEC to yield population-weighted canton-level estimates of %CF and other covariates.
Potential Confounders
The outcome in this study is aggregated counts of under-5 LRI mortalities per canton and study period; given that the unit of analysis is canton-period, potential confounders can only be those that vary from year to year and across cantons and that covary with both the outcome (count of under-5 LRI mortalities) and the exposure (%CF). We focused on the domains of urbanization, improved infrastructure, and socioeconomic development as potential drivers of both increased clean CF access and use and improved child health at the canton level. For example, these factors could improve the availability of clean CFs (i.e., LPG cylinder refill distribution networks), household economics to increase the affordability of clean CFs, and labor market participation, which in turn could increase the relative importance of using time-saving clean CFs. At the same time, these factors could increase the availability of health care resources (e.g., antenatal care, vaccines) and improve nutrition. After adjustment, we assume that variation in %CF is random with respect to other risk factors for under-5 LRI mortality, implying that we provide an unbiased estimate of the association between 5-y canton-period %CF and average under-5 LRI mortality.
We assembled a consistent set of variables from a variety of nationally representative surveys and surveys representative of cantons that intend to cover and serve as proxies for these domains, including household conditions, sociodemographics, and health care access and usage. For covariates related to urbanization, infrastructure, and socioeconomic development, we assigned estimates from the 1990, 2001, and 2010 decennial census to the 1988–1992, 1999–2003, and 2008–2012 study periods, respectively, and combined all ENEMDU surveys from 2015 to 2019 to establish canton-level estimates for the 2015–2019 study period. We combined three additional surveys, namely the Living Conditions Survey [Encuesta de condiciones de vida (ECV)], the Maternal and Child Health Survey [Encuesta demografica de salud maternal e infantil (ENDEMAIN)], and the National Survey of Health and Nutrition [Encuesta nacional de salud y nutricion (ENSANUT)], on fecundity, women’s health, and infant health and used them to estimate canton-period child vaccination status and antenatal care usage. In particular, we used ENDEMAIN 1989, ENDEMAIN 1994, and ECV 1995 for the 1988–1992 period; ECV 1998, ECV 1999, ENDEMAIN 1999, and ENDEMAIN 2004 for the 1999–2003 period; ECV 2006, ENSANUT 2012, and ECV 2014 for the 2008–2012 period; and ENSANUT 2018 for the 2015–2019 period.44–47 As with canton-period %CF, for the census-derived estimates, we simply divided the number of relevant responses by the total number of canton-period observations; for the other surveys, we combined all relevant observations and estimated, using the provided expansion factors.
Table 1 summarizes the potential confounders considered. These potential confounders included the percentage of households in a canton that were rural, percentage having available grid electrification, household building materials (e.g., the percentage of households with a dirt floor), household water and sanitation practices (e.g., the percentage of households with municipal piped water into the home), adult women’s literacy, girls’ school attendance rates, the percentage of households in which an Indigenous language was spoken, childhood vaccination rates, average age of mothers at birth, and antenatal care usage. We used principal components analysis (PCA) to separately construct indices for a) household materials; b) household hygiene and water practices; and c) under-5 vaccination rates. Canton-period indices were produced by subtracting a given canton-period estimate from the overall parameter mean, dividing by the scaling factor, multiplying by the first principal component, and then summing across all component variables. The household materials index was composed of the percentage of households whose homes use the highest-quality roof, wall, and floor materials. The household hygiene index was composed of the percentage of households with the highest-quality household water source, household toilet and solid waste disposal, household trash removal, and household exclusive shower. The vaccine index was composed of the percentage of children under 5 y old receiving the appropriate number of doses for the tuberculosis vaccine; the trivalent diphtheria, pertussis, and tetanus vaccine; the measles vaccine; and the polio vaccine. We tested the correlations between all potential confounding variables and the exposure and outcome over time and space (see Supplemental Information, Section 3).
Table 1 Descriptive statistics of cantonal under-5 lower respiratory infection mortality, clean fuel use, and covariates in Ecuador from 1988–1992 to 2015–2019.
Overall (n=676 cantons) 1988–1992 (n=169 cantons) 1999–2003 (n=169 cantons) 2008–2012 (n=169 cantons) 2015–2019 (n=169 cantons)
Under-5 lower respiratory infection mortalities [mean (SD)]a 6.18 (21.3) 12.71 (33.2) 5.59 (17.5) 3.79 (15.5) 2.48 (10.4)
Total under-5 population [mean (SD)]b 9,482 (25,733) 8,880 (23,333) 9,398 (25,397) 9,872 (27,273) 9,779 (26,958)
Under-5 lower respiratory infection mortalities, per 100,000 under-5 population [mean (SD)] 58.95 (111.70) 137.13 (184.86) 50.48 (70.41) 28.88 (33.19) 17.54 (23.59)
Proportion of households primarily using a clean-burning cooking fuel [mean (SD)]c 0.71 (0.25) 0.41 (0.18) 0.70 (0.19) 0.83 (0.14) 0.91 (0.10)
Proportion of households that are rural [mean (SD)] 0.63 (0.22) 0.68 (0.22) 0.64 (0.22) 0.62 (0.23) 0.59 (0.22)
Proportion of households not connected to electricity grid [mean (SD)] 0.15 (0.20) 0.39 (0.22) 0.19 (0.15) 0.00 (0.01) 0.03 (0.03)
Materials index [mean (SD)]d 0.00 (1.42) −0.98 (1.52) −0.37 (1.28) 0.42 (1.11) 0.93 (0.88)
Household hygiene index [mean (SD)]d 0.00 (1.85) 1.86 (1.19) 0.56 (1.18) −0.32 (1.14) −2.10 (1.13)
Proportion of adult women who are literate [mean (SD)] 0.83 (0.09) 0.79 (0.10) 0.85 (0.08) 0.88 (0.06) 0.81 (0.09)
Proportion of girls under 18 y old attending school [mean (SD)] 0.83 (0.10) 0.74 (0.06) 0.75 (0.05) 0.89 (0.03) 0.94 (0.04)
Proportion of households where an Indigenous language is spoken [mean (SD)] 0.07 (0.15) 0.06 (0.12) 0.08 (0.16) 0.08 (0.15) 0.09 (0.18)
Proportion of children under 5 with three doses of the pneumococcal conjugate vaccine [mean (SD)]d 0.22 (0.27) 0.00 (0.00) 0.00 (0.00) 0.27 (0.15) 0.60 (0.16)
Vaccine index [mean (SD)]e 0.00 (1.72) −0.32 (2.35) 0.39 (1.19) 0.74 (0.98) −0.82 (1.57)
Average age of mother at delivery in years [mean (SD)] 25.5 (1.2) 26.3 (1.0) 25.8 (0.9) 25.1 (0.8) 25.2 (1.4)
Proportion of pregnant women receiving formal antenatal care [mean (SD)] 0.86 (0.15) 0.74 (0.14) 0.82 (0.14) 0.92 (0.15) 0.95 (0.07)
Median number of antenatal care visits, if any [mean (SD)] 5.88 (1.56) 5.07 (1.43) 5.19 (1.36) 6.25 (1.43) 7.01 (1.13)
Mean ambient PM2.5 (μg/m3) [mean (SD)] 16.6 (2.8) NA 14.8 (2.1) 17.8 (2.7) 17.2 (2.7)
Note: PM, particulate matter; PM2.5, PM with aerodynamic diameter ≤2.5μm; SD, standard deviation.
a Under-5 lower respiratory mortalities represent the yearly average of the years covered in the time period (1990–1992, 1999–2003, 2008–2012, 2015–2019). Therefore, it is possible for a canton-period estimate to not be a whole number.
b We estimate under-5 population in the 1988–1992, 1999–2001, and 2008–2012 time periods by counting the number of children under age 5 y in the 1990, 2001, and 2010 censuses, respectively. Because there has not been a survey with national coverage since 2010, we rely on age-specific population estimates produced by the Ecuadorian National Statistical Agency (INEC) that are based on the most recent census, the national birth and death registries, and data on migration and immigration, among other factors. These can be found freely at https://sni.gob.ec/proyecciones-y-estudios-demograficos. We average the yearly estimates from 2015 to 2019 to produce the canton-period estimates.
c Cooking fuel options included: piped/centralized gas, gas cylinders, electricity, kerosene (locally referred to as kerex), firewood, coal, and gasoline. Clean fuel options included piped gas, gas cylinders, and electricity.
d Indices are produced using the first component from principal components analysis. Canton-period indices are produced by subtracting a given canton-period estimate from the overall parameter mean, dividing by the scaling factor, and multiplying by the first principal component. Then, all parameters are summed to produce the index. The household materials index is comprised of roof, wall, and floor materials as specified in d-f; positive values indicate higher quality materials. The household hygiene index is comprised of the household water source, household toilet and solid waste disposal, household trash removal, and household exclusive shower; more negative values indicate more hygienic practices. The vaccine index is composed of all vaccines other than the pneumococcal conjugate vaccine–3.
e The pneumococcal conjugate vaccine–3 did not exist prior to the 2010 time period in Ecuador. Given that there was no similar vaccine, we assigned a 0% coverage value to all cantons in the 1988–1992 and 1999–2003 periods. We do not have data on which of the multiple pneumococcal conjugate vaccines were administered in the 2008–2012 and 2015–2019 periods in Ecuador.
Including all available potential confounders in our model could risk multicollinearity (leading to unstable coefficient estimates based on the inclusion or exclusion of variables) or overspecification of the model, leading to inflated standard errors. We developed a parsimonious model with limited correlation between variables (Figure S1), while still retaining important potential confounders in each of the relevant domains. Namely, we did not include multiple measures of household building materials or household water and sanitation practices in our preferred models because of multicollinearity. Our preferred model adjusted for the percentage of households in a canton that were rural; percentage of households that were not electrified via the grid; an index of household materials; percentage of households with a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; percentage of adult women who were literate; percentage of girls under 18 y old who reported attending school; percentage of households in which an individual in the household or the respondent spoke an Indigenous language; an index of vaccines administered among children under 5 y old; percentage of children under 5 y old receiving three doses of the pneumococcal conjugate vaccine; percentage of women who received formal antenatal care prior to delivery; and the median number of antenatal care visits among women receiving any antenatal care. Table S4 describes potential confounders included in alternative specifications.
We observed some implausible and missing canton-period variable estimates. Given our limited number of canton-periods overall, we sought to address these cases and to include all available data to maximize our power to detect an association between %CF and under-5 LRI mortality. Some cantons were relatively small, and some surveys contained relatively few observations (like vaccinations, antenatal care, and those in the 2015–2019 period). As a result, there were some covariates for which the canton-period estimates were implausibly zero (n=4, 0.6% of all canton-period observations for all potential covariates). We approached these cases in one of two ways. When the canton-period was in 1999–2003 or 2008–2012, we linearly interpolated between the preceding and following canton-period estimates. When the canton-period was either the first or last period, we replaced the zero estimate with the closest canton-period estimate for that covariate. There were some cases in which there were no observations for a given variable in a canton-period (0.3% of all canton-period observations). These primarily occurred for questions related to vaccinations or antenatal care visits in smaller cantons. In those cases, we provided the canton with the province-level average values across the relevant period.
Statistical Analyses
First, we modeled the association between %CF and under-5 LRI mortality linearly in generalized linear models (GLMs), presenting the association as a mortality rate ratio (MRR) per 10 percentage point increase in canton-period clean fuel use. We used quasi-Poisson regression to account for overdispersion in the under-5 LRI mortality data (allowing the variance of the outcome to be greater than its mean) and included canton-level under-5 population as an offset term. Beyond the potential confounding variables, we also included canton fixed effects to control for potential unobserved spatial confounding and fixed effects for study period (1988–2002, 1999–2003, 2008–2012, and 2015–2019) to account for potential unobserved temporal confounding (i.e., trends in development not captured by the measured potential confounders). Our use of fixed effects assumed that association between %CF and under-5 LRI mortality is the same for all cantons and periods but allowed the intercepts to vary without imposing any distribution on the estimated intercepts. Models were run using the “fixest” package in R (version 4.2.2; R Development Core Team)48; standard errors were clustered at the canton level.
Next, we relaxed the assumption of linearity in the relationship between %CF and under-5 LRI mortality using generalized additive models (GAMs). This model included a penalized spline for %CF, fixed effects for canton and period, and the aforementioned potential confounding variables. The optimal number of degrees of freedom for the curve was selected using the generalized cross-validation criterion.49 We presented a canton- and period-averaged exposure–response relationship relative to the mean of %CF and estimated MRRs relative to increases of 10 percentage points at different points of the %CF distribution (i.e., from 45% to 55% and from 75% to 85%) to characterize the shape of the detected association.
Visual inspection of the nonlinear model output appeared to indicate a threshold in the relationship between %CF and under-5 LRI mortality. To further explore the possibility of such a “breakpoint,” we conducted a segmented regression using the “segmented” package in R,50 accounting for the same confounders and fixed effects for canton and period as our preferred specification. Such a breakpoint would imply a change in the magnitude of the association between clean fuel use and under-5 LRI at a certain level of %CF, which could provide a policy-relevant target for canton clean fuel penetration. Previous work hypothesized that a critical level of clean fuel adoption may be needed to result in health benefits,51 and, thus, we sought to investigate whether such a phenomenon could be observed in our empirical analysis. The segmented regression model assumes a piecewise linear relationship between %CF and under-5 LRI mortality and can detect a breakpoint in generalized linear models (GLMs). Results from the fitted GAM indicated that this assumption of linearity on either side of a threshold was reasonable. A range of initial breakpoint values were tested based on the GAM exposure–response plot. If a breakpoint were detected, the model provided the breakpoint and coefficient estimates for both sides.
Throughout this study, we considered p<0.05 as a threshold for statistical significance.
Estimating Averted under-5 LRI Mortalities from Increased Clean Fuel Use
We estimated the change in under-5 LRI mortalities over the study period attributable to changes in %CF. To enable year-by-year accrual of health benefits over the full study period, we constructed a cantonal data set for each year since 1990 by linearly interpolating %CF and all covariates between the middle years of each period (1990, 2001, 2010, and 2017); covariates in 2018 and 2019 were assigned 2017 values. Then, we used the exposure–response relationship modeled in the preferred GAM specification to predict the expected number of under-5 LRI mortalities in each canton-year based on the %CF in that canton-year (LRI%CF_current_year). We then made the same prediction based on %CF in 1990 but with the contemporaneous linearly interpolated canton-year covariates (i.e., when using 1992 under-5 LRI counts, we used 1992 covariates but 1990 %CF), providing a counterfactual estimate of the number of under-5 LRI mortalities if %CF had remained fixed at 1990 levels (LRI%CF_1990). Therefore, subtracting LRI%CF_current_year from LRI%CF_1990 estimated the averted under-5 LRI mortalities attributable to changes in %CF in that canton-year; we then summed together these canton-year estimates to yield the full number of averted under-5 LRI mortalities attributable to changes in %CF between 1990 and 2019.
We additionally estimated the total declines in under-5 LRI mortalities across the full study period to determine the proportion attributable to changes in %CF (LRI Decline%CF). To do this, we predicted yearly under-5 LRI mortalities holding all covariates and %CF fixed at their 1990 levels but retaining increases in under-5 population between 1990 and 2019 (LRI%CF_Covariates_1990). Hence, LRI Decline%CF was calculated by Equation 1: (1) LRI Decline %CF=(LRI%CF_1990 – LRI%CF_current_year)(LRI%CF_Covariates_1990 – LRI%CF_current_year).
Associations by Sex, Study Period, and Region
We conducted analyses to assess sex-, study period–, and region-specific associations. Analyses mirrored our preferred GLM and GAM specifications. In sex-stratified analyses, we grouped the counts of under-5 LRI mortalities by sex and used sex-specific under-5 population offsets per canton-period. We stratified the study sample by study period (1988–1992, 1999–2003, 2008–2012, and 2015–2019). In this analysis, we did not include fixed effects for study period or canton but did adjust for all other confounders. We used the Cochran’s Q-test to assess effect modification on the multiplicative scale.52 For region, we interacted %CF with a dummy variable for region of the country (the Amazonian region, the Andean region, and the Coastal region), retaining fixed effects for study period and canton and covariates from the preferred specification.
Clean Fuel Use and Ambient Air Pollution
We investigated the associations between canton clean CF use, ambient air pollution, and under-5 LRI mortality. First, we used ambient PM2.5 concentrations for South America derived from satellite-retrieved aerosol optical depth, chemical transport modeling, and ground-based measurements at a 0.1o×0.1o resolution (roughly 1.1km×1.1km), available since 1998, to estimate mean canton ambient PM2.5 concentrations in the three most recent study periods (1999–2003, 2008–2012, 2015–2019).53 Within canton polygons, we estimated calendar-year mean PM2.5 concentrations and then averaged across years of the period. We linearly modeled the association between canton %CF and mean canton ambient PM2.5 concentrations in an empty model with only fixed effects for canton and period and then in an adjusted model using the potential confounding variables from our primary specification, minus those related to children’s health and health care. In an additional specification, we included ambient PM2.5 concentrations in both our empty and preferred adjusted model of the association between %CF and under-5 LRI mortality.
Sensitivity Analyses and Robustness Checks
We conducted a range of additional analyses to assess the sensitivity and robustness of our results to alternative specifications. We conducted four regressions with alternative potential confounder combinations. We also tested all combinations of potential confounders in quasi-Poisson GLMs (excluding combinations that included indices and their components). In another specification, we allowed for nonlinear confounding relationships using penalized splines with two knots for covariates that indicated nonlinearities when tested one by one in adjusted models (percentage of households that are rural and percent of households with grid electricity). We also tested alternative model specifications that a) allowed an additional degree of freedom for the spline assessing the association between %CF and under-5 LRI; b) modeled the outcome in a negative binomial vs. quasi-Poisson regression; c) used random intercepts vs. fixed effects for canton to use information from both within and between cantons for estimates, whereas coefficients from the fixed effects approach are effectively within-canton estimates; d) included regional fixed effects in the main specification; e) excluded the Galapagos Islands, which may have meaningfully different health, socioeconomic, or fuel use conditions due to their isolation; f) excluded the cantons that contain Ecuador’s two most populous cities (Quito and Guayaquil) because they may be influential in the results because of higher variance in the outcome (see plot of residuals from the main model in Figure S2); and g) analyzed the association between %CF and under-5 LRI mortality at province level rather than the canton level. Finally, we assessed the possibility that incomplete mortality registry data might confound the relationship between clean CF use and under-5 LRI mortality. To do so we estimated the association between mortality registry completeness, obtained from Peralta et al.,54 and %CF at the province level from 2001–2013 in a linear model with province and period fixed effects.
Results
Study Sample Characteristics
Our data show that clean CF increased and under-5 LRI mortalities declined over the study period (Figure 1; Excel Table S1). The mean canton-level %CF in the first period (1988–1992) was 41% (median and interquartile range: 39%, 26% to 55%). In the final period (2015–2019), %CF had increased to 91% (median and interquartile range: 95%, 87% to 97%). Nationwide, %CF increased from 59% to 95% over the study period. Between 1990 and 2019, we observed 179,976 under-5 mortalities, of which 29,897 were attributable to LRIs. In the first study period (1988–1992), we observed an average of 10,962 total under-5 mortalities each year, of which 2,146 were attributable to LRIs. Between 2015 and 2019, the final study period, we observed an average of 3,941 under-5 mortalities each year, of which 401 were attributable to LRIs. We also observed improvements in cantonal wealth, sanitation, education, and health care access and usage over the study periods (Table 1; Table S5).
Figure 1. Clean fuel use, under-5 lower respiratory infection mortality, and under-5 population in 1988–1992, 1999–2003, 2008–2012, and 2015–2019. Left panel shows canton-level primary clean fuel use. Middle panel shows canton-level rates of under-5 lower respiratory infection mortality per 100,000 under-5 population. Right panel shows under-5 population. Thicker borders represent modern-day provinces, and thinner borders represent cantonal borders in 1990 (n=173). The Galapagos islands are shown in an inset in the bottom right—they are otherwise found 560 mi west of the western coast of Ecuador. Cantons not included in the analysis are shown in gray (n=4) (see Section 2). See Table 1 for summaries of period-specific data and Excel Table S1 for raw data.
Figure 1 is a set of twelve maps of Ecuador and its cantons. On the left, the four maps depict the clean fuel use in 1988 to 1992, 1999 to 2003, 2008 to 2012, and 2015 to 2019. A scale that depicts primary clean fuel use ranges from 0 to 100 percent in increments of 25. Over time, primary clean fuel use increases across the country. At the center, the four maps depict the under-5 lower respiratory infection mortality rate per 100,000 under-5 population in 1988 to 1992, 1999 to 2003, 2008 to 2012, and 2015 to 2019. A scale that depicts under-5 lower respiratory infection mortality ranges from 5 to 10 in increments of 5, 10 to 25 in increments of 15, 20 to 50 in increments of 25, 50 to 100 in increments of 50, and 100 to 250 in increments of 150. Over time, under-5 lower respiratory infection mortality rates decline across the country. On the right, the four maps depict the under-5 population in 1988 to 1992, 1999 to 2003, 2008 to 2012, and 2015 to 2019. A scale that depicts under-5 population ranges from 1000 to 10000 in increments of 9000, 10000 to 50000 in increments of 40000, and 50000 to 250000 in increments of 200000. Over time, under-5 population increases in most cantons; there is substantial heterogeneity across the countries with two cantons having more than 250,000 population but most falling between 10,000 and 50,000.
Exposure–Response Relationship
Results from our generalized linear quasi-Poisson regression models showed a negative association between %CF and under-5 LRI with a MRR estimate of 0.79 [95% confidence interval (CI): 0.72, 0.87] per 10 percentage point increase in %CF in the unadjusted models that only included fixed effects for canton and time period and 0.90 (95% CI: 0.79, 1.02) per 10 percentage point increase in %CF in the preferred adjusted specification (Figure S9). Using the same preferred adjusted specification, we also found a nonlinear relationship between %CF and under-5 LRI mortality, from which we estimated an MRR of 0.94 (95% CI: 0.86, 1.04) for an increase in %CF from 45% to 55% and an MRR of 0.82 (95% CI: 0.79, 0.85) for an increase in %CF from 75% to 85% (Figure 2; Excel Table S2). Segmented regression analyses detected a threshold of 61% clean fuel use (95% CI: 52%, 70%), with a nonstatistically significant relationship below the %CF threshold (MRR: 0.99 per 10 percentage point increase in %CF, 95% CI: 0.88, 1.10) and a significant negative relationship above the threshold (MRR: 0.81 per 10 percentage point increase in %CF, 95% CI: 0.72, 0.92). Figure S3 shows the cantons that reached 61% clean fuel use in each study period; overall, 32% of all observations are below the 61% threshold.
Figure 2. Adjusted nonlinear association between canton-level clean fuel use and under-5 LRI mortality rate. The top panel shows the MRR of under-5 LRI mortality spline response function and 95% confidence interval from the generalized additive model relative to the mean of %CF (71%), holding all else constant. Annotated MRRs estimate an increase in %CF from 45% to 55% and from 75% to 85%, respectively, holding all else constant from models like the main model, but with the lower value as the reference as opposed to the mean. The annotated threshold (dashed vertical line at 61%) is estimated from segmented regressions based on linear associations, rather than the nonlinear association shown on this plot. The bottom panel is a histogram showing the distribution of n=668 canton-period %CF estimates. This preferred specification adjusted for percent of households in a canton that are rural; percent of households that are not grid electrified; an index of household materials; household has a modern toilet connected to the municipal sewers or a septic tank, a cesspool, or a latrine; adult women’s literacy; under 18 y of age girls’ school attendance rate; an individual in the household or the respondent speaks an Indigenous language; an index of vaccines administered among children under 5 y of age; coverage of the pneumococcal conjugate vaccine (three doses) among children under 5 y old; percent of women that received formal antenatal care prior to delivery; and the median number of antenatal care visits if attended. See Excel Table S2a for effect estimates and Excel Table S2b for canton-period %CF estimates. Note: CF, cooking fuel; LRI, lower respiratory infection; MRR, mortality rate ratio.
Figure 2 is a set of one ribbon plus line graph (top) and one histogram (bottom) stacked on top of each other, with a shared x-axis. The shared x-axis is the percent of households in a canton-period that cook primarily with a clean fuel. The x-axis ranges from 20% to 100%, with increments labeled at 25%, 50%, 75%, and 100%. The bottom histogram shows the distribution of canton-period observations of clean fuel use; most observations are in the range from 50% to 100%. The ribbon and line graph show the under-5 lower respiratory infection mortality rate ratio at the given level of clean fuel use, relative to mean of clean cooking fuel use (71 percent). The ribbon, which represents the 95% confidence intervals, is wide at lower levels of clean fuel use and is much narrower above about 50% clean fuel use. The line curves upward at lower levels of clean fuel use, crossing 1 (the zero effect estimate) at about 20% clean fuel use, peaking at about 1.3 around 40% clean fuel use, and declining roughly linearly after that point. The graph is annotated to illustrate three findings. First, the mortality rate ratio for an increase in a canton-period from 45% clean fuel use to 55% is 0.94, with 95 percent confidence interval from 0.86 to 1.04. Second, the mortality rate ratio for an increase in a canton-period from 75% clean fuel use to 85% is 0.82, with 95 percent confidence interval from 0.79 to 0.85. Together, this shows the nonlinearity of the response and the narrowing of the confidence interval at higher levels. Additionally, the graph is annotated to show a vertical line at clean cooking 61 percent, which is the detected breakpoint.
Sex-, Study Period–, and Region-Specific Associations
There was no difference in the linear or nonlinear association between %CF and under-5 LRI across sex-stratified subsets (MRR for females per 10 percentage point increase in %CF: 0.90, 95% CI: 0.79, 1.02; MRR for males: 0.91, 95% CI: 0.82, 1.02; Cochran’s Q-test p=0.90) (Table S6; Figure S4). The negative association between %CF and under-5 LRI mortality was stronger in more recent periods, with no significant linear association observed in the first period (MRR 1.03; 95% CI: 0.91, 1.17) and MRRs between 0.83 and 0.65 observed in the subsequent three time periods (Cochran’s Q-test p<0.01) (Table S6; Figure S5). Given the increasing proportion of cantons reaching 60% of households primarily using a clean CF over the study periods (Figure S3), these period-specific results are generally consistent with the observed threshold effect. The similarity of nonlinear associations at high levels of %CF suggests that differences are driven in part by the range in %CF in each period (Figure S6). In the Andean and Coastal regions, we observed linear and nonlinear associations between %CF and under-5 LRI mortality that were similar to those in the main model (Table S6; Figure S6; Cochran’s Q-test p=0.86). There was no observed association between %CF and under-5 LRI mortality in the Amazonian region (Figure S6).
Estimated Averted under-5 LRI Mortalities
We estimated that increases in clean fuel use were associated with 7,343 averted under-5 LRI mortalities (95% CI: 2,555; 12,131) between the first (1988–1992) and final period (2015–2019). Increases in %CF were estimated to have averted under-5 LRI mortalities in 94% of cantons; estimates for total under-5 LRI averted were significantly different from 0 in 41% of cantons. The averted under-5 LRI mortalities attributable to increased %CF account for 19% (95% CI: 7%, 31%) of all declines in under-5 LRI mortality observed during study period (Table S7; Figure S7), with spatial heterogeneity observed (Figure 3; Excel Table S3).
Figure 3. Percentage of all reduced under-5 LRI mortality from 1990 to 2019 attributable to increased clean fuel use at the canton level. The Galapagos islands are shown in an inset in the bottom right—they are otherwise found 560 mi west of the western coast of Ecuador. Cantons not included in the analysis are shown in gray (n=4). See Excel Table S3 for raw data.
Figure 3 is a map of Ecuador that depicts the percentage of all reduced under-5 lower respiratory infection mortality from 1990 to 2019 attributable to increased clean fuel use. In the image, the map is broken into cantonal borders from the year 1990. The colors are scales such that the percentage of total mortalities averted attributable to increases clean fuel use range from 0 to greater than 30 percent in increments of 5. There is a wide range of values, with some cantons at or near 0 percent and others above 30 percent. Pooled across all cantons, the overall value we estimate is 19 percent.
Ambient Air Pollution and Clean Fuel Use
Canton average ambient PM2.5 concentrations increased in Ecuador over the final three study periods from an average of 14.8 μg/m3 in 1999–2003 to 17.2 μg/m3 in 2015–2019 (Table 1; Figure S8). An increase of 10 percentage points in canton %CF was associated with a 0.25 μg/m3 (95% CI: 0.14, 0.36) reduction and a 0.20 μg/m3 (95% CI: 0.08, 0.33) reduction in ambient PM2.5 in the empty and adjusted models, respectively. In a three-period model with only canton and period fixed effects, a 1-μg/m3 increase in canton average ambient PM2.5 was associated with an MRR of 1.29 (95% CI: 0.98, 1.69). Adding canton average ambient PM2.5 somewhat attenuated the association between %CF and under-5 LRI mortality in unadjusted and adjusted three-period models and was not itself significantly associated with under-5 LRI mortality (Figure S8). An increase of 10 percentage points in canton %CF was associated with an MRR of 0.77 (95% CI: 0.62, 0.94) in a three-period model with only canton and period fixed effects; when ambient PM2.5 was included the MRR for %CF was 0.82 (95% CI: 0.72, 0.93). In the adjusted model, an increase of 10 percentage points in canton %CF was associated with an MRR of 0.88 (95% CI: 0.72, 1.08); when ambient PM2.5 was included the MRR for %CF was 0.90 (95% CI: 0.74, 1.10).
Sensitivity Analyses and Robustness Checks
Our results were robust to alternative specifications. The linear and nonlinear associations modeled for a selection of alternative potential confounding variable specifications demonstrated consistency in the magnitude and shape of the association between %CF and under-5 LRI mortality, especially above ∼60% clean fuel use (the determined breakpoint from segmented regressions) (Figures S9 and S10). Segmented regressions applied to these alternative specifications estimated nearly identical thresholds and MRRs similar to those from the preferred specification (Table S8). Estimates of total averted under-5 LRI mortalities attributable to increased %CF between 1990 and 2019 were similar across alternative specifications, ranging from 6,236 (95% CI: 2,400; 10,072) to 9,061 (95% CI: 3,813; 14,308) deaths averted (Table S7). Canton %CF was negatively associated with under-5 LRI mortality in 99% of 73,818 models with different potential confounders, and the CI for MRRs did not cross 1.00 in 70% of models (Figure S11; Table S9). Results were robust to allowing nonlinear confounding relationships (Figure S12), to modeling the outcome as a negative binomial distribution (Figure S13), to using random intercepts rather than fixed effects for cantons (Figure S14), to including a fixed effect for region of the country (Figure S15), to omitting the Galapagos islands (Figure S16), and to omitting the cantons that contain Guayaquil and Quito (Figure S17). We also observed negative linear and nonlinear associations between %CF and under-5 LRI mortality at the province level in both unadjusted and adjusted models (Figure S18). We observed no significant association between changes in %CF and mortality registry completeness at the province level from 2001 to 2013 (Figure S19).
Discussion
Well-documented, large-scale clean household energy transitions are uncommon in the modern era. We capitalized on one such cooking energy transition to empirically estimate the health benefits of widespread clean CF adoption and use. Using publicly available mortality data, we found a robust, nonlinear association between clean fuel use and under-5 LRI mortality at the canton level over the last 30 y in Ecuador. Notably, we observed statistically significant declines in under-5 LRI mortality associated with increased clean fuel use only when >60% of households in a canton cooked primarily with a clean fuel (LPG or electricity). In total, we estimate that increased clean fuel use averted 7,340 under-5 mortalities from LRIs (95% CI: 2,560; 12,130), accounting for 19% of observed declines in under-5 LRI mortalities over the same time frame.
To date, there have been few studies estimating the health benefits of large-scale transitions to clean CF use. One study found that an 80% reduction in kerosene use, replaced with LPG, between 2008 and 2012 in Indonesia yielded a decrease of 1 percentage point in infant mortality rate.55 Other studies have estimated the theoretical cost-effectiveness of potential clean cookstove programs to improve health using existing exposure–response associations between PM2.5 exposure and health outcomes, exposure contrast scenarios between baseline traditional stove use and clean cookstove use, and underlying population demographics and disease rates.17,56–65 These theoretical studies offer general guidance by assessing potential benefits and costs of clean cooking transitions, but they have several key limitations, including: a) assumptions that personal air pollution exposure scenarios, which are typically based on few, if any, in-country measurements, are consistent across time and space; b) use of relatively fixed background disease data that fail to capture spatiotemporal trends; and c) use of modeled disease data. Considering these limitations, our study offers advancement by using observed data on household fuel choice, household economics and demographics, and cause-coded deaths over three decades to establish a context-specific empirical relationship between increased clean fuel use and child mortality.
Our observation of a threshold effect suggests that nearly complete community-wide interventions may be needed to adequately achieve personal air pollution exposure reductions that yield health benefits. One potential interpretation of the threshold is that when approximately 60% of a canton uses clean fuels, there is sufficient community adoption to decrease household contributions to community-level air pollution, such that personal exposure is meaningfully reduced by the combination of reduced exposure at both household and community scales and, thus, health benefits accrue. Previous studies in which clean cooking interventions were provided to only a few households in a community observed smaller-than-expected exposure reductions perhaps because of persistently elevated community-level air pollution concentrations, potentially due to the remainder of households in the community using biomass for their household needs.10–12,18,66 The threshold may also represent the point at which many households a) use LPG nearly exclusively, b) have sufficiently phased out traditional biomass stove use, and/or c) have attained and sustained personal exposure reductions beneficial to health. The observed threshold could also result from a pattern whereby relatively earlier clean fuel adopters were at lower risk of under-5 LRI mortality than later adopters were, and, thus, as %CF increases in cantons where >60% of households already primarily cook with a clean fuel, the benefits of clean fuel use are finally observed. Given these uncertainties and the relatively few observations below the threshold in our own data, it is possible that other locations or time periods would either have a threshold effect at a different level of clean fuel adoption and use or have no threshold effect at all.
Our investigation of ambient PM2.5 levels is noteworthy for two reasons. First, the association we found persists—and is only slightly attenuated—after accounting for ambient PM2.5 concentrations, suggesting a significant independent relationship between clean fuel use and under-5 LRI mortality at the canton level. Second, the negative association between clean fuel use and ambient PM2.5 concentrations provides suggestive evidence that household biomass burning for cooking and heating contributes to ambient air pollution, extending previous findings that have primarily been based on emissions inventories and chemical transport modeling or highly localized air pollution measurement studies.24,67–69 Although modeled ambient PM2.5 concentrations have increased in Ecuador by ∼2.5 μg/m3 between 1999 and 2019 (Table 1), we estimate that increased clean fuel use has been associated with a reduction in ambient PM2.5 concentrations of ∼0.5 μg/m3 over the same time period. This association represents an unaccounted-for externality of widespread clean fuel scale-up and could imply additional benefits for investments in expanding the use of clean CFs.
In this study we consider gas to be a clean CF; however, evidence suggests that cooking with gas can still increase indoor air pollution, especially nitrogen dioxide (NO2).70,71 Given that elevated NO2 concentrations are associated with negative respiratory health outcomes,72 it is worth considering the potential limitations of a transition from biomass to gas. Still, in comparison with cooking with polluting fuels like firewood and even considering the emissions from gas cooking, transitioning to gas and phasing out polluting biomass fuels is likely to be beneficial, with studies observing reduced levels of PM2.5, carbon monoxide, and NO2 in such transitions.12,18,19,73 However, to our knowledge, there have been no studies directly comparing transitions from polluting biomass fuels to gas vs. electric cooking, which can be considered a cleaner alternative to gas because it produces no emissions at the point of use.
Limitations
This study relies on publicly available administrative data for all analyses. Although such data were not collected with the intention of being used for epidemiological analyses and have less precision than other sources of prospectively collected data that more directly measure outcomes, the findings from this study suggest that they may have high utility for retrospective analyses of countrywide changes in health and indicators of environmental exposures in countries with consistent and extensive administrative data collection mechanisms. Taking advantage of such data, which were previously collected, validated, and repeated throughout time, facilitates analyses of the type performed here. A related factor is that we lack direct measures of economic indicators (e.g., canton-level GDP, percentage of residents living below the poverty line), which have not been collected in a manner that facilitates canton-level estimates in each of our study periods using publicly available data.
A key limitation of our study is the lack of individual-level data on CF status and other risk factors that can be matched with available mortality data, thus limiting us to an ecological analysis. Therefore, it is important to consider that although a transition from using polluting fuels for cooking to clean fuels is a household-level change, in this study we are conducting an area-level analysis, and thus we can make no inference about the individual household-level impacts of such a transition on under-5 LRI mortality risk.
An additional limitation of our analysis is that we lack data on secondary CF use. Existing evidence suggests that fuel stacking (i.e., the use of multiple fuel types to meet all cooking needs) is common, especially when a clean CF has been recently acquired. Although there are no nationwide CF stacking data in Ecuador, our previous work indicates that biomass use secondary to LPG may be common in rural Andean and Coastal regions.35 Nevertheless, we also found a high contrast in air pollution exposure dependent on whether the household primarily cooked with a clean fuel or firewood, suggesting substantial health-risk reduction when LPG is used primarily instead of firewood.74
The lack of publicly available mortality data prior to 1990 limited our analysis to a time period when Ecuador’s population had already begun to transition toward use of clean fuel. Fewer canton-period observations at the lowest ends of clean fuel use have resulted in wide CIs for the association between %CF and under-5 LRI mortality rate. It is plausible that the lack of data at the lowest ends of %CF reduced our power to detect an association. Thus, our analysis was limited in its ability to fully capture the health benefits of the Government of Ecuador’s investment in cooking gas subsidies because increases in clean fuel use had already occurred by 1990.
Although the mortality registry intends to capture all deaths in Ecuador, there may be some data missingness. Nevertheless, the mortality registry broadly agrees with Global Burden of Diseases, Injuries, and Risk Factors (GBD) estimates, which aim to estimate true morbidity and mortality by statistically correcting for reporting errors and biases.2 For example, the GBD estimates that in 1990 there were 2,223 under-5 LRI mortalities in comparison with 2,250 under-5 LRI mortalities observed in our data and 795 under-5 LRI mortalities in 2017 in comparison with 405 observed in our data in 2017.1 Given our use of canton and period fixed effects, incomplete mortality records could present a problem in our estimated association only if reporting differences covaried over time and across cantons with changes in clean fuel use; however, we found no association between changes in a measure of mortality registry completeness and %CF at the province level from 2001 to 2013. In any case, our estimates would be biased toward the null in the event that underreporting of under-5 LRI mortalities is not associated with our exposure.
Although under-5 LRI mortality is the leading cause of child mortality in Ecuador, the median number of cases per year in each canton-period was 1.3, and 48% of observations had ≤1 case per year. This relatively low sample size and relatively low variation in the outcome may lead to wide CIs and limit our ability to detect an association between %CF and under-5 LRI mortality. Nevertheless, we observe a consistent negative association between %CF and under-5 LRI mortality across a range of specifications, including when aggregating cantons to provinces.
Conclusions
Modeled evidence from global burden of disease studies suggests that reduced HAP has been the leading contributing factor to recent observed declines in under-5 LRI mortality worldwide. Existing evidence suggests that transitions to clean-burning CFs that reduce air pollution exposure could significantly reduce under-5 mortality, but real-world evidence estimating the impacts of such transitions is limited. Nevertheless, clean CFs are being adopted by biomass-using households around the world because of widespread programmatic efforts by governments and other organizations. Using data on mortality and CF use across 30 y of clean fuel scale-up in Ecuador, our results, despite the limitations of ecological studies, provide among the first empirical observations of the benefits of increased clean CF use at a nationwide scale over several decades. These findings are relevant to other regions with similarly increased clean CF use and to regions that are currently developing and implementing large clean cooking policies. Providing estimates of child health improvement from these transitions may inspire greater evidence-based investment in clean CFs.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors gratefully acknowledge support from the United States National Institute of Environmental Health Sciences (NIEHS) P30 ES009089, F31 ES031833 to C.F.G., and R01 ES028805, R01 ES030616, and R01 ES029943 to M.A.K. The authors also acknowledge support from the National Heart, Lung, and Blood Institute K23 HL135349 to A.G.L.
All data that support the findings presented in this article are publicly available from the Instituto Nacional de Estadistica y Censos (https://www.ecuadorencifras.gob.ec/estadisticas/). Code and data necessary to replicate the findings presented in this article are available at https://doi.org/10.7910/DVN/6XYZLM.
C.F.G. worked on conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, visualization, and manuscript production (original draft writing, review, and editing). M.L.B. reviewed and edited the manuscript. M.A.K. conducted funding acquisition, methodology, and manuscript writing, review, and editing. A.G.L. conducted funding acquisition and worked on manuscript writing, review, and editing. A.P. worked on funding acquisition; methodology; and writing, review, and editing. S.B.S. performed data curation; investigation; and manuscript writing, review, and editing. E.T. worked on funding acquisition and manuscript preparation (writing, review, and editing). A.V. performed manuscript review and editing. D.W.J. worked on conceptualization, funding acquisition, investigation, methodology, and manuscript writing, review, and editing.
Middle authors are listed alphabetically.
==== Refs
References
1. GBD 2019 Risk Factors Collaborators. 2020. Global burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 396 (10258 ):1223–1249, PMID: , 10.1016/S0140-6736(20)30752-2.33069327
2. GBD 2016 Lower Respiratory Infections Collaborators. 2018. Estimates of the global, regional, and national morbidity, mortality, and aetiologies of lower respiratory infections in 195 countries, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet Infect Dis 18 (11 ):1191–1210, PMID: , 10.1016/S1473-3099(18)30310-4.30243584
3. GBD 2016 Lower Respiratory Infections Collaborators. 2020. Quantifying risks and interventions that have affected the burden of lower respiratory infections among children younger than 5 years: an analysis for the Global Burden of Disease Study 2017. Lancet Infect Dis 20 (1 ):60–79, PMID: , 10.1016/S1473-3099(19)30410-4.31678026
4. Lee KK, Bing R, Kiang J, Bashir S, Spath N, Stelzle D, et al. 2020. Adverse health effects associated with household air pollution: a systematic review, meta-analysis, and burden estimation study. Lancet Glob Health 8 (11 ):e1427–e1434, PMID: , 10.1016/S2214-109X(20)30343-0.33069303
5. Mortimer K, Balmes JR. 2018. Cookstove trials and tribulations: what is needed to decrease the burden of household air pollution? Ann Am Thorac Soc 15 (5 ):539–541, PMID: , 10.1513/AnnalsATS.201710-831GH.29466681
6. Smith KR, McCracken JP, Weber MW, Hubbard A, Jenny A, Thompson LM, et al. 2011. Effect of reduction in household air pollution on childhood pneumonia in Guatemala (RESPIRE): a randomised controlled trial. Lancet 378 (9804 ):1717–1726, PMID: , 10.1016/S0140-6736(11)60921-5.22078686
7. Burnett RT, Pope CA, Ezzati M, Olives C, Lim SS, Mehta S, et al. 2014. An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure. Environ Health Perspect 122 (4 ):397–403, PMID: , 10.1289/ehp.1307049.24518036
8. Burnett R, Chen H, Szyszkowicz M, Fann N, Hubbell B, Pope CA, et al. 2018. Global estimates of mortality associated with long-term exposure to outdoor fine particulate matter. Proc Natl Acad Sci USA 115 (38 ):9592–9597, PMID: , 10.1073/pnas.1803222115.30181279
9. Keller JP, Katz J, Pokhrel AK, Bates MN, Tielsch J, Zeger SL. 2020. A hierarchical model for estimating the exposure-response curve by combining multiple studies of acute lower respiratory infections in children and household fine particulate matter air pollution. Environ Epidemiol 4 (6 ):e119, PMID: , 10.1097/EE9.0000000000000119.33778354
10. Kaali S, Jack D, Lee AG. 2020. The conundrum of cleaner cookstove interventions: necessary but insufficient? Am J Respir Crit Care Med 203 (11 ):1336–1338, PMID: , 10.1164/rccm.202012-4353ED.33357118
11. Pope D, Bruce N, Dherani M, Jagoe K, Rehfuess E. 2017. Real-life effectiveness of ‘improved’ stoves and clean fuels in reducing PM2.5 and CO: systematic review and meta-analysis. Environ Int 101 :7–18, PMID: , 10.1016/j.envint.2017.01.012.28285622
12. Pope D, Johnson M, Fleeman N, Jagoe K, Duarte R, Maden M, et al. 2021. Are cleaner cooking solutions clean enough? A systematic review and meta-analysis of particulate and carbon monoxide concentrations and exposures. Environ Res Lett 16 (8 ):083002, 10.1088/1748-9326/ac13ec.
13. Dherani M, Pope D, Mascarenhas M, Smith KR, Weber M, Bruce N. 2008. Indoor air pollution from unprocessed solid fuel use and pneumonia risk in children aged under five years: a systematic review and meta-analysis. Bull World Health Organ 86 (5 ):390–398C, PMID: , 10.2471/blt.07.044529.18545742
14. Kinney PL, Asante K-P, Lee AG, Ae-Ngibise KA, Burkart K, Boamah-Kaali E, et al. 2021. Prenatal and postnatal household air pollution exposures and pneumonia risk. Chest 160 (5 ):1634–1644, PMID: , 10.1016/j.chest.2021.06.080.34298005
15. Po JYT, FitzGerald JM, Carlsten C. 2011. Respiratory disease associated with solid biomass fuel exposure in rural women and children: systematic review and meta-analysis. Thorax 66 (3 ):232–239, PMID: , 10.1136/thx.2010.147884.21248322
16. Quinn AK, Adjei IA, Ae-Ngibise KA, Agyei O, Boamah-Kaali EA, Burkart K, et al. 2021. Prenatal household air pollutant exposure is associated with reduced size and gestational age at birth among a cohort of Ghanaian infants. Environ Int 155 :106659, PMID: , 10.1016/j.envint.2021.106659.34134048
17. Steenland K, Pillarisetti A, Kirby M, Peel J, Clark M, Checkley W, et al. 2018. Modeling the potential health benefits of lower household air pollution after a hypothetical liquified petroleum gas (LPG) cookstove intervention. Environ Int 111 :71–79, PMID: , 10.1016/j.envint.2017.11.018.29182949
18. Chillrud SN, Ae-Ngibise KA, Gould CF, Owusu-Agyei S, Mujtaba M, Manu G, et al. 2021. The effect of clean cooking interventions on mother and child personal exposure to air pollution: results from the Ghana Randomized Air Pollution and Health Study (GRAPHS). J Expo Sci Environ Epidemiol 31 (4 ):683–698, PMID: , 10.1038/s41370-021-00309-5.33654272
19. Liao J, Kirby M, Pillarisetti A, Piedrahita R, Balakrishnan K, Sambandam S, et al. 2021. LPG stove and fuel intervention among pregnant women reduce fine particle air pollution exposures in three countries: pilot results from the HAPIN trial. Environ Pollut 291 :118198, PMID: , 10.1016/j.envpol.2021.118198.34740288
20. Islam MM, Wathore R, Zerriffi H, Marshall JD, Bailis R, Grieshop AP. 2022. Assessing the effects of stove use patterns and kitchen chimneys on indoor air quality during a multiyear cookstove randomized control trial in rural India. Environ Sci Technol 56 (12 ):8326–8337, PMID: , 10.1021/acs.est.1c07571.35561333
21. Shen G, Hays MD, Smith KR, Williams C, Faircloth JW, Jetter JJ. 2018. Evaluating the performance of household liquefied petroleum gas cookstoves. Environ Sci Technol 52 (2 ):904–915, PMID: , 10.1021/acs.est.7b05155.29244944
22. Islam MM, Wathore R, Zerriffi H, Marshall JD, Bailis R, Grieshop AP. 2021. In-use emissions from biomass and LPG stoves measured during a large, multi-year cookstove intervention study in rural India. Sci Total Environ 758 :143698, PMID: , 10.1016/j.scitotenv.2020.143698.33321364
23. Chowdhury S, Chafe ZA, Pillarisetti A, Lelieveld J, Dey S. 2019. The Contribution of Household Fuels to Ambient Air Pollution in India. https://static1.squarespace.com/static/59f2a2038a02c746600f6bbb/t/5cfe7c5e8ce3aa0001af5cae/1560181855319/HAP-AAP+CCAPC+brief-May+2019+%281%29.pdf [accessed 19 June 2019].
24. Conibear L, Butt EW, Knote C, Arnold SR, Spracklen DV. 2018. Residential energy use emissions dominate health impacts from exposure to ambient particulate matter in India. Nat Commun 9 (1 ):617, PMID: , 10.1038/s41467-018-02986-7.29434294
25. Health Effects Institute, GBD MAPS Working Group. 2018. Burden of Disease Attributable to Major Air Pollution Sources in India. https://www.healtheffects.org/publication/gbd-air-pollution-india [accessed 20 October 2020].
26. Guo H, Kota SH, Chen K, Sahu SK, Hu J, Ying Q, et al. 2018. Source contributions and potential reductions to health effects of particulate matter in India. Atmos Chem Phys 18 (20 ):15219–15229, 10.5194/acp-18-15219-2018.
27. Lelieveld J, Evans JS, Fnais M, Giannadaki D, Pozzer A. 2015. The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature 525 (7569 ):367–371, PMID: , 10.1038/nature15371.26381985
28. Chowdhury S, Dey S, Guttikunda S, Pillarisetti A, Smith KR, Di Girolamo L. 2019. Indian annual ambient air quality standard is achievable by completely mitigating emissions from household sources. Proc Natl Acad Sci USA 116 (22 ):10711–10716, PMID: , 10.1073/pnas.1900888116.30988190
29. International Energy Agency. 2020. SDG7: Data and Projections. https://www.iea.org/reports/sdg7-data-and-projections/access-to-clean-cooking [accessed 16 April 2021].
30. Quinn A, Bruce N, Puzzolo E, Dickinson K, Sturke R, Jack DW, et al. 2018. An analysis of efforts to scale up clean household energy for cooking around the world. Energy Sustain Dev 46 :1–10, PMID: , 10.1016/j.esd.2018.06.011.30886466
31. Shankar AV, Quinn AK, Dickinson KL, Williams KN, Masera O, Charron D, et al. 2020. Everybody stacks: lessons from household energy case studies to inform design principles for clean energy transitions. Energy Policy 141 :111468, PMID: , 10.1016/j.enpol.2020.111468.32476710
32. Troncoso K, Soares da Silva A. 2017. LPG fuel subsidies in Latin America and the use of solid fuels to cook. Energy Policy 107 :188–196, 10.1016/j.enpol.2017.04.046.
33. Guta D, Baumgartner J, Jack D, Carter E, Shen G, Orgill-Meyer J, et al. 2022. A systematic review of household energy transition in low and middle income countries. Energy Res Soc Sci 86 :102463, 10.1016/j.erss.2021.102463.
34. Gould CF, Schlesinger S, Toasa AO, Thurber M, Waters WF, Graham JP, et al. 2018. Government policy, clean fuel access, and persistent fuel stacking in Ecuador. Energy Sustain Dev 46 :111–122, PMID: , 10.1016/j.esd.2018.05.009.30294142
35. Gould CF, Schlesinger SB, Molina E, Bejarano ML, Valarezo A, Jack DW. 2020. Household fuel mixes in peri-urban and rural Ecuador: explaining the context of LPG, patterns of continued firewood use, and the challenges of induction cooking. Energy Policy 136 :111053, PMID: , 10.1016/j.enpol.2019.111053.32675905
36. Instituto Nacional de Estadística y Censos. 2017. Decennial Census. https://anda.inec.gob.ec/anda/index.php/catalog [accessed 10 May 2018].
37. Pollard SL, Williams KN, O’Brien CJ, Winiker A, Puzzolo E, Kephart JL, et al. 2018. An evaluation of the fondo de inclusión social energético program to promote access to liquefied petroleum gas in Peru. Energy Sustain Dev 46 :82–93, PMID: , 10.1016/j.esd.2018.06.001.30364502
38. Aronsky D, Haug PJ, Lagor C, Dean NC. 2005. Accuracy of administrative data for identifying patients with pneumonia. Am J Med Qual off Qual 20 (6 ):319–328, PMID: , 10.1177/1062860605280358.16280395
39. Guevara RE, Butler JC, Marston BJ, Plouffe JF, File TM, Breiman RF. 1999. Accuracy of ICD-9-CM codes in detecting community-acquired pneumococcal pneumonia for incidence and vaccine efficacy studies. Am J Epidemiol 149 (3 ):282–289, PMID: , 10.1093/oxfordjournals.aje.a009804.9927225
40. Watzlaf VJM, Garvin JH, Moeini S, Anania-Firouzan P. 2007. The effectiveness of ICD-10-CM in capturing public health diseases. Perspect Health Inf Manag 4 :6, PMID: .18066356
41. Anderson RN, Miniño AM, Hoyert DL, Rosenberg HM. 2001. Comparability of cause of death between ICD-9 and ICD-10: preliminary estimates. Natl Vital Stat Rep 49 (2 ):1–32, PMID: .11381674
42. Mikkelsen L, Phillips DE, AbouZahr C, Setel PW, de Savigny D, Lozano R, et al. 2015. A global assessment of civil registration and vital statistics systems: monitoring data quality and progress. Lancet 386 (10001 ):1395–1406, PMID: , 10.1016/S0140-6736(15)60171-4.25971218
43. Shupler M, Hystad P, Birch A, Miller-Lionberg D, Jeronimo M, Arku RE, et al. 2020. Household and personal air pollution exposure measurements from 120 communities in eight countries: results from the PURE-AIR study. Lancet Planet Health 4 (10 ):e451–e462, PMID: , 10.1016/S2542-5196(20)30197-2.33038319
44. Instituto Nacional de Estadística y Censos. Encuesta de Condiciones de Vida. https://www.ecuadorencifras.gob.ec/encuesta-de-condiciones-de-vida-ecv/ [accessed 7 January 2023].
45. Instituto Nacional de Estadística y Censos. ENEMDU Anual: Encuesta Nacional de Empleo, Desempleo y Subempleo. https://www.ecuadorencifras.gob.ec/enemdu-anual/ [accessed 7 January 2023].
46. Instituto Nacional de Estadística y Censos. Encuesta Nacional de Salud, Salud Reproductiva y Nutrición. https://www.ecuadorencifras.gob.ec/encuesta-nacional-de-salud-salud-reproductiva-y-nutricion-ensanut-2012/ [accessed 7 January 2023].
47. Instituto Nacional de Estadistica y Censos. Encuesta de Salud Materna e Infantil. https://catalog.ihsn.org/catalog/1880 [accessed 7 January 2023].
48. Bergé L. 2018. Efficient Estimation of Maximum Likelihood Models with Multiple Fixed-Effects: The R Package FENmlm. https://www.semanticscholar.org/paper/Efficient-estimation-of-maximum-likelihood-models-R-Berg%C3%A9/c620e4c12c6d58d1244db85a825e55f339754089 [accessed 17 January 2022].
49. Wood SN. 2017. Generalized Additive Models: An Introduction with R. 2nd ed. Boca Raton, FL: Chapman and Hall/CRC.
50. Muggeo VMR. 2008. Segmented: an R package to fit regression models with broken-line relationships. R News 8/1 :20–25.
51. Desai MA. 2016. Multiscale Drivers of Global Environmental Health. https://escholarship.org/uc/item/4337q6b1 [accessed 26 May 2021].
52. Kaufman JS, MacLehose RF. 2013. Which of these things is not like the others? Cancer 119 (24 ):4216–4222, PMID: , 10.1002/cncr.28359.24022386
53. van Donkelaar A, Hammer MS, Bindle L, Brauer M, Brook JR, Garay MJ, et al. 2021. Monthly global estimates of fine particulate matter and their uncertainty. Environ Sci Technol 55 (22 ):15287–15300, PMID: , 10.1021/acs.est.1c05309.34724610
54. Peralta A, Benach J, Borrell C, Espinel-Flores V, Cash-Gibson L, Queiroz BL, et al. 2019. Evaluation of the mortality registry in Ecuador (2001–2013)–social and geographical inequalities in completeness and quality. Population Health Metrics 17 :1–2, 10.1186/s12963-019-0183-y.30654828
55. Imelda. 2018. Indoor air pollution and infant mortality: a new approach. Am Econ Assoc Pap Proc 108 :416–421, 10.1257/pandp.20181119.
56. Aunan K, Alnes LWH, Berger J, Dong Z, Ma L, Mestl HES, et al. 2013. Upgrading to cleaner household stoves and reducing chronic obstructive pulmonary disease among women in rural China—a cost-benefit analysis. Energy Sustain Dev 17 (5 ):489–496, 10.1016/j.esd.2013.06.002.
57. Dagnachew AG, Hof AF, Lucas PL, van Vuuren DP. 2020. Scenario analysis for promoting clean cooking in sub-Saharan Africa: costs and benefits. Energy 192 :116641, 10.1016/j.energy.2019.116641.
58. Hutton G, Rehfuess E, Tediosi F. 2007. Evaluation of the costs and benefits of interventions to reduce indoor air pollution. Energy Sustain Dev 11 (4 ):34–43, 10.1016/S0973-0826(08)60408-1.
59. Jeuland M, Pattanayak SK, Bluffstone R. 2015. The economics of household air pollution. Annu Rev Resour Econ 7 (1 ):81–108, 10.1146/annurev-resource-100814-125048.
60. Jeuland M, Tan Soo JS, Shindell D. 2018. The need for policies to reduce the costs of cleaner cooking in low income settings: implications from systematic analysis of costs and benefits. Energy Policy 121 :275–285, 10.1016/j.enpol.2018.06.031.
61. Lacey FG, Henze DK, Lee CJ, Donkelaar AV, Martin RV. 2017. Transient climate and ambient health impacts due to national solid fuel cookstove emissions. Proc Natl Acad Sci USA 114 (6 ):1269–1274, PMID: , 10.1073/pnas.1612430114.28115698
62. Malla MB, Bruce N, Bates E, Rehfuess E. 2011. Applying global cost-benefit analysis methods to indoor air pollution mitigation interventions in Nepal, Kenya and Sudan: insights and challenges. Energy Policy 39 (12 ):7518–7529, 10.1016/j.enpol.2011.06.031.
63. Mehta S, Shahpar C. 2004. The health benefits of interventions to reduce indoor air pollution from solid fuel use: a cost-effectiveness analysis. Energy Sustain Dev 8 (3 ):53–59, 10.1016/S0973-0826(08)60466-4.
64. Pillarisetti A, Jamison DT, Smith KR. 2017. Household Energy Interventions and Health and Finances in Haryana, India: An Extended Cost-Effectiveness Analysis. In: Disease Control Priorities, Third Edition (Volume 7): Injury Prevention and Environmental Health. Mock CN, Nugent R, Kobusingye O, Smith KR, eds. Washington, DC: World Bank, 223–237, 10.1596/978-1-4648-0522-6_ch12.
65. Rosenthal J, Quinn A, Grieshop AP, Pillarisetti A, Glass RI. 2018. Clean cooking and the SDGs: integrated analytical approaches to guide energy interventions for health and environment goals. Energy Sustain Dev 42 :152–159, PMID: , 10.1016/j.esd.2017.11.003.29861575
66. Quansah R, Semple S, Ochieng CA, Juvekar S, Armah FA, Luginaah I, et al. 2017. Effectiveness of interventions to reduce household air pollution and/or improve health in homes using solid fuel in low-and-Middle income countries: a systematic review and meta-analysis. Environ Int 103 :73–90, PMID: , 10.1016/j.envint.2017.03.010.28341576
67. Apte JS, Pant P. 2019. Toward cleaner air for a billion Indians. Proc Natl Acad Sci USA 116 (22 ):10614–10616, PMID: , 10.1073/pnas.1905458116.31101719
68. Rooney B, Zhao R, Wang Y, Bates KH, Pillarisetti A, Sharma S, et al. 2019. Impacts of household sources on air pollution at village and regional scales in India. Atmospheric Chem Phys 19 (11 ):7719–7742, 10.5194/acp-19-7719-2019.
69. Fleming LT, Weltman R, Yadav A, Edwards RD, Arora NK, Pillarisetti A, et al. 2018. Emissions from village cookstoves in Haryana, India, and their potential impacts on air quality. Atmospheric Chem Phys 18 (20 ):15169–15182, 10.5194/acp-18-15169-2018.
70. Kephart JL, Fandiño-Del-Rio M, Williams KN, Malpartida G, Lee A, Steenland K, et al. 2021. Nitrogen dioxide exposures from LPG stoves in a cleaner-cooking intervention trial. Environ Int 146 :106196, PMID: , 10.1016/j.envint.2020.106196.33160161
71. Nicole W. 2014. Cooking up indoor air pollution: emissions from natural gas stoves. Environ Health Perspect 122 (1 ):A27, PMID: , 10.1289/ehp.122-A27.24380970
72. World Health Organization. 2021. WHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide. https://apps.who.int/iris/handle/10665/345329 [accessed 11 July 2022].
73. Fandiño-Del-Rio M, Kephart JL, Williams KN, Shade T, Adekunle T, Steenland K, et al. 2022. Household air pollution concentrations after liquefied petroleum gas interventions in rural Peru: findings from a one-year randomized controlled trial followed by a one-year pragmatic crossover trial. Environ Health Perspect 130 (5 ):57007, PMID: , 10.1289/EHP10054.35549716
74. Gould CF, Schlesinger SB, Molina E, Bejarano ML, Valarezo A, Jack DW. 2020. Long-standing LPG subsidies, cooking fuel stacking, and personal exposure to air pollution in rural and peri-urban Ecuador. J Expo Sci Environ Epidemiol 30 (4 ):707–720, PMID: , 10.1038/s41370-020-0231-5.32415299
|
PMC010xxxxxx/PMC10069757.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37011136
EHP10486
10.1289/EHP10486
Research
Self-Reported Primary Cooking Fuels Use and Risk of Chronic Digestive Diseases: A Prospective Cohort Study of 0.5 Million Chinese Adults
Wen Qiaorui 1 *
Liu Tanxin 1 *
Yu Yuelin 1 *
Zhang Yunjing 1
Yang Yingzi 1
Zheng Rongshou 2
Li Liming 1
https://orcid.org/0000-0003-1569-3244
Chen Ru 2
Wang Shengfeng 1
1 Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing, China
2 National Cancer Center, National Clinical Research Center for Cancer, Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China
Address correspondence to Shengfeng Wang, Department of Epidemiology and Biostatistics, School of Public Health, Peking University Health Science Center, Beijing 100191, China. Email: [email protected]. And, Ru Chen, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 430022, China. Email: [email protected]
3 4 2023
4 2023
131 4 04700212 10 2021
22 2 2023
02 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Household air pollution (HAP) from inefficient combustion of solid fuels is a major health concern worldwide. However, prospective evidence on the health impacts of solid cooking fuels and risks of chronic digestive diseases remains scarce.
Objectives:
We explored the effects of self-reported primary cooking fuels on the incidence of chronic digestive diseases.
Methods:
The China Kadoorie Biobank recruited 512,726 participants 30–79 years of age from 10 regions across China. Information on primary cooking fuels at the current and previous two residences was collected via self-reporting at baseline. Incidence of chronic digestive diseases was identified through electronic linkage and active follow-up. Cox proportional hazards regression models were used to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations of self-reported long-term cooking fuel patterns and weighted duration of self-reported solid cooking fuel use with chronic digestive diseases incidence. Linear trend was tested by assigning the medians of weighted duration in each group and then taking those as continuous variables in the models. Subgroup analyses were undertaken across the baseline characteristics of participants.
Results:
During 9.1±1.6 y of follow-up, 16,810 new cases of chronic digestive diseases were documented, among which 6,460 were diagnosed as cancers. Compared with long-term cleaner fuel use, self-reported long-term use of solid cooking fuels (i.e., coal, wood) was associated with elevated risks of chronic digestive diseases (HR=1.08; 95% CI: 1.02, 1.13), including nonalcoholic fatty liver disease (NAFLD) (HR=1.43; 95% CI: 1.10, 1.87), hepatic fibrosis/cirrhosis (HR=1.35; 95% CI: 1.05, 1.73), cholecystitis (HR=1.19; 95% CI: 1.07, 1.32), and peptic ulcers (HR=1.15; 95% CI: 1.00, 1.33). The longer the weighted duration of self-reported solid cooking fuel use, the higher the risks of chronic digestive diseases, hepatic fibrosis/cirrhosis, peptic ulcers, and esophageal cancer (pTrend<0.05). The aforementioned associations were modified by sex and body mass index (BMI). Positive associations of always solid cooking fuel use with chronic digestive disease, hepatic fibrosis/cirrhosis, NAFLD, and cholecystitis were observed among women but not men. The longer the weighted duration of self-reported solid cooking fuel use, the higher the risk of NAFLD among those with a BMI ≥28 kg/m2.
Discussion:
Long-term self-reported solid cooking fuels use was associated with higher risks of chronic digestive diseases. The positive association of HAP from solid cooking fuels with chronic digestive diseases indicates for an imminent promotion of cleaner fuels as public health interventions. https://doi.org/10.1289/EHP10486
Supplemental Material is available online (https://doi.org/10.1289/EHP10486).
* These authors contributed equally to this work.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Globally, close to 2.8 billion people are exposed to, and 3.8 million annual premature deaths are caused by, household air pollution (HAP), which is largely attributed to the inefficient combustion of solid fuels (i.e., wood, coal, charcoal, biomass).1 Situations are even worse in rural areas in both China and abroad, where the use of solid fuels persists despite the popularization of gaseous fuels.2 Incomplete combustion of solid fuels (mostly coal and biomass) emit various toxic pollutants, such as particulate matter (PM), carbon monoxide (CO), or nitrogen dioxide (NO2), and conditions are worse in poorly ventilated households.3,4
Increased attention has been paid to the adverse effects of HAP on human health. Multiple chemical compounds released from the incomplete combustion of solid fuels can be absorbed through the digestive tract.5 Accumulating experimental evidence has suggested that household air pollutants [e.g., fine PM (PM ≤2.5μm in aerodynamic diameter; PM2.5), arsenic, polycyclic aromatic hydrocarbons (PAHs)] can induce oxidative damage,6 liver inflammation and genotoxicity,7–9 increase intestinal permeability,10 alter gut microbiomes,11 and intestinal immunity,12 potentially contributing to a range of digestive diseases. Residential or traffic-related PM2.5 and nitrogen oxides (NOx) have been associated with the risk of peptic ulcer diseases (PUDs; including peptic ulcer bleeding, duodenal ulcers, gastric ulcers),13,14 nonalcoholic fatty liver disease (NAFLD),15 and liver cancer.16,17
In addition, existing studies have hinted that chronic digestive diseases might share specific genetic and metabolic features, such as hormone fluctuation.18–20 However, previous epidemiologic studies primarily focused on ambient air pollution, with cross-sectional or case–control designs. In terms of the impacts of HAP from solid fuels combustion on digestive system, to our knowledge, merely one prospective cohort study in China has explored the association between using solid fuels for cooking and the risks of chronic liver disease mortality.21 Huge gaps remain in the association between HAP caused by cooking fuels and several major types of chronic digestive diseases. It is vital to consider different time periods between exposures and incidence of chronic digestive diseases. Therefore, understanding the incident risk of chronic digestive disease is of great etiologic importance.
The China Kadoorie Biobank (CKB) is one of the largest prospective cohorts on a general population globally, having recruited ∼0.5 million adults from 10 geographically diverse regions in China. Under the framework of CKB, the present study explored the association of self-reported solid fuels used for cooking with the incidence of chronic digestive diseases [i.e., chronic hepatitis, hepatic fibrosis/cirrhosis, NAFLD, cholelithiasis, cholecystitis, esophagitis, gastroesophageal reflux disease (GERD), PUD, and digestive cancers] among Chinese adults.
Methods
Study Population
Details of the CKB design have been reported previously and are thus briefly summarized in this paper.22 The baseline survey of the CKB study was carried out between June 2004 and July 2008, enrolling 512,726 participants 30–79 years of age from 10 geographically diverse regions (5 rural areas: Henan, Gansu, Sichuan, Zhejiang, and Hunan; 5 urban areas: Harbin, Qingdao, Suzhou, Liuzhou, and Haikou) across China (Figure S1).22 The study regions were selected based on local cooperative willingness and health care capacity, population stability, quality of death registration, establishment of disease registries, disease patterns, and potential exposure to risk factors. For the baseline survey, laptop-based questionnaires administered by trained staff were used to collect information on demographic and socioeconomic status, lifestyle habits, medical history, and current medication. About 5% (25,000) surviving participants were randomly selected for resurvey every 4–5 y after completion of the baseline survey (first during May 2008–October 2008; second during 2013–2014; third during 2020–2021), using the same methods as in the baseline survey, to understand how those health-related factors change over time and to correct for potential regression dilution bias.23,24 The weighted coefficient of kappa was used to measure the consistency in information collected at baseline and during resurveys, ranging from 0.6 to 0.9 on similar contents.
Approval was obtained from the ethical review committee of the Chinese Center for Disease Control and Prevention and the Oxford Tropical Research Ethics Committee, University of Oxford. All participants provided written informed consent before enrollment.
Assessment of Primary Cooking Fuel
At baseline, participants were asked about their cooking frequency (daily, weekly, monthly, never/rarely, no cooking facility) and their living duration in the three most recent (if any) residence where they had lived for at least 1 y. From participants who reported cooking at least monthly, further information was collected on primary cooking fuel type (coal, wood, gas, electricity, other unspecified) and equipment of cookstove ventilation (all stoves were equipped with chimney/extractor, not all, none) in each residence. If more than one type of fuels were used in any residence, only the one used most frequently was recorded. Those who reported regular cooking (daily or weekly) were categorized based on their self-reported long-term primary cooking fuel use in all residences, namely, “always solid fuels” (defined as using coal or wood), “always cleaner fuels” (defined as using gasoline or electricity) and “switched from solid to clean fuels.” This definition on fuel-type classification has been applied in previous studies.21,25–29 Moreover, self-reported always solid fuel users were further classified into three groups, namely, “always coal,” “always wood,” “mix of coal and wood.”21 The weighted kappa coefficient between self-reported cooking fuels use at baseline and at resurvey was 0.61.25
Assessment of Covariates
All covariates were gathered at baseline. Trained staff members asked participants about their sociodemographic information (age, sex, education, annual household income, occupation), lifestyle habits (alcohol consumption, smoking status, consumption of preserved vegetables, physical activity), living environment (environmental tobacco smoke), medical history (cirrhosis/chronic hepatitis, gallstone/gallbladder disease, PUD, doctor-diagnosed cancer), family history of cancer (mother, father, siblings) by using laptop-based questionnaires. Preserved-vegetable consumption was gathered by asking participants about their intake frequency (never/rarely, monthly, 1–3 d/wk, 4–6 d/wk, or daily) during the past 12 months. The total daily physical activity level was calculated by multiplying the metabolic equivalent tasks (METs) value for a particular type of physical activity by hours spent on that activity per day and summing the metabolic equivalent hours per day (MET-h/d) for all activities.30 Environmental tobacco smoke was collected as the frequency of exposure to tobacco smoke at home, in the workplace, or in public places, and possible answers were never or rarely, less than once a week, 1–2 d/wk, 3–5 d/wk, daily, or nearly every day. Standing height (in meters) and body weight (in kilograms) were measured using standard instruments and protocols with daily calibration. A 10-mL blood sample was collected from each participant by trained medical staff, and tested for hepatitis B surface antigen (HBsAg) (ACON Biotech).
Follow-Up and Outcome Ascertainment
The vital status of study participants was monitored annually (baseline until 31 December 2015) through official residential records and disease and death certificates reported to the regional Center for Disease Control, supplemented by trained staff with medical education reviewing the medical records and national health insurance system if necessary. Electronic linkage with the national health insurance system was achieved for ∼99% of participants, recording details of all hospitalized events including disease description, diagnostic procedure, and the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code.31 Active validation of vital status was conducted annually for those who were not included in the health insurance system via local residential and administrative records. Less than 1% of CKB participants (n=4,875) were lost to follow-up before the end of the study.25,32 Standardized verbal autopsies were performed to determine the most probable cause for the small number of deaths (<5%) that occurred without previous medical attention.21,33 Any deaths occurring among participants were coded using ICD-10 codes by trained staff blinded to baseline information.
The primary outcomes of this study were incidence of chronic digestive diseases, including chronic hepatitis (ICD-10 code B18), hepatic fibrosis/cirrhosis (ICD-10 code K74), NAFLD (ICD-10 code K76), cholelithiasis (ICD-10 code K80), cholecystitis (ICD-10 code K81), esophagitis (ICD-10 code K20), GERD (ICD-10 code K21), PUD (ICD-10 codes K25–K28), and digestive system cancers [esophageal cancer (ICD-10 code C15), gastric cancer (ICD-10 code C16), colorectal cancer (ICD-10 codes C18–C20), liver cancer (ICD-10 code C22), pancreatic cancer (ICD-10 code C25)] (Table S1). Participants who did not have the above events were censored upon death, loss to follow-up, or 31 December 2015, whichever occurred first.
Statistical Analysis
Participants were excluded if they had a) unreliable self-reported information (participants whose accumulated period of the three most recent residence was greater than their age, n=3,791); b) self-reported diagnosis of hepatic cirrhosis or chronic hepatitis (n=6,161), gallstone/gallbladder disease (which might increase the risk of cholelithiasis and cholecystitis) (n=30,064),34 PUD (n=17,402) or any cancer (n=2,187) prior to the baseline survey; c) missing data on body mass index (BMI) (n=2); d) reported no exposure to any types of the cooking fuels listed (n=3,252); e) reported no cooking facilities in each residence (n=41,485, to control for confounding bias caused by the HAP from cooking oil fumes, from cooking regardless of fuel type used)35; or f) switched from cleaner to solid fuels (n=940). A total of 407,422 participants were left in the primary analysis.
Baseline characteristics of the study population are presented as means±standard deviation (SD) for continuous variables, and counts (percentages) for categorical variables across self-reported long-term primary cooking fuel use (always cleaner fuels, always solid fuels, switched from solid to clean fuels, never cooked regularly). In addition, analysis was preformed to compare the baseline characteristics between participants included in and excluded from the main analysis in this study. Missing data were not included in the related analysis.
Cox proportional hazards regression models were applied to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between self-reported solid cooking fuels (taking always cleaner fuels users as the reference group) and incidence of chronic digestive diseases (including cancers), with time in the study as the timescale. HRs (95% CIs) were adjusted for established risk factors, including age at baseline (5-y intervals), sex, 10 study regions, study date, education level (nonformal education, primary school, middle school, high school and above), household income [<10,000, 10,000–19,999, 20,000–34,999, ≥35,000 Chinese yuan renminbi (CNY) per year], occupation (agriculture or related workers, factory worker, administrator or manager, professional or technical, sales or service workers, retired, housewife/husband, self-employed, unemployed, others or not stated), alcohol consumption (never/rarely, occasional, ex-drinker or reduced intake, weekly regular), smoking status (never smoker, occasional smoker, ex-regular smoker, current smoker), physical activity (in metabolic equivalent of tasks hours per day; continuous), BMI (in kilograms per meter squared, continuous), consumption of preserved vegetables (never/occasionally, 1–2 d/wk, 3–5 d/wk, daily or almost every day), environmental tobacco smoke (<1 day/wk, 1–5 d/wk, daily or almost every day), equipment of cookstove ventilation (all stoves, not all stoves, none) and HBsAg test result (negative, positive). Unclear and missing data on HBsAg test results were taken as negative when included as a covariate. The proportional hazard assumption was checked by comparing the HRs of the first and second half of the study period, and no violation was observed.36
The weighted duration of self-reported solid cooking fuel use was calculated by taking cooking frequency and duration at each residence into consideration simultaneously. The weight coefficients for cooking frequency were 1 for those who cooked daily, 1/7 for those who cooked weekly, 1/30 for those who cooked monthly, and 0 for those who never/rarely cooked or did not have a cooking facility. The weight coefficients for cooking fuels were 1 for solid fuels (coal or wood) and 0 for cleaner fuels (gas or electricity). The association of long-term self-reported primary solid cooking fuels with incidence of chronic digestive diseases (including cancers) were examined by collapsing the weighted duration of self-reported solid cooking fuel use into three groups (0, <20, and ≥20),21 and testing the linear trend by assigning the medians of weighted duration in each group and then taking those as continuous variables in the models.
Subgroup analyses were undertaken to examine the potential effect modification of age (<50, ≥50 y), region (rural, urban), sex (men, women), smoking status (never, ever), alcohol consumption (never, ever), preserved-vegetable consumption (never, ever), BMI (in kilograms per meter squared; <18.5, 18.5–24, 24–28, ≥28), and HBsAg status (negative, positive). The likelihood ratio test was used to test effect modification by comparing models with and without the cross-product term of solid cooking fuel use and the abovementioned potential modifiers.
Sensitivity analyses were conducted by a) excluding participants who developed chronic digestive diseases within the first year or the first 2 y of follow-up separately to test the robustness of primary analysis, in the likelihood of reverse causation; b) calculating the weighted duration of exposure by using weight coefficients of 0.5 for cooking weekly and 1.0 for cooking daily, consistent with previous CKB studies26,37; or c) calculating the proportion of the individuals’ past cooking fuel exposure captured by asking about the previous three residences, assuming they started cooking using a stove at the age of 10, 12, 14, and 16 y, respectively, and additionally adjusting for the proportion among all participants.
Several sensitivity analyses were conducted seperately. a) We explored the mediating effect of hormone fluctuation on the association between HAP and selected chronic digestive disease, using pregnancy as the proxy, and estimated the total effect of HAP on selected chronic digestive disease, the effect of HAP on pregnancy, and the effect of pregnancy on selected chronic digestive disease by logistic regression (Figure S2). We used the stepwise method to test whether pregnancy played a role of mediator between HAP and incidence of selected chronic digestive disease.38 b) An interaction analysis of BMI (in kilograms per meter squared; <24, 24–28, ≥28) was conducted with further stratification by age (<50, ≥50 y) and sex (men, women), to explore the confounding effect of age and sex on the interaction of BMI and HAP on incidence of chronic digestive diseases. c) We quantitatively illustrated the extent to which that an unmeasured confounder needed to have with the exposure and the outcome on the HR scale to fully account for an observed exposure–outcome association, above and beyond the measured covariates by E-value analyses: E=HR+sqrt [HR×(HR−1)]. For factors that show inverse associations with diseases outcomes, E-values were calculated using 1/HR instead of HR. The E-values were the minimum strength of association that an unmeasured confounder needed to have with the exposure and the outcome on the HR scale to fully account for an observed exposure–outcome association, above and beyond the measured covariates.39 Finally, d) we estimated the association of the average regional annual PM2.5 levels (one of the major ambient air pollutants) of 10 study regions at the baseline time point vs. PM2.5 of households using cleaner fuels with chronic digestive diseases by linear regression to reveal the effect of ambient air pollution on the risks of chronic digestive diseases, given that the accurate location of participants were unavailable in the CKB public database.
This article is organized according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines (Table S2). Two-sided p-values were used and p<0.05 was considered as statistical significance. All analyses were performed using R (version 4.1.1; R Development Core Team).
Results
Baseline Characteristics
Among the 407,422 participants, at baseline, the mean age±SD was 52.0±10.6 y, and 64% (n=261,570) were women. A total of 85.0% (n=346,348) participants reported cooking at least monthly at baseline, of whom 23.3% (n=80,555) always used cleaner fuels, 50.6% (n=175,261) always used solid fuels, and 26.1% (n=90,532) had switched from solid to cleaner fuels. Compared with self-reported cleaner fuel users, those who always used solid fuels were more likely to be older women, reside in rural regions, be less educated, have lower household income, be exposed to environmental tobacco smoke, and be less likely to have a stove equipped with ventilation (Table 1). Similar conditions were found among excluded participants (Table S3). As for the self-reported duration of solid fuel use, the median (interquartile range; IQR) and mean±SD were 24 (15–36) and 26.05±15.08 y. As for the weighted duration of self-reported solid cooking fuel use, which incorporated cooking frequency and self-reported duration of solid fuel use, the median (IQR) and mean±SD were 18 (6–29) and 19.32±14.75.
Table 1 Baseline characteristics of 407,422 participants from the China Kadoorie Biobank, grouped by long-term cooking fuels use [mean±SD or n (%)].
Characteristics Self-reported primary cooking fuels All participants
Always cleaner Always solid Solid to cleaner Never cooked regularly
Participants 80,555 (19.8) 175,261 (43.0) 90,532 (22.2) 61,074 (15.0) 407,422 (100.0)
Age (y) 48.7±10.1 52.0±10.6 54.3±10.1 52.8±11.0 52.0±10.6
Sex
Men 33,122 (41.1) 39,461 (22.5) 20,851 (23.0) 53,172 (87.1) 146,606 (36.0)
Women 47,433 (58.9) 135,800 (77.5) 69,681 (77.0) 7,902 (12.9) 260,816 (64.0)
Residence
Urban 70,495 (87.5) 20,052 (11.4) 75,485 (83.4) 21,798 (35.7) 187,830 (46.1)
Rural 10,060 (12.5) 155,209 (88.6) 15,047 (16.6) 39,276 (64.3) 219,592 (53.9)
Education
Educated 76,127 (94.5) 125,332 (71.5) 73,622 (81.3) 54,660 (89.5) 329,741 (80.9)
No formal education 4,428 (5.5) 49,929 (28.5) 16,910 (18.7) 6,414 (10.5) 77,681 (19.1)
Household income (CNY/y)
<10,000 74,832 (42.7) 7,422 (9.2) 14,381 (15.9) 11,717 (19.2) 108,352 (26.6)
10,000–19,999 52,691 (30.0) 21,636 (26.9) 27,424 (30.3) 16,551 (27.1) 118,302 (29.0)
20,000–34,999 31,669 (18.1) 28,496 (35.4) 27,502 (30.4) 17,286 (28.3) 104,953 (25.8)
≥35,000 16,069 (9.2) 23,001 (28.6) 21,225 (23.4) 15,520 (25.4) 75,815 (18.6)
Smoking status
Men
Ever 20,864 (63.0) 27,650 (70.1) 12,934 (62.0) 35,796 (67.3) 97,244 (60.1)
Never 12,258 (37.0) 11,811 (29.9) 7,917 (38.0) 17,376 (32.7) 49,362 (39.9)
Women
Ever 1,090 (2.3) 4,472 (3.3) 2,305 (3.3) 190 (2.4) 8,057 (3.1)
Never 46,343 (97.7) 131,328 (96.7) 67,376 (96.7) 7,712 (97.6) 252,759 (96.9)
Alcohol consumption
Men
Ever 27,886 (84.2) 28,574 (72.4) 17,083 (81.9) 40,543 (76.2) 114,086 (77.8)
Never 5,236 (15.8) 10,887 (27.6) 3,768 (18.1) 12,629 (23.8) 32,520 (22.2)
Women
Ever 22,182 (46.8) 45,991 (33.9) 23,567 (33.8) 2,692 (34.1) 94,432 (36.2)
Never 25,251 (53.2) 89,809 (66.1) 46,114 (66.2) 5,210 (65.9) 166,384 (63.8)
Consumption of preserved vegetables
Never 17,327 (21.5) 29,863 (17.0) 15,211 (16.8) 12,208 (20.0) 74,609 (18.3)
Occasionally 22,232 (27.6) 62,134 (35.5) 22,592 (25.0) 21,370 (35.0) 128,328 (31.5)
Weekly 40,996 (50.9) 83,264 (47.5) 52,729 (58.2) 27,496 (45.0) 204,485 (50.2)
Environmental tobacco smoke (d/wk)
<1 35,130 (43.6) 57,624 (32.9) 40,213 (44.4) 20,896 (34.2) 153,863 (37.8)
1–5 18,207 (22.6) 33,788 (19.3) 17,682 (19.5) 15,901 (26.0) 85,578 (21.0)
6–7 27,218 (33.8) 83,846 (47.8) 32,637 (36.1) 24,277 (39.8) 167,978 (41.2)
Cookstove ventilation equipment
All stoves 23,553 (29.3) 15,238 (8.7) 8,981 (9.9) 7,138 (11.7) 54,910 (13.5)
Some stoves 53,354 (66.2) 106,111 (60.5) 77,027 (85.1) 40,103 (65.7) 276,595 (67.9)
None stove 3,648 (4.5) 53,912 (30.8) 4,524 (5.0) 13,833 (22.6) 75,917 (18.6)
Physical activity (MET-h/d) 20.1±12.4 22.0±13.6 19.4±13.3 21.9±16.3 21.0±13.8
BMI (kg/m2)
<18 9,875 (5.6) 2,092 (2.6) 2,405 (2.7) 2,535 (4.1) 16,907 (4.1)
18–24 96,702 (55.2) 37,568 (46.6) 40,745 (45.0) 33,097 (54.2) 208,112 (51.1)
24–28 52,669 (30.1) 30,245 (37.6) 34,712 (38.3) 20,078 (32.9) 137,704 (33.8)
≥28 16,015 (9.1) 10,650 (13.2) 12,670 (1.0) 5,364 (8.8) 44,699 (11.0)
Seropositive HBsAga
Positive 2,363 (3.0) 4,256 (2.4) 2,651 (3.0) 1,812 (3.0) 11,082 (2.8)
Negative 76,908 (96.4) 166,442 (97.0) 85,905 (96.0) 57,872 (96.5) 387,127 (96.5)
Unclear 475 (0.6) 969 (0.6) 903 (1.0) 312 (0.5) 2,659 (0.7)
Note: BMI, body mass index; CNY, Chinese yuan renminbi; HBsAg, hepatitis B virus surface antigen; MET-h/d, metabolic equivalent of tasks hours per day; SD, standard deviation.
a 6,554 (1.6%) participants had a missing value on the seropositive HBsAg test result and were not included in the calculation of percentages. Data were complete for other variables.
Self-Reported Exposure to Solid Cooking Fuels and Risks of Chronic Digestive Diseases
During a follow-up of 9.1±1.6 y, 16,810 new cases of chronic digestive disease were documented, the top five among which were cholecystitis (n=6,396), cholelithiasis (n=6,294), PUD (n=2,813), hepatic fibrosis/cirrhosis (n=992), and NAFLD (n=722). Compared with those who reported always using cleaner fuels, participants reporting long-term solid fuel use for cooking had elevated risks of incident chronic digestive diseases (HR=1.08; 95% CI: 1.02, 1.13), NAFLD (HR=1.43; 95% CI: 1.10, 1.87), hepatic fibrosis/cirrhosis (HR=1.35; 95% CI: 1.05, 1.73), cholecystitis (HR=1.19; 95% CI: 1.07, 1.32), and PUD (HR=1.15; 95% CI: 1.00, 1.33) but not of any digestive cancers (Table 2). Similar results were found when categorizing by weighted duration of self-reported solid cooking fuel use. Participants with ≥20 weighted duration of solid cooking fuels exposure had 5% (HR=1.05; 95% CI: 1.01, 1.10) higher risk of chronic digestive diseases, 40% (HR=1.40; 95% CI: 1.14, 1.71) higher risk of hepatic fibrosis/cirrhosis, 15% (HR=1.15; 95% CI: 1.02, 1.30) higher risk of PUD, and 42% (HR=1.42; 95% CI: 1.20, 1.68) higher risk of esophageal cancer (Figure 1, Table S4). The longer the weighted duration, the higher the risk observed for the above associations (pLinear trend<0.05; Figure 1, Table S4).
Table 2 The association between long-term using pattern of primary cooking fuels and incidence of chronic digestive diseases among 407,422 Chinese adults after a mean 9.1-y follow-up, derived from Cox proportional hazards models.
Categories Always cleanera Always solidb Solid to cleaner Never cooked regularlyc
Participants 80,555 175,261 90,532 61,074
All chronic digestive diseases
Cases 3,155 11,406 4,375 3,670
Person-years 714,257.87 1,551,236.93 802,433.18 529,447.45
HR (95% CI)d 1.00 (Ref) 1.08 (1.02, 1.13) 1.02 (0.97, 1.07) 1.04 (0.98, 1.10)
Chronic viral hepatitise
Cases 180 283 133 123
Person-years 724,940.42 1,593,318.32 817,851.99 542,520.09
HR (95% CI)d 1.00 (Ref) 1.14 (0.85, 1.52) 1.04 (0.81, 1.34) 0.96 (0.73, 1.27)
Hepatic fibrosis/cirrhosise
Cases 166 454 192 180
Person-years 725,202.43 1,593,052.72 817,855.62 542,505.98
HR (95% CI)d 1.00 (Ref) 1.35 (1.05, 1.73) 1.23 (0.98, 1.55) 1.00 (0.78, 1.29)
Nonalcoholic fatty liver diseasee
Cases 128 335 125 134
Person-years 725,127.73 1,592,783.32 817,839.09 542,305.69
HR (95% CI)d 1.00 (Ref) 1.43 (1.10, 1.87) 0.91 (0.70, 1.19) 1.24 (0.95, 1.61)
Cholelithiasis
Cases 746 3,499 1,335 714
Person-years 722,724.2 1,580,318.77 813,136.3 539,982.08
HR (95% CI)d 1.00 (Ref) 0.93 (0.84, 1.02) 1.01 (0.92, 1.11) 0.97 (0.87, 1.09)
Cholecystitis
Cases 650 3,908 1,016 822
Person-years 722,608.26 1,576,658.04 813,558.76 538,717.82
HR (95% CI)d 1.00 (Ref) 1.19 (1.07, 1.32) 1.15 (1.04, 1.28) 1.16 (1.03, 1.30)
Esophagitis
Cases 51 148 58 71
Person-years 725,499.46 1,593,695.03 818,207.44 542,696.38
HR (95% CI)d 1.00 (Ref) 1.25 (0.84, 1.87) 1.06 (0.71, 1.58) 1.28 (0.85, 1.94)
Gastroesophageal reflux disease
Cases 50 116 60 38
Person-years 725,515.29 1,594,004.85 818,228.84 542,885.94
HR (95% CI)d 1.00 (Ref) 1.08 (0.69, 1.71) 0.79 (0.53, 1.18) 1.00 (0.61, 1.63)
Peptic ulcer diseases
Cases 466 1,239 570 538
Person-years 723,696.82 1,589,092.13 816,080.29 540,541.13
HR (95% CI)d 1.00 (Ref) 1.15 (1.00, 1.33) 1.05 (0.92, 1.19) 1.07 (0.92, 1.23)
All digestive cancer
Cases 1,040 2,624 1,356 1,440
Person-years 723,410.95 1,589,015.04 815,383.61 540,037.96
HR (95% CI)d 1.00 (Ref) 1.00 (0.91, 1.1) 0.94 (0.86, 1.03) 0.97 (0.89, 1.07)
Esophageal cancer
Cases 82 838 112 372
Person-years 725,552.31 1,592,494.4 818,204.34 542,213.09
HR (95% CI)d 1.00 (Ref) 1.26 (0.95, 1.67) 0.97 (0.72, 1.31) 0.96 (0.72, 1.27)
Gastric cancer
Cases 309 660 419 430
Person-years 724,958.81 1,593,039.54 817,498.89 542,127.53
HR (95% CI)d 1.00 (Ref) 1.02 (0.86, 1.21) 0.97 (0.83, 1.14) 1.01 (0.86, 1.20)
Colorectal cancer
Cases 424 624 571 335
Person-years 724,525.9 1,592,683.24 816,770.96 542,009.7
HR (95% CI)d 1.00 (Ref) 0.88 (0.74, 1.03) 0.95 (0.83, 1.09) 0.92 (0.78, 1.09)
Liver cancere
Cases 186 445 222 286
Person-years 725,415.21 1,593,826.37 818,090.07 542,685.62
HR (95% CI)d 1.00 (Ref) 1.14 (0.91, 1.42) 0.98 (0.79, 1.21) 1.16 (0.93, 1.43)
Pancreatic cancer
Cases 88 178 98 90
Person-years 725,620.03 1,594,204.36 818,332.82 542,940.08
HR (95% CI)d 1.00 (Ref) 0.98 (0.71, 1.37) 0.71 (0.52, 0.97) 1.00 (0.72, 1.4)
Note: BMI, body mass index; CI, confidence interval; HR, hazard ratio; Ref, reference.
a Cleaner fuels refer to gas or electricity.
b Solid fuels refer to coal or wood.
c Never cooked regularly: individuals who reported cooking monthly or less frequently throughout the recall period.
d HRs (95% CIs) were adjusted for age at baseline, sex, study centers, study date, education level, occupation, household income, alcohol consumption, smoking status, physical activity, BMI, consumption of preserved vegetables, environmental tobacco smoke, cookstove with ventilation, family history of cancer for digestive cancers, and hepatitis B virus surface antigen test result for liver diseases.
e 6,554 (1.6%) participants had a missing value on the seropositive hepatitis B virus surface antigen test result and were classified as being in the negative group.
Figure 1. Association between weighted duration of solid cooking fuel use and chronic digestive diseases incidence among 407,422 Chinese adults, after a mean 9.1-y follow-up, derived from Cox proportional hazards models. (A) All chronic digestive diseases, (B) chronic viral hepatitis, (C) hepatic fibrosis or cirrhosis, (D) nonalcoholic fatty liver disease, (E) cholelithiasis, (F) cholecystitis, (G) esophagitis, (H) gastroesophageal reflux disease, (I) peptic ulcer diseases, (J) all digestive cancer, (K) esophageal cancer, (L) gastric cancer, (M) colorectal cancer, (N) liver cancer, and (O) pancreatic cancer. The x-axis is the weighted duration of self-reported solid fuel use. The y-axis is the hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between weighted duration of self-reported solid fuel use and chronic diseases incidence, adjusted for age at baseline, sex, study centers, study date, education level, occupation, household income, alcohol consumption, smoking status, physical activity, body mass index, consumption of preserved vegetables, environmental tobacco smoke, cookstove with ventilation, family history of cancer for digestive cancers, and HBsAg test result for liver diseases (6,554 participants had a missing value on the seropositive HBsAg test result, accounting for 1.6%, and were classified as being in the negative group). The boxes represent HRs, and the vertical lines represent 95% CIs. The corresponding numerical data are listed in Table S4. Note: HBsAg, hepatitis B virus surface antigen.
Figures 1A to 1O are box and whisker plots titled all chronic digestive diseases, chronic viral hepatitis, hepatic fibrosis or cirrhosis, nonalcoholic fatty liver disease, cholelithiasis, cholecystitis, esophagitis, gastroesophageal reflux disease, peptic ulcer diseases, all digestive cancer, esophageal cancer, gastric cancer, colorectal cancer, liver cancer, and pancreatic cancer, plotting hazard ratios (95 percent confidence intervals), ranging from 0.5 to 1.5 in increments of 0.5 (y-axis) across weighted duration of self-reported solid cooking fuel use, ranging from 0 to 20 in increments of 20 (x-axis) for lowercase italic p trend, respectively.
When further categorizing for solid cooking fuel types, long-term wood users had higher risks of chronic digestive diseases (HR=1.10; 95% CI: 1.03, 1.17), hepatic fibrosis/cirrhosis (HR=1.42; 95% CI: 1.02, 1.98), NAFLD (HR=1.42; 95% CI: 1.03, 1.97), cholecystitis (HR=1.20; 95% CI: 1.06, 1.36), and PUD (HR=1.24; 95% CI: 1.04, 1.49), whereas elevated risk was observed only for total chronic digestive diseases (HR=1.09; 95% CI: 1.00, 1.19) among long-term coal users, compared with self-reported cleaner fuel users (Table 3).
Table 3 The association between long-term solid cooking fuel types among always solid fuel users and incidence of chronic digestive diseases among 255,816 Chinese adults, derived from Cox proportional hazards models.
Categories Always cleanera Always coal Always wood Mix of coal and wood
Participants 80,555 71,076 75,349 28,836
All chronic digestive diseases
Cases 3,155 3,169 6,218 2,019
Person-years 714,257.87 641,706.45 653,139.09 256,391.4
HR (95% CI)b 1.00 (Ref) 1.09 (1.00, 1.19) 1.10 (1.03, 1.17) 1.14 (1.05, 1.23)
Chronic viral hepatitisc
Cases 180 122 121 40
Person-years 724,940.42 651,505.2 678,285.9 263,527.23
HR (95% CI)b 1.00 (Ref) 1.01 (0.64, 1.58) 1.19 (0.80, 1.79) 0.95 (0.57, 1.58)
Hepatic fibrosis/cirrhosisc
Cases 166 172 236 46
Person-years 725,202.43 651,367.44 678,149.01 263,536.27
HR (95% CI)b 1.00 (Ref) 1.31 (0.90, 1.91) 1.42 (1.02, 1.98) 0.86 (0.55, 1.34)
Nonalcoholic fatty liver diseasec
Cases 128 50 228 57
Person-years 725,127.73 651,781.24 677,478.15 263,523.94
HR (95% CI)b 1.00 (Ref) 1.34 (0.88, 2.04) 1.42 (1.03, 1.97) 1.27 (0.83, 1.96)
Cholelithiasis
Cases 746 814 1,844 841
Person-years 722,724.2 648,664.42 671,287.95 260,366.4
HR (95% CI)b 1.00 (Ref) 1.03 (0.89, 1.19) 0.93 (0.83, 1.04) 1.08 (0.94, 1.25)
Cholecystitis
Cases 650 653 2,607 648
Person-years 722,608.26 649,510.95 665,945.68 261,201.41
HR (95% CI)b 1.00 (Ref) 1.13 (0.96, 1.32) 1.20 (1.06, 1.36) 1.13 (0.97, 1.31)
Esophagitis
Cases 51 47 93 8
Person-years 725,499.46 651,769.04 678,284.73 263,641.26
HR (95% CI)b 1.00 (Ref) 1.15 (0.53, 2.46) 1.34 (0.82, 2.19) 0.78 (0.31, 1.94)
Gastroesophageal reflux disease
Cases 50 33 49 34
Person-years 725,515.29 651,820.11 678,583.19 263,601.55
HR (95% CI)b 1.00 (Ref) 0.97 (0.50, 1.88) 0.96 (0.53, 1.73) 1.62 (0.83, 3.17)
Peptic ulcer diseases
Cases 466 293 719 227
Person-years 723,696.82 650,719.68 675,530.53 262,841.93
HR (95% CI)b 1.00 (Ref) 1.02 (0.82, 1.28) 1.24 (1.04, 1.49) 1.29 (1.02, 1.62)
All digestive cancer
Cases 1,040 1,221 1,115 288
Person-years 723,410.95 649,209.73 676,707.7 263,097.61
HR (95% CI)b 1.00 (Ref) 1.05 (0.90, 1.22) 0.94 (0.83, 1.07) 0.95 (0.80, 1.13)
Esophageal cancer
Cases 82 590 188 60
Person-years 725,552.31 650,489.13 678,446.21 263,559.06
HR (95% CI)b 1.00 (Ref) 1.21 (0.81, 1.79) 1.20 (0.83, 1.73) 1.09 (0.70, 1.70)
Gastric cancer
Cases 309 233 370 57
Person-years 724,958.81 651,455.13 678,009.24 263,575.17
HR (95% CI)b 1.00 (Ref) 1.05 (0.78, 1.4) 1.06 (0.85, 1.34) 0.98 (0.69, 1.4)
Colorectal cancer
Cases 424 238 288 98
Person-years 724,525.9 651,278.45 677,997.13 263,407.65
HR (95% CI)b 1.00 (Ref) 1.06 (0.82, 1.37) 0.82 (0.65, 1.03) 0.89 (0.66, 1.21)
Liver cancerc
Cases 186 182 205 58
Person-years 725,415.21 651,656.21 678,568.41 263,601.75
HR (95% CI)b 1.00 (Ref) 1.20 (0.86, 1.69) 0.98 (0.73, 1.32) 1.05 (0.70, 1.55)
Pancreatic cancer
Cases 88 38 120 20
Person-years 725,620.03 651,858.85 678,699.18 263,646.33
HR (95% CI)b 1.00 (Ref) 0.62 (0.34, 1.13) 0.81 (0.54, 1.22) 0.67 (0.36, 1.26)
Note: BMI, body mass index; CI, confidence interval; HR, hazard ratio; Ref, reference.
a Cleaner fuels refer to gas or electricity.
b HRs (95% CIs) were adjusted for age at baseline, sex, study centers, study date, education level, occupation, household income, alcohol consumption, smoking status, physical activity, BMI, consumption of preserved vegetables, environmental tobacco smoke, cookstove with ventilation, family history of cancer for digestive cancers, and hepatitis B virus surface antigen test result for liver diseases.
c 6,554 (1.6%) participants had a missing value on the seropositive hepatitis B virus surface antigen test result and were classified as being in the negative group.
In the subgroup analyses, the risk of total chronic digestive diseases attributed to long-term solid fuels was observed among the elderly (≥50 years of age), rural residents, women, ever drinkers, but not among their counterparts (pInteraction<0.05; Figure 2, Table S5). In addition, women solid fuel users were associated with a 72% (HR=1.72; 95% CI: 1.13, 2.62) higher risk of hepatic fibrosis/cirrhosis, 57% (HR=1.57; 95% CI: 1.10, 2.24) higher risk of NAFLD, and 21% (HR=1.21; 95% CI: 1.06, 1.38) higher risk of cholecystitis, whereas no significant association was observed among men (Tables S6 and S7). Similar results were found with weighted duration of self-reported solid fuel use. Longer weighted duration of exposure to solid cooking fuels was associated higher risks of hepatic fibrosis/cirrhosis among women, whereas no such association was observed among men. (Tables S8 and S9).
Figure 2. Adjusted hazard ratios (HRs) derived from Cox proportional hazards models for chronic digestive diseases associated with long-term solid fuels use after a mean 9.1-y follow-up among 407,422 Chinese adults, stratified by baseline characteristics. Analyses were conducted among 407,422 participants, taking always cleaner fuel users as the reference group, but only HRs (95% CIs) for always solid fuel users are presented. The boxes represent HRs, and the horizontal lines represent 95% CIs. HRs (95% CIs) were adjusted for age at baseline, sex, study centers, study date, education level, occupation, household income, alcohol consumption, smoking status, physical activity, body mass index, consumption of preserved vegetables, environmental tobacco smoke, cookstove with ventilation, family history of cancer for digestive cancers, and HBsAg test result for liver diseases (6,554 participants had a missing value on the seropositive HBsAg test result, accounting for 1.6%, and were classified as being in the negative group). The corresponding numerical data are listed in Table S5. Note: BMI, body mass index; CI, confidence interval; HBsAg, hepatitis B virus surface antigen; HR, hazard ratio.
Figure 2 is a forest plot, plotting characteristics, ranging as (bottom to top) Hepatitis B surface antigen, including negative and positive; Body mass index (kilograms per meter squared), including less than 18.5, 18.5 to 24, 24 to 28 and greater than or equal to 28; Regular drinking (women), including never and ever; Regular drinking (men), including never and ever; Smoking status (women), including never and ever; Smoking status (men), including never and ever; Sex, including men and women; Region, including rural and urban; Age (years), including less than 50 and greater than or equal to 50; and Overall (y-axis) across hazard ratios (95 percent confidence intervals), ranging from 0.50 to 2.00 in increments of 0.50 (x-axis) for lowercase italic p for interaction.
Sensitivity and Post Hoc Analysis
The associations of long-term solid fuels with chronic digestive diseases remained stable after excluding participants who developed chronic digestive diseases in the first, as well as the first 2 y, of follow-up (Tables S10 and S11). The association between weighted duration of solid cooking fuel use and chronic digestive diseases were consistent when applying another weighted coefficient for cooking frequency (Table S12). The associations of long-term use pattern of primary cooking fuels with chronic digestive diseases were stable when additionally adjusting for the proportion of past cooking fuel exposure captured by asking about the previous three residences, assuming they started cooking using a stove at the age of 10, 12, 14, and 16 y, respectively (Table S13). Mediation analysis suggested pregnancy might as a potential mediator for hepatic fibrosis/cirrhosis (Table S14). When stratified by BMI, solid fuel users with a BMI ≥28 kg/m2 had a higher risk of NAFLD (pInteraction<0.001; Table S15). Moreover, the longer the weighted duration of self-reported solid fuel use, the higher the risk of NAFLD that was observed among those with BMI ≥28 kg/m2 (pTrend=0.026, pInteraction<0.001; Table S16). Such interaction persisted after further stratification by age and sex (Table S17). Although solid fuel users with BMI <24 kg/ m2 had a higher risk of cholecystitis (pInteraction<0.001; Table S15), the risk did not increase with the longer weighted duration of self-reported solid fuel use (Table S16). Solid fuel users who had ever consumed preserved vegetables were more likely to develop NAFLD than those who had never consumed preserved vegetables (pInteraction=0.008; Table S15). However, the risk of NAFLD did not increase with the longer weighted duration of self-reported solid fuel use (Table S16).
To quantitatively illustrate the robustness of our results, we performed E-value analysis. The E-values and corresponding lower confidence limits for HRs are presented in Tables S18 and S19. Another post hoc analysis was conducted to estimate the average annual PM2.5 levels vs. number of households using clean fuels with chronic digestive diseases (as a proxy of ambient pollution) in the 10 study regions, with a regression coefficient of −0.013 (95% CI: −0.08, 0.05) (Table S20).
Discussion
This large-scale prospective study showed that HAP caused by solid cooking fuels, especially wood, was associated with higher risks of incident various chronic digestive diseases, including hepatic fibrosis/cirrhosis, NAFLD, cholecystitis, and PUD. The longer the weighted duration of self-reported solid cooking fuel use, the greater the likelihood of developing chronic digestive diseases, hepatic fibrosis/cirrhosis, PUD, and esophageal cancer. Moreover, sex and BMI were effect modifiers.
A cross-sectional study in Ningbo, China, demonstrated that exposure to cooking oil fumes was associated with a higher prevalence of fatty liver disease.35 Previous studies based on the CKB cohort have investigated the association of solid cooking fuels with the incidence of and death from several diseases (Table S21),21,25–29,37,40–42 among which only Chan et al. considered diseases of the digestive system and they observed a positive association between solid cooking fuel and major chronic liver disease mortality.21 Considering that disease mortality is influenced by numerous factors, including medication, surgery, and other treatments, which were not collected in the CKB study, this study focused on incidence instead, aiming to investigate the time period that contributed to the disease etiology and to eliminate the potential bias of further exposure related to death.43 In addition, we further extended our analysis to the scope of digestive systems, and we separately analyzed the incident risk of different types of chronic digestive diseases.
Our study first revealed that those reported long-term solid fuel use for cooking may have excess risks of hepatic fibrosis/cirrhosis and NAFLD, providing important evidence on adverse impacts of HAP on chronic digestive diseases. Despite the paucity of research on HAP, increasing evidence has suggested the link of ambient air pollutants with chronic digestive diseases. The Framingham Heart Study revealed that traffic-related air pollution was associated with hepatic steatosis.44 Another large cohort from the Rome Longitudinal Study found a positive association between long-term exposure to air pollutants (PM and NOx) and incident cirrhosis.45 Evidence from in vivo experimental studies have suggested different mechanisms through which air pollution could promote chronic digestive diseases.45–47 Murine models have indicated the causative role for wood smoke PM in oxidative liver damage, genotoxicity, endoplasmic reticulum stress response.48,49 Moreover, Tan et al. showed that ambient air PM could activate Kupffer cells, producing cytokines and resident hepatic macrophages and exacerbating NAFLD through the production of pro-inflammatory cytokines.50 Impaired glycogen storage, glucose intolerance, and insulin resistance could promote the progression of nonalcoholic steatohepatitis, cirrhosis, and hepatocellular carcinoma.51–53 In addition to PM, the PAHs fraction of airborne particles was reported to cause hematogenic damage and liver fibrosis.54 Given that ambient and household air pollution are composed of similar chemical compounds, we assume that the association of HAP from solid fuels combustion with hepatic fibrosis/cirrhosis and NAFLD are biologically plausible.
To the best of our knowledge, this is the first large-scale prospective study investigating the association of HAP with cholecystitis and PUD. Some studies in vivo and in vitro have investigated the association between air pollution and PUD, with inconsistent conclusions. A time-stratified case-crossover study in Canada did not observe significant impacts of air pollution on upper gastrointestinal bleeding due to PUD.55 However, another case-crossover study in Hong Kong found that short-term elevation in ambient NO2 might trigger PUD and increase the risk of PUD emergency department admission.13 Several studies in China of the same design further supported the association between ambient air pollutants and the increase in hospitalizations for PUD.56–58 In terms of the long-term effect of air pollutants, a prospective study involving 60,273 adults 65 years of age in Hong Kong measured the annual mean concentrations of PM2.5 for residential regions from 1998 to 2011 and reported an estimated 18% (95% CI: 1.02, 1.36) higher risk for PUD hospitalization per 10 μg/m3.59 Although gaps in the pathophysiological mechanism made it difficult to estimate the etiologically relevant time window precisely,43 this study suggested long-term effects of air pollutants on the digestive system. Some studies have suggested that air pollutants may have direct toxic effects on epithelial cells after they were inhaled and transported. Another hypothesis was that air pollutants would increase intestinal permeability, alter the gut microbiota, and promote inflammation, thus contributing toward the development of PUD.10–12
An interesting phenomenon is that we found a longer weighted duration of self-reported solid cooking fuel exposure, which was a more comprehensive assessment, to be associated with a higher risk of esophageal cancer incidence, whereas no significant elevated risk was observed when comparing always solid fuel users with always cleaner fuel users. Given that the length of natural history and etiologically relevant time window varies for different diseases, we postulate that long-term, but not short-term, exposure to solid cooking fuel combustion increases the risk of esophageal cancer. The Golestan Cohort Study has reported that lifetime duration (20 y) of exclusive biomass burning for cooking purpose was associated with an elevated risk of esophageal cancer (HR=1.89; 95% CI: 1.02, 3.50).60 In addition, a meta-analysis,61 including 16 case–control studies and a cohort study using external controls,62 suggested that using biomass fuel might increase the risk of esophageal squamous cell carcinoma. This association is biologically plausible, considering that household combustion of biomass fuels is classified as a probably carcinogenic substance according to the International Agency for Research on Cancer (IARC) Working Group evaluation3,63 and that the inhaled combustion fumes and particles might damage esophageal mucosa through a process called mucociliary clearance. We also acknowledge the potential for chance findings, but the power of comparing those with ≥20 weighted duration of exposure to those with zero reached up to 99%, thus the present conclusions could not be altered. Moreover, for some carcinogens acting on the first stages of carcinogenesis, several decades may elapse before their effects become visible.64 If this is the case for air pollutants emitted from solid cooking fuel combustion, participants with only short latencies may not be able to show an association.
Consistent with the study by Chan et al. on chronic liver disease,21 the present study found long-term use of wood for cooking, rather than coal, to be associated with higher risks of chronic digestive diseases, compared with long-term cleaner fuel use. Inefficient combustion emits a complex mixture of chemical compounds that could induce systemic effects on chronic digestive diseases. One possible explanation for the different impacts of the two fuels is that burning wood may generate more harmful pollutants that can be inhaled and absorbed into the human body, inducing oxidative stress and systemic inflammation and leading to digestive diseases.65 Previous literature has found that farmers who mainly used biomass fuels for cooking and heating tended to have higher urinary levels of PAH metabolites, benzene, and CO than workers from coal factories, where no workers used biomass fuels, after controlling for diet and other lifestyle factors.66 In other words, biomass might expose those long-term users to additional PAHs, which play an important role in toxicity and oxidative damage.67 Direct measurements of HAP from two solid fuels are warranted to elucidate the underlying mechanisms of pathogenesis.
This study highlights the priority of transition from solid fuels to cleaner alternatives. Despite the national fuels transition program in rural regions since 1982,68 disproportionate inequality in fuel types persists in the CKB population, especially among those who are older and of lower socioeconomic status.69 Previous literature in Chinese households also reported that income, education, region, energy price, and fuels accessibility were the most influential factors driving fuel transition.70 These findings may be applied to other developing countries with a similar economic status and solid fuels usage. Actually, solid fuels use, ventilation, temperature, and many other factors could influence the exact exposure level for individuals. Personal portable samplers are expected to have a more accurate assessment on daily inhalation exposure and have gained popularity in more recent studies. The CKB study itself has carried out a pilot study to collect time-resolved data using static and wearable devices in one urban and two rural Chinese communities and found that the overall measured mean PM2.5 levels were 2- to 3-fold higher in the cool than warm season and in rural areas (60.4–78.6 μg/m3 for the warm season; 114.8–160.5 μg/m3 for cool season) than urban sites (40.7 μg/m3 for warm season; 78.2 μg/m3 for cool season).71 Meanwhile, dozens of previous studies (Table S22) reported personal exposures to main air pollutants emitted from cooking fuel combustion in China, including PM2.5, CO, SO2, black carbon, and benzo[a]pyrene.72,73 The range of the aforementioned pollutants’ concentrations varied widely; however, at the least, the mean reported concentrations of PM2.5 (up to 369 μg/m3) completely exceeded the WHO Guidelines for Indoor Air Quality (10 μg/m3).
Moreover, this study suggests that sex and BMI would modify the association between solid fuels use and chronic digestive diseases. Women with long-term exposure to solid fuels combustion were more likely to have excess risk of chronic diseases compared with men. This indicates that HAP would be a gender issue, in accordance with previous studies in sub-Saharan households74 and in Nepal,75,76 where females were more likely to be responsible for cooking in a family,35 especially among those families with low income and education level,77 which also had a greater likelihood of using polluting cooking fuels.78 In addition, the difference could also be considered as a proxy of biological hormones. As reviewed by Zia et al.,79 women’s ovarian hormones, such as estrogen and progesterone, are reported as being one of the main causes of the observed sex differences in upper gastrointestinal motility. In our study, the significant mediating effect of pregnancy on the pathway from HAP to the incidence of digestive diseases also implied this point. Furthermore, PAH-DNA adducts, as the presence of a biologically effective dose of genotoxic compounds from combustion of solid fuels, were found to be higher among women exposed to wood emissions.80 We hypothesize that women are susceptible to the impacts of HAP and thus have higher risks of digestive diseases.
In addition, we found a longer-weighted duration of solid cooking fuel use to be associated with a higher risk of NAFLD among those with BMI ≥28 kg/m2 but not among those with BMI <28 kg/m2. Obesity, as a result of poor diet, is an independent risk factor for NAFLD. Excess free fatty acids in obese individuals has been reported to induce the production of inflammatory factors in hepatocytes and to lead to hepatic inflammation,81,82 which is also the pathological pathway of how air pollution leads to NAFLD.65 Additional studies are needed to corroborate our findings and to elucidate the mechanisms underlying the complex relationship between the aforementioned risk factors. Cholecystitis appears to be more likely to occur in solid fuel users with a BMI <24 kg/m2, consistent with Lee’s finding that BMI was significantly negatively correlated with the severity of cholecystitis.83
The major strengths of this study include prospective design, large sample size, diverse study regions across 10 areas of China, complete follow-up, and the comparison of coal and wood burning on chronic digestive diseases. However, several limitations merit discussion. First, information on solid fuels was self-reported, which might entail recall bias and misclassification, as delineated in previous CKB studies.21,26,32,33 We excluded participants who reported a greater accumulated period of the three most recent residences than their age, and the weighted kappa coefficient between baseline and resurvey also has shown acceptable reproducibility for cooking fuels use (0.61).25 Second, only the primary cooking fuel type was collected at each residence, limiting us to exploring coexisting interaction between different fuel types (coal, wood, and cleaner fuels). We acknowledge that collecting only primary cooking fuel type at the three most recent residences could not completely represent every individual’s past cooking fuel exposure, and this bias would be escalated among older population, especially for those with more than three site changes. It is speculated that under the circumstance of transition to cleaner fuels in mainland China,72,84,85 the potential missing measurement of exposure for those participants in early life stage is more likely to be solid fuels exposure, and such bias might cause an underestimation of the positive association between self-reported solid fuel use and chronic digestive diseases. Information on cooking fuel use in this study was collected at baseline without update during follow up. This bias might underestimate the association between solid fuel exposure and the risk of chronic digestive disease and will not overturn the significant associations in this study. Besides, we did not document the personal exposures to main air pollutants emitted from cooking fuel and suggested further studies to adopt this strategy. Third, we lacked information on some potential confounders, such as stove type, housing type, size of household, heating fuel, lighting fuel, population density, dietary habits, ambient sources of exposure, Helicobacter pylori infection, exposure to aflatoxin,21,86 and detailed information on income, smoking, alcohol consumption, and geographic location due to data unavailability. Although we excluded participants with a history of PUD in the primary analysis (as a proxy for H. pylori infection) and adjusted for socioeconomic status and 10 study areas, which might relate to aflatoxin,21 the possibility of residual confounding cannot be ruled out. Even though post hoc analysis was done to guarantee our results, confounder associations with solid fuel use and specific chronic digestive diseases still have the potentiality to move the confidence interval to the null. Fourth, outcome information of the CKB study was gathered by electronic linkage to hospitalization records, local disease and death registries, and active follow-up, which might lead to the underreporting of outpatients and underestimate the association.87 Fifth, the CKB cohort was not designed to be a representative of the nationwide population.22
Therefore, cautions should be taken in generalizing the results in this study to other populations. In spite of this, the CKB study population included a large number of participants from diverse regions and can undeniably provide evidence for future scientific conclusion.
HAP poses a major health threat worldwide. The present study provides evidence on the association between HAP from self-reported solid fuels use for cooking and incidence of chronic digestive diseases, including hepatic fibrosis/cirrhosis, NAFLD, cholecystitis, peptic ulcer, and esophageal cancer, which were associated with about 1,472,000, 169,000, 125,000, 236,000, and 498,000 deaths each year according to the Global Burden of Disease Study 2019.88,89 Although the underlying pathogenesis remains currently unknown, this study reinforces the importance of public health interventions in promoting cleaner fuels. Moreover, further study with more precise adjustment for confounders, direct measurement of personal exposure of indoor solid fuel combustion, and dose–response is needed to better characterize the risk presented by the use of different fuel types.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
T.L. and S.W. designed the study. S.W. acquired the data. R.C., Y.Z., and T.L. analyzed the data. Q.W., T.L., and Y.Y. drafted the manuscript. All authors contributed to the interpretation of data and revised the article. All authors read and approved the final article.
We thank the Chinese Center for Disease Control and Prevention, Chinese Ministry of Health, National Health and Family Planning Commission of China, and 10 provincial/regional Health Administrative Departments. The most important acknowledgment is to the participants in the study and the members of the survey teams in each of the 10 regional centers, as well as to the project development and management teams based at Beijing, Oxford, and the 10 regional centers.
The CKB baseline survey and first resurvey was supported by grants (2016YFC0900500, 2016YFC0900501, 2016YFC0900504) from the National Key Research and Development Program of China (recipient L.L.), grants from the Kadoorie Charitable Foundation in Hong Kong and grants (088158/Z/09/Z, 104085/Z/14/Z, 104085/Z/14/Z) from Wellcome Trust in the UK (recipient Z.C.). The funders had no role in the study design, data collection, data analysis and interpretation, writing of the report or decision to submit the article for publication. The present study was supported by the National Natural Science Foundation (grant 81502884) (recipient S.W.).
The data sets used and analyzed during the present study are available from the corresponding author on reasonable request.
==== Refs
References
1. Stoner O, Lewis J, Martínez IL, Gumy S, Economou T, Adair-Rohani H. 2021. Household cooking fuel estimates at global and country level for 1990 to 2030. Nat Commun 12 (1 ):5793, PMID: , 10.1038/s41467-021-26036-x.34608147
2. Zhang L, Li H, Chen T, Liao H. 2022. Health effects of cooking fuel transition: a dynamic perspective. Energy 251 :123907, 10.1016/j.energy.2022.123907.
3. IARC (International Agency for Research on Cancer). 2013. Air Pollution and Cancer. IARC Scientific Publication 161. Straif K, Cohen A, Samet J, eds. https://publications.iarc.fr/_publications/media/download/3692/e67ee17589f8759dc6d811c03c7b061a11827ee3.pdf [accessed 21 March 2023].
4. Hou X, Mao Z, Song X, Kang N, Zhang C, Li R, et al. 2022. Kitchen ventilation alleviated adverse associations of domestic fuel use and long-duration cooking with platelet indices as biomarkers of cardiovascular diseases. Sci Total Environ 834 :155341, PMID: , 10.1016/j.scitotenv.2022.155341.35452724
5. Sheikh M, Poustchi H, Pourshams A, Etemadi A, Islami F, Khoshnia M, et al. 2019. Individual and combined effects of environmental risk factors for esophageal cancer based on results from the Golestan Cohort Study. Gastroenterology 156 (5 ):1416–1427, PMID: , 10.1053/j.gastro.2018.12.024.30611753
6. Wang Z, Li Z, Ye Y, Xie L, Li W. 2016. Oxidative stress and liver cancer: etiology and therapeutic targets. Oxid Med Cell Longev 2016 :7891574, PMID: , 10.1155/2016/7891574.27957239
7. Bishayee A. 2014. The role of inflammation and liver cancer. Adv Exp Med Biol 816 :401–435, PMID: , 10.1007/978-3-0348-0837-8_16.24818732
8. Kim JW, Park S, Lim CW, Lee K, Kim B. 2014. The role of air pollutants in initiating liver disease. Toxicol Res 30 (2 ):65–70, PMID: , 10.5487/TR.2014.30.2.065.25071914
9. VoPham T. 2019. Environmental risk factors for liver cancer and nonalcoholic fatty liver disease. Curr Epidemiol Rep 6 (1 ):50–66, PMID: , 10.1007/s40471-019-0183-2.31080703
10. Mutlu EA, Engen PA, Soberanes S, Urich D, Forsyth CB, Nigdelioglu R, et al. 2011. Particulate matter air pollution causes oxidant-mediated increase in gut permeability in mice. Part Fibre Toxicol 8 :19, PMID: , 10.1186/1743-8977-8-19.21658250
11. Kish L, Hotte N, Kaplan GG, Vincent R, Tso R, Gänzle M, et al. 2013. Environmental particulate matter induces murine intestinal inflammatory responses and alters the gut microbiome. PLoS One 8 (4 ):e62220, PMID: , 10.1371/journal.pone.0062220.23638009
12. Salim SY, Jovel J, Wine E, Kaplan GG, Vincent R, Thiesen A, et al. 2014. Exposure to ingested airborne pollutant particulate matter increases mucosal exposure to bacteria and induces early onset of inflammation in neonatal IL-10-deficient mice. Inflamm Bowel Dis 20 (7 ):1129–1138, PMID: , 10.1097/MIB.0000000000000066.24859295
13. Tian LW, Qiu H, Sun SZ, Tsang H, Chan KP, Leung WK. 2017. Association between emergency admission for peptic ulcer bleeding and air pollution: a case-crossover analysis in Hong Kong’s elderly population. Lancet Planet Health 1 (2 ):E74–E81, PMID: , 10.1016/S2542-5196(17)30021-9.29851584
14. Yu Z, Mao X, Tang M, Chen Y, Wu M, Jin M, et al. 2021. Association between past exposure to fine particulate matter (PM2.5) and peptic ulcer: a cross-sectional study in eastern China. Chemosphere 265 :128706, PMID: , 10.1016/j.chemosphere.2020.128706.33139052
15. Hsieh S, Leaderer BP, Feldstein AE, Santoro N, McKay LA, Caprio S, et al. 2018. Traffic-related air pollution associations with cytokeratin-18, a marker of hepatocellular apoptosis, in an overweight and obese paediatric population. Pediatr Obes 13 (6 ):342–347, PMID: , 10.1111/ijpo.12228.28730729
16. Pan WC, Wu CD, Chen MJ, Huang YT, Chen CJ, Su HJ, et al. 2016. Fine particle pollution, alanine transaminase, and liver cancer: a Taiwanese prospective cohort study (REVEAL-HBV). J Natl Cancer Inst 108 (3 ):djv341, PMID: , 10.1093/jnci/djv341.26561636
17. Pedersen M, Andersen ZJ, Stafoggia M, Weinmayr G, Galassi C, Sørensen M, et al. 2017. Ambient air pollution and primary liver cancer incidence in four European cohorts within the ESCAPE project. Environ Res 154 :226–233, PMID: , 10.1016/j.envres.2017.01.006.28107740
18. Pietzner M, Stewart ID, Raffler J, Khaw KT, Michelotti GA, Kastenmüller G, et al. 2021. Plasma metabolites to profile pathways in noncommunicable disease multimorbidity. Nat Med 27 (3 ):471–479, PMID: , 10.1038/s41591-021-01266-0.33707775
19. Backman JD, Li AH, Marcketta A, Sun D, Mbatchou J, Kessler MD, et al. 2021. Exome sequencing and analysis of 454,787 UK biobank participants. Nature 599 (7886 ):628–634, PMID: , 10.1038/s41586-021-04103-z.34662886
20. Pietzner M, Wheeler E, Carrasco-Zanini J, Cortes A, Koprulu M, Wörheide MA, et al. 2021. Mapping the proteo-genomic convergence of human diseases. Science 374 (6569 ):eabj1541, PMID: , 10.1126/science.abj1541.34648354
21. Chan KH, Bennett DA, Kurmi OP, Yang L, Chen Y, Lv J, et al. 2020. Solid fuels for cooking and tobacco use and risk of major chronic liver disease mortality: a prospective cohort study of 0.5 million Chinese adults. Int J Epidemiol 49 (1 ):45–55, PMID: , 10.1093/ije/dyz216.31650183
22. Chen Z, Chen J, Collins R, Guo Y, Peto R, Wu F, et al. 2011. China Kadoorie Biobank of 0.5 million people: survey methods, baseline characteristics and long-term follow-up. Int J Epidemiol 40 (6 ):1652–1666, PMID: , 10.1093/ije/dyr120.22158673
23. Rosner B, Willett WC, Spiegelman D. 1989. Correction of logistic regression relative risk estimates and confidence intervals for systematic within-person measurement error. Stat Med 8 (9 ):1051–1069, PMID: , 10.1002/sim.4780080905.2799131
24. Fibrinogen Studies Collaboration. 2006. Regression dilution methods for meta-analysis: assessing long-term variability in plasma fibrinogen among 27,247 adults in 15 prospective studies. Int J Epidemiol 35 (6 ):1570–1578, PMID: , 10.1093/ije/dyl233.17148467
25. Li J, Qin C, Lv J, Guo Y, Bian Z, Zhou W, et al. 2019. Solid fuel use and incident COPD in Chinese adults: findings from the China Kadoorie Biobank. Environ Health Perspect 127 (5 ):57008, PMID: , 10.1289/EHP2856.31095433
26. Yu K, Qiu G, Chan KH, Lam KBH, Kurmi OP, Bennett DA, et al. 2018. Association of solid fuel use with risk of cardiovascular and all-cause mortality in rural China. JAMA 319 (13 ):1351–1361, PMID: , 10.1001/jama.2018.2151.29614179
27. Chan KH, Kurmi OP, Bennett DA, Yang L, Chen Y, Tan Y, et al. 2019. Solid fuel use and risks of respiratory diseases. A cohort study of 280,000 Chinese never-smokers. Am J Respir Crit Care Med 199 (3 ):352–361, PMID: , 10.1164/rccm.201803-0432OC.30235936
28. Yu K, Lv J, Qiu G, Yu C, Guo Y, Bian Z, et al. 2020. Cooking fuels and risk of all-cause and cardiopulmonary mortality in urban China: a prospective cohort study. Lancet Glob Health 8 (3 ):e430–e439, PMID: , 10.1016/S2214-109X(19)30525-X.31972151
29. Chan KH, Yan M, Bennett DA, Guo Y, Chen Y, Yang L, et al. 2021. Long-term solid fuel use and risks of major eye diseases in China: a population-based cohort study of 486,532 adults. PLoS Med 18 (7 ):e1003716, PMID: , 10.1371/journal.pmed.1003716.34324491
30. Ainsworth BE, Haskell WL, Herrmann SD, Meckes N, Bassett DR Jr, Tudor-Locke C, et al. 2011. 2011 Compendium of Physical Activities: a second update of codes and MET values. Med Sci Sports Exerc 43 (8 ):1575–1581, PMID: , 10.1249/MSS.0b013e31821ece12.21681120
31. WHO (World Health Organization). 2016. International Statistical Classification of Diseases and Related Health Problems, 10th Revision. https://apps.who.int/iris/handle/10665/246208 [accessed 21 March 2023].
32. Li X, Yu C, Guo Y, Bian Z, Shen Z, Yang L, et al. 2019. Association between tea consumption and risk of cancer: a prospective cohort study of 0.5 million Chinese adults. Eur J Epidemiol 34 (8 ):753–763, PMID: , 10.1007/s10654-019-00530-5.31152367
33. Wang L, Jin G, Yu C, Lv J, Guo Y, Bian Z, et al. 2020. Cancer incidence in relation to body fatness among 0.5 million men and women: findings from the China Kadoorie Biobank. Int J Cancer 146 (4 ):987–998, PMID: , 10.1002/ijc.32394.31115907
34. Sanders G, Kingsnorth AN. 2007. Gallstones. BMJ 335 (7614 ):295–299, PMID: , 10.1136/bmj.39267.452257.AD.17690370
35. Lin J, Ni S, Shi Q, Xiong Z, Kang J, Sun X, et al. 2020. Environmental exposure to cooking oil fumes and fatty liver disease. Ann Palliat Med 9 (6 ):3810–3817, PMID: , 10.21037/apm-20-1730.33183027
36. Chan WC, Millwood IY, Kartsonaki C, Du H, Guo Y, Chen Y, et al. 2021. Spicy food consumption and risk of gastrointestinal-tract cancers: findings from the China Kadoorie Biobank. Int J Epidemiol 50 (1 ):199–211, PMID: , 10.1093/ije/dyaa275.33484129
37. Cheng ES, Chan KH, Weber M, Steinberg J, Young J, Canfell K, et al. 2022. Solid fuel, secondhand smoke, and lung cancer mortality: a prospective cohort of 323,794 Chinese never-smokers. Am J Respir Crit Care Med 206 (9 ):1153–1162, PMID: , 10.1164/rccm.202201-0114OC.35616543
38. Iacobucci D. 2012. Mediation analysis and categorical variables: the final frontier. J Consum Psychol 22 (4 ):582–594, 10.1016/j.jcps.2012.03.006.
39. VanderWeele TJ, Ding P. 2017. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med 167 (4 ):268–274, PMID: , 10.7326/M16-2607.28693043
40. Chair SY, Choi KC, Cao X, Cheng HY, Chau JPC, Liu T, et al. 2021. Association between household solid fuel use for cooking and sleep disturbance in rural China: findings from the China Kadoorie Biobank data. Sleep Med 83 :13–20, PMID: , 10.1016/j.sleep.2021.04.029.33990061
41. Liu T, Chen R, Zheng R, Li L, Wang S. 2021. Household air pollution from solid cooking fuel combustion and female breast cancer. Front Public Health 9 :677851, PMID: , 10.3389/fpubh.2021.677851.34422742
42. Qiu S, Chen X, Chen X, Luo G, Guo Y, Bian Z, et al. 2022. Solid fuel use, socioeconomic indicators and risk of cardiovascular diseases and all-cause mortality: a prospective cohort study in a rural area of Sichuan, China. Int J Epidemiol 51 (2 ):501–513, PMID: , 10.1093/ije/dyab191.34536011
43. White E, Hunt JR, Casso D. 1998. Exposure measurement in cohort studies: the challenges of prospective data collection. Epidemiol Rev 20 (1 ):43–56, PMID: , 10.1093/oxfordjournals.epirev.a017971.9762508
44. Li W, Dorans KS, Wilker EH, Rice MB, Long MT, Schwartz J, et al. 2017. Residential proximity to major roadways, fine particulate matter, and hepatic steatosis the Framingham Heart Study. Am J Epidemiol 186 (7 ):857–865, PMID: , 10.1093/aje/kwx127.28605427
45. Orioli R, Solimini AG, Michelozzi P, Forastiere F, Davoli M, Cesaroni G. 2020. A cohort study on long-term exposure to air pollution and incidence of liver cirrhosis. Environ Epidemiol 4 (4 ):e109, PMID: , 10.1097/EE9.0000000000000109.33778350
46. Arciello M, Gori M, Maggio R, Barbaro B, Tarocchi M, Galli A, et al. 2013. Environmental pollution: a tangible risk for NAFLD pathogenesis. Int J Mol Sci 14 (11 ):22052–22066, PMID: , 10.3390/ijms141122052.24213605
47. Li W, Wilker EH, Dorans KS, Rice MB, Schwartz J, Coull BA, et al. 2016. Short-term exposure to air pollution and biomarkers of oxidative stress: the Framingham Heart Study. J Am Heart Assoc 5 (5 ):e002742, PMID: , 10.1161/JAHA.115.002742.27126478
48. Danielsen PH, Loft S, Jacobsen NR, Jensen KA, Autrup H, Ravanat JL, et al. 2010. Oxidative stress, inflammation, and DNA damage in rats after intratracheal instillation or oral exposure to ambient air and wood smoke particulate matter. Toxicol Sci 118 (2 ):574–585, PMID: , 10.1093/toxsci/kfq290.20864625
49. Laing S, Wang G, Briazova T, Zhang C, Wang A, Zheng Z, et al. 2010. Airborne particulate matter selectively activates endoplasmic reticulum stress response in the lung and liver tissues. Am J Physiol Cell Physiol 299 (4 ):C736–C749, PMID: , 10.1152/ajpcell.00529.2009.20554909
50. Tan HH, Fiel MI, Sun Q, Guo J, Gordon RE, Chen LC, et al. 2009. Kupffer cell activation by ambient air particulate matter exposure may exacerbate non-alcoholic fatty liver disease. J Immunotoxicol 6 (4 ):266–275, PMID: , 10.3109/15476910903241704.19908945
51. Conklin DJ. 2013. From lung to liver: how does airborne particulate matter trigger NASH and systemic insulin resistance? J Hepatol 58 (1 ):8–10, PMID: , 10.1016/j.jhep.2012.10.008.23069475
52. Xu X, Yavar Z, Verdin M, Ying Z, Mihai G, Kampfrath T, et al. 2010. Effect of early particulate air pollution exposure on obesity in mice: role of p47phox. Arterioscler Thromb Vasc Biol 30 (12 ):2518–2527, PMID: , 10.1161/ATVBAHA.110.215350.20864666
53. Zheng Z, Xu X, Zhang X, Wang A, Zhang C, Hüttemann M, et al. 2013. Exposure to ambient particulate matter induces a NASH-like phenotype and impairs hepatic glucose metabolism in an animal model. J Hepatol 58 (1 ):148–154, PMID: , 10.1016/j.jhep.2012.08.009.22902548
54. Meiß R, Heinrich U, Offermann M, Themann H. 1982. Long-term study of liver damage following subcutaneous injection of airborne particle extracts and polycyclic aromatic hydrocarbon fractions. Int Arch Occup Environ Health 49 (3–4 ):305–314, PMID: , 10.1007/BF00377939.7068242
55. Quan S, Yang H, Tanyingoh D, Villeneuve PJ, Stieb DM, Johnson M, et al. 2015. Upper gastrointestinal bleeding due to peptic ulcer disease is not associated with air pollution: a case-crossover study. BMC Gastroenterol 15 :131, PMID: , 10.1186/s12876-015-0363-6.26467538
56. Tsai SS, Chiu HF, Yang CY. 2019. Ambient air pollution and hospital admissions for peptic ulcers in Taipei: a time-stratified case-crossover study. Int J Environ Res Public Health 16 (11 ):1916, PMID: , 10.3390/ijerph16111916.31151209
57. Wu M, Lu J, Yang Z, Wei F, Shen P, Yu Z, et al. 2021. Ambient air pollution and hospital visits for peptic ulcer disease in China: a three-year analysis. Environ Res 196 :110347, PMID: , 10.1016/j.envres.2020.110347.33130162
58. Wu MY, Tang ML, Yu ZB, Mao X, Chen Y, Wang J, et al. 2021. Association between short-term exposure to air pollution and peptic ulcer bleeding: a case-crossover study in China. Atmos Environ 256 :118438, 10.1016/j.atmosenv.2021.118438.
59. Wong CM, Tsang H, Lai HK, Thach TQ, Thomas GN, Chan KP, et al. 2016. STROBE-long-term exposure to ambient fine particulate air pollution and hospitalization due to peptic ulcers. Medicine (Baltimore) 95 (18 ):e3543, PMID: , 10.1097/MD.0000000000003543.27149464
60. Sheikh M, Poustchi H, Pourshams A, Khoshnia M, Gharavi A, Zahedi M, et al. 2020. Household fuel use and the risk of gastrointestinal cancers: the Golestan Cohort Study. Environ Health Perspect 128 (6 ):67002, PMID: , 10.1289/EHP5907.32609005
61. Okello S, Akello SJ, Dwomoh E, Byaruhanga E, Opio CK, Zhang R, et al. 2019. Biomass fuel as a risk factor for esophageal squamous cell carcinoma: a systematic review and meta-analysis. Environ Health 18 (1 ):60, PMID: , 10.1186/s12940-019-0496-0.31262333
62. Hogstedt C, Jansson C, Hugosson M, Tinnerberg H, Gustavsson P. 2013. Cancer incidence in a cohort of Swedish chimney sweeps, 1958–2006. Am J Public Health 103 (9 ):1708–1714, PMID: , 10.2105/AJPH.2012.300860.23327283
63. IARC. 2010. Household use of solid fuels and high-temperature frying. IARC Monogr Eval Carcinog Risks Hum 95 :1–430, PMID: .20701241
64. Day NE, Brown CC. 1980. Multistage models and primary prevention of cancer. J Natl Cancer Inst 64 (4 ):977–989, PMID: .6929006
65. Xu MX, Ge CX, Qin YT, Gu TT, Lou DS, Li Q, et al. 2019. Prolonged PM2.5 exposure elevates risk of oxidative stress-driven nonalcoholic fatty liver disease by triggering increase of dyslipidemia. Free Radic Biol Med 130 :542–556, PMID: , 10.1016/j.freeradbiomed.2018.11.016.30465824
66. Fan R, Li J, Chen L, Xu Z, He D, Zhou Y, et al. 2014. Biomass fuels and coke plants are important sources of human exposure to polycyclic aromatic hydrocarbons, benzene and toluene. Environ Res 135 :1–8, PMID: , 10.1016/j.envres.2014.08.021.25261857
67. Deng Q, Dai X, Feng W, Huang S, Yuan Y, Xiao Y, et al. 2019. Co-exposure to metals and polycyclic aromatic hydrocarbons, microRNA expression, and early health damage in coke oven workers. Environ Int 122 :369–380, PMID: , 10.1016/j.envint.2018.11.056.30503314
68. Smith KR, Shuhua G, Kun H, Daxiong Q. 1993. One hundred million improved cookstoves in China: how was it done? World Dev 21 (6 ):941–961, 10.1016/0305-750X(93)90053-C.
69. Chan KH, Lam KBH, Kurmi OP, Guo Y, Bennett D, Bian Z, et al. 2017. Trans-generational changes and rural–urban inequality in household fuel use and cookstove ventilation in China: a multi-region study of 0.5 million adults. Int J Hyg Environ Health 220 (8 ):1370–1381, PMID: , 10.1016/j.ijheh.2017.09.010.28986011
70. Chen Y, Shen H, Zhong Q, Chen H, Huang T, Liu J, et al. 2016. Transition of household cookfuels in China from 2010 to 2012. Appl Energy 184 :800–809, 10.1016/j.apenergy.2016.07.136.
71. Chan KH, Xia X, Ho KF, Guo Y, Kurmi OP, Du H, et al. 2021. Regional and seasonal variations in household and personal exposures to air pollution in one urban and two rural Chinese communities: a pilot study to collect time-resolved data using static and wearable devices. Environ Int 146 :106217, PMID: , 10.1016/j.envint.2020.106217.33129001
72. Zhu X, Yun X, Meng W, Xu H, Du W, Shen G, et al. 2019. Stacked use and transition trends of rural household energy in mainland China. Environ Sci Technol 53 (1 ):521–529, PMID: , 10.1021/acs.est.8b04280.30512946
73. Du W, Li X, Chen Y, Shen G. 2018. Household air pollution and personal exposure to air pollutants in rural China—a review. Environ Pollut 237 :625–638, PMID: , 10.1016/j.envpol.2018.02.054.29525629
74. Meng T, Florkowski WJ, Sarpong DB, Chinnan M, Resurreccion AVA. 2021. Cooking fuel usage in sub-Saharan urban households. Energies 14 (15 ):4629, 10.3390/en14154629.
75. Acharya B, Marhold K. 2019. Determinants of household energy use and fuel switching behavior in Nepal. Energy 169 :1132–1138, 10.1016/j.energy.2018.12.109.
76. Giri M, Goswami B. 2018. Determinants of household’s choice of fuel for cooking in developing countries: evidence from Nepal. J Dev Policy Pract 3 (2 ):137–154, 10.1177/2455133318769392.
77. Ghimire S, Mishra SR, Sharma A, Siweya A, Shrestha N, Adhikari B. 2019. Geographic and socio-economic variation in markers of indoor air pollution in Nepal: evidence from nationally-representative data. BMC Public Health 19 (1 ):195, PMID: , 10.1186/s12889-019-6512-z.30764804
78. WHO. 2014. WHO Indoor Air Quality Guidelines: Household Fuel Combustion. Geneva, Switzerland: World Health Organization. https://apps.who.int/iris/rest/bitstreams/632228/retrieve [accessed 22 March 2023].
79. Zia JK, Heitkemper MM. 2016. Upper gastrointestinal tract motility disorders in women, gastroparesis, and gastroesophageal reflux disease. Gastroenterol Clin North Am 45 (2 ):239–251, PMID: , 10.1016/j.gtc.2016.02.003.27261896
80. Mumford JL, Lee X, Lewtas J, Young TL, Santella RM. 1993. DNA adducts as biomarkers for assessing exposure to polycyclic aromatic hydrocarbons in tissues from Xuan Wei women with high exposure to coal combustion emissions and high lung cancer mortality. Environ Health Perspect 99 :83–87, PMID: , 10.1289/ehp.939983.8319664
81. Milić S, Lulić D, Štimac D. 2014. Non-alcoholic fatty liver disease and obesity: biochemical, metabolic and clinical presentations. World J Gastroenterol 20 (28 ):9330–9337, PMID: .25071327
82. Polyzos SA, Kountouras J, Mantzoros CS. 2019. Obesity and nonalcoholic fatty liver disease: from pathophysiology to therapeutics. Metabolism 92 :82–97, PMID: , 10.1016/j.metabol.2018.11.014.30502373
83. Lee HK, Han HS, Min SK. 2009. The association between body mass index and the severity of cholecystitis. Am J Surg 197 (4 ):455–458, PMID: , 10.1016/j.amjsurg.2008.01.029.18760398
84. Carter E, Yan L, Fu Y, Robinson B, Kelly F, Elliott P, et al. 2019. Household transitions to clean energy in a multiprovincial cohort study in China. Nat Sustain 3 (1 ):42–50, 10.1038/s41893-019-0432-x.
85. Tao S, Ru MY, Du W, Zhu X, Zhong QR, Li BG, et al. 2018. Quantifying the rural residential energy transition in China from 1992 to 2012 through a representative national survey. Nat Energy 3 (7 ):567–573, 10.1038/s41560-018-0158-4.
86. IEA (International Energy Agency). 2021. Tracking SDG7: the Energy Progress Report 2021. https://www.iea.org/reports/tracking-sdg7-the-energy-progress-report-2021 [accessed 22 March 2023].
87. Bross I. 1954. Misclassification in 2 × 2 tables. Biometrics 10 (4 ):478–486, 10.2307/3001619.
88. Murray CJL. 2022. The Global Burden of Disease Study at 30 years. Nat Med 28 (10 ):2019–2026, PMID: , 10.1038/s41591-022-01990-1.36216939
89. IHME (Institute for Health Metrics and Evaluation). 2019. GBD compare. https://www.healthdata.org/data-visualization/gbd-compare [accessed 17 October 2022].
|
PMC010xxxxxx/PMC10069759.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37011135
EHP11661
10.1289/EHP11661
Research
Acute Effects of Ambient Air Pollution on Asthma Emergency Department Visits in Ten U.S. States
https://orcid.org/0000-0003-3807-6927
Bi Jianzhao 1
D’Souza Rohan R. 2
Moss Shannon 2
Senthilkumar Niru 3
Russell Armistead G. 3
Scovronick Noah C. 4
Chang Howard H. 2
Ebelt Stefanie 4
1 Department of Environmental & Occupational Health Sciences, University of Washington, Seattle, Washington, USA
2 Department of Biostatistics and Bioinformatics, Emory University, Atlanta, Georgia, USA
3 School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, Georgia, USA
4 Gangarosa Department of Environmental Health, Emory University, Atlanta, Georgia, USA
Address correspondence to Jianzhao Bi, Department of Environmental & Occupational Health Sciences, University of Washington, 4225 Roosevelt Way NE, Seattle, WA 98105 USA. Email: [email protected]
3 4 2023
4 2023
131 4 04700301 6 2022
05 2 2023
02 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Previous studies of short-term ambient air pollution exposure and asthma morbidity in the United States have been limited to a small number of cities and/or pollutants and with limited consideration of effects across ages.
Objectives:
To estimate acute age group-specific effects of fine and coarse particulate matter (PM), major PM components, and gaseous pollutants on emergency department (ED) visits for asthma during 2005–2014 across the United States.
Methods:
We acquired ED visit and air quality data in regions surrounding 53 speciation sites in 10 states. We used quasi-Poisson log-linear time-series models with unconstrained distributed exposure lags to estimate site-specific acute effects of air pollution on asthma ED visits overall and by age group (1–4, 5–17, 18–49, 50–64, and 65+ y), controlling for meteorology, time trends, and influenza activity. We then used a Bayesian hierarchical model to estimate pooled associations from site-specific associations.
Results:
Our analysis included 3.19 million asthma ED visits. We observed positive associations for multiday cumulative exposure to all air pollutants examined [e.g., 8-d exposure to PM2.5: rate ratio of 1.016 with 95% credible interval (CI) of (1.008, 1.025) per 6.3-μg/m3 increase, PM10–2.5: 1.014 (95% CI: 1.007, 1.020) per 9.6-μg/m3 increase, organic carbon: 1.016 (95% CI: 1.009, 1.024) per 2.8-μg/m3 increase, and ozone: 1.008 (95% CI: 0.995, 1.022) per 0.02-ppm increase]. PM2.5 and ozone showed stronger effects at shorter lags, whereas associations of traffic-related pollutants (e.g., elemental carbon and oxides of nitrogen) were generally stronger at longer lags. Most pollutants had more pronounced effects on children (<18 y old) than adults; PM2.5 had strong effects on both children and the elderly (>64 y old); and ozone had stronger effects on adults than children.
Conclusions:
We reported positive associations between short-term air pollution exposure and increased rates of asthma ED visits. We found that air pollution exposure posed a higher risk for children and older populations. https://doi.org/10.1289/EHP11661
Supplemental Material is available online (https://doi.org/10.1289/EHP11661).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Ambient air pollution poses great risks to human health. Exposure to ambient air pollution contributes 4 million deaths with 140 million disability-adjusted life-years (DALYs) each year.1,2 Despite a reduction of air pollution among developed countries over the past decades, there has been a substantial increase in exposure levels in many other parts of the world.3,4 Global age-standardized exposure levels to ambient particulate matter (PM) and ozone (O3), two major air pollutants regulated worldwide, increased 41% and 3% over the period 1990–2017, respectively.1 Over 90% of the world’s population lives in places where air quality concentrations exceed the World Health Organization’s ambient air quality guideline limits.5 There is substantial epidemiological evidence regarding adverse health effects of air pollution exposure, especially for respiratory and cardiovascular outcomes, even at low concentrations.6
Asthma is a lifelong inflammatory disease of the airways that is characterized by reversible airflow obstruction and bronchospasm with recurrent symptoms of coughing, wheezing, shortness of breath, and chest tightness. In severe cases, the symptoms may be triggered multiple times per week or even within a day. Asthma is the most prevalent chronic respiratory disease, and the most common chronic disease among children, affecting 260 million people and causing 460 thousand deaths worldwide in 2019.2 In the United States, asthma affected 25 million people and was associated with more than 3,500 deaths in 2019.7 There is an estimated $81 billion annual economic cost associated with asthma, including medical costs and indirect costs of loss of work and school days.8
Accumulating epidemiological evidence points to the influence of short-term exposure to ambient air pollution on exacerbation of asthma—particularly among children and the elderly—reflected by increased emergency department use.9–12 Traffic-related air pollution, e.g., certain particulate matter (PM) components, oxides of nitrogen (NOx), and carbon monoxide (CO), may play a particular role.13 However, meta-analyses and systematic reviews on the topic have reported high heterogeneity in studies investigating these associations.9 Differential characteristics (e.g., air pollution concentration, composition, and exposure, as well as outcome definition), study population (e.g., population susceptibility, access to health care), and analytical methods employed across studies are possible driving factors behind the observed heterogeneity.14,15 In the United States, for example, air pollution–asthma morbidity studies have been limited to a small number of cities16–21 and pollutants.22–26 Studies have also typically been restricted to only hospitalization outcomes27–29 or only the young population.30–35 There remains a need to comprehensively examine associations between short-term exposure to air pollution and asthma-related morbidity, particularly using a multisite design that employs consistent exposure and outcome assessment methods across locations and that thus enables a rigorous assessment of pooled effects as well as possible effect heterogeneity.
To address these gaps, we conducted a multisite time-series analysis to investigate associations between short-term exposure to ambient air pollution [PM2.5 (fine particulate matter less than 2.5μm in aerodynamic diameter) with its major components, PM10–2.5 (coarse particulate matter between 2.5 and 10μm in aerodynamic diameter), and gaseous pollutants] and risk of asthma emergency department (ED) visits, including those resulting in hospital admission, during the period 2005–2014. Study sites were selected based on locations of federal PM speciation stations in 10 U.S. states (Arizona, California, Georgia, Maryland, Missouri, Nevada, New Jersey, New York, North Carolina, and Utah). We further investigated effects of air pollution exposure at different lag times (single-day lags from lag 0 to lag 7 and distributed lags 0–2, 3–6, and 0–7) and compared effects across age groups (1–4, 5–17, 18–49, 50–64, and 65+ y) and regions (the eastern and western United States).
Data and Methods
ED Visits and Hospital Admissions
We acquired patient-level ED visit data as part of the Environmental Exposures and Health Across the Nation (ENVISION) study.20,36 The ENVISION database includes outpatient and inpatient billing records from individual states, with key variables for each patient record including date of visit, whether the visit resulted in an admission to the hospital (ED admission), International Classification of Disease Ninth Revision (ICD-9) diagnosis codes, patient age, and ZIP code of patient residence.
The ED visit data for this analysis were from 10 U.S. states for 2005–2014: Arizona (Department of Health Services, 2010–2014), California (Department of Health Care Access and Information, 2005–2014), Georgia (Georgia Hospital Association, 2011–2014), Maryland (Department of Health, 2005–2014), Missouri (Department of Health and Senior Services, 2005–2014), Nevada (Division of Health Care Financing and Policy, 2009–2014), New Jersey (Department of Health, Center for Health Statistics & Informatics, 2005–2014), New York (Department of Health, 2005–2014), North Carolina (North Carolina Hospital Discharge Database, 2007–2014), and Utah (Department of Health, 2005–2014). We included patient records with a primary diagnosis of asthma (ICD-9=493). The ED visit data were checked for implausible values, facility closures, abnormal distributional trends, and missingness. Implausible values were set to missing, and facility indicators were created to indicate days of hospital operation. Missing values for essential variables, admission date, and patient ZIP codes were excluded from the analysis. Additionally, any visit without at least one ICD diagnosis code was excluded from the analysis. Missingness exclusions were minimal across the 10 states. Categorization of demographic variables were verified against the original data source and data dictionaries; categorical variable levels and processing decisions were standardized across the 10 states of ED visit data.
We restricted the study population to patients with a residential ZIP code within 30 mi of 53 Air Quality System (AQS) PM speciation monitoring sites in the 10 states (Figure S1). Our criteria to select AQS sites to be considered are as follows: a) having available air quality measurements for >120 d during the period 2005–2014; b) including at least one ZIP-code geographical centroid within 15 mi; c) having available primary ED visits (including those resulting in hospital admission) for asthma within 30 mi for >365 d during the period 2005–2014; d) having at least 10% of days with nonzero ED visits over the period 2005–2014; e) having the longest gap of ED visits shorter than 365 d during 2005–2014; and f) having the maximum daily ED visit counts ≥5. We aggregated the patient-level ED visits by day to obtain time-series of daily ED visit counts for the 30-mi area surrounding each monitoring site.
The Emory University institutional review board approved this study and granted an exemption from informed consent requirements, given the minimal risk nature of the study and the infeasibility of obtaining informed consent from individual patients.
Air Pollution and Meteorological Data
Because the AQS monitors only measured PM2.5 components every third or sixth day, we applied daily air pollution exposure estimates from a data fusion approach that combines ground observations with chemical transport model simulations from the Community Multiscale Air Quality (CMAQ) model (referred to as CMAQ-fused data) as our primary exposure data.37 The CMAQ-fused data provided estimates at a spatial resolution of 12km. Data from the CMAQ-fused grid cell overlapping each AQS site location were assigned to the corresponding ED visit time series. The CMAQ-fused air pollutants included PM (daily mean PM2.5 and PM10–2.5), major PM2.5 components [daily mean elemental carbon (EC), organic carbon (OC), nitrate (NO3−), and sulfate (SO42−)], and major gaseous pollutants [daily 8-h maximum ozone (O3), 1-h maximum oxides of nitrogen (NOx), nitrogen dioxide (NO2), sulfur dioxide (SO2), and CO]. PM10–2.5 concentrations were generated by subtracting PM2.5 from PM10 concentrations. Negative PM10–2.5 concentrations that accounted for 3% of all PM10–2.5 data, resulting from uncertainty in the CMAQ-fused data at very low coarse particle pollution conditions, were converted to 0. We analyzed PM2.5 components individually instead of combinations of multiple components to avoid introducing large uncertainty to exposure estimates resulting from collinearity among components. We used Pearson correlation coefficients to describe the correlations among the CMAQ-fused air pollutants. The main advantage of using the CMAQ-fused data for this analysis was that the data provided temporally continuous estimates of exposure for all pollutants of interest, which allowed for estimating cumulative health associations over multiple lagged days. This exposure dataset was especially advantageous for assessment of PM2.5 components that were not measured temporal continuously in the AQS network.
We also acquired available daily ground observations of criteria air pollutants (PM2.5, O3, NO2, SO2, and CO) at the AQS sites as our secondary exposure data. In sensitivity analyses, we compared the robustness of health associations estimated based on the CMAQ-fused and AQS data to evaluate the reliability of the CMAQ-fused data.
Meteorological factors, especially air temperature and humidity, are critical confounders of the associations between short-term exposure to air pollution and respiratory outcomes.38,39 We obtained hourly meteorological parameters—ground-level air temperature (T) and dew-point temperature (DPT)—from the Meteorological Assimilation Data Ingest System (MADIS) Aviation Routine Weather Report (METAR) database (https://madis.ncep.noaa.gov/). Because the METAR sites did not overlap the AQS sites, we used ordinary kriging on the hourly scale to obtain temporally continuous T and DPT at the AQS sites. We then calculated daily 1-h maximum T and 24-h mean DPT based on the hourly values. Finally, we converted the continuous daily maximum T to discrete integers (in degrees Kelvin).
Time-Series and Bayesian Hierarchical Modeling
We adopted a two-stage modeling approach to estimate relative risks of short-term exposure to air pollution on ED visits for asthma. In the first stage, we built AQS site-specific quasi-Poisson log-linear models to estimate associations at single-day lags (lag 0 to lag 7; lag 0 is the same day, lag 1 is the previous day, etc.) and cumulative associations with unconstrained distributed lags 0–2 (3-d immediate cumulative exposure), 3–6 (4-d delayed cumulative exposure), and 0–7 (8-d prolonged cumulative exposure). Our models assumed linear air pollution effects due to the relatively small ranges of short-term exposure levels in the United States as well as limited evidence of deviation from linearity found in prior studies.10 The distributed lag model is expressed as40: (1) log [E (Yt)]=α+∑q=dDβt−qAPt−q+(confounders),
where E(Yt) is the mean ED visit count on day t; APt−q is an air pollutant’s concentration q days before day t; and the sum of βt−q is the main parameter of interest for distributed lagged associations (i.e., lags 0–2, 3–6, or 0–7). The use of distributed lags was motivated by the evidence showing that the adverse respiratory effects of air pollution exposure can occur over multiple days.41,42 The single-day lag model had a similar structure with only one pollution term on a specific day and a corresponding β coefficient.
The site-specific models included nonlinear meteorology effects (moving averages of daily maximum T and mean DPT over the lag periods for the distributed lag models; single-day mean DPT at the same lags for the single-day lag models) specified as natural cubic splines with 4 degrees of freedom (df). Discrete integers of daily maximum T (in degrees Kelvin) were included as an additional ordinal covariate. The models also included daily indicator variables for day-of-week (Monday through Sunday), holidays (0=nonholidays, 1=federal and Federal Reserve Board holidays), and for each hospital in the 30-mi area surrounding each AQS site, indicating whether or not it contributed visits (to account for changes to ED visit totals attributable to hospital data availability). Long-term time trends were controlled using natural cubic splines with 12 knots per year (i.e., monthly knots). In addition, the models were controlled for ED visit counts for influenza to adjust for viral-induced asthma in flu seasons.43
The time-series design treats individuals within the same geographical area (in our case, the 30-mi area surrounding each AQS site) as the “at-risk population.” This “population” serves as its own control over the multiday exposure windows. Individual-level characteristics averaged across the entire population (e.g., socioeconomic status, diet, and physical activity) are expected to change minimally over the course of several days and thus are not considered to be potential confounders.44
In the second stage, we built a Bayesian hierarchical model (BHM) to pool site-specific log relative risks [RRs, per interquartile range (IQR) increase in pollution concentration] derived from the first stage to estimate a pooled RR and its 95% CI. The BHM is given by: (2) βi^=θi+ϵi θi∼N(μ,τ2) ϵi∼N(0,σi^2),
where θi is the unobserved true log RR at site i and ϵi is the random deviation of the risk that is independent across sites. Site-specific log RR is denoted by βi^ with standard error σi^. We assumed θi follows a normal distribution with mean μ, the pooled risk, and variance τ2, the between-site heterogeneity. We also assumed the priors of μ and τ2 follow noninformative normal and inverse-gamma distributions, respectively. The 95% CI was calculated based on the posterior mean and standard deviation (SD) of the pooled log RR. The pooled log RR and 95% CI were then exponentiated. Our statistical analyses were conducted based on the R (version 4.0.2; R Development Core Team) packages “dlnm” (version 2.4.6)45 and “R2jags” (version 0.6-1).
We conducted the 2-stage modeling process for all 11 air pollutants (PM2.5, PM10–2.5, EC, OC, nitrate, sulfate, O3, NO2, NOx, SO2, and CO) to estimate their single-day lag and distributed-lag effects. We also conducted stratification analyses by age group (1–4, 5–17, 18–49, 50–64, and 65+ y) and by region (east and west; Figure S1).
To examine robustness of the single-pollutant effect estimates, we built several two-pollutant models using single-day lags (lag 0 to lag 7). The first two-pollutant model included PM2.5 and O3, which estimates the effects of both pollutants after adjusting for each other. The second type of two-pollutant models targeted traffic-related pollutants—EC, NO2, NOx, and CO—to derive their effect estimates individually, while controlling for PM2.5 or O3 (i.e., treating PM2.5 or O3 as the second pollution term).
Sensitivity Analyses
We performed several sensitivity analyses to further examine robustness of our findings. For meteorological adjustment, we tested different degrees of freedom (from 2 to 6) of the natural cubic splines based on 8-d cumulative exposure. For long-term time trends, we tested natural cubic splines with varying knots (from 0.7 to 1.3 knots per month) based on 8-d cumulative exposure. Additionally, we adopted the negative outcome control approach to detect unmeasured confounding.46,47 In our case, the negative outcome control was tomorrow’s (lag −1) air pollution concentrations as an indicator in the single-day lag modeling framework. Furthermore, we examined the use of different definitions of influenza ED visits by the Armed Forces Health Surveillance Center (AFHSC) in adjusting for viral-induced asthma in flu seasons based on 8-d cumulative exposure.48 Finally, we built single-day lag models using the AQS observations for several criteria air pollutants—PM2.5, O3, NO2, SO2, and CO—and compared the effect estimates with those estimated based on the CMAQ-fused data at the same time points (i.e., for days when there were no AQS data, the corresponding CMAQ-fused data were excluded as well).
Results
Summary Statistics
There were 3,190,333 ED visits for asthma over our study period 2005–2014. The mean daily ED visit count with primary asthma diagnosis was 874 (SD=210) across all locations (Table 1). The five age groups (1–4, 5–17, 18–49, 50–64, and 65+ y) accounted for, respectively, 14.9%, 23.4%, 39.0%, 13.6%, and 7.3% of the total ED visits. The remaining 1.8% were ED visits made by patients with an age below 1 y or unknown age.
Table 1 Summary statistics for daily emergency department visit counts with primary asthma diagnosis and visit counts for influenza in 10 U.S. states (within 30-mi areas around 53 AQS sites) over the study period 2005–2014.
Parameter Mean SD Minimum Maximum IQR
ED visits for asthma
All ages, daily visit counts 874 210 402 1,762 284
Ages 1–4 y, daily visit counts 130 44 35 287 61
Ages 5–17 y, daily visit counts 204 83 49 627 111
Ages 18–49 y, daily visit counts 341 70 173 760 91
Ages 50–64 y, daily visit counts 119 30 49 260 41
Ages 65+ y, daily visit counts 63 17 21 166 22
ED visits for influenza
All ages, daily visit counts 139 251 0 1,949 129
Ages 1–4 y, daily visit counts 17 34 0 334 16
Ages 5–17 y, daily visit counts 29 67 0 780 27
Ages 18–49 y, daily visit counts 59 99 0 768 55
Ages 50–64 y, daily visit counts 13 25 0 208 12
Ages 65+ y, daily visit counts 12 29 0 361 10
Note: AQS, Air Quality System; ED, emergency department; IQR, interquartile range; SD, standard deviation.
According to the CMAQ-fused air pollution concentrations (Table 2), the average of site-specific daily PM2.5 concentration was 10.5 μg/m3 (SD=5.5 μg/m3) with a maximum (of site-specific daily concentrations) of 53.5 μg/m3 [25th, 75th percentile=(36.1 μg/m3,63.1 μg/m3)]. O3 had an average of 0.04 ppm (SD=0.01 ppm) with a maximum of 0.10 ppm [25th, 75th percentile=(0.09 ppm, 0.10 ppm, respectively)]. NO2, NOx, SO2, and CO had averages of site-specific daily concentrations of 22.6 ppb, 48.7 ppb, 3.9 ppb, and 0.6 ppm, respectively.
Table 2 Summary statistics for daily, site-specific air pollution and meteorological parameters at 53 AQS sites over the study period 2005–2014. An overall summary statistic (mean, SD, minimum, or maximum) of a parameter is a summary of 53 site-specific statistics. The 25th and 75th percentiles correspond to the percentiles of 53 site-specific statistics.
Parameter Overall mean (25th, 75th percentile) Overall SD (25th, 75th percentile) Overall minimum (25th, 75th percentile) Overall maximum (25th, 75th percentile)
PM
24-h avg PM2.5 (μg/m3) 10.5 (9.6, 11.2) 5.5 (4.3, 6.1) 1.6 (1.2, 2.1) 53.5 (36.1, 63.1)
24-h avg PM10–2.5 (μg/m3) 12.0 (7.0, 14.4) 7.0 (4.2, 8.3) 0.1 (0.0, 0.0) 76.8 (34.7, 77.0)
24-h avg EC (μg/m3) 0.7 (0.5, 0.8) 0.4 (0.3, 0.6) 0.0 (0.0, 0.1) 5.4 (2.7, 6.7)
24-h avg OC (μg/m3) 3.1 (2.5, 3.7) 2.3 (1.6, 2.9) 0.3 (0.1, 0.4) 34.1 (14.3, 45.7)
24-h avg nitrate (μg/m3) 1.9 (1.0, 2.0) 2.4 (1.1, 2.8) 0.0 (0.0, 0.0) 37.0 (12.4, 47.1)
24-h avg sulfate (μg/m3) 2.0 (1.3, 2.7) 1.8 (0.9, 2.5) 0.1 (0.0, 0.1) 24.3 (9.8, 27.8)
Gaseous pollutants
8-h max O3 (ppm) 0.04 (0.04, 0.04) 0.01 (0.01, 0.01) 0.01 (0.00, 0.01) 0.10 (0.09, 0.10)
1-h max NO2 (ppb) 22.6 (17.0, 27.5) 9.6 (7.9, 11.4) 2.9 (1.3, 4.3) 70.8 (55.0, 84.5)
1-h max NOx (ppb) 48.7 (32.7, 60.2) 38.1 (21.6, 50.3) 3.4 (1.6, 4.6) 323.7 (179.6, 442.1)
1-h max SO2 (ppb) 3.9 (1.8, 5.5) 3.8 (1.6, 4.8) 0.1 (0.0, 0.1) 42.8 (19.0, 50.9)
1-h max CO (ppm) 0.6 (0.5, 0.8) 0.4 (0.2, 0.5) 0.1 (0.1, 0.1) 3.1 (2.1, 3.7)
Meteorology
1-h max temperature (K) 294.1 (291.2, 296.7) 8.7 (7.9, 10.2) 269.7 (263.3, 278.0) 317.8 (314.2, 319.6)
24-h avg dew point temperature (K) 279.3 (278.1, 281.3) 7.9 (5.6, 10.0) 253.3 (249.3, 257.0) 294.7 (292.1, 297.0)
Note: AQS, Air Quality System; avg, average; CO, carbon monoxide; EC, elemental carbon; max, maximum; OC, organic carbon; PM, particulate matter; SD, standard deviation.
Figure S2 shows the mean site-specific Pearson correlation coefficients of daily concentrations among the CMAQ-fused air pollutants. Most correlation coefficients among air pollutants were below 0.5. NO2 and NOx had a correlation coefficient of 0.78 because NO2 is the most prevalent form of NOx. Traffic-related air pollutants had high correlations, especially between NOx and CO (correlation coefficient=0.75); EC was moderately correlated with other traffic-related pollutants (NOx, NO2, and CO; correlation coefficients around 0.5).
Relative Risks for All Ages
Table 3 shows the 10-state pooled RRs and 95% CIs for associations between IQR increases in 3-d immediate (distributed lags 0–3), 4-d delayed (distributed lags 3–6), and 8-d prolonged (distributed lags 0–7) cumulative exposure to individual air pollutants and asthma ED visits for the entire study population across all age groups. We observed that, in general, increases in cumulative exposure to air pollutants were positively associated with increased rates of asthma ED visits [e.g., 8-d prolonged exposure to PM2.5: 1.016 (95% CI: 1.008, 1.025) per 6.3 μg/m3 increase, PM10–2.5: 1.014 (95% CI: 1.007, 1.020) per 9.6 μg/m3 increase, OC: 1.016 (95% CI: 1.009, 1.024) per 2.8 μg/m3 increase, NO2: 1.025 (95% CI: 1.012, 1.039) per 18.9 ppb increase, and O3: 1.008 (95% CI: 0.995, 1.022) per 0.02 ppm increase]. For most of the pollutants, as expected, 8-d prolonged exposure had stronger effects than 3-d immediate exposure. PM10–2.5 and nitrate had similar effects for three cumulative exposure windows. PM2.5 and sulfate had the weakest effects for the 4-d delayed exposure. O3 had the strongest effect for the 3-d immediate exposure.
Table 3 Pooled effects of cumulative air pollution exposure on asthma emergency department visits across all age groups (n=3,190,333).
Pollutant IQR Pooled RR (95% CI)
Distributed lags 0–2
PM PM2.5 6.3μg/m3 1.013 (1.007, 1.020)
PM10–2.5 9.6 μg/m3 1.011 (1.007, 1.016)
Major PM2.5 components EC 0.5 μg/m3 1.005 (1.001, 1.008)
OC 2.8 μg/m3 1.009 (1.005, 1.012)
Nitrate 1.7 μg/m3 1.002 (0.999, 1.005)
Sulfate 1.8 μg/m3 1.008 (1.005, 1.012)
Gaseous pollutants O3 0.02 ppm 1.022 (1.014, 1.030)
NOx 47.6 ppb 0.996 (0.991, 1.002)
NO2 18.9 ppb 1.004 (0.995, 1.012)
CO 0.5 ppm 1.003 (0.998, 1.009)
SO2 4.2 ppb 1.001 (0.997, 1.005)
Distributed lags 3–6
PM PM2.5 6.3 μg/m3 1.007 (1.002, 1.011)
PM10–2.5 9.6 μg/m3 1.014 (1.008, 1.021)
Major PM2.5 components EC 0.5 μg/m3 1.006 (1.000, 1.012)
OC 2.8 μg/m3 1.010 (1.005, 1.016)
Nitrate 1.7 μg/m3 1.001 (0.998, 1.005)
Sulfate 1.8 μg/m3 1.001 (0.996, 1.006)
Gaseous pollutants O3 0.02 ppm 1.002 (0.991, 1.012)
NOx 47.6 ppb 1.010 (1.005, 1.016)
NO2 18.9 ppb 1.019 (1.011, 1.027)
CO 0.5 ppm 1.007 (0.999, 1.016)
SO2 4.2 ppb 1.004 (0.999, 1.009)
Distributed lags 0–7
PM PM2.5 6.3 μg/m3 1.016 (1.008, 1.025)
PM10–2.5 9.6 μg/m3 1.014 (1.007, 1.020)
Major PM2.5 components EC 0.5 μg/m3 1.010 (1.002, 1.018)
OC 2.8 μg/m3 1.016 (1.009, 1.024)
Nitrate 1.7 μg/m3 1.004 (0.999, 1.009)
Sulfate 1.8 μg/m3 1.008 (1.001, 1.015)
Gaseous pollutants O3 0.02 ppm 1.008 (0.995, 1.022)
NOx 47.6 ppb 1.013 (1.005, 1.020)
NO2 18.9 ppb 1.025 (1.012, 1.039)
CO 0.5 ppm 1.013 (1.001, 1.026)
SO2 4.2 ppb 1.007 (0.999, 1.014)
Note: CI, credible interval; CO, carbon monoxide; EC, elemental carbon; IQR, interquartile range; OC, organic carbon; PM, particulate matter; RR, rate ratio.
Figure 1 and Table S1 show the pooled RRs and 95% CIs for associations between increases in single-day exposure (from lag 0 to lag 7) to individual air pollutants and increased rates of asthma ED visits for the entire study population across all age groups. OC, traffic-related pollutants (EC, NOx, NO2, and CO), and SO2 (the latter mostly from the burning of fossil fuels in power plants and other industrial facilities) had a similar single-day lag pattern in that the effects were generally weaker for exposure at shorter lags (lag 1 to lag 2) than at longer lags (lag 4 to lag 6) before the ED visit. The same-day (lag 0) exposure had similar effects as exposure at lag 4 to lag 6 for these pollutants. O3, PM2.5, and sulfate, on the other hand, had stronger effects at shorter lags. The effects of exposure to PM10–2.5 and nitrate had relatively uniform distributions across lags.
Figure 1. Pooled effects of single-day air pollution exposure (lag 0 to lag 7) on asthma emergency department visits across all age groups (n=3,190,333); units: per 6.3-μg/m3 increase in PM2.5, per 9.6-μg/m3 increase in PM10–2.5, per 0.5-μg/m3 increase in EC, per 2.8-μg/m3 increase in OC, per 1.7-μg/m3 increase in nitrate, per 1.8-μg/m3 increase in sulfate, per 0.02-ppm increase in O3, per 47.6-ppb increase in NOx, per 18.9-ppb increase in NO2, per 0.5-ppm increase in CO, and per 4.2-ppb increase in SO2; corresponding numeric data is located in Table S1. Note: CO, carbon monoxide; EC, elemental carbon; OC, organic carbon; PM, particulate matter.
Figure 1 eleven error bar graphs, plotting rate ratio, ranging from 0.99 to 1.02 in increments of 0.01 for ozone; 0.990 to 1.005 in increments of 0.005 for oxides of nitrogen; 0.99 to 1.01 in increments of 0.01 for nitrogen dioxide; 0.995 to 1.010 in increments of 0.005 for Carbon monoxide; 0.998 to 1.004 in increments of 0.002 for sulfur dioxide; 1.000 to 1.010 in increments of 0.005 for particulate matter begin subscript 2.5 end subscript; 1.000 to 1.015 in increments of 0.005 for particulate matter begin subscript 10 to 2.5 end subscript; 1.000 to 1.008 in increments of 0.004 for elemental carbon; 1.000 to 1.008 in increments of 0.004 for organic carbon; 0.998 to 1.004 in increments of 0.002 for nitrate; and 1.000 to 1.005 in increments of 0.005 for sulfate (y-axis) across lag day, ranging from Lag 0 to Lag 7 in unit increments (x-axis).
To quantify heterogeneity in risk estimates across sites, Table S2 shows the estimated SDs of the site-specific log RRs (τ in Equation 2) associated with each air pollutant; τ is a measure reflecting how site-specific log RRs varied around the pooled estimate per IQR increase in pollution concentration. It is notable that the values for CO, PM2.5, NO2, O3, and EC were relatively high, potentially indicating larger heterogeneity in their effects on asthma ED visits across sites in comparison with other pollutants. The observed heterogeneity across sites may be associated with differences in exposure, population susceptibility, access to health care, etc.
Age Group–Specific Relative Risks
Figure 2 and Table S3 show the pooled RRs and 95% CIs for associations between IQR increases in 3-d immediate (distributed lags 0–3), 4-d delayed (distributed lags 3–6), and 8-d prolonged (distributed lags 0–7) cumulative exposure to individual air pollutants and increased rates of asthma ED visits for subpopulations by age group. We observed that for NOx, NO2, CO, SO2, PM10–2.5, nitrate, and sulfate, the effects on younger individuals age 17 y and under were more pronounced than adult age groups. As expected, 8-d prolonged exposure to these pollutants generally had stronger effects than shorter-term exposure. For PM2.5, EC, and OC, the effects were stronger on both young and old populations than on other adults. For O3, the effects were generally stronger on adults age 18 y and above, whereas the effects of 3-d immediate exposure were at a similar level across age groups.
Figure 2. Pooled effects of cumulative air pollution exposure on asthma emergency department visits for individual age groups (n for age group 1–4y=473,890, n for age group 5–17y=746,797, n for age group 18–49y=1,246,191, n for age group 50–64y=433,997, n for age group 65+y=231,384); units: per 6.3-μg/m3 increase in PM2.5, per 9.6-μg/m3 increase in PM10–2.5, per 0.5-μg/m3 increase in EC, per 2.8-μg/m3 increase in OC, per 1.7-μg/m3 increase in nitrate, per 1.8-μg/m3 increase in sulfate, per 0.02-ppm increase in O3, per 47.6-ppb increase in NOx, per 18.9-ppb increase in NO2, per 0.5-ppm increase in CO, and per 4.2-ppb increase in SO2; corresponding numeric data is located in Table S3. Note: CO, carbon monoxide; EC, elemental carbon; OC, organic carbon; PM, particulate matter.
Figure 2 is an error bar graph, plotting Rate ratio, ranging from 0.92 to 1.04 in increments of 0.04 for ozone, 0.99 to 1.05 in increments of 0.03 for oxides of nitrogen; 0.96 to 1.12 in increments of 0.04 for nitrogen dioxide; 0.98 to 1.04 in increments of 0.02 for Carbon monoxide; 0.98 to 1.03 in increments of 0.01 for sulfur dioxide; 0.99 to 1.03 in increments of 0.01 for particulate matter begin subscript 2.5 end subscript; 0.99 to 1.04 in increments of 0.01 for particulate matter begin subscript 10 to 2.5 end subscript; 0.99 to 1.03 in increments of 0.01 for elemental carbon; 0.99 to 1.03 in increments of 0.01 for organic carbon; 1.00 to 1.02 in increments of 0.01 for nitrate; and 0.99 to 1.03 in increments of 0.01 for sulfate (y-axis) across Age group, ranging from 01 to 04, 05 to 17, 18 to 49, 50 to 64, and 65 plus (x-axis) for distributed lags, including lags 0 to 2, lags 3 to 6, and lags 0 to 7.
Region-Specific Relative Risks
Figure 3 and Table S4 show the pooled RRs and 95% CIs for associations between IQR increases in 3-d immediate (distributed lags 0–3), 4-d delayed (distributed lags 3–6), and 8-d prolonged (distributed lags 0–7) cumulative exposure to individual air pollutants and increased rates of asthma ED visits for subpopulations by region (east and west). We observed that for EC, nitrate, O3 (except for 3-d immediate exposure), NOx, NO2, and CO, the effects were stronger in the east. SO2 had stronger effects in the west. Other pollutants generally had similar effects across regions.
Figure 3. Pooled effects of cumulative air pollution exposure on asthma emergency department visits in different regions (n for west=1,036,115, n for east=2,154,218); units: per 6.3-μg/m3 increase in PM2.5, per 9.6-μg/m3 increase in PM10–2.5, per 0.5-μg/m3 increase in EC, per 2.8-μg/m3 increase in OC, per 1.7-μg/m3 increase in nitrate, per 1.8-μg/m3 increase in sulfate, per 0.02-ppm increase in O3, per 47.6-ppb increase in NOx, per 18.9-ppb increase in NO2, per 0.5-ppm increase in CO, and per 4.2-ppb increase in SO2; corresponding numeric data is located in Table S4. Note: CO, carbon monoxide; EC, elemental carbon; OC, organic carbon; PM, particulate matter.
Figure 3 is a set of eleven error bar graphs. On the top, the six graphs are titled particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 to 2.5 end subscript, elemental carbon, organic carbon, nitrate, sulfate, plotting rate ratio, ranging from 0.99 to 1.03 in increments of 0.01 (y-axis) across Region, including West and East (x-axis) for distributed lags, including lags 0 to 2, lags 3 to 6, and lags 0 to 7, respectively. At the bottom, the five graphs are titled ozone, oxides of nitrogen, nitrogen dioxide, carbon monoxide, sulfur dioxide, plotting rate ratio, ranging from 0.98 to 1.04 in increments of 0.02 (y-axis) across Region, including West and East (x-axis) for distributed lags, including lags 0 to 2, lags 3 to 6, and lags 0 to 7, respectively.
Two-Pollutant Models
PM2.5 and O3 had consistent effects with similar single-day lag patterns in both single- and two-pollutant model settings (Figure S3). A similar lag structure, in which the effects were stronger at shorter lags (except for lag 0), was observed in both model settings for O3.
After adjusting for PM2.5, the effects of all traffic-related air pollutants (EC, NO2, NOx, and CO) at lag 0 to lag 4 became smaller or negative (RRs<1.0), whereas the effects at longer lags (lag 5 to lag 7) were more consistent with those in the single-pollutant setting (Figure S4). In contrast, after adjusting for O3, the effects of these pollutants were consistent with those in the single-pollutant setting across lags. A possible reason for the shift of effects is that the traffic-related pollutants had higher correlations with PM2.5 than with O3.
Sensitivity Analyses
The sensitivity analyses did not alter the overall conclusions. The relative risk estimates of asthma ED visits associated with short-term exposure to air pollution were consistent with different degrees of freedom (df=2–6) of natural cubic splines of daily maximum T and mean DPT (Figures S5 and S6). The relative risk estimates were also consistent with different monthly knots (df=0.7–1.3) (Figure S7). Our models generated similar relative risk estimates when controlling for different definitions of influenza activity (Figure S8). Except for O3, all other pollutants showed no significant associations between tomorrow’s (lag −1) pollution concentrations and today’s (lag 0) ED visits when controlling for today’s pollution concentrations, and the lag 0 RRs and 95% CIs remained about the same before and after controlling for tomorrow’s pollution concentrations (Figure S9). Last, the estimated single-day effects of air pollution exposure on asthma ED visits were consistent between using the CMAQ-fused and AQS data as exposure estimates (Figure S10 and Table S5).
Discussion
Using a multisite time-series design spanning 10 U.S. states, this analysis observed positive associations between increases in short-term exposure to multiple ambient air pollutants, including PM2.5, PM10–2.5, major PM2.5 components (EC, OC, sulfate, and nitrate), and gaseous pollutants (O3, NOx, NO2, SO2, and CO), and increased rates of ED visits for asthma. We observed differential effects of air pollution on asthma across age groups, in that NOx, NO2, CO, SO2, PM10–2.5, nitrate, and sulfate had stronger effects on children and adolescents than adult age groups; PM2.5, EC, and OC had strong effects on both the young and older populations; and O3 had stronger effects on adult age groups than children and adolescents in general. This analysis provides robust evidence with regard to adverse effects of short-term air pollution exposure on asthma-related outcomes.
This analysis improves on prior short-term air pollution-asthma morbidity studies in four ways. First, our study domain covers multiple U.S. states with a large number of asthma ED visits, and thus our study population may be more representative of the U.S. population than prior studies. Second, we assessed associations across age groups from children age 1–4 y to the elderly population age 65+ y. Third, we examined a wide spectrum of air pollutants that included criteria air pollutants and PM2.5 components. Finally, our use of temporally continuous exposure estimates allowed both single-day and multiday cumulative effect analyses. This analysis has thus provided a comprehensive and systematic view of the effects of air pollution on asthma morbidity across various regions of the United States, allowing for rigorous assessment of pooled effects and effect heterogeneity.
The air pollutants this analysis focused on are major ambient pollutants mostly from anthropogenic sources, with extensive epidemiological evidence regarding their adverse effects on acute cardiovascular and other respiratory disease outcomes.49–51 The anthropogenic sources of PM, a mixture of solid particles and liquid droplets, include motorized vehicles, industrial processes, power generation, agriculture, road dust, and residential wood burning.52 Wildland fires, as a natural source, also have played an increasing role in heavy, episodic PM pollution in recent decades.53,54 O3 is a secondary pollutant formed by photochemical reactions between NOx and volatile organic compounds (VOCs) as major precursors.55 The two precursors are mainly from mobile sources and industrial processes in urban areas, whereas wildland fires and biogenic emissions are important natural sources of VOCs.56 In regard to other gases, SO2 is primarily produced by the burning of fossil fuels that contain sulfur during energy production and industrial processes, and CO is principally emitted from fossil fuel combustion.57
Our findings are generally consistent with previous work demonstrating acute asthmatic effects of short-term exposure to air pollution across all age groups, with a similar magnitude of effect.9,10,12,17,25,30,58 For example, based on ED data from 17 U.S. states, Strosnider et al.25 estimated that an increase of 0.02 ppm in daily 8-h maximum O3 was associated with 1.055-fold higher rate of asthma ED visits (95% CI: 1.048, 1.063); and an increase of 10 μg/m3 in daily mean PM2.5 was associated with 1.038-fold higher rate of asthma ED visits (95% CI: 1.030, 1.045). A meta-analysis with effect estimates mostly from countries in North America and Europe reported pooled relative risks (per 10 μg/m3 increase) of 1.008 (95% CI: 1.005, 1.011; number of effect sizes=27) for daily 8-h maximum or 24-h mean O3, 1.014 (95% CI: 1.008, 1.020; n=22) for daily mean NO2, and 1.010 (95% CI: 1.001, 1.020; n=23) for daily mean SO2 associated with asthma-related ED visits.10 However, the pooled associations between asthma ED visits and daily 1-h maximum NO2 and SO2 reported by Zheng et al.10 were close to the null (0.999 for NO2 and 1.003 for SO2). In contrast, we observed positive associations (RRs>1.0) for both pollutants, possibly due to the higher statistical power and a more consistent quality of our exposure and health data. Additionally, the single-day lag patterns of certain air pollutants observed in our analysis, especially PM2.5 and O3, are consistent with a previous finding: Strickland et al.30 found a tendency that warm-season PM2.5 and O3 concentrations were associated with higher rates of pediatric asthma ED visits at shorter lags in Atlanta, Georgia. Our analysis also revealed a unique lag-specific pattern that the effects of traffic-related air pollutants (NO2, NOx, and CO in particular) tended to peak at lag 0 followed by protective effects at lag 1, with another peak at lag 4 to lag 6. This pattern may reflect two different types of asthma-related outcomes in which lag 0 was associated with more acute outcomes. The lack of detailed diagnosis information in electronic hospital billing records limited our ability to fully interpret the observed pattern, and this pattern is worth further explorations in studies with more detailed asthma-related health data.
Our age group–stratified analysis adds to the evidence that children may be more susceptible to air pollution (especially NOx, NO2, CO, SO2, PM10–2.5, nitrate, and sulfate) with higher risks of acute asthma-related outcomes.9,10,11,58 Possible explanations of childhood vulnerability to air pollution include: a) immature growth of airways and underdeveloped host defense capacity,10 b) a higher ratio of inhaled air volume to body weight than that found in adults,9 and c) higher physical activity levels that may potentially be associated with increased time spent outdoors and increased exposure to ambient air pollution than adults.59 Children age 1–4 y may experience transient wheeze, and asthma diagnoses can be challenging.30 ED visits for asthma selected based on the ICD-9 code may actually identify a mix of asthma and wheeze for this young population. However, it is still valuable to report their effects of air pollution exposure because this age range accounted for 15% of total ED visits in our study population and ED visits indicate severe symptoms that require emergency care. Our results can help identify and prevent potential asthma triggers and benefit long-term control of the disease among children.
Our analysis also shows that exposure to some air pollutants (PM2.5, EC, OC, and O3) were associated with strong adverse effects on the elderly population. Despite asthma being usually considered a disease of young people, older people suffer disproportionately from asthma; people age 65+ y have the highest rate of asthma mortality in the United States.60 There are at least two phenotypes among older patients with asthma: asthma of early onset that has persisted into older adulthood (“longstanding”) and asthma that starts in middle age or older (“late-onset”).61 A plausible mechanism of the effects of air pollution is that short-term exposure to air pollution may amplify inflammatory responses of remodeled airways in the elderly.11,58 Asthma in the elderly is underdiagnosed and undertreated61; with an ever-increasing elderly population worldwide, identifying risk factors of asthma may benefit detection and proper management of the disease in old age, with a great impact from the public health perspective.
Our analysis showed evidence of differential effects of air pollution on asthma across regions, where EC, nitrate, O3, NOx, NO2, and CO had stronger effects in the eastern states, and SO2 had stronger effects in the western states. Differential air pollution characteristics (e.g., pollution concentration, composition, and exposure), population susceptibility, and access to health care may play a role in the observed differences. The regional difference in the health effects of air pollution exposure is an important topic that is worth additional research with spatially more complete health data (e.g., ZIP code–level ED visits and hospitalization data covering the entire state, as opposed to our current data surrounding AQS sites).
There are several plausible biological mechanisms that air pollution triggers asthma exacerbations: oxidative stress from reactive oxygen species, airway remodeling and inflammation, and sensitization to aeroallergens.62–64 Pulmonary inflammation may also serve as an indirect cause of asthma exacerbations through its impacts on host defenses and respiratory viral infections.65,66 Specifically, PM, a mixture of various chemical species, imposes more complex impacts on asthmatic airways and may cause oxidative stress and airway remodeling and hyper-responsiveness.67 O3 and NO2 have been demonstrated in generating free radicals that impair the function and structure of airways and releasing inflammatory mediators.68 SO2 has been found to promote airway inflammation and eosinophilia.69 CO may serve as a proxy of air pollutants from incomplete combustion, whereas its biological mechanisms associated with asthma exacerbations are less certain.11,70 Although the CO concentrations observed in this analysis were low in comparison with the current regulatory standards, no known safe threshold has been found for CO exposure.57 Our recent exposure modeling study demonstrated substantial CO concentration variations at a finer, intracity scale in the United States.71 This current analysis further provided evidence of the adverse effects of CO, either as a primary pollutant or a proxy of other pollutants, at low exposure levels.
Our analysis has several strengths. The analysis was based on a multistate electronic hospital billing records data set in which ED diagnosis data were subject to standardized preprocessing and quality assurance procedures that aimed to provide a consistent quality across states. This large-scale ED data set allowed sufficient statistical power to estimate robust effect estimates for both all-age and age group–stratified analyses. In contrast, the vast majority of prior studies on this topic confined to much smaller study areas.72,73 In addition, we used temporally continuous exposure estimates available for major PM components and multiple criteria gaseous pollutants.37 The temporal continuity improved the comparability of single-day effect estimates and allowed the analysis of multiday (distributed-lag) exposure. Also, the use of continuous exposure time series minimized the potential bias being introduced due to dropping missing/incomplete records.74 Because the improved exposure estimates produced by Senthilkumar et al.37 are also spatially complete, future analyses may be expanded spatially by linking ED data at locations not represented by the AQS monitoring network, particularly in nonurban environments. Last, our analysis rigorously controlled for potential confounders of the air pollution–asthma association through the adjustment for meteorology (air temperature and humidity), influenza activity, long-term time trend, and other important indictors. It is worth mentioning that our analysis is among the first to account for influenza-related confounding for respiratory disease outcomes. Future analyses may further consider temporal activities of other respiratory viruses, including respiratory syncytial virus and SARS-CoV-2.
We acknowledge several limitations to this analysis. First, although we adopted improved air pollution exposure estimates by bias-correcting chemical transport model simulations using ground-level air pollution observations, exposure misclassification was still expected to exist, and the levels of misclassification might differ among different air pollutants.37 The potential exposure misclassification, if nondifferential, may result in bias toward the null and thus underestimated asthma-related effect estimates.75–77 Second, our use of hospital billing records may be subject to ascertainment bias.78 However, the potential ascertainment bias is less likely to be correlated with air pollution concentrations and to bias the magnitude of the observed asthma-related effect estimates. Third, as we assessed associations between asthma ED visits and multiple air pollutants, some statistically significant associations may have been observed by chance. Further, the statistically significant negative association observed in the negative outcome control analysis for O3, which was not expected, may indicate the potential residual confounding specifically for the pollutant. However, this association may also be due to the large number of pollutants investigated. Last, we assumed the effects of air pollution exposure on asthma-related outcomes to be linear in this analysis. The assumption of nonlinear exposure–response relationships is worth additional investigations, which is beyond the scope of this study.
In summary, this analysis assessed the risk of ED visits for asthma associated with short-term exposure to PM2.5, PM10–2.5, major PM2.5 components, and gaseous pollutants in 10 U.S. states over a 10-y period. We reported adverse effects of all air pollutants investigated on asthma-related outcomes and differential effects of air pollution that posed a higher risk to children and older populations. This analysis is among the first to investigate acute effects of air pollution on asthma across the entire age range at the U.S. national scale with a comprehensive set of pollutants, emphasizing the urgent need of further reduction of air pollution to avert increase in asthma morbidity.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This research was supported by funding from the National Institute of Environmental Health Sciences (NIEHS) of the National Institutes of Health (NIH) under award numbers R01ES027892 and P30ES019776. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. We are grateful for the support of the health data sources listed below and their contributing hospitals. The data used to produce this publication were acquired from the: Arizona Department of Health Services (years 2010–2014); California Office of Statewide Planning and Development, now California Department of Health Care Access and Information (years 2005–2014); Georgia Hospital Association (years 2011–2014); Maryland Department of Health, Health Services Cost Review Commission (years 2005–2014); Missouri Department of Health and Senior Services (years 2005–2014); Nevada Division of Health Care Financing and Policy (DJCFP), released through the Center for Health Information Analysis (CHIA) of the University of Nevada, Las Vegas (years 2009–2014); New Jersey Department of Health, Center for Health Statistics & Informatics, Trenton, New Jersey (years 2005–2014); New York State Department of Health, Statewide Planning and Research Cooperative System (years 2005–2014); North Carolina Hospital (inpatient, ambulatory surgery/outpatient, emergency room) Discharge Database (Truven Health Analytics, years 2007–2014) from the Cecil G. Sheps Center for Health Services Research and the North Carolina Division of Health Service Regulation; and Utah Department of Health, Office of Health Care Statistics (years 2005–2014).
The contents of this publication, including data analysis, interpretation, conclusions derived, and the views expressed herein are solely those of the authors and do not represent the conclusions or official views of data sources listed above. Authorization to release this information does not imply endorsement of this study or its findings by any of these data sources. The data sources, their employees, officers, and agents make no representation, warranty, or guarantee as to the accuracy, completeness, currency, or suitability of the information provided here.
==== Refs
References
1. GBD (Global Burden of Diseases) 2017 Risk Factor Collaborators. 2018. Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 392 (10159 ):1923–1994, PMID: , 10.1016/S0140-6736(18)32225-6.30496105
2. GBD 2019 Diseases and Injuries Collaborators. 2020. Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 396 (10258 ):1204–1222, PMID: , 10.1016/S0140-6736(20)30925-9.33069326
3. Butt EW, Turnock ST, Rigby R, Reddington CL, Yoshioka M, Johnson JS, et al. 2017. Global and regional trends in particulate air pollution and attributable health burden over the past 50 years. Environ Res Lett 12 (10 ):104017. 10.1088/1748-9326/aa87be.
4. Fuller GW, Font A. 2019. Keeping air pollution policies on track. Science 365 (6451 ):322–323, PMID: , 10.1126/science.aaw9865.31346054
5. World Health Organization. 2016. Ambient Air Pollution: A Global Assessment of Exposure and Burden of Disease. Geneva, Switzerland : World Health Organization.
6. Health Effects Institute. 2015. HEI Strategic Plan for Understanding the Health Effects of Air Pollution 2015–2020. Boston, MA: Health Effects Institute.
7. U.S. Centers for Disease Control and Prevention . 2019 . 2019 National Health Interview Survey (NHIS) Data. https://www.cdc.gov/asthma/nhis/2019/data.htm [accessed 28 March 2023].
8. Nurmagambetov T, Kuwahara R, Garbe P. 2018. The economic burden of asthma in the United States, 2008–2013. Ann Am Thorac Soc 15 (3 ):348–356, PMID: , 10.1513/AnnalsATS.201703-259OC.29323930
9. Zhang S, Li G, Tian L, Guo Q, Pan X. 2016. Short-term exposure to air pollution and morbidity of COPD and asthma in East Asian area: a systematic review and meta-analysis. Environ Res 148 :15–23, PMID: , 10.1016/j.envres.2016.03.008.26995350
10. Zheng XY, Orellano P, Lin HL, Jiang M, Guan WJ. 2021. Short-term exposure to ozone, nitrogen dioxide, and sulphur dioxide and emergency department visits and hospital admissions due to asthma: a systematic review and meta-analysis. Environ Int 150 :106435, PMID: , 10.1016/j.envint.2021.106435.33601224
11. Zheng XY, Ding H, Jiang LN, Chen SW, Zheng JP, Qiu M, et al. 2015. Association between air pollutants and asthma emergency room visits and hospital admissions in time series studies: a systematic review and Meta-Analysis. PLoS One 10 (9 ):e0138146, PMID: , 10.1371/journal.pone.0138146.26382947
12. Orellano P, Quaranta N, Reynoso J, Balbi B, Vasquez J. 2017. Effect of outdoor air pollution on asthma exacerbations in children and adults: systematic review and multilevel meta-analysis. PLoS One 12 (3 ):e0174050, PMID: , 10.1371/journal.pone.0174050.28319180
13. Delfino RJ, Wu J, Tjoa T, Gullesserian SK, Nickerson B, Gillen DL. 2014. Asthma morbidity and ambient air pollution: effect modification by residential traffic-related air pollution. Epidemiology 25 (1 ):48–57, PMID: , 10.1097/EDE.0000000000000016.24240657
14. Hooper LG, Kaufman JD. 2018. Ambient air pollution and clinical implications for susceptible populations. Ann Am Thoracic Soc 15 (suppl 2 ):S64–S68, PMID: , 10.1513/AnnalsATS.201707-574MG.29676646
15. Baxter LK, Sacks JD. 2014. Clustering cities with similar fine particulate matter exposure characteristics based on residential infiltration and in-vehicle commuting factors. Sci Total Environ 470–471 :631–638, PMID: , 10.1016/j.scitotenv.2013.10.019.24176711
16. Gass K, Klein M, Sarnat SE, Winquist A, Darrow LA, Flanders WD, et al. 2015. Associations between ambient air pollutant mixtures and pediatric asthma emergency department visits in three cities: a classification and regression tree approach. Environ Health 14 :58, PMID: , 10.1186/s12940-015-0044-5.26123216
17. Alhanti BA, Chang HH, Winquist A Mulholland JA, Darrow LA, Sarnat SE. 2016. Ambient air pollution and emergency department visits for asthma: a multi-city assessment of effect modification by age. J Expo Sci Environ Epidemiol 26 (2 ):180–188, PMID: , 10.1038/jes.2015.57.26350981
18. Castner J, Guo L, Yin Y. 2018. Ambient air pollution and emergency department visits for asthma in Erie County, New York 2007–2012. Int Arch Occup Environ Health 91 (2 ):205–214, PMID: , 10.1007/s00420-017-1270-7.29043427
19. Chang HH, Pan A, Lary DJ, Waller LA, Zhang L, Brackin BT, et al. 2019. Time-series analysis of satellite-derived fine particulate matter pollution and asthma morbidity in Jackson, MS. Environ Monit Assess 191 (suppl 2 ):280, PMID: , 10.1007/s10661-019-7421-4.31254082
20. Bi J, D’Souza RR, Rich DQ, Hopke PK, Russell AG, Liu Y, et al. 2020. Temporal changes in short-term associations between cardiorespiratory emergency department visits and PM2.5 in Los Angeles, 2005 to 2016. Environ Res 190 :109967, PMID: , 10.1016/j.envres.2020.109967.32810677
21. Byrwa-Hill BM, Venkat A, Presto AA, Rager JR, Gentile D, Talbott E. 2020. Lagged association of ambient outdoor air pollutants with asthma-related emergency department visits within the Pittsburgh region. Int J Environ Res Public Health 17 (22 ):8619, PMID: , 10.3390/ijerph17228619.33233547
22. Sacks JD, Rappold AG, Davis JA, Richardson DB, Waller AE, Luben TJ. 2014. Influence of urbanicity and county characteristics on the association between ozone and asthma emergency department visits in North Carolina. Environ Health Perspect 122 (5 ):506–512, PMID: , 10.1289/ehp.1306940.24569869
23. Malig BJ, Pearson DL, Chang YB, Broadwin R, Basu R, Green RS, et al. 2016. A time-stratified case–crossover study of ambient ozone exposure and emergency department visits for specific respiratory diagnoses in California (2005–2008). Environ Health Perspect 124 (6 ):745–753, PMID: , 10.1289/ehp.1409495.26647366
24. Gharibi H, Entwistle MR, Ha S, Gonzalez M, Brown P, Schweizer D, et al. 2019. Ozone pollution and asthma emergency department visits in the Central Valley, California, USA, during June to September of 2015: a time-stratified case-crossover analysis. J Asthma 56 (10 ):1037–1048, PMID: , 10.1080/02770903.2018.1523930.30299181
25. Strosnider HM, Chang HH, Darrow LA, Liu Y, Vaidyanathan A, Strickland MJ. 2019. Age-specific associations of ozone and fine particulate matter with respiratory emergency department visits in the United States. Am J Respir Crit Care Med 199 (7 ):882–890, PMID: , 10.1164/rccm.201806-1147OC.30277796
26. Cisneros R, Gharibi H, Entwistle MR, Tavallali P, Singhal M, Schweizer D, et al. 2021. Nitrogen dioxide and asthma emergency department visits in California, USA during cold season (November to February) of 2005 to 2015: a time-stratified case-crossover analysis. Sci Total Environ 754 :142089–33254941, PMID: , 10.1016/j.scitotenv.2020.142089.33254941
27. Goodman JE, Zu K, Loftus CT, Tao G, Liu X, Lange S. 2017. Ambient ozone and asthma hospital admissions in Texas: a time-series analysis. Asthma Res Pract 3 :6, PMID: , 10.1186/s40733-017-0034-1.28794889
28. Zu K, Liu X, Shi L, Tao G, Loftus CT, Lange S, et al. 2017. Concentration-response of short-term ozone exposure and hospital admissions for asthma in Texas. Environ Int 104 :139–145, PMID: , 10.1016/j.envint.2017.04.006.28434561
29. Baek J, Kash BA, Xu X, Benden M, Roberts J, Carrillo G. 2020. Effect of ambient air pollution on hospital readmissions among the pediatric asthma patient population in South Texas: a case-crossover study. Int J Environ Res Public Health 17 (13 ):4846, PMID: , 10.3390/ijerph17134846.32640508
30. Strickland MJ, Darrow LA, Klein M, Flanders WD, Sarnat JA, Waller LA, et al. 2010. Short-term associations between ambient air pollutants and pediatric asthma emergency department visits. Am J Respir Crit Care Med 182 (3 ):307–316, PMID: , 10.1164/rccm.200908-1201OC.20378732
31. Li S, Batterman S, Wasilevich E, Wahl R, Wirth J, Su FC, et al. 2011. Association of daily asthma emergency department visits and hospital admissions with ambient air pollutants among the pediatric Medicaid population in Detroit: time-series and time-stratified case-crossover analyses with threshold effects. Environ Res 111 (8 ):1137–1147, PMID: , 10.1016/j.envres.2011.06.002.21764049
32. Gleason JA, Bielory L, Fagliano JA. 2014. Associations between ozone, PM2.5, and four pollen types on emergency department pediatric asthma events during the warm season in New Jersey: a case-crossover study. Environ Res 132 :421–429, PMID: , 10.1016/j.envres.2014.03.035.24858282
33. Xiao Q, Liu Y, Mulholland JA, Russell AG, Darrow LA, Tolbert PE, et al. 2016. Pediatric emergency department visits and ambient air pollution in the U.S. State of Georgia: a case-crossover study. Environ Health 15 (1 ):115, PMID: , 10.1186/s12940-016-0196-y.27887621
34. O’Lenick CR, Winquist A, Mulholland JA, Friberg MD, Chang HH, Kramer MR, et al. 2017. Assessment of neighbourhood-level socioeconomic status as a modifier of air pollution-asthma associations among children in Atlanta. J Epidemiol Community Health 71 (2 ):129–136, PMID: , 10.1136/jech-2015-206530.27422981
35. Huang W, Schinasi LH, Kenyon CC, Moore K, Melly S, Hubbard RA, et al. 2021. Effects of ambient air pollution on childhood asthma exacerbation in the Philadelphia metropolitan region, 2011–2014. Environ Res 197 :110955, PMID: , 10.1016/j.envres.2021.110955.33676951
36. Thomas N, Ebelt ST, Newman AJ, Scovronick N, D’Souza RR, Moss SE, et al. 2021. Time-series analysis of daily ambient temperature and emergency department visits in five US cities with a comparison of exposure metrics derived from 1-km meteorology products. Environ Health 20 (1 ):55, PMID: , 10.1186/s12940-021-00735-w.33962633
37. Senthilkumar N, Gilfether M, Chang HH, Russell AG, Mulholland J. 2022. Using land use variable information and a random forest approach to correct spatial mean bias in fused CMAQ fields for particulate and gas species. Atmos Environ 274 (40 ):118982, 10.1016/j.atmosenv.2022.118982.
38. Buckley JP, Samet JM, Richardson DB. 2014. Commentary: does air pollution confound studies of temperature? Epidemiology 25 (2 ):242–245, PMID: , 10.1097/EDE.0000000000000051.24487206
39. Song X, Wang S, Hu Y, Yue M, Zhang T, Liu Y, et al. 2017. Impact of ambient temperature on morbidity and mortality: an overview of reviews. Sci Total Environ 586 :241–254, PMID: , 10.1016/j.scitotenv.2017.01.212.28187945
40. Schwartz J. 2000. The distributed lag between air pollution and daily deaths. Epidemiology 11 (3 ):320–326, PMID: , 10.1097/00001648-200005000-00016.10784251
41. Croft DP, Zhang W, Lin S, Thurston SW, Hopke PK, Masiol M, et al. 2019. The association between respiratory infection and air pollution in the setting of air quality policy and economic change. Ann Am Thorac Soc 16 (3 ):321–330, PMID: .30398895
42. Bi J, Barry V, Weil EJ, Chang HH, Ebelt S. 2021. Short-term exposure to fine particulate air pollution and emergency department visits for kidney diseases in the Atlanta metropolitan area. Environ Epidemiol 5 (4 ):e164, PMID: , 10.1097/EE9.0000000000000164.34414347
43. Glezen WP. 2006. Asthma, influenza, and vaccination. J Allergy Clin Immunol 118 (6 ):1199–1206. quiz 1207-8, PMID: , 10.1016/j.jaci.2006.08.032.17157648
44. Bhaskaran K, Gasparrini A, Hajat S, et al. 2013. Time series regression studies in environmental epidemiology. Int J Epidemiol 42 (4 ):1187–1195, PMID: , 10.1093/ije/dyt092.23760528
45. Gasparrini A, Armstrong B, Kenward MG. 2010. Distributed lag non-linear models. Stat Med 29 (21 ):2224–2234, PMID: , 10.1002/sim.3940.20812303
46. Lipsitch M, Tchetgen Tchetgen E, Cohen T. 2010. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology 21 (3 ):383–388, PMID: , 10.1097/EDE.0b013e3181d61eeb.20335814
47. Flanders WD, Klein M, Darrow LA, Strickland MJ, Sarnat SE, Sarnat JA, et al. 2011. A method for detection of residual confounding in time-series and other observational studies. Epidemiology 22 (1 ):59–67, PMID: , 10.1097/EDE.0b013e3181fdcabe.21068669
48. Eick-Cost AA, Hunt DJ. 2015. Assessment of ICD-9-based case definitions for influenza-like illness surveillance. MSMR 22 (9 ):2–7, PMID: .26418885
49. Bourdrel T, Bind M-A, Béjot Y, Morel O, Argacha JF. 2017. Cardiovascular effects of air pollution. Arch Cardiovasc Dis 110 (11 ):634–642, PMID: , 10.1016/j.acvd.2017.05.003.28735838
50. Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K, et al. 2017. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 389 (10082 ):1907–1918, PMID: , 10.1016/S0140-6736(17)30505-6.28408086
51. Tiotiu AI, Novakova P, Nedeva D, Chong-Neto HJ, Novakova S, Steiropoulos P, et al. 2020. Impact of air pollution on asthma outcomes. Int J Environ Res Public Health 17 (17 ):6212, PMID: , 10.3390/ijerph17176212.32867076
52. U.S. EPA (U.S. Environmental Protection Agency). 2019. Integrated Science Assessment (ISA) for Particulate Matter (Final Report, December 2019). Washington, DC: U.S. Environmental Protection Agency.
53. Jolly WM, Cochrane MA, Freeborn PH, Holden ZA, Brown TJ, Williamson GJ, et al. 2015. Climate-induced variations in global wildfire danger from 1979 to 2013. Nature Commun 6 :7537, PMID: , 10.1038/ncomms8537.26172867
54. McClure CD, Jaffe DA. 2018. US particulate matter air quality improves except in wildfire-prone areas. Proc Nat Acad Sci U S A 115 (31 ):7901–7906. PMID: 30012611, 10.1073/pnas.1804353115.
55. Krupa SV, Manning WJ. 1988. Atmospheric ozone: formation and effects on vegetation. Environ Pollut 50 (1 ):101–137, PMID: , 10.1016/0269-7491(88)90187-X.15092655
56. Kansal A. 2009. Sources and reactivity of NMHCs and VOCs in the atmosphere: a review. J Hazard Mater 166 (1 ):17–26, PMID: , 10.1016/j.jhazmat.2008.11.048.19136203
57. U.S. EPA. 2010. Integrated Science Assessment for Carbon Monoxide. Washington, DC: US Environmental Protection Agency.
58. Lim H, Kwon HJ, Lim JA, Choi JH, Ha M, Hwang SS, et al. 2016. Short-term effect of fine particulate matter on children’s hospital admissions and emergency department visits for asthma: a systematic review and meta-analysis. J Prev Med Public Health 49 (4 ):205–219, PMID: , 10.3961/jpmph.16.037.27499163
59. Gasana J, Dillikar D, Mendy A, Forno E, Ramos Vieira E. 2012. Motor vehicle air pollution and asthma in children: a meta-analysis. Environ Res 117 :36–45, PMID: , 10.1016/j.envres.2012.05.001.22683007
60. Pate CA, Zahran HS, Qin X, Johnson C, Hummelman E, Malilay J. 2021. Asthma surveillance – United States, 2006–2018. MMWR Surveill Summ 70 (5 ):1–32, PMID: , 10.15585/mmwr.ss7005a1.34529643
61. Hanania NA, King MJ, Braman SS, Saltoun C, Wise RA, Enright P, et al. 2011. Asthma in the elderly: current understanding and future research needs–a report of a National Institute on Aging (NIA) workshop. J Allergy Clin Immunol 128 (3 Suppl ):S4–24, PMID: , 10.1016/j.jaci.2011.06.048.21872730
62. Auerbach A, Hernandez ML. 2012. The effect of environmental oxidative stress on airway inflammation. Curr Opin Allergy Clin Immunol 12 (2 ):133–139, PMID: , 10.1097/ACI.0b013e32835113d6.22306553
63. Esposito S, Tenconi R, Lelii M, Preti V, Nazzari E, Consolo S, et al. 2014. Possible molecular mechanisms linking air pollution and asthma in children. BMC Pulmonary Medicine 14 (1 ):31, PMID: , 10.1186/1471-2466-14-31.24581224
64. Pfeffer PE, Mudway IS, Grigg J. 2021. Air pollution and asthma: mechanisms of harm and considerations for clinical interventions. Chest 159 (4 ):1346–1355, PMID: , 10.1016/j.chest.2020.10.053.33461908
65. Botturi K, Langelot M, Lair D, Pipet A, Pain M, Chesne J, et al. 2011. Preventing asthma exacerbations: what are the targets? Pharmacol Ther 131 (1 ):114–129, PMID: , 10.1016/j.pharmthera.2011.03.010.21440000
66. Brugha R, Grigg J. 2014. Urban air pollution and respiratory infections. Paediatr Respir Rev 15 (2 ):194–199, PMID: , 10.1016/j.prrv.2014.03.001.24704510
67. Stanek LW, Brown JS, Stanek J, Gift J, Costa DL. 2010. Air pollution toxicology—a brief review of the role of the science in shaping the current understanding of air pollution health risks. Toxicol Sci 120 (suppl 1 ):S8–S27, PMID: , 10.1093/toxsci/kfq367.21147959
68. Bayram H, Sapsford RJ, Abdelaziz MM, Khair OA. 2001. Effect of ozone and nitrogen dioxide on the release of proinflammatory mediators from bronchial epithelial cells of nonatopic nonasthmatic subjects and atopic asthmatic patients in vitro. J Allergy Clin Immunol 107 (2 ):287–294, PMID: , 10.1067/mai.2001.111141.11174195
69. Cai C, Xu J, Zhang M, Chen XD, Li L, Wu J, et al. 2008. Prior SO2 exposure promotes airway inflammation and subepithelial fibrosis following repeated ovalbumin challenge. Clin Exp Allergy 38 (10 ):1680–1687, PMID: , 10.1111/j.1365-2222.2008.03053.x.18631350
70. Ji H, Biagini Myers JM, Brandt EB, Brandt EB, Brokamp C, Ryan PH, et al. 2016. Air pollution, epigenetics, and asthma. Allergy Asthma Clin Immunol 12 :51, PMID: , 10.1186/s13223-016-0159-4.27777592
71. Bi J, Zuidema C, Clausen D, Kirwa K, Young MT, Gassett AJ, et al. 2022. Within-city variation in ambient carbon monoxide concentrations: leveraging low-cost monitors in a spatiotemporal modeling framework. Environ Health Perspect 130 (9 ):97008, PMID: , 10.1289/EHP10889.36169978
72. Goodman JE, Zu K, Loftus CT, Lynch HN, Prueitt RL, Mohar I, et al. 2018. Short-term ozone exposure and asthma severity: weight-of-evidence analysis. Environ Res 160 :391–397, PMID: , 10.1016/j.envres.2017.10.018.29059621
73. Madaniyazi L, Xerxes S. 2021. Outdoor air pollution and the onset and exacerbation of asthma. Chronic Dis Transl Med 7 (2 ):100–106, PMID: , 10.1016/j.cdtm.2021.04.003.34136769
74. Junger WL, Ponce de Leon A. 2015. Imputation of missing data in time series for air pollutants. Atmos Environ 102 :96–104, 10.1016/j.atmosenv.2014.11.049.
75. Zeger SL, Thomas D, Dominici F, Samet JM, Schwartz J, Dockery D, et al. 2000. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ Health Perspect 108 (5 ):419–426, PMID: , 10.1289/ehp.00108419.10811568
76. Goldman GT, Mulholland JA, Russell AG, Srivastava A, Strickland MJ, Klein M, et al. 2010. Ambient air pollutant measurement error: characterization and impacts in a time-series epidemiologic study in Atlanta. Environ Sci Technol 44 (19 ):7692–7698, PMID: , 10.1021/es101386r.20831211
77. Dionisio KL, Baxter LK, Chang HH. 2014. An empirical assessment of exposure measurement error and effect attenuation in bipollutant epidemiologic models. Environ Health Perspect 122 (11 ):1216–1224, PMID: , 10.1289/ehp.1307772.25003573
78. Schneider A, Gantner L, Maag I, Borst MM, Wensing M, Szecsenyi J. 2005. Are ICD-10 codes appropriate for performance assessment in asthma and COPD in general practice? Results of a cross sectional observational study. BMC Health Serv Res 5 (1 ):11, PMID: , 10.1186/1472-6963-5-11.15683548
|
PMC010xxxxxx/PMC10075082.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37017430
EHP11989
10.1289/EHP11989
Research
Long-Term Exposure to Air Pollution and COVID-19 Vaccine Antibody Response in a General Population Cohort (COVICAT Study, Catalonia)
https://orcid.org/0000-0002-9605-0461
Kogevinas Manolis 1 2 3 4
Karachaliou Marianna 1
Espinosa Ana 1 2 3 4
Aguilar Ruth 1
Castaño-Vinyals Gemma 1 2 3 4
Garcia-Aymerich Judith 1 2 3
Carreras Anna 5
Cortés Beatriz 5
Pleguezuelos Vanessa 6
Papantoniou Kyriaki 7
Rubio Rocío 1
Jiménez Alfons 1 2
Vidal Marta 1
Serra Pau 8
Parras Daniel 8
Santamaría Pere 8 9
Izquierdo Luis 1 10
Cirach Marta 1
Nieuwenhuijsen Mark 1 2 3
Dadvand Payam 1 2 3
Straif Kurt 1
Moncunill Gemma 1 10
de Cid Rafael 5 *
Dobaño Carlota 1 10 *
Tonne Cathryn 1 2 3 *
1 Barcelona Institute for Global Health, Barcelona, Spain
2 CIBER Epidemiologia y Salud Pública, Madrid, Spain
3 Universitat Pompeu Fabra, Barcelona, Spain
4 Hospital del Mar Medical Research Institute, Barcelona, Spain
5 Genomes for Life-GCAT lab Group, Germans Trias i Pujol Research Institute, Badalona, Spain
6 Banc de Sang i Teixits, Barcelona, Spain
7 Department of Epidemiology, Center of Public Health, Medical University of Vienna, Vienna, Austria
8 Institut d’Investigacions Biomèdiques August Pi Sunyer, Barcelona, Spain
9 Department of Microbiology, Immunology and Infectious Diseases, Snyder Institute for Chronic Diseases, Cumming School of Medicine, University of Calgary, Calgary, Alberta, Canada
10 CIBER Enfermedades Infecciosas, Barcelona, Spain
Address correspondence to Manolis Kogevinas, NCDs & Environment Group, Barcelona Institute for Global Health (ISGlobal), 88 Doctor Aiguader Rd., 08003 Barcelona, Spain. Telephone: 34-93 214 7332. Email: [email protected]
05 4 2023
4 2023
131 4 04700110 8 2022
05 2 2023
22 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Ambient air pollution has been associated with COVID-19 disease severity and antibody response induced by infection.
Objectives:
We examined the association between long-term exposure to air pollution and vaccine-induced antibody response.
Methods:
This study was nested in an ongoing population-based cohort, COVICAT, the GCAT-Genomes for Life cohort, in Catalonia, Spain, with multiple follow-ups. We drew blood samples in 2021 from 1,090 participants of 2,404 who provided samples in 2020, and we included 927 participants in this analysis. We measured immunoglobulin M (IgM), IgG, and IgA antibodies against five viral-target antigens, including receptor-binding domain (RBD), spike-protein (S), and segment spike-protein (S2) triggered by vaccines available in Spain. We estimated prepandemic (2018–2019) exposure to fine particulate matter [PM ≤2.5μm in aerodynamic diameter (PM2.5)], nitrogen dioxide (NO2), black carbon (BC), and ozone (O3) using Effects of Low-Level Air Pollution: A Study in Europe (ELAPSE) models. We adjusted estimates for individual- and area-level covariates, time since vaccination, and vaccine doses and type and stratified by infection status. We used generalized additive models to explore the relationship between air pollution and antibodies according to days since vaccination.
Results:
Among vaccinated persons not infected by SARS-CoV-2 (n=632), higher prepandemic air pollution levels were associated with a lower vaccine antibody response for IgM (1 month post vaccination) and IgG. Percentage change in geometric mean IgG levels per interquartile range of PM2.5 (1.7 μg/m3) were −8.1 (95% CI: −15.9, 0.4) for RBD, −9.9 (−16.2, −3.1) for S, and −8.4 (−13.5, −3.0) for S2. We observed a similar pattern for NO2 and BC and an inverse pattern for O3. Differences in IgG levels by air pollution levels persisted with time since vaccination. We did not observe an association of air pollution with vaccine antibody response among participants with prior infection (n=295).
Discussion:
Exposure to air pollution was associated with lower COVID-19 vaccine antibody response. The implications of this association on the risk of breakthrough infections require further investigation. https://doi.org/10.1289/EHP11989
Supplemental Material is available online (https://doi.org/10.1289/EHP11989).
* These authors contributed equally to this work.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Air pollution has been associated with COVID-19 disease initially in ecological studies and later in cohort studies using individual data.1–6 There are some differences between findings of individual-based studies concerning specific pollutants7 and association with clinical disease,8 but overall results are consistent in showing that long-term exposure to air pollution is associated with COVID-19 disease and severity of the disease. We have previously shown that prepandemic exposure to air pollution in Catalonia was associated with a 20%–50% increased risk of COVID-19 disease and with higher risk for severe COVID-19.9 Even though potential biases, particularly selection bias and confounding, were important concerns in early studies,10 the most recent evidence, including populations evaluated when SARS-CoV-2 testing became massively available, indicates a positive association between long-term exposure to air pollution and COVID-19 hospitalizations and severity. Biases identified in early studies may still be present in more recent studies, but they are likely less important. In the study in Catalonia, we also observed positive associations between air pollution and immunoglobulin G (IgG) and IgA levels to specific viral antigens induced by SARS-CoV-2 infection.9
Air pollution has been associated with multiple health outcomes, including lung cancer, cardiovascular and respiratory diseases, metabolic diseases and diabetes, and mental health, as well as increased risk of several respiratory viral and bacterial infections, including influenza and respiratory syncytial virus.11 Air pollutants have been shown to impair immune responses, induce oxidative stress, and stimulate proinflammatory cytokine release, thereby favoring multiple diseases.12–14 The negative impact of chronic inflammation on vaccines efficacy has been seen mainly in the elderly and in chronic inflammatory conditions.15 In relation to COVID-19, air pollutants may alter several immune pathways also mediated by epigenetic regulation16 that are involved in the development and severity of the disease and could also affect vaccine efficacy.
There exists evidence, not always consistent, on the association of post-vaccination antibody levels to exposure to immunotoxicants, such as polychlorinated biphenyls (PCBs), per- and polyfluoroalkyl substances (PFAS), and metals,17–19 although studies on the effects of exposure to air pollutants on post-vaccination antibody levels in children or adults are largely lacking. A study in sera of 6-y-old children in Germany20 identified lower antibody IgG titers against tetanus toxoid in children living in higher air pollution areas. There is previous evidence that air pollutants, such as polycyclic aromatic hydrocarbons and secondhand smoke, affect vaccine response in humans and animals.21 To our knowledge, a single, fairly small study in China evaluated COVID-19 vaccine antibody response in relation to air pollution and identified lower neutralizing antibody titers of an inactivated SARS-CoV-2 vaccine among participants with higher air pollution levels; they also found higher levels of markers of chronic inflammation in the same participants.22
A number of factors have been associated with the COVID-19 vaccine response, predominantly previous infection and type of vaccine, but also age, sex, chronic disease, and smoking.23–25 Several other factors have been associated with other vaccines in children and adults, including diet, predominantly in relation to malnutrition, mental health, and some lifestyle factors.26 SARS-CoV-2 elicits robust humoral immune responses, including production of virus-specific IgM, IgA, and IgG. IgM and IgA isotypes dominate the early antibody response to SARS-CoV-2, and IgA contributes to virus neutralization at mucosal sites.27–29 In serum, the three isotypes display neutralizing activity, with IgM and IgG1 (the predominant subclass of IgG) being the most important contributors.30
In this study, we examined the COVID-19 vaccine antibody response in a general population cohort in Catalonia in relation to prepandemic air pollution levels. By late spring 2021, the majority of the Catalan population had received a first vaccine dose and some had received a second. We measured IgM, IgA, and IgG antibodies against viral antigens elicited by vaccines administered in Spain. Our primary outcome was based on levels of IgG to spike proteins in noninfected vaccinated persons.
Materials and Methods
Study Design and Participants
The COVID-19 cohort in Catalonia (the COVICAT study) evaluates the health impact of the COVID-19 pandemic on the population in Catalonia, Spain, and builds on five preexisting adult cohort studies.9 We limited this analysis to the vaccine response in the largest cohort within COVICAT, the GCAT-Genomes for Life cohort.31 The GCAT cohort started participant recruitment in 2015 and includes middle-aged (40–65 years of age at baseline) residents of Catalonia. Most enrolled participants were blood donors invited through a public agency, the Blood and Tissue Bank (BST). The last prepandemic follow-up was done in 2018–2019 (n=9,308). In early summer 2020 post-lockdown, participants completed an online COVICAT questionnaire or responded to a computer-assisted telephone questionnaire; a random sample of participants also provided blood samples up to mid-November 2020.9,32 Residential address at the time of the prepandemic questionnaire was geocoded.
We recontacted eligible participants a year later in spring 2021 after COVID-19 vaccine administration began in Spain (Figure S1). Participants were asked to respond to a questionnaire (online or via telephone) and provide a blood sample. Blood sampling in 2021 was offered to 2,404 participants, including all participants with a seropositive or undetermined serostatus in 2020 (response rate, N=507, 44.0%) and to a random sample of seronegative participants in 2020 (response rate, N=575, 46.2%). People were aware of their 2020 serology results. A total of 1,090 participants provided blood samples and completed the questionnaire. We excluded from this analysis individuals who were not vaccinated (n=120), those vaccinated with Janssen COVID-19 vaccine due to the small number of people receiving this (one dose) vaccine (n=16), one participant <40 years of age, and 26 participants with incomplete information (questionnaire 2021, serology 2020, or air pollution data). This left 927 participants for this analysis.
Ethical approval was obtained by the ethics committees at the Hospital Universitari Germans Trias i Pujol (CEI no. PI-20-182) and the Parc de Salut Mar (CEIM-PS MAR, no. 2020/9,307/I). All participants provided informed consent and had consented to be recontacted during the first follow-up.
Air Pollution Exposure
We linked participants’ prepandemic addresses to estimates of long-term exposure to the following air pollutants: particulate matter with an aerodynamic diameter of ≤2.5μm (PM2.5), black carbon (BC), nitrogen dioxide (NO2), and ozone (O3). Air pollution estimates for the period 2018–2019 were based on models developed by the Effects of Low-Level Air Pollution: A Study in Europe (ELAPSE) project (http://www.elapseproject.eu/), which have been described in detail.33 We applied Europe-wide hybrid land-use regression models incorporating air pollution monitoring data, satellite observations, dispersion model estimates, land use, and traffic variables as predictors. We based the models for PM2.5, NO2, and O3 (warm season) on 2010 measurements in the AirBase database that is maintained by the European Environmental Agency. We used European Study of Cohorts for Air Pollution Effects (ESCAPE) monitoring data to develop models of BC.34 The model was evaluated using 5-fold hold-out validation in random subsets of the monitoring data stratified by type of measurement and region of Europe. Models explained 66% of measured spatial variation for PM2.5 in annual average concentrations in hold-out validation; the corresponding fraction for BC was 52%; for NO2, 58%; and for O3, 63%. Participants were assigned the annual average 2010 concentration based on predicted surfaces (100×100m) from the ELAPSE model. We then applied a temporal correction to estimate exposures for the years 2018 and 2019 following protocols for temporal extrapolation developed in the ESCAPE project. Although 2010 models have been validated, we did not have validation against measurement data from 2018–2019. We used daily time-series data from the official routine monitoring network and calculated the ratios between the 2018–2019 period and 2010 for NO2, nitrogen oxides (NOx), PM2.5, and O3. Because BC is not measured at routine monitoring stations, we used NOx to temporally adjust for BC values given that it is a primary combustion pollutant from traffic emissions with similar pollutant behavior. We used the average of 2018–2019 levels for each pollutant as our main exposure metrics.
Samples and Serology
Blood samples collected at both follow-up periods (2020 and 2021) were processed within 24 h of collection and frozen, and anti-SARS-CoV-2 antibody levels in plasma were analyzed in one batch at the ISGlobal Immunology laboratory in Barcelona. The levels [median fluorescence intensity (MFI)] of IgG, IgM, and IgA were assessed by high-throughput multiplex quantitative suspension array technology against a panel of five SARS-CoV-2 antigens: the spike full length protein (S) and the receptor-binding domain (RBD) (both fused with C-terminal 6xHis and StrepTag purification sequences and purified from supernatant of lentiviral-transduced CHO-S cells cultured under a fed-batch system35), the subregion S2 (purchased from SinoBiological), the nucleocapsid (N) full length (FL), and the specific C-terminal (Ct) region (both expressed in E. coli and His tag-purified).36 Assay performance was previously established as 100% specificity and 95.78% sensitivity for seropositivity 14 d after symptoms onset (15). Antigen-coupled microspheres were added to a 384-well μClear flat bottom plate (Greiner Bio-One) in multiplex (2,000 microspheres per analyte per well) in a volume of 90μL of Luminex buffer [1% bovine serum albumin (BSA), 0.05% Tween 20, 0.05% sodium azide in phosphate-buffered saline (PBS)] using the 384 channels Integra Viaflo semi-automatic device (96/384, 384 channel pipette). Hyperimmune pools were used as positive controls prepared at 2-fold, eight serial dilutions from 1:12.5. Prepandemic samples were used as negative controls to estimate the cutoff of seropositivity. Ten microliters of each dilution of the positive control, negative controls, and test samples (prediluted 1:50 in 96 round-bottom well plates), were added to a 384-well plate using an Assist Plus Integra device with a 12-channel Voyager pipette (final test sample dilution of 1:500 for all isotypes, and a second dilution at 1:5,000 for IgG to assess the response to S proteins in vaccinated participants, avoiding signal saturation). To quantify IgM and IgA, test samples and controls were pretreated with antihuman IgG (Gullsorb) at 1:10 dilution, to avoid IgG interferences. Technical blanks consisting of Luminex buffer and microspheres without samples were added in 4 wells to control for nonspecific signals. Plates were incubated for 1 h at room temperature in agitation (Titramax 1000) at 900 rpm and protected from light. Then, the plates were washed three times with 200μL/well of PBS-T (0.05% Tween 20 in PBS), using BioTek 405 TS (384-well format). Twenty-five microliters of goat antihuman IgG-phycoerythrin (PE) (GTIG-001; Moss Bio) diluted 1:400, goat antihuman IgA-PE (GTIA-001; Moss Bio) 1:200, or goat antihuman IgM-PE (GTIM-001; Moss Bio) 1:200 in Luminex buffer were added to each well and incubated for 30 min as before. Plates were washed and microspheres resuspended with 80μL of Luminex buffer, covered with an adhesive film, and sonicated 20 s on a sonicator bath platform before acquisition on the Flexmap three-dimensional reader. At least 50 microspheres per analyte and per well were acquired, and MFIs were reported for each isotype–antigen combination. Assay positivity cutoffs specific for each isotype–antigen combination were calculated as 10 to the mean plus 3 standard deviations (SDs) of log10-transformed MFI of 128 prepandemic controls. Results were defined as undetermined when the MFI levels for a given isotype–antigen combination were between the positivity threshold and an upper limit at 10 to the mean plus 4.5 SD of the log10-transformed MFIs of prepandemic samples, and no other isotype–antigen combination was above the positivity cutoff. We defined overall serostatus by isotype and by antigen. Results for IgM are informative for the evaluation of short-term effects in antibody response and although measured in the whole study population are shown only for those participants having one dose at the time of the 2021 study visit, sampled within 1 month post vaccination.
Vaccination and SARS-CoV-2 Infection
We retrieved data on the number of doses, date of administration, and trade names of vaccines for each participant from the electronic health records of the Epidemiological Surveillance Emergency Service of Catalonia of the Department of Health. Participants had received the following vaccines: Comirnaty (BNT162b2, mRNA; BioNTech-Pfizer), Spikevax (mRNA-1273; Moderna), and Vaxzevria (ChAdOx1 nCoV-19; AstraZeneca).
We detected a heterologous prime-boost approach in 11 people (Vaxzevria as a first shot followed by Comirnaty). Thus, participants’ vaccine type is categorized according to the type of their first dose.
We used a two-part strategy to identify participants infected with SARS-CoV-2 prior to the 2021 study visit a) positive viral detection test (polymerase chain reaction or antigen test) prior to sample collection in 2021, self-reported in study questionnaires, or identified through record linkage with the SARS-CoV-2 test registry from the Epidemiological Surveillance Emergency Service of Catalonia of the Department of Health,37 and b) seropositivity based on our antibody data using the following criteria: seropositivity in the prevaccination 2020 serology sample, or seropositivity to N-antigen in 2021 sample, given that the available vaccines do not contain or produce N-antigen.
Covariates
Information on basic characteristics (age, sex, and educational level) was available from earlier contacts and verified in the COVICAT questionnaire (available in Spanish at http://www.gcatbiobank.org/media/upload/arxius/COVICAT/encuesta%20COVICAT.pdf). In this analysis, we used self-reported information on several variables, including educational level, smoking, alcohol consumption, medical history diet, symptoms related to COVID-19, height, and weight. Medical history included prior diagnosis of any chronic disease, asking for a list of several major diseases that required a visit to the doctor or medical treatment in the last 6 months, such as cardiovascular (hypertension, heart attack, angina pectoris), respiratory (asthma, chronic obstructive pulmonary disease), diabetes, kidney, immune-related (autoimmune diseases, HIV, or other immunodeficiency), digestive, or gynecological diseases, as well as cancer, mental health diseases (anxiety, depression, or other diseases), and addictions, along with an open question on any other disease. Diet was assessed with the 14-point Mediterranean Diet Adherence Screener (MEDAS) used in the Prevención con Dieta Mediterránea (PREDIMED) study that has been validated against a classic food frequency questionnaire.38 Self-reported symptoms related to COVID-19 included fever, cough, dyspnea, fatigue, headache, muscle/joint pain, loss of odor/taste, nausea, vomiting, and rash. We collected changes in residential address from the prepandemic period through the 2021 follow-up. We linked prepandemic residential addresses to census tract-level deprivation index based on the 2011 census39; the index uses six indicators, specifically, unemployment, manual work, temporary employment, insufficient education at >16 years of age, young age, and dwellings without access to internet. We also linked residential address with population density and degree of urbanization of the census tract of residence using information from the 2011 census.40
Statistical Analysis
In all analyses, we used antibody levels that were log10 transformed due to their skewed distribution. We applied linear regression models to assess the association between air pollution levels and the log10-transformed antibody levels, and results were expressed as percentage change in the geometric mean and 95% confidence intervals (CIs). Air pollutants were modeled continuously and estimates per interquartile range (IQR) of each pollutant were reported. Antibody responses to vaccination measured through IgM levels included only participants sampled within 1 month post first dose vaccination, whereas analyses on IgG/IgA levels included all participants irrespective of sampling time post vaccination. We considered the following variables as potential confounders: age (continuous), sex (male, female), highest attained educational level (primary or less, secondary, university), and socioeconomic status according to area of residency (in quantiles). Time since last vaccination (<31, 31–60, 61–90, 91–120, >120 d), type of vaccine and number of doses are strongly related to vaccine antibody response, and we also included those variables in all models. We finally adjusted for factors that have been associated with response to COVID-19 or other vaccines, including smoking, diet, prior chronic diseases, and mental health.23–26 We expected an attenuated effect among those with previous SARS-CoV-2 infection and therefore defined a priori stratified analysis by infection status. Evaluation of effect modification by infection status through likelihood ratio tests indicated pInteraction<0.1 for all air pollutants and the S and S2 spike antigens. In the infected strata, we adjusted models for severity of the infection based on self-reported symptoms and from hospital records (0 symptoms, 1–3 symptoms, ≥4 symptoms, hospitalized). Participants with missing covariates were excluded from the complete-case analysis models. We used generalized additive models to explore the relationship between days since vaccination and IgG antibody levels among participants without prior infection according to air pollution levels (with low defined as below the median vs. high defined as above the median of the distribution for each pollutant). We performed all statistical analyses using Stata/SE (version 16; StataCorp LLC.).
Results
Study Population
The flow chart in Figure S1 shows the participants contacted, those recruited, and those excluded. Of the 1,090 participants recruited who gave blood samples, 927 participants were included in this analysis (Figure S1). The mean age was 57 y (range: 44–72 y), and 58% were female (Table 1). We evaluated differences in air pollution exposure, sociodemographic, and clinical variables between the 1,090 persons who gave blood samples in 2021 (participants) and the 1,314 who were contacted but did not attend the study visit in 2021 (nonparticipants). The average NO2 levels for participants and nonparticipants was similar (36.8 μg/m3 vs. 36.6 μg/m3, respectively) and for PM2.5 were nearly identical (16.6 μg/m3) (Table S1). Participants were slightly older (55.8 years of age) than nonparticipants (55.3 years of age). There were minor, nonstatistically significant differences for socioeconomic status/education, sex, and prior infection based on serological tests in 2020 (Table S1).
Table 1 Description of the study population, COVICAT study, Catalonia, Spain (n=927).
Characteristic Mean±SD or n (%)
Age (y) 57.5±6.9
Sex
Male 389 (42.0)
Female 538 (58.0)
Quintiles of deprivation indexa
1 (least deprived) 185 (20.0)
2 181 (19.5)
3 192 (20.7)
4 190 (20.5)
5 (most deprived) 179 (19.3)
Educational level
University 453 (48.9)
Secondary 382 (41.2)
Primary or less 92 (9.9)
Type of vaccine
Comirnaty–Pfizer/BioNTech 422 (45.5)
Spikevax–Moderna 113 (12.2)
Vaxzevria–AstraZeneca 392 (42.3)
Doses (n)
1 319 (34.4)
2 608 (65.6)
Evidence of previous infection at time of serology
No 632 (68.2)
Yes 295 (31.8)
Seropositivity 2021
No 19 (2.0)
Yes 908 (98)
Note: COVICAT, COVID-19 cohort in Catalonia; SD, standard deviation.
a 1 (−2.27, −1.44), 2 (−1.44, −1,09), 3 (−1.09, −0.81), 4 (−0.81, −0.28), 5 (−0.28, 1.81).
Air Pollution Exposure
Pollution levels at residence had the following correlations: 0.8 between NO2 and PM2.5, ∼0.8 for NO2 and BC and 0.7 for PM2.5 and BC, and ∼−0.8 for NO2 and PM2.5 with O3 (Table 2). Mean exposure during 2018–2019 in the study population was 35.1 μg/m3 for NO2 and 16.4 μg/m3 for PM2.5 (Table 3).
Table 2 Spearman correlations of air pollution concentrations (2018–2019 average) at residence (n=927), COVICAT study, Catalonia, Spain.
NO2 PM2.5 BC O3
NO2 1
PM2.5 0.799 1
BC 0.789 0.739 1
O3 −0.808 −0.780 −0.686 1
Note: BC, black carbon; COVICAT, COVID-19 cohort in Catalonia; NO2, nitrogen dioxide; O3, ozone; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter.
Table 3 Distribution of air pollution concentrations (2018–2019 average) by type of vaccine, COVICAT cohort study, Catalonia, Spain.
Pollutant (μg/m³) Mean±SD Geometric mean (95% CI) 50th percentile 25th–75th percentile
All vaccines (n=927)
NO2 35.1±8.9 33.7 (33.0, 34.4) 36.7 30.0–40.7
PM2.5 16.4±1.4 16.3 (16.2, 16.4) 16.6 15.7–17.3
BC 1.8±0.4 1.8 (1.8, 1.8) 1.9 1.7–2.1
O3 64.2±6.6 63.9 (63.5, 64.3) 62.5 60.4–66.1
Pfizer (n=422)
NO2 34.9±9.0 33.5 (32.5, 34.5) 36.5 29.6–40.5
PM2.5 16.3±1.5 16.2 (16.1, 16.4) 16.6 15.5–17.3
BC 1.8±0.4 1.8 (1.8, 1.8) 1.9 1.7–2.1
O3 64.5±6.9 64.2 (63.5, 64.8) 62.6 60.7–66.1
Moderna (n=113)
NO2 34.3±8.9 32.6 (30.5, 34.9) 35.5 29.5–40.5
PM2.5 16.2±1.3 16.2 (15.9, 16.4) 16.1 15.5–17.2
BC 1.8±0.4 1.7 (1.6, 1.8) 1.8 1.6–2.1
O3 64.9±6.7 64.6 (63.4, 65.8) 63.3 60.4–69.4
AstraZeneca (n=392)
NO2 35.6±8.7 34.3 (33.3, 35.3) 37.5 31.7–40.9
PM2.5 16.5±1.3 16.4 (16.3, 16.6) 16.7 15.8–17.4
BC 1.9±0.4 1.8 (1.8, 1.8) 1.9 1.7–2.1
O3 63.8±6.3 63.5 (62.9, 64.1) 62.4 60.1–65.2
Note: BC, black carbon; CI, confidence interval; COVICAT, COVID-19 cohort in Catalonia; NO2, nitrogen dioxide; O3, ozone; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; SD, standard deviation.
Vaccination and Association of Air Pollutants with Type of Vaccine
Among vaccinated participants, 319 (34.4%) had one dose and 605 (65.6%) had completed two doses (Table 1). The first dose was either Comirnaty–Pfizer/Bionetch (45.5%) or Vaxzevria–AstraZeneca- (42.3%), and a smaller fraction of people were vaccinated with Spikevax–Moderna (12.2%). Vaxzevria was administered as a first dose followed by Comirnaty in 11 people. Median time since the last vaccination was 28 d (IQR: 14–54 d, minimum: 1 d, maximum: 160 d). Median time between the first and second dose was 21 d for Comirnaty (IQR: 21–22 d, minimum: 14 d, maximum: 42 d), 28 d for Spikevax (IQR: 28–29 d, minimum: 27 d, maximum: 36 d), and 83 d for Vaxzevria (IQR: 78–95 d, minimum: 63 d, maximum: 121 d).
There was no clear association of air pollutant levels with the type of vaccine administered (Table 3). Recipients of the AstraZeneca vaccine had only slightly higher average exposure levels of NO2 (35.6μg/m³) and PM2.5 (16.5μg/m³) compared with those receiving the Pfizer (34.9μg/m³ and 16.3μg/m³, respectively) and Moderna vaccines (34.3μg/m³ and 16.2μg/m³, respectively).
Association of Air Pollutants with Vaccine Antibody Response
Among participants without prior infection, antibody responses to vaccination measured through IgM levels (participants within 1 month post first dose vaccination) and IgG levels (any time post vaccination, all participants) were negatively associated with long-term air pollution; no associations were observed for IgA (Table 4). The decrease in IgM levels was between 5% and 14% per IQR increase in NO2 and PM2.5, and it was statistically significant for most antigens. The decrease in IgM was slightly smaller for BC, whereas no associations were observed for O3. For IgG, IQR increases in NO2, PM2.5, and BC were associated with statistically significant decreases of the S and S2 antigens. An IQR increase in PM2.5 was associated with an ∼8%–10% decrease in IgG levels; for RBD, percentage change =−8.1 (95% CI: −15.9, 0.4); for S, percentage change =−9.9 (95% CI: −16.2, −3.1); and for S2, percentage change =−8.4 (95% CI: −13.5, 3.0). Similar associations were observed for BC, whereas for NO2 the decrease in IgG antibody response was slightly smaller (Table 4). An inverse association for IgG was observed for exposure to O3. Estimates for IgG and air pollutants adjusted only for age and sex are shown in Table S2. Associations adjusted for the same variables as in the Table 4 models but also including lifestyle factors (smoking and alcohol consumption), the diet MEDAS score, prior mental health diseases, and prior medical history of chronic diseases gave similar estimates (Table S3) as those shown in Table 4 for more limited adjustments. The direction of the air pollution–antibody response associations was similar in women and men, but they were stronger in men (Table S4). There were no statistically significant interactions (pInteraction<0.1) observed by sex for NO2, PM2.5, or BC, although there were for O3 exposure. Finally, we evaluated whether the association of air pollution with vaccine response depended on type of vaccine (Table S5). Results were similar for the three main vaccines administered.
Table 4 Association of air pollution with antibody levels induced after vaccination for IgM, IgA, and IgG among COVICAT cohort participants without prior infection (n=632).
Pollutants/Spike Antigensa IgM
n=357
(one dose only)b IgA
n=632 IgG
n=632
% change (95% CI)c % change (95% CI)c % change (95% CI)c
NO2
RBD −10.8 (−19.3, −1.4) 3.2 (−6.0, 13.3) −1.8 (−10.2, 7.4)
S −12.2 (−22.9, 0.0) 1.6 (−9.9, 14.5) −7.3 (−13.9, −0.2)
S2 −5.3 (−14.2, 4.4) 3.0 (−7.8, 15.1) −6.7 (−11.9, −1.1)
PM2.5
RBD −10.8 (−19.6, −1.0) 2.4 (−6.7, 12.3) −8.1 (−15.9, 0.4)
S −14.3 (−25.1, −1.9) 1.5 (−9.9, 14.4) −9.9 (−16.2, −3.1)
S2 −6.0 (−15.1, 4.0) 3.0 (−7.8, 15.0) −8.4 (−13.5, −3.0)
BC
RBD −5.6 (−13.8, 3.3) 1.0 (−7.2, 10.0) −7.6 (−14.8, 0.3)
S −7.8 (−18.1, 3.7) 0.3 (−10.1, 11.9) −10.0 (−15.8, −3.7)
S2 −4.1 (−12.2, 4.7) 5.9 (−4.3, 17.2) −8.0 (−12.7, −3.1)
O3
RBD 6.3 (−1.1, 14.3) −2.8 (−9.3, 4.1) 4.6 (−2.1, 11.7)
S 7.6 (−2.0, 18.2) 0.9 (−7.6, 10.2) 6.9 (1.2, 12.8)
S2 2.1 (−4.8, 9.5) 0.1 (−7.7, 8.6) 4.6 (0.3, 9.1)
Note: Percentage change (95% CI) per IQR of air pollutants from linear regression for the log10 MFI. %, percentage; BC, black carbon; CI, confidence interval; COVICAT, COVID-19 cohort in Catalonia; Ig, immunoglobulin; IQR, interquartile range; MFI, median fluorescence intensity; NO2, nitrogen dioxide; O3, ozone; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RBD, receptor-binding domain; S, spike-protein; S2, segment spike-protein.
a Pollutants: NO2 (IQR: 10.7), PM2.5 (IQR: 1.7), BC (IQR: 0.4), and O3 (IQR: 5.7); viral-target antigens: RBD, S, and S2.
b Restricted to 357 persons sampled within 1 month post first dose vaccination.
c Adjusted for age (continuous), sex, educational level (university, secondary, primary or less), quintiles of deprivation index, type of vaccine (Comirnaty, Spikevax, Vaxzevria), number of doses and time since last vaccine (<31, 31–60, 61–90, 91–120, >120 days). Adjustment for IgM does not include time since last vaccine.
Among participants without prior infection, differences in IgG antibody response between high and low air pollution exposures (defined as above or below the median exposure for NO2 and PM2.5) were small but persisted over time since vaccination after both the first (Figure 1) and the second dose (Figure 2). After the first dose, the peak in IgG RBD and S responses occurred later, and lower levels were observed in people exposed to high levels of PM2.5 and NO2 (Figure 1). After the second dose, we identified a faster decrease of IgG levels with time in noninfected participants with high air pollution exposure (Figure 2).
Figure 1. Vaccine responses in time since first vaccination by exposure to high or low air pollution levels among participants in the COVICAT study without prior infection and one vaccine dose (n=319). Generalized additive models exploring the relationship between days since vaccination and antibody IgG levels induced after vaccination, by high (red, plus) or low (blue, circle) NO2 and PM2.5 air pollution levels. High and low air pollutant levels were defined as above or below the median (NO2: 36.7; PM2.5: 16.6). IgG levels were determined for three viral-target antigens: RBD, S, and S2. Data used in the figure are available in Excel Tables S1 and S2. Note: COVICAT, COVID-19 cohort in Catalonia; Ig, immunoglobulin; MFI, median fluorescence intensity; NO2, nitrogen dioxide; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RBD, receptor-binding domain; S, spike-protein; S2, segment spike-protein.
Figure 1 is a set of six scatter plots titled Participants with 1 dose and no prior infection. On the top, the first three plots are titled Nitrogen dioxide, plotting Log10 median fluorescence intensity values, ranging from 2.5 to 5 in increments of 0.5; 2 to 6 in unit increments, and 3.5 to 5.5 in increments of 0.5 (y-axis) across days since vaccination, ranging from 0 to 100 in increments of 20 (x-axis) for Immunoglobulin G receptor-binding domain, Immunoglobulin G spike-protein, and Immunoglobulin G spike-protein 2. At the bottom, the three plots titled particulate matter begin subscript 2.5 end subscript, plotting Log10 median fluorescence intensity values, ranging from 2.5 to 5 in increments of 0.5; 2 to 6 in unit increments, and 3.5 to 5.5 in increments of 0.5 (y-axis) across days since vaccination, ranging from 0 to 100 in increments of 20 (x-axis) for Immunoglobulin G receptor-binding domain, Immunoglobulin G spike-protein, and Immunoglobulin G spike-protein 2.
Figure 2. Vaccine responses in time since second vaccination by exposure to high or low air pollution levels among participants in the COVICAT study without prior infection and two vaccine doses (n=608). Generalized additive models exploring the relationship between days since vaccination and antibody IgG levels induced after vaccination, by high (red, plus) or low (blue, circle) NO2 and PM2.5 air pollution levels. High and low air pollutant levels were defined as above or below the median (NO2: 36.7; PM2.5: 16.6). IgG levels were determined for three viral-target antigens: RBD, S, and S2. Data used in the figure are available in Excel Tables S3 and S4. Note: COVICAT, COVID-19 cohort in Catalonia; Ig, immunoglobulin; MFI, median fluorescence intensity; NO2, nitrogen dioxide; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RBD, receptor-binding domain; S, spike-protein; S2, segment spike-protein.
Figure 2 is a set of six scatter plots titled Participants with two doses and no prior infection. On the top, the first three plots are titled Nitrogen dioxide, plotting Log10 median fluorescence intensity values, ranging from 2.5 to 5 in increments of 0.5; 3.5 to 5.5 in increments of 0.5, and 3.5 to 5.5 in increments of 0.5 (y-axis) across days since vaccination, ranging from 0 to 150 in increments of 50 (x-axis) for Immunoglobulin G receptor-binding domain, Immunoglobulin G spike-protein, and Immunoglobulin G spike-protein 2. At the bottom, the three plots titled particulate matter begin subscript 2.5 end subscript, plotting Log10 median fluorescence intensity values, ranging from 2.5 to 5 in increments of 0.5; 3.5 to 5.5 in increments of 0.5, and 3.5 to 5.5 in increments of 0.5 (y-axis) across days since vaccination, ranging from 0 to 150 in increments of 50 (x-axis) for Immunoglobulin G receptor-binding domain, Immunoglobulin G spike-protein, and Immunoglobulin G spike-protein 2.
Among infected participants, there were no associations observed between air pollution and antibody response to vaccines for any of the air pollutants examined (Table 5). The largest decrease in IgG levels per IQR increase in PM2.5 was observed for RBD (−8.7%) but results were not statistically significant.
Table 5 Association of air pollution with antibody levels induced after vaccination for IgM, IgA, and IgG among COVICAT cohort participants with prior infection (n=295).
Pollutants/spike antigensa IgM (n=140) (one dose only)b IgA (n=295) IgG (n=295)
% change (95% CI)c % change (95% CI)c % change (95% CI)c
NO2
RBD 2.3 (−11.8, 18.7) −2.5 (−21.4, 20.9) −1.0 (−17.9, 19.4)
S −0.1 (−17.7, 21.4) 2.6 (−18.9, 29.9) 0.8 (−13.7, 17.7)
S2 −5.0 (−18.2, 10.3) 14.9 (−5.3, 39.5) 2.4 (−8.6, 14.7)
PM2.5
RBD 6.1 (−8.2, 22.7) −14.8 (−31.0, 5.2) −8.7 (−24.1, 9.7)
S −0.4 (−17.6, 20.5) −10.9 (−29.3, 12.4) −5.2 (−18.6, 10.5)
S2 −5.1 (−18.0, 9.8) 6.4 (−12.2, 28.8) −1.8 (−12.1, 9.9)
BC
RBD −0.2 (−12.4, 13.8) −10.4 (−25.7, 8.0) −1.0 (−15.9, 16.5)
S −3.9 (−19.0, 14.0) −5.8 (−23.3, 15.7) 1.0 (−11.8, 15.6)
S2 −8.0 (−19.3, 4.8) 7.5 (−9.3, 27.3) 0.8 (−8.7, 11.3)
O3
RBD −3.9 (−14.4, 7.9) 8.0 (−7.6, 26.3) 1.8 (−11.2, 16.6)
S −2.1 (−15.9, 13.9) 3.2 (−13.0, 22.5) 1.0 (−9.7, 13.1)
S2 2.2 (−9.0, 14.9) −3.0 (−15.7, 11.8) −0.5 (−8.3, 8.1)
Note: Percentage change (95% CI) per IQR of air pollutants from linear regression for the log10 MFI. %, percentage; BC, black carbon; CI, confidence interval; COVICAT, COVID-19 cohort in Catalonia; Ig, immunoglobulin; IQR, interquartile range; MFI, median fluorescence intensity; NO2, nitrogen dioxide; O3, ozone; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RBD, receptor-binding domain; S, spike-protein; S2, segment spike-protein.
a Pollutants: NO2 (IQR: 10.7), PM2.5 (IQR: 1.7), BC (IQR: 0.4), and O3 (IQR: 5.7); viral-target antigens: RBD, S, and S2.
b Restricted to 140 persons sampled within 1 month post first dose vaccination.
c Adjusted for age (continuous), sex, educational level (university, secondary, primary or less), quintiles of deprivation index, type of vaccine (Comirnaty, Spikevax, Vaxzevria), number of doses and time since last vaccine (<31, 31–60, 61–90, 91–120, >120 d), disease severity (0 symptoms, 1–3 symptoms, ≥4 symptoms, hospitalized). Adjustment for IgM does not include time since last vaccine.
Discussion
We examined whether long-term exposure to air pollution is associated with antibody responses to COVID-19 vaccines in a prospective cohort study. Our study resulted in several key findings. First, we identified that exposure to PM2.5, NO2, and BC was associated with a 5%–10% decrease in vaccine antibody responses among individuals without prior infection after adjusting for confounders. The decrease was shown both for early responses measured through IgM and late responses measured through IgG. Second, lower antibody response among participants with air pollution exposure above the median persisted over several months following vaccination. In this study, we did not address whether the observed decrease in antibody responses was associated with risk of breakthrough infections and their severity.
Air pollutants have been shown to impair immune response, including effects on severe COVID-19, although there is very limited evidence evaluating the association of long-term exposure to air pollution on vaccines response.22 Wider effects of air pollution indicate alterations across multiple classes of immune cells affecting various diseases, including respiratory infections, exacerbations of asthma and chronic obstructive pulmonary disease, and cardiovascular disease.12–14 Chronic inflammation, such as induced by air pollution, has been associated with a negative effect on vaccine efficacy.15 Finally, our findings on air pollution exposure are consistent with evidence regarding pollution more broadly, including evidence that persistent organic pollutants, such as PCBs, reduce vaccine response in children.17–19
We identified an association of exposure to air pollution with vaccine antibody response among participants without prior infection. Moreover, people exposed to air pollution had a later peak in antibody responses after the first dose and they had persistently lower levels of antibodies in time. The combination in the same multiplex assay of antigens only present in the virus (N) and in both the virus and the vaccines (S) allowed us to make a distinction between participants who were infected and those who were not along with data from viral detection tests and prior serology in the beginning of the pandemic. We had shown in an earlier analysis based on the COVICAT data9 that higher levels of air pollution were associated with a higher risk of severe COVID-19 and a higher antibody response to the infection and that previous infections are associated with higher vaccine antibody responses.23 Thus, the effect of pollutants on vaccine responses could differ and be masked among those infected. We therefore adjusted the models on the association of air pollution with vaccine antibody response among those infected for disease severity. As the pandemic and vaccination campaigns have evolved, a higher proportion of the population has immunity developed through a combination of infection and vaccine, and further research should investigate the role of long-term exposure to air pollution on this hybrid immunity.
The identification of small differences in the response of the three S antigens are difficult to evaluate given that they are highly correlated. Decreases in antibody levels were consistent for NO2, PM2.5, and BC. Positive associations for O3 are most likely due to the strong inverse correlation between O3 and NO2 levels.
For COVID-19 vaccines, similarly to influenza and other vaccines, research on immunological responses has largely focused on IgG responses while other isotypes are generally neglected. The inclusion of the three isotypes is a strength of this study. Virus-specific IgM are produced early following infection/vaccination, followed by IgA and IgG virus-specific antibody production.29 The identification of differences in early responses measured through IgM reinforce our findings concerning IgG. We identified a negative effect of air pollution particularly on IgG and to a much lesser extent for IgA levels. We had hypothesized that air pollution would have a negative impact also for IgA. Unfortunately, we could not assess direct effects of air pollution on IgA production and distribution at mucosal sites, particularly in the respiratory tract. Systemic IgA and mucosal IgA may not necessarily correlate given that they are under different regulatory mechanisms. Recent studies, however, have identified IgA levels following intramuscular COVID-19 vaccination also in saliva,41,42 as well as a correlation between serum and saliva IgA levels.42 These data highlight the need for assessing antigen-specific mucosal IgA levels in future studies.
Key strengths of our study are the use of prospectively collected individual-level data through individual contact and electronic registers and the use of repeated serological testing for a wide range of antigens. Sources of bias that are concerns in previous studies linking air pollution and COVID-19 disease10 are less relevant for the outcomes we studied. Detection bias that may affect studies on SARS-CoV-2, was not an issue in this study, which was based on complete serological testing of the study population. The availability of extensive individual information, both prepandemic and at two different post-pandemic time points, allows for extensive control of potential confounding by lifestyle factors (e.g., tobacco smoking) and contextual variables (e.g., socioeconomic status) and also for factors that have been shown in vaccine trials for other diseases to be associated with vaccine response (e.g., nutrition). In our study, the degree of confounding by these variables was minimal. The two main limitations of our study are the low response rates among participants providing blood samples for a second time and the lack of long-term clinical follow-up data to associate alterations in vaccine antibody response with clinical effects. Our main outcome is based on levels of IgG to S proteins, a biomarker, which eliminates the possibility of self-selection based on the outcome. Still, immune response could be a correlate of other factors associated with exposure to air pollution, such as socioeconomic status, and bias from nonresponse would occur if individuals with more symptoms were more likely to participate and if participation was related to prepandemic air pollution levels. We showed that participants were very similar to those not participating in terms of exposure, vaccination factors, and prior clinical symptoms, which probably eliminates or substantially diminishes the probability of this bias. In addition, we showed that the type of vaccine administered was also not associated with air pollution exposure. Finally, the population studied was mostly mid-age. Given well-known differences in vaccine response by age,26 the generalization of these findings particularly to older ages may be limited.
Conclusions and Potential Implications
Our study identifies an effect of air pollution on COVID-19 vaccine immune response. Participants exposed to higher levels of fine particles (PM2.5), NO2, and BC had ∼10% lower antibody responses to S antigens that are elicited by the vaccines. This finding strengthens the evidence on the multiple immune pathways through which air pollution affects multiple diseases, including infections and chronic diseases. Whether this decrease in antibody response has observable effects on future risk of COVID-19 risk should be evaluated with longer-term prospective data. Similarly, our findings open the possibility of air pollution affecting immunization for other diseases. Overall our findings add to the knowledge on the adverse effects of air pollution that are identified even in the relatively low levels observed in western Europe and urge for stricter control of exposure as recommended by the World Health Organization.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We are grateful to all the GCAT volunteers who participated in the study and to all the Blood and Tissue Bank workers for sample recruitment. We acknowledge E. Prados, L. Mayer, and J. Chi for their assistance with the antibody analyses.
We acknowledge support from the Incentius a l’Avaluació de Centres CERCA (in_CERCA); EIT HEALTH BP2020-20873-Certify.Health; Fundació Privada Daniel Bravo Andreu; Spanish Ministry of Science & Innovation (PID2019-110810RB-I00 grant); the Spanish State Research Agency and Ministry of Science and Innovation through the “Centro de Excelencia Severo Ochoa 2019–2023” Program (CEX2018-000806-S), the Instituto de Salud Carlos III (PI17/01555) and the Generalitat de Catalunya through the CERCA Program. This study makes use of data generated by the GCAT-Genomes for Life. Cohort study of the Genomes of Catalonia, Fundació IGTP. IGTP is part of the CERCA Program/Generalitat de Catalunya. GCAT is supported by Acción de Dinamización del ISCIII-MINECO and the Department of Health of the Generalitat of Catalunya (ADE 10/00026); the Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR) (2017-SGR 529). G.M. is supported by RYC2020-029886-I/AEI/10.13039/501100011033, co-funded by European Social Fund. B.C. is supported by ISCIII national grant PI18/01512. R.R. is supported by the Health Department, Catalan Government (PERIS SLT017/20/000224). The full list of the investigators who contributed to the generation of the GCAT data is available from http://www.genomesforlife.com. Data are available from the authors.
==== Refs
References
1. Bowe B, Xie Y, Gibson AK, Cai M, van Donkelaar A, Martin RV, et al. 2021. Ambient fine particulate matter air pollution and the risk of hospitalization among COVID-19 positive individuals: cohort study. Environ Int 154 :106564, PMID: , 10.1016/j.envint.2021.106564.33964723
2. Chen Z, Sidell MA, Huang BZ, Chow T, Eckel SP, Martinez MP, et al. 2022. Ambient air pollutant exposures and COVID-19 severity and mortality in a cohort of COVID-19 patients in southern California. Am J Respir Crit Care Med 206 (4 ):440–448, PMID: , 10.1164/rccm.202108-1909OC.35537137
3. Yu Z, Bellander T, Bergström A, Dillner J, Eneroth K, Engardt M, et al. 2022. Association of short-term air pollution exposure with SARS-CoV-2 infection among young adults in Sweden. JAMA Netw Open 5 (4 ):e228109, PMID: , 10.1001/jamanetworkopen.2022.8109.35442452
4. Nobile F, Michelozzi P, Ancona C, Cappai G, Cesaroni G, Davoli M, et al. 2022. Air pollution, SARS-CoV-2 incidence and COVID-19 mortality in Rome—a longitudinal study. Eur Respir J 60 (3 ):2200589, PMID: , 10.1183/13993003.00589-2022.35896215
5. English PB, Von Behren J, Balmes JR, Boscardin J, Carpenter C, Goldberg DE, et al. 2022. Association between long-term exposure to particulate air pollution with SARS-CoV-2 infections and COVID-19 deaths in California, U.S.A. Environ Adv 9 :100270, PMID: , 10.1016/j.envadv.2022.100270.35912397
6. Kim H, Samet JM, Bell ML. 2022. Association between short-term exposure to air pollution and COVID-19 mortality: a population-based case-crossover study using individual-level mortality registry confirmed by medical examiners. Environ Health Perspect 130 (11 ):117006, PMID: , 10.1289/EHP10836.36367781
7. Chen C, Wang J, Kwong J, Kim J, van Donkelaar A, Martin RV, et al. 2022. Association between long-term exposure to ambient air pollution and COVID-19 severity: a prospective cohort study. CMAJ 194 (20 ):E693–E700, PMID: , 10.1503/cmaj.220068.35609912
8. Sheridan C, Klompmaker J, Cummins S, James P, Fecht D, Roscoe C. 2022. Associations of air pollution with COVID-19 positivity, hospitalisations, and mortality: observational evidence from UK Biobank. Environ Pollut 308 :119686, PMID: , 10.1016/j.envpol.2022.119686.35779662
9. Kogevinas M, Castaño-Vinyals G, Karachaliou M, Espinosa A, de Cid R, Garcia-Aymerich J, et al. 2021. Ambient air pollution in relation to SARS-CoV-2 infection, antibody response, and COVID-19 disease: a cohort study in Catalonia, Spain (COVICAT study). Environ Health Perspect 129 (11 ):117003, PMID: , 10.1289/EHP9726.34787480
10. Villeneuve PJ, Goldberg MS. 2020. Methodological considerations for epidemiological studies of air pollution and the SARS and COVID-19 coronavirus outbreaks. Environ Health Perspect 128 (9 ):95001, PMID: , 10.1289/EHP7411.32902328
11. Domingo JL, Rovira J. 2020. Effects of air pollutants on the transmission and severity of respiratory viral infections. Environ Res 187 :109650, PMID: , 10.1016/j.envres.2020.109650.32416357
12. Gangwar RS, Bevan GH, Palanivel R, Das L, Rajagopalan S. 2020. Oxidative stress pathways of air pollution mediated toxicity: recent insights. Redox Biol 34 :101545, PMID: , 10.1016/j.redox.2020.101545.32505541
13. Glencross DA, Ho TR, Camiña N, Hawrylowicz CM, Pfeffer PE. 2020. Air pollution and its effects on the immune system. Free Radic Biol Med 151 :56–68, PMID: , 10.1016/j.freeradbiomed.2020.01.179.32007522
14. Suzuki T, Hidaka T, Kumagai Y, Yamamoto M. 2020. Environmental pollutants and the immune response. Nat Immunol 21 (12 ):1486–1495, PMID: , 10.1038/s41590-020-0802-6.33046888
15. Cianci R, Franza L, Massaro MG, Borriello R, De Vito F, Gambassi G. 2020. The interplay between immunosenescence and microbiota in the efficacy of vaccines. Vaccines (Basel) 8 (4 ):636, PMID: , 10.3390/vaccines8040636.33147686
16. Bulka CM, Enggasser AE, Fry RC. 2022. Epigenetics at the intersection of COVID-19 risk and environmental chemical exposures. Curr Environ Health Rep 9 (3 ):477–489, PMID: , 10.1007/s40572-022-00353-9.35648356
17. Grandjean P, Andersen EW, Budtz-Jørgensen E, Nielsen F, Mølbak K, Weihe P, et al. 2012. Serum vaccine antibody concentrations in children exposed to perfluorinated compounds. JAMA 307 (4 ):391–397, PMID: , 10.1001/jama.2011.2034.22274686
18. Zhang X, Xue L, Deji Z, Wang X, Liu P, Lu J, et al. 2022. Effects of exposure to per- and polyfluoroalkyl substances on vaccine antibodies: a systematic review and meta-analysis based on epidemiological studies. Environ Pollut 306 :119442, PMID: , 10.1016/j.envpol.2022.119442.35568291
19. Zeng Z, Ngai S, Wang Q, Liang W, Huo X. 2021. Early-life exposure to widespread environmental toxicants and children’s health risks: a focus on the post-vaccination antibody potency or immunoglobulin levels. Sci Total Environ 781 :146714, 10.1016/j.scitotenv.2021.146714.
20. Idel H, Stiller-Winkler R, Malin EM, Krämer U. 1994. Antibodies to tetanus toxoid in children from areas with different heavy levels of air pollution [in German]. Zentralbl Hyg Umweltmed 195 (5–6 ):457–462, PMID: .7916873
21. Franza L, Cianci R. 2021. Pollution, inflammation, and vaccines: a complex crosstalk. Int J Environ Res Public Health 18 (12 ):6330, PMID: , 10.3390/ijerph18126330.34208042
22. Zhang S, Chen S, Xiao G, Zhao M, Li J, Dong W, et al. 2022. The associations between air pollutant exposure and neutralizing antibody titers of an inactivated SARS-CoV-2 vaccine. Environ Sci Pollut Res Int 29 (9 ):13720–13728, PMID: , 10.1007/s11356-021-16786-y.34599446
23. Moncunill G, Aguilar R, Ribes M, Ortega N, Rubio R, Salmerón G, et al. 2022. Determinants of early antibody responses to COVID-19 mRNA vaccines in a cohort of exposed and naïve healthcare workers. EBioMedicine 75 :103805, PMID: , 10.1016/j.ebiom.2021.103805.35032961
24. Ward H, Whitaker M, Flower B, Tang SN, Atchison C, Darzi A, et al. 2022. Population antibody responses following COVID-19 vaccination in 212,102 individuals. Nat Commun 13 (1 ):907, PMID: , 10.1038/s41467-022-28527-x.35173150
25. Karachaliou M, Moncunill G, Espinosa A, Castaño-Vinyals G, Rubio R, Vidal M, et al. 2022. SARS-CoV-2 infection, vaccination, and antibody response trajectories in adults: a cohort study in Catalonia. BMC Med 20 (1 ):347, PMID: , 10.1186/s12916-022-02547-2.36109713
26. Zimmermann P, Curtis N. 2019. Factors that influence the immune response to vaccination. Clin Microbiol Rev 32 (2 ):e00084-18, PMID: , 10.1128/CMR.00084-18.30867162
27. Sterlin D, Mathian A, Miyara M, Mohr A, Anna F, Claër L, et al. 2021. IgA dominates the early neutralizing antibody response to SARS-CoV-2. Sci Transl Med 13 (577 ):eabd2223, PMID: , 10.1126/scitranslmed.abd2223.33288662
28. Zhang J, Zhang H, Sun L. 2022. Therapeutic antibodies for COVID-19: is a new age of IgM, IgA and bispecific antibodies coming? MAbs 14 (1 ):2031483, PMID: , 10.1080/19420862.2022.2031483.35220888
29. Hope JL, Bradley LM. 2021. Lessons in antiviral immunity. Science 371 (6528 ):464–465, PMID: , 10.1126/science.abf6446.33510014
30. Klingler J, Weiss S, Itri V, Liu X, Oguntuyo KY, Stevens C, et al. 2021. Role of immunoglobulin M and A antibodies in the neutralization of severe acute respiratory syndrome coronavirus 2. J Infect Dis 223 (6 ):957–970, PMID: , 10.1093/infdis/jiaa784.33367897
31. Obón-Santacana M, Vilardell M, Carreras A, Duran X, Velasco J, Galván-Femenía I, et al. 2018. GCAT|Genomes for Life: a prospective cohort study of the genomes of Catalonia. BMJ Open 8 (3 ):e018324, PMID: , 10.1136/bmjopen-2017-018324.29593016
32. Karachaliou M, Moncunill G, Espinosa A, Castaño-Vinyals G, Jiménez A, Vidal M, et al. 2021. Infection induced SARS-CoV-2 seroprevalence and heterogeneity of antibody responses in a general population cohort study in Catalonia Spain. Sci Rep 11 (1 ):21571, PMID: , 10.1038/s41598-021-00807-4.34732749
33. de Hoogh K, Chen J, Gulliver J, Hoffmann B, Hertel O, Ketzel M, et al. 2018. Spatial PM2.5, NO2, O3 and BC models for western Europe—evaluation of spatiotemporal stability. Environ Int 120 :81–92, PMID: , 10.1016/j.envint.2018.07.036.30075373
34. Eeftens M, Beelen R, de Hoogh K, Bellander T, Cesaroni G, Cirach M, et al. 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PMcoarse in 20 European study areas; results of the ESCAPE project. Environ Sci Technol 46 (20 ):11195–11205, PMID: , 10.1021/es301948k.22963366
35. Ortega N, Ribes M, Vidal M, Rubio R, Aguilar R, Williams S, et al. 2021. Seven-month kinetics of SARS-CoV-2 antibodies and role of pre-existing antibodies to human coronaviruses. Nat Commun 12 (1 ):4740, PMID: , 10.1038/s41467-021-24979-9.34362897
36. Dobaño C, Santano R, Jiménez A, Vidal M, Chi J, Rodrigo Melero N, et al. 2021. Immunogenicity and crossreactivity of antibodies to the nucleocapsid protein of SARS-CoV-2: utility and limitations in seroprevalence and immunity studies. Transl Res 232 :60–74, PMID: , 10.1016/j.trsl.2021.02.006.33582244
37. Catalan Ministry of Health. 2019. Catalan Ministry of Health Catalan epidemiological surveillance system. https://salutpublica.gencat.cat/ca/ambits/vigilancia_salut_publica/XVEC/ [accessed 4 February 2023].
38. Schröder H, Fitó M, Estruch R, Martínez-González MA, Corella D, Salas-Salvadó J, et al. 2011. A short screener is valid for assessing Mediterranean diet adherence among older Spanish men and women. J Nutr 141 (6 ):1140–1145, PMID: , 10.3945/jn.110.135566.21508208
39. Duque I, Domínguez-Berjón MF, Cebrecos A, Prieto-Salceda MD, Esnaola S, Calvo Sánchez M, et al. 2021. Índice de privación en españa por sección censal en 2011. Gac Sanit 35 (2 ):113–122, PMID: , 10.1016/j.gaceta.2019.10.008.32014314
40. INE (Instituto Nacional de Estadística). 2011. Population and Housing Census. https://www.ine.es/en/censos2011_datos/cen11_datos_inicio_en.htm [accessed 4 February 2023].
41. Darwich A, Pozzi C, Fornasa G, Lizier M, Azzolini E, Spadoni I, et al. 2022. BNT162b2 vaccine induces antibody release in saliva: a possible role for mucosal viral protection? EMBO Mol Med 14 (5 ):e15326, PMID: , 10.15252/emmm.202115326.35393790
42. Sheikh-Mohamed S, Isho B, Chao GYC, Zuo M, Cohen C, Lustig Y, et al. 2022. Systemic and mucosal IgA responses are variably induced in response to SARS-CoV-2 mRNA vaccination and are associated with protection against subsequent infection. Mucosal Immunol 15 (5 ):799–808, PMID: , 10.1038/s41385-022-00511-0.35468942
|
PMC010xxxxxx/PMC10075304.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37018010
EHP9838
10.1289/EHP9838
Research
Association of Ambient Temperature with Mortality in Resident and Multiethnic Transient Populations in a Desert Climate, 2006–2014
https://orcid.org/0000-0002-3380-7092
Yezli Saber 1 2 3
Khan Altaf H. 4
https://orcid.org/0000-0002-8762-5403
Yassin Yara M. 1
https://orcid.org/0000-0001-5075-7392
Khan Anas A. 1 5
Alotaibi Badriah M. 1
https://orcid.org/0000-0003-3089-5936
Bouchama Abderrezak 3
1 Global Centre for Mass Gatherings Medicine, Ministry of Health, Riyadh, Saudi Arabia
2 Biostatistics, Epidemiology and Scientific Computing Department, King Faisal Specialist Hospital and Research Centre, Riyadh, Saudi Arabia
3 Experimental Medicine Department, King Abdullah International Medical Research Center/King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia
4 Department of Biostatistics and Bioinformatics, King Abdullah International Medical Research Center/King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia
5 Department of Emergency Medicine, College of Medicine, King Saud University, Riyadh, Saudi Arabia
Address correspondence to Saber Yezli, Global Centre for Mass Gatherings Medicine, Public Health Directorate, Ministry of Health, Riyadh, Saudi Arabia. Telephone: +96611401555, ext. 2073. Email: [email protected]. And, Abderrezak Bouchama, King Abdullah International Medical Research Center (KAIMRC), King Saud Bin Abdulaziz University for Health Sciences (KSAU-HS), Ministry of National Guard – Health Affairs (MNGHA), Riyadh, Saudi Arabia. Telephone: +966 1 429 9999, ext. 94544. Email: [email protected]
5 4 2023
4 2023
131 4 04700417 6 2021
23 2 2023
06 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Although the association between ambient temperature and mortality in local populations is evident, this relationship remains unclear in transient populations (e.g., due to immigration, mass gatherings, or displacement). The holy city of Mecca annually shelters two populations comprising its residents and the transitory Hajj pilgrims (>2 million people from >180 countries). Both live side by side in a hot desert climate, rendering the development of evidence-based heat-protective measures challenging.
Objectives:
We aimed to characterize the ambient temperature–mortality relationship and burden for the Mecca resident and Hajj transient populations, which have distinct levels of adaptation to ambient temperature.
Methods:
We analyzed daily air temperature and mortality data for Mecca residents and pilgrims over nine Hajj seasons between 2006 and 2014, using a fitted standard time-series Poisson model. We characterized the temperature–mortality relationship with a distributed lag nonlinear model with 10 d of lag. We determined the minimum mortality temperature (MMT) and attributable deaths for heat and cold for the two populations.
Results:
The median average daily temperature during the Hajj seasons was 30°C (19°C–37°C). There were 8,543 and 10,457 nonaccidental deaths reported during the study period among Mecca residents and pilgrims, respectively. The MMT was 2.5°C lower for pilgrims in comparison with the MMT for Mecca residents (23.5°C vs. 26.0°C). The temperature–mortality relationship shape varied from inverted J to U shape for the Mecca and pilgrim populations, respectively. Neither hot nor cold temperatures had a statistically significant association with mortality in Mecca residents. In contrast, for pilgrims, elevated temperatures were associated with significantly high attributable mortality of 70.8% [95% confidence interval (CI): 62.8, 76.0]. The effect of heat on pilgrims was immediate and sustained.
Discussion:
Our findings indicate that pilgrims and Mecca residents exposed to the same hot environmental conditions exhibited distinct health outcomes. This conclusion suggests that a precision public health approach may be warranted to protect against high environmental temperature during mass gatherings of diverse populations. https://doi.org/10.1289/EHP9838
Supplemental Material is available online (https://doi.org/10.1289/EHP9838).
No conflicts of interest to declare.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Ambient temperature has been etiologically linked with a wide range of human morbidities and mortalities,1,2 particularly when it reaches both extremes.3,4 Extreme temperatures can lead to life-threatening heatstroke and profound hypothermia during heat waves and cold spells, respectively, mainly among vulnerable populations, such as frail people, older people, and children.5,6 Similarly, extreme temperatures can exacerbate preexisting chronic conditions, amplifying the risks of mortality and morbidity, especially among patients with cardiovascular and respiratory diseases.2,4,7
Epidemiological studies have also consistently reported an association between daily environmental temperature and mortality besides episodes of extreme changes.1,8 This relationship has been demonstrated in most populations studied under different climates.9–11 An optimal threshold or minimum mortality temperature (MMT) was determined in many cities and countries worldwide, with the lowest and highest being average daily temperatures (ADTs) of 12°C and 32°C, respectively.9 Mortality rates increase at temperatures outside the local MMT, for all-cause mortality and for specific causes, such as cardiovascular and respiratory diseases.1,2 More recently, the MMT was found to correlate with the most frequent ambient temperature in a given location,9 supporting the hypothesis that the MMT is a good indicator of human adaptability to the local climate.
Mortality risk attributable to ambient temperatures has been reported for many of the world’s major cities.1 However, to our knowledge, this was not investigated for the city of Mecca in the Kingdom of Saudi Arabia (KSA), characterized by extremely hot summers (often reaching >48°C), and relatively warm winters.12 Mecca is the holiest city for the ∼1.9 billion Muslims worldwide and is the site of the Hajj religious pilgrimage.13 Each year, Hajj is attended by >2 million pilgrims with different ethnicities and adaptation to heat, originating from up to 180 countries with varied climates.14,15 The pilgrim population lives side by side with the locally adapted residents of Mecca for up to several weeks.14 Moreover, because the lunar calendar dictates the Hajj, it moves by 11 d every year; hence, the two populations are confronted with different seasons, including summer.15 This unique setup poses a challenging public health dilemma defining optimum temperature and the threshold for heat and cold waves, thereby establishing precision mitigation measures that protect these diverse populations.15
Using 10 y of Mecca daily ambient temperature and mortality data, our study aimed to determine the temperature-mortality relationship for both Mecca residents and Hajj pilgrims. Our goal was to identify the temperature threshold of minimum mortality and quantify the total mortality burden attributed to nonoptimal ambient temperature for both populations.
Methods
Study Site
Mecca is located in a narrow desert valley surrounded by mountains in the Hejaz region of the KSA. The city is 70km from the Red Sea, 277m above sea level, and characterized by a hot desert climate, with an ADT that ranges from 16°C to 42°C and relative humidity of 30% to 40%.14 Mecca experiences warm to hot temperatures all year round, and ADTs are rarely below 30°C, even during winter.12 Mecca is inhabited by approximately 2 million people but experiences a significant biyearly influx of millions of local and international pilgrims during the Hajj and Ramadan seasons.14 The latter are dictated by the lunar calendar; hence, the influx of pilgrims may occur at different seasons across the year.
Study Design and Data Sources
This is an observational study of historical records of daily mortality and daily average ambient temperature in Mecca, KSA, collected between 1 February 2006 and 1 November 2014. We obtained the mortality data from the Saudi Ministry of Health, which included daily counts of death and the Hajj status (resident/pilgrim). The data also had the cause of death classified by the International Classification of Diseases 10th Revision (ICD-10). We defined nonaccidental mortality as mortality from all causes other than injury, poisoning, and certain other consequences of external causes (ICD-10: S and T, except T67.x and T68.x) and external causes of morbidity (ICD-10: V, W, X, and Y).
We extracted daily ambient temperature data, including minimum, mean, and maximum temperatures, from the National Oceanic and Atmospheric Administration (NOAA) database from the Mecca weather station (GHCND: SAM00041030), at 39.833 longitude and 21.483 latitude.
Statistical Methods
We analyzed all data using the R software (version 4.0.2; R Development Core Team) and the distributed lag nonlinear model (DLNM) package.16 We summarized and reported descriptive statistics of mortality data using percentages and frequencies and used the mean, median, minimum, and maximum to describe the ADT. We also used graphical exploration of the ADT over time to explore seasonality and potential long-term trends. Similarly, we examined the time series of mortality over the study period through graphical exploration to investigate the crude distribution of seasonality and short and long-term trends.
Data sets and covariates.
The Hajj season represents the 11th and 12th months of the Islamic lunar calendar and indicates when the first pilgrims start arriving in Mecca for Hajj until they leave the city. To construct a consistent comparison of the relationship between temperature and mortality among Mecca residents and pilgrims, the analysis was conducted over the same time period for both groups. As such, using the original Mecca mortality data, we generated two separate data subsets, one for Mecca residents and another for pilgrims. This separation was achieved by restricting the original data set to nonaccidental mortality among Mecca residents or pilgrims during only the Hajj season of each year. Given that the Hajj season shifts by ≈11 d earlier each Gregorian calendar year (Table S1), only limited months of the Georgian calendar from 2006 to 2014 were represented in the data subsets. Both Mecca residents and pilgrims subsets included the following relevant variables: ADT, daily nonaccidental death count, day of the week (DOW), day of year, year, and Hajj-Ritual-Day. DOW was used to control for the effect of the weekday on daily mortality. Hajj-Ritual-Day covariate is on a binary scale and indicates the Hajj rituals days. During these 5 d, a significant variation in the dynamics between the Mecca and pilgrim populations occur because all pilgrims will be located in the same area of Mecca city conducting the Hajj rituals, away from most Mecca residents.
Temperature–mortality relationship.
We used a previously described statistical modeling approach to define the complex relationship between temperature and mortality.1 We applied a time-series regression model to investigate the relationship between ADT and daily mortality counts for Mecca residents and pilgrims separately. For each population, we modeled the relationship using Quasi-Poisson regression coupled with a DLNM to estimate population-specific temperature–mortality association as relative risk (RR).1,17 Both Mecca residents’ and pilgrims’ time splines were not continuous but were composed of multiple equally spaced and ordered series of multiple Hajj seasons in different years. Therefore, we used seasonal chunks analysis according to a recently reported methodology18 to model the relationship between ADT and daily mortality counts for both populations. For the seasonal chunks analysis, we also included natural splines for day of the year (DOY) and time to describe the potential seasonal effect within each Hajj season, and the long-time trend, respectively.
The regression equation for the model for Mecca residents and pilgrims is given below: (1) Yt∼quasi-Poisson(μt)Log (μt)=α+β1×bs(Temperature)+β2×DOW+β3×ns(Time, 9)+β4×Hajj-Ritual-Day+β5×ns(DOY, 2),
where t is the day of the observation, Yt is the daily all-cause nonaccidental death count on day t, α is the intercept, and βs are coefficients of slopes for all the covariates in the models. Time is the number of days in the beginning of the study period. To adjust for the long-term trend and seasonality, we parameterized the Time variable using a natural cubic smoothing spline (ns) (whose second derivatives at two end points are zero) and 3 knots and 2 boundary knots.
We parameterized the temperature variables for both data subsets in the DLNM framework (R package: https://cran.r-project.org/web/packages/dlnm/) using a quadratic B-Spline function (fu=“bs”) with 4 degrees of freedom (df). To generate crossbasis matrix for both data subsets, we applied “equalknots” and “logknots” functions as given in DLNM R package. We used a basis function of degree 2 (degree of the piecewise polynomial) with 4 df in “equalknots” functions for the variable Temperature, whereas lag time is based on natural splines with three equally spaced knots were specified for the lag time. As most pilgrims leave Mecca after a 2 wk stay in the city, we studied the lag effect of temperature on Mecca residents and pilgrims over a 10-d lag period.
We tested the goodness of fit for the models via graphical plots and compared the models’ parameter estimates and standard errors for both the quasi-Poisson and Negative-Binomial models as well testing different choice of degrees of freedom and autocorrelation in the residuals. We used “foraway” and “boot” an R package to explore the models’ fit based on deviance residuals, normal Q-Q lot, Cook statistics vs. standardized leverage, and influential point to identify the observations with high points. In DLNM framework we used 9 df for the time spline in the final model for both Mecca residents and pilgrims.
Estimates of minimum mortality temperatures and attributable risk measures.
The MMT was derived from the best linear unbiased prediction of the overall cumulative exposure–response association for each population. The MMT is the optimum temperature at which mortality risk is lowest, and we deemed it the reference for calculating the attributable risk by recentering the quadratic B-spline that models the exposure–response association. The MMT values for Mecca residents and pilgrims were determined as the temperatures at which mortality is minimized in the estimated temperature–mortality association curves.1,19
We used the fitted model to estimate fraction of death attributable to nonoptimal temperature during the study period as described previously.1,17 For each day of the series, for each population, we used the overall cumulative RR corresponding to each day’s temperature to compute the attributable deaths and fraction of attributable deaths as described by Gasparrini and Leone.17 For each population, we computed the attributable fraction (AF), their bootstrapped CIs, and attributable number (AN) as a risk measure in forward and backward lagged distribution perspectives. Further, we calculated the components attributable to cold and heat by adding the subsets corresponding to days with temperatures lower or higher than the MMT.17 We defined mortality attributable to extreme low and extreme high temperatures as the cumulative mortality below or above the 2.5th and 97.5th temperature percentiles, respectively.1
Ethics Approval and Consent to Participate
The study was approved by the King Fahad Medical City Ethics Committee and the institutional review board (IRB; log#:17-141E) and conducted in accordance with the Ethics Committee’s guidelines. Consent was waived because this is an observational study that retrospectively analyzed aggregate deidentified mortality data.
Results
Ambient Temperature and Nonaccidental Death
Ambient temperature.
During the study period, the Hajj seasons fell during the cool months of the year (September–January) with an overall median ADT of 30°C (min 19°C–max 37°C). Each year, Hajj took place during increasingly warmer weather, with the median ADT for each season increasing from 24°C in 2006 to 34°C in 2014 (Figure 1).
Figure 1. Distribution of Mecca city average daily temperature per Hajj season during the study period. Box plots show the median, mean (x), minimum and maximum values, and interquartile range of temperature in Mecca for each Hajj season during the study period (2006–2014). Data can be found in Excel Table S1.
Figure 1 is a box and whiskers plot, plotting temperature degree Celsius, ranging from 15 to 40 in increments of 5 (y-axis) across Hajj year, ranging from 2006 to 2014 (x-axis).
Nonaccidental death.
We analyzed 19,000 nonaccidental deaths that occurred in Mecca during the Hajj seasons between 2006 and 2014. Of these, 8,543 (44.9%) were among Mecca residents, and 10,457 (55.1%) were among pilgrims. Deceased individuals were older for pilgrims compared to Mecca residents [mean (±SD) age of 63.6 y (±13.3) and 54.6 y (26.3) respectively].
Ambient temperature–daily mortality association.
The exposure–response association between daily mortality, ADT, and time is displayed in Figure 2. The panels on the right show that the MMT was 26.0°C at the 20th percentile for Mecca residents and 23.5°C at the 6th percentile for pilgrims. Likewise, they display the extreme low and high temperatures defined by the 2.5th and 97.5th percentiles of temperature distribution. Extreme cold and hot temperature were 22°C and 36°C, respectively.
Figure 2. Relationship between ambient temperature and nonaccidental mortality among Mecca residents and Hajj pilgrims for the Hajj seasons: 2006–2014. Left panels: 3D visualization of the relationship between average daily temperature, mortality risk and time (days), for Mecca residents (upper panel) and pilgrims (lower panel). Right panels: overall cumulative mortality exposure–response relationship with 95% CI (shaded gray) and temperatures distribution for Mecca residents (upper panel) and pilgrims (lower panel), using Quasi-Poisson regression coupled with a DLNM with 10 d of lag. Solid gray vertical lines are MMTs, and dashed gray vertical lines are the 2.5th and 97.5th percentiles. Temperature frequency represents the number of cumulative days at that temperature over the entire study period. Data can be found in Excel Table S2. Note: CI, confidence interval; DLNM, distributed lag nonlinear model; MMT, minimum mortality temperature; RR, relative risk.
Figure 2 is a set of two two-dimensional function area graphs and two histograms plus line graphs. The two-dimensional function area graphs, plotting relative risk, ranging from 1.0 to 3.0 in increments of 0.5 and 1.0 to 5.0 in increments of 1.0 (y-axis) across temperature (degree Celsius), ranging from 35 to 20 in decrements of 5 and Lag (days), ranging from 10 to 0 in decrements of 2 (x-axis). The two histograms plus line graphs, plotting relative risks, ranging from 0.5 to 7.5 in increments of 0.5 and 0.5 to 9.5 in increments of 3.0, 9.5 to 13.0 in increments of 3.5, 13.0 to 25.0 in increments of 4.0 (y-axis) across temperature (degree Celsius), ranging from 20 to 35 in increments of 5 (x-axis). The temperature frequency ranges between 0 and 50.
For Mecca residents, the association between the risk of death and temperature increased slowly and linearly for hot temperatures. However, it amplified rapidly at cold temperatures, giving an inverted J-shape to the exposure–response relationship. For pilgrims, the risk increased sharply and nonlinearly as a function of temperatures above and below the MMT, resulting in a U-shaped exposure–response relationship.
Left panels of Figure 2 indicate that the modeled relative mortality risk for Mecca residents and pilgrims varied against both the daily mean temperature range and the lag days. The relationship between ATDs, time, and risk of death was complex and nonlinear for both Mecca residents and pilgrims. For both populations, the main effect of cold was delayed and long-lasting (Figure 3). On the other hand, for heat, the increase in the risk of death was immediate and vanished within 10 d.
Figure 3. Lag–response curves for different key temperatures for Mecca residents and Hajj pilgrims for the Hajj seasons: 2006–2014. Lag–response associations between key temperatures and RR of all-cause nonaccidental mortality for Mecca residents (top) and pilgrims (bottom), using Quasi-Poisson regression coupled with a DLNM with 10 d of lag. The reference values were the MMT for each population (26.0°C for Mecca residents, 23.5°C for pilgrims) over the Hajj seasons. For each population, the key temperatures are the maximum and minimum recorded temperatures and extreme hot and cold cut off points (the 2.5th and 97.5th percentiles, respectively). Data can be found in Excel Table S3. Note: DLNM, distributed lag nonlinear model; MMT, minimum mortality temperature; RR, relative risk.
Figure 3 is a set of two line graphs, plotting relative risk, ranging from 0.9 to 1.4 in increments of 0.2 and 0.5 to 3 in increments of 0.5 (y-axis) across Lay (days), ranging from 0 to 10 in increments of 2 (x-axis) for temperatures, including 19 degrees Celsius, 22 degrees Celsius, 36 degrees Celsius, and 37 degrees Celsius.
For Mecca residents (Figure 3, top panel), the effect of cold (19°C and 22°C) appeared after 2 d of the onset, was most noticeable at days 8–9, and persisted through day 10. On the other hand, the effect of elevated temperatures (36°C and 37°C) was more immediate, appearing from day of onset, and was most noticeable on days 2–3. The impact continued for a further few days and disappeared beyond day 6, where a potential harvesting effect was observed. The estimated lag-specific increased mortality rate associated with low and elevated temperatures reached up to 23.9% and 10.8%, respectively depending on the temperature and lag time. The effect intensified and lasted longer at extreme temperatures (19°C and 37°C).
For pilgrims (Figure 3, bottom panel), the effect of low temperature (19°C and 22°C) was observed on the 2nd day from the onset, was most noticeable at days 3–4, and did not resolve within 10 d. The effect of elevated temperature (36°C and 37°C) was immediate, most noticeable 1 d from onset, and rapidly declined by day 2. The effect however persisted for another 7 d, only vanishing by day 9 from onset. The estimated lag-specific increased rate of mortality associated with low and elevated temperatures reached up to 48.8% and 132.2% respectively, depending on the temperature and lag. In general, the effect intensified as temperatures became more extreme.
Temperature-Attributable Fraction of Death
Table 1 displays the estimated overall and cold and hot attributable fractions of death for Mecca residents and pilgrims. Heat accounted for most of the attributable fraction of mortality for both populations, whereas cold did not seem to account for any significant burden. Moreover, mortality attributable to heat (16.0%; 95% CI: −5.5, 30.4) for Mecca residents was not significant.
Table 1 Estimated total mortality fraction (percentage) attributable to temperature reported as overall, hot, and cold components with 95% empirical CIs for Mecca residents and pilgrims.
Number of nonaccidental deaths (2006–2014) MMTa (percentile) Extreme cold–hot temperaturesb Attributable mortality
Overall Cold Hot
Hajj pilgrims 10,457 23.5°C (6th) 22°C–37°C 73.0 (62.8, 78.3) 2.2 (−4.2, 4.1) 70.8 (62.8, 76.0)
Mecca residents 8,543 26.0°C (20th) 22°C–37°C 16.5 (−5.9, 31.4) 0.5 (−0.4, 0.6) 16.0 (−5.5, 30.4)
Note: CI, confidence interval; MMT, minimum mortality temperature: .
a The MMT was derived from the best linear unbiased prediction of the overall cumulative exposure–response association for each population as the temperature at which mortality is minimized in the estimated temperature-mortality association curves.
b Respectively, the 2.5th and 97.5th percentiles of the temperature range in the data subset for each population.
Sensitivity analysis.
The overall RRs and CIs did not vary significantly at different temperature and degrees of freedom for Mecca residents (Table S2). However, for pilgrims, the model was sensitive to the choice of degree of freedom, particularly for heat (Table S3). In the residual autocorrelation analysis, the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots of the residuals showed that for the Mecca residents model, there was no significant autocorrelation (Figure S1). However, for pilgrims there was a significant autocorrelation at lags 1–5 (Figure S2). We tested the sensitivity of the results to the residual autocorrelation in the model for pilgrims by adding terms for the lag 1 residuals in a subsequent model as an additional covariate. The results (Table S4) show that the point estimates and their CIs were not significantly different from those from the original model. The diagnostic plots for both the quasi-Poisson and Negative-Binomial models for Mecca residents (Figure S3) and pilgrims (Figure S4) were similar. In general, for both models, residual vs. fitted showed some mild conical shape, and in the Q-Q plots some points deviated from the reference line at the top of the plots. Scale-location and residual vs. leverage plots showed some outlying points. Estimated model’s coefficient parameters, standard errors as well as their ratios for quasi-Poisson and Negative Binomial models for Mecca residents (Table S5) and pilgrims (Table S6) show that coefficients for both models are very similar, whereas the standard errors seem to be lower in Negative-Binomial in comparison with quasi-Poisson model.
Discussion
Our study investigated the association between environmental temperature and all-cause nonaccidental mortality of the regular inhabitants of the holy city of Mecca and the transient pilgrims’ population over nine Hajj seasons. The setup of our study is unique because it offered the possibility to analyze how ambient temperature affects the health of two distinct populations in terms of adaptation to the local climate, under the same environmental conditions, and on such a large scale. We made the following observations: a) The MMT, an indicator of human adaptability to local climate, was 2.5°C lower for pilgrims in comparison with that of Mecca residents; b) the temperature–mortality relationships differed between the two populations, taking inverted J and U shapes for Mecca residents and pilgrims, respectively, thus indicating vulnerability at cold temperatures for Mecca residents and at both extremes of ambient temperatures for pilgrims; and c) the estimated fraction of death attributable to elevated temperatures was statistically significant for pilgrims (reaching 70.8%) but not for Mecca residents. Overall, our findings suggest that pilgrims were the population more vulnerable to high temperatures. This vulnerability may be attributable to the lack of acclimatization and adaptations to the local climate and other factors inherent in the pilgrims’ population and the Hajj and its rituals.14
The MMT has been shown to vary between populations in different regions and cities, being higher in warmer locations than in colder ones.1,9 Most studies report that approximately the 75th percentile of temperature distribution is the relative optimum temperature for human beings, although it can significantly fluctuate between locations (39th–99th).1,19 In this study, we found that the MMT for Mecca residents was 26.0°C. This MMT is lower than the 30°C observed in regions with similar arid and semiarid climates1 and the recent theoretical estimations of MMTs for the Arabian Peninsula region.9 However, it is worth noting that our analysis was limited to the Hajj seasons (2 months of each year) only. Nevertheless, the Mecca residents’ MMT is very close to the most frequent ambient temperature in the city over the study period (27°C), which is in contrast with the pilgrims’ MMT, which was much lower, at 23.5°C. Given that the MMT for pilgrims was close to the minimum recorded temperature, uncertainty around the MMT estimation could be significant.20 In general, for a given location, the MMT is very close to the local most frequent temperature9 and represents the optimum temperature for the resident population. As such, deviations outside the MMT are associated with human mortality.1 Therefore, it is not surprising that the observed significant deviation outside the MMT contributed to the high mortality associated with heat displayed by pilgrims but not by Mecca residents. Overall, our results indicate that MMT can reflect human adaptability to a local climate and concurs with a previous study9 that it can serve as a tool to identify vulnerable populations, including the extraordinarily diverse people who attend the Hajj pilgrimage.
In the current study, there was no significant association between temperature and mortality among Mecca residents, similar to what was reported for Kuwait city,21,22 also characterized by a desert climate, and for hot cities in the United States.23 In the latter, the findings were attributed to physiological acclimatization and personal adaptation strategies (e.g., air conditioning, housing structure, clothing type) among residents, as well as the population not being exposed to very low temperatures during the study period.23 These factors could also explain our results among Mecca residents. On the other hand, temperature was significantly associated with mortality among pilgrims, driven mainly by heat. The estimated fraction of heat-attributable mortality among pilgrims (70.8%) is at least 40 times higher than those reported among local populations in Australia, Brazil, Canada, China, Italy, Japan, South Korea, Spain, Sweden, Taiwan, Thailand, the UK, and the United States,1 as well as Switzerland, the Netherlands, Ireland, India, and South Africa.2 This fraction is quantified for the short period of the Hajj season and therefore should be interpreted in this context. Nevertheless, lack of acclimatization and adaptations to the local climate may possibly contribute to the distinct heat–mortality association we observed among pilgrims. However, other factors, such as sociodemographic mix, the effect of travel between different climatic zones, living conditions, and physical stress associated with the Hajj rituals, may have also been at play.14
The lag structure effect of temperature on Mecca residents was in accordance with other studies exhibiting a delayed and long-lasting association for cold8,19 and immediate impact followed by a decline in risk of mortality over many days for heat.24,25 For heat, the RR declined to below 1.0 post day 6 from onset, consistent with the short-term mortality displacement.24,26 For pilgrims, the association with cold was also delayed and long-lasting, whereas that of heat was more pronounced and immediate, reaching its peak 1 d from onset. Although the RR declined by day 2, the association with heat was sustained for a further week. Many pilgrims are older with underlying health conditions,27 representing a primary finite pool of highly susceptible individuals depleted by the initial temperature stress. A possible explanation for the prolonged effect of heat is that pilgrims are generally a “temperature-vulnerable” population that is very diverse ethnically and geographically, with different adaptations to temperature, and originating from countries that might have very different climates.14 Moreover, there have been many reported risk factors that can potentiate the pilgrims’ vulnerability to variations in ambient temperature.8,28,29 For example, a large proportion of pilgrims are older, with comorbidities including diabetes, hypertension, and cardiovascular diseases and taking multiple medications (e.g., beta-blockers and nonsteroidal anti-inflammatory drugs) that can interfere with thermoregulation.14 Other identified risk factors are inherent in the Hajj, due to its characteristics (e.g., crowding, physically demanding rituals), pilgrims’ behavior (e.g., lack of awareness of the seriousness of the risk of heat exposure, performing rituals during the peak sunshine hours, lack of sun protection, inadequate sleep, little food, and suboptimal hydration), and frequency of infection (e.g., respiratory tract infections).14,15 Pilgrims are also committed to physically demanding rituals, undertaken at specific times, dates, and locations, often in the outside environment.14
Our findings imply that mitigation measures should be primarily directed toward protection against hot weather. Such protection is particularly important because during the study period the Hajj had not entered the hot seasons; thus, the observed ambient temperature-attributable deaths among pilgrims are likely to correspond to a minimum point estimate. Historically, the incidence of life-threatening heatstroke increased from 22 to 250 cases per 100,000 pilgrims when the Hajj last entered the hot season.30 Climate data also predict that global warming will increase the frequency and intensity of extreme danger heatwaves worldwide,3 including in Mecca and during Hajj.31 Increase temperature variation driven by global warming may also increase heat-attributable mortality among Mecca residents as death from heat increases with increasing temperature variation during hot periods, including in hot cities.23 The likelihood of mortality increase is because, although there is evidence for acclimatization to higher mean temperatures among populations,10,11,23 acclimatization to increase in temperature variation is less likely.23
Our study had several limitations. First is the lack of information on humidity and dew point, which have been shown to modify the effect of air temperature on mortality.32 Second, air temperature and mortality studies assume that the number and the demographic structure of the population at risk are fairly constant over time, subject to normal short- and long-term trends. In our study, this assumption is violated for the pilgrims analysis because the underlying demographics in each season can vary. Third, the data analyzed covered only 2 months of each year; hence, the series was not continuous. Fourth, the pilgrims model was sensitive to the choice of degree of freedom, particularly for heat. Finally, in addition to death, temperature extremes can increase disease burden and lead to excess emergency room visits and hospitalizations.2 In our study, we could not assess the full health impact of nonoptimal temperature due to a lack of some information, such as emergency visits and hospitalizations. Future studies should consider collecting and analyzing this information to determine the full impact of nonoptimal temperatures.
Conclusions
To the best of our knowledge, our study is the first attempt to investigate the relationship between ambient temperature and all-cause nonaccidental mortality in the holy city of Mecca and among the transient pilgrim population. Our results indicate a nonlinear exposure response, with a significant association of heat with risk of mortality among pilgrims. Our findings will help direct more targeted public health interventions to reduce the burden of illness and death related to nonoptimal temperatures for pilgrims and Mecca residents. In particular, stricter implementation of heat-illness prevention strategies, including focused health education and improving the adherence of pilgrims to preventive measures. In addition, our study may help determine the definitions of a heat wave for both populations, which are not yet characterized. This determination may be important to guide the development of heat-warning systems that would allow a proportionate mobilization of the human and capacity resources to face heat waves, potentially minimizing the associated mortality and morbidity for the several million visitors from around the world to the holy city of Mecca. In the global context, our results may inform the need to determine specific population-based heat wave/cold spell thresholds and warning systems for mobile populations (e.g., migrants, mass gatherings attendees, and displaced populations) across climatic zones to minimize the effect of ambient temperature on health. Moreover, Hajj pilgrims come to the hot environment of Mecca from more than 180 countries, thus reflecting the world’s population exposed to a hot climate. Therefore, our work also contributes to our collective understanding of the potential impact of rising global temperatures on populations worldwide in the short and long terms.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
S.Y., A.B., B.M.A. and A.H.K. conceptualized the study and its design. S.Y., A.B., and A.H.K. were involved in the literature search. S.Y. and B.M.A. were involved in data acquisition. S.Y., Y.M.Y., and A.H.K. were involved in data curation. S.Y. and A.H.K. were involved in data analysis. S.Y., A.H.K., Y.M.Y. were involved in visualization. All authors were involved in the interpretation of the study results. S.Y., A.B., and A.H.K. were involved in writing the first draft of the manuscript. All authors were involved in reviewing and editing the original manuscript and reviewed and approved the final manuscript. S.Y., A.B., and A.H.K. verified the underlying data. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.
Anonymized data will be available from the corresponding author (S.Y.) on reasonable request and with an appropriate data-sharing agreement, subject to review and following approval of a study proposal by the Saudi Ministry of Health General Department of Research and Studies ([email protected]).
==== Refs
References
1. Gasparrini A, Guo Y, Hashizume M, Lavigne E, Zanobetti A, Schwartz J, et al. 2015. Mortality risk attributable to high and low ambient temperature: a multicountry observational study. Lancet 386 (9991 ):369–375, PMID: , 10.1016/S0140-6736(14)62114-0.26003380
2. Cheng J, Xu Z, Bambrick H, Su H, Tong S, Hu W. 2019. Impacts of exposure to ambient temperature on burden of disease: a systematic review of epidemiological evidence. Int J Biometeorol 63 (8 ):1099–1115, PMID: , 10.1007/s00484-019-01716-y.31011886
3. Gasparrini A, Guo Y, Sera F, Vicedo-Cabrera AM, Huber V, Tong S, et al. 2017. Projections of temperature-related excess mortality under climate change scenarios. Lancet Planet Health 1 (9 ):e360–e367, PMID: , 10.1016/S2542-5196(17)30156-0.29276803
4. Han J, Liu S, Zhang J, Zhou L, Fang Q, Zhang J, et al. 2017. The impact of temperature extremes on mortality: a time-series study in Jinan, China. BMJ Open 7 (4 ):e014741, PMID: , 10.1136/bmjopen-2016-014741.28465307
5. Berko J, Ingram DD, Saha S, Parker JD. 2014. Deaths Attributed to Heat, Cold, and Other Weather Events in the United States, 2006–2010. National Health Statistics Reports, No. 76, PMID: , https://stacks.cdc.gov/view/cdc/24418 [accessed 4 April 2023]. 25073563
6. Bouchama A, Knochel JP. 2002. Heat stroke. N Engl J Med 346 (25 ):1978–1988, PMID: , 10.1056/NEJMra011089.12075060
7. Moghadamnia MT, Ardalan A, Mesdaghinia A, Keshtkar A, Naddafi K, Yekaninejad MS. 2017. Ambient temperature and cardiovascular mortality: a systematic review and meta-analysis. PeerJ 5 :e3574, PMID: , 10.7717/peerj.3574.28791197
8. Chen R, Yin P, Wang L, Liu C, Niu Y, Wang W, et al. 2018. Association between ambient temperature and mortality risk and burden: time series study in 272 main Chinese cities. BMJ 363 :k4306, PMID: , 10.1136/bmj.k4306.30381293
9. Yin Q, Wang J, Ren Z, Li J, Guo Y. 2019. Mapping the increased minimum mortality temperatures in the context of global climate change. Nat Commun 10 (1 ):4640, PMID: , 10.1038/s41467-019-12663-y.31604931
10. Arbuthnott K, Hajat S, Heaviside C, Vardoulakis S. 2016. Changes in population susceptibility to heat and cold over time: assessing adaptation to climate change. Environ Health 15 (suppl 1 ):33, PMID: , 10.1186/s12940-016-0102-7.26961541
11. Keatinge WR, Donaldson GC, Cordioli E, Martinelli M, Kunst AE, Mackenbach JP, et al. 2000. Heat related mortality in warm and cold regions of Europe: observational study. BMJ 321 (7262 ):670–673, PMID: , 10.1136/bmj.321.7262.670.10987770
12. Abdou AEA. 2014. Temperature trend on Makkah, Saudi Arabia. Atmos Climate Sc 4 (3 ):457–481, 10.4236/acs.2014.43044.
13. Tagliacozzo E, Toorawa SM. 2016. The Hajj: Pilgrimage in Islam. Cambridge, England: Cambridge University Press.
14. Yezli S. 2021. Risk factors for heat-related illnesses during the Hajj mass gathering: an expert review. Rev Environ Health 38 (1 ):33, PMID: , 10.1515/reveh-2021-0097.34714988
15. Yezli S, Khan A, Bouchama A. 2019. Summer hajj pilgrimage in the era of global warming: a call for vigilance and better understanding of the risks. J Travel Med 26 (7 ):taz069, PMID: , 10.1093/jtm/taz069.31518432
16. Gasparrini A. 2011. Distributed lag linear and non-linear models in R: the package dlnm. J Stat Softw 43 (8 ):1–20, PMID: , 10.18637/jss.v043.i08.22003319
17. Gasparrini A, Leone M. 2014. Attributable risk from distributed lag models. BMC Med Res Methodol 14 :55, PMID: , 10.1186/1471-2288-14-55.24758509
18. Gasparrini A. 2021. Distributed Lag Linear and Non-Linear Models for Time Series Data. https://cran.r-project.org/web/packages/dlnm/vignettes/dlnmTS.pdf [accessed 23 February 2022].
19. Guo Y, Gasparrini A, Armstrong B, Li S, Tawatsupa B, Tobias A, et al. 2014. Global variation in the effects of ambient temperature on mortality: a systematic evaluation. Epidemiology 25 (6 ):781–789, PMID: , 10.1097/EDE.0000000000000165.25166878
20. Tobias A, Armstrong B, Gasparrini A. 2017. Brief report: investigating uncertainty in the minimum mortality temperature: methods and application to 52 Spanish cities. Epidemiology 28 (1 ):72–76, PMID: , 10.1097/EDE.0000000000000567.27748681
21. Vicedo-Cabrera AM, Scovronick N, Sera F, Royé D, Schneider R, Tobias A, et al. 2021. The burden of heat-related mortality attributable to recent human-induced climate change. Nat Clim Chang 11 (6 ):492–500, PMID: , 10.1038/s41558-021-01058-x.34221128
22. Alahmad B, Shakarchi AF, Khraishah H, Alseaidan M, Gasana J, Al-Hemoud A, et al. 2020. Extreme temperatures and mortality in Kuwait: who is vulnerable? Sci Total Environ 732 :139289, PMID: , 10.1016/j.scitotenv.2020.139289.32438154
23. Braga AL, Zanobetti A, Schwartz J. 2001. The time course of weather-related deaths. Epidemiology 12 (6 ):662–667, PMID: , 10.1097/00001648-200111000-00014.11679794
24. Hajat S, Armstrong BG, Gouveia N, Wilkinson P. 2005. Mortality displacement of heat-related deaths: a comparison of Delhi, Sao Paulo, and London. Epidemiology 16 (5 ):613–620, PMID: , 10.1097/01.ede.0000164559.41092.2a.16135936
25. Ha J, Shin Y, Kim H. 2011. Distributed lag effects in the relationship between temperature and mortality in three major cities in South Korea. Sci Total Environ 409 (18 ):3274–3280, PMID: , 10.1016/j.scitotenv.2011.05.034.21683987
26. Guo Y, Gasparrini A, Armstrong BG, Tawatsupa B, Tobias A, Lavigne E, et al. 2017. Heat wave and mortality: a multicountry, multicommunity study. Environ Health Perspect 125 (8 ):087006, PMID: , 10.1289/EHP1026.28886602
27. Yezli S, Mushi A, Almuzaini Y, Balkhi B, Yassin Y, Khan A. 2021. Prevalence of diabetes and hypertension among Hajj pilgrims: a systematic review. Int J Environ Res Public Health 18 (3 ):1155, PMID: , 10.3390/ijerph18031155.33525524
28. Dang TN, Seposo XT, Duc NHC, Thang TB, An DD, Hang LTM, et al. 2016. Characterizing the relationship between temperature and mortality in tropical and subtropical cities: a distributed lag non-linear model analysis in Hue, Viet Nam, 2009–2013. Glob Health Action 9 :28738, PMID: , 10.3402/gha.v9.28738.26781954
29. Leon LR, Bouchama A. 2015. Heat stroke. Compr Physiol 5 (2 ):611–647, PMID: , 10.1002/cphy.c140017.25880507
30. Ghaznawi HI, Ibrahim MA. 1987. Heatstroke and heat exhaustion in pilgrims performing the Haj (annual pilgrimage) in Saudi Arabia. Ann Saudi Med 7 :323–326, 10.5144/0256-4947.1987.323.
31. Kang S, Pal JS, Eltahir EAB. 2019. Future heat stress during Muslim pilgrimage (Hajj) projected to exceed “extreme danger” levels. Geophys Res Lett 46 (16 ):10094, 10.1029/2019GL083686.
32. Zeng J, Zhang X, Yang J, Bao J, Xiang H, Dear K, et al. 2017. Humidity may modify the relationship between temperature and cardiovascular mortality in Zhejiang province, China. Int J Environ Res Public Health. 14 (11 ):1383, PMID: , 10.3390/ijerph14111383.29135955
|
PMC010xxxxxx/PMC10075311.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37018009
EHP10549
10.1289/EHP10549
Research
Urinary Phthalate Metabolites and Slow Walking Speed in the Korean Elderly Environmental Panel II Study
https://orcid.org/0000-0003-2717-8370
Yoon Jeonggyo 1 2
https://orcid.org/0000-0002-8688-5174
García-Esquinas Esther 3 4
https://orcid.org/0000-0003-4802-010X
Kim Junghoon 5
https://orcid.org/0000-0001-5452-7965
Kwak Jung Hyun 6
https://orcid.org/0000-0002-5539-7653
Kim Hongsoo 7 8 9
https://orcid.org/0000-0001-8726-9288
Kim Sungroul 10 11
Kim Kyoung-Nam 12
Hong Yun-Chul 13
https://orcid.org/0000-0003-3228-8179
Choi Yoon-Hyeong 2 14
1 Department of Community, Environment and Policy, Mel and Enid Zuckerman College of Public Health, University of Arizona, Tucson, Arizona, USA
2 Department of Preventive Medicine, Gachon University College of Medicine, Incheon, Korea
3 Department of Chronic Diseases Epidemiology, National Center for Epidemiology, Instituto de Salud Carlos III (ISCIII), Madrid, Spain
4 Ciber of Epidemiology and Public Health (CIBERESP), Madrid, Spain
5 Department of Sports Medicine, Graduate School of Sports Convergence, Korea Maritime and Ocean University, Busan, Korea
6 Department of Food and Nutrition, Gangneung-Wonju National University, Gangneung, Gangwon-do, Korea
7 Department of Public Health Science, Graduate School of Public Health; Seoul National University, Seoul, Korea
8 Institute of Health & Environment, Seoul National University, Seoul, Korea
9 Institute of Aging, Seoul National University, Seoul, Korea
10 Department of Environmental Health Sciences, Soonchunhyang University, Asan, Korea
11 Department of ICT Environmental Health System, Graduate School, Soonchunhyang University (BK21Four), Asan, Korea
12 Department of Preventive Medicine and Public Health, Ajou University School of Medicine, Suwon, Korea
13 Department of Preventive Medicine, Seoul National University College of Medicine, Seoul, Korea
14 School of Health and Environmental Science, College of Health Science, Korea University, Seoul, Korea
Address correspondence to Yoon-Hyeong Choi, School of Health and Environmental Science, College of Health Science, Korea University 145 Anam-ro, Seongbuk-gu, Seoul, Republic of Korea 02841. Email: [email protected]
5 4 2023
4 2023
131 4 04700526 10 2021
06 1 2023
13 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Previous epidemiological studies have suggested that phthalate exposure may contribute to neurocognitive and neurobehavioral disorders and decreased muscle strength and bone mass, all of which may be associated with reduced physical performance. Walking speed is a reliable assessment tool for measuring physical performance in adults age 60 y and older.
Objective:
We investigated associations between urinary phthalate metabolites and slowness of walking speed in community-dwelling adults ages 60–98 y.
Methods:
We analyzed 1,190 older adults [range, 60–98 y of age; mean±standard deviation (SD) , 74.81±5.99] from the Korean Elderly Environmental Panel II study and measured repeatedly up to three times between 2012 and 2014. Phthalate exposure was estimated using the following phthalate metabolites in urine samples: mono-(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), mono-(2-ethyl-5-oxohexyl) phthalate (MEOHP), mono-n-butyl phthalate (MnBP), mono-(2-ethyl-5-carboxypentyl) phthalate (MECPP), and mono-benzyl phthalate (MBzP). Slowness was defined as a walking speed of <1.0meter/second. We used logistic and linear regression models to evaluate the association between each urinary phthalate metabolite and slowness or walking-speed change. We also used Bayesian kernel machine regression (BKMR) to examine overall mixture effects on walking speed.
Results:
At enrollment, MBzP levels were associated with an increased odds of slowness [odds ratio (OR) per doubling increase: 1.15, 95% confidence interval (CI): 1.02, 1.30; OR for the highest vs. lowest quartile: 2.20 (95% CI: 1.12, 4.35) with p-trend across quartiles=0.031]. In longitudinal analyses, MEHHP levels showed an increased risk of slowness [OR per doubling increase: 1.15 (95% CI: 1.02, 1.29), OR for the highest vs. lowest quartile: 1.47 (95% CI: 1.04, 2.06), p- trend=0.035]; whereas those with higher MnBP showed a reduced risk of slowness [OR per doubling increase: 0.84 (95% CI: 0.74, 0.96), OR in the highest (vs. lowest) quartile: 0.64 (95% CI: 0.47, 0.87), p-trend=0.006]. For linear regression models, MBzP quartiles were associated with slower walking speed (p-trend=0.048) at enrollment, whereas MEHHP quartiles were associated with slower walking speed, and MnBP quartiles were associated with faster walking speed in longitudinal analysis (p-trend=0.026 and <0.001, respectively). Further, the BKMR analysis revealed negative overall trends between the phthalate metabolite mixtures and walking speed and DEHP group (MEHHP, MEOHP, and MECPP) had the main effect of the overall mixture.
Discussion:
Urinary concentrations of prevalent phthalates exhibited significant associations with slow walking speed in adults ages 60–98 y. https://doi.org/10.1289/EHP10549
Supplemental Material is available online (https://doi.org/10.1289/EHP10549).
None of the authors have any conflict of interest regarding the content of this article.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Walking speed is one of the quickest, most reliable, and simple measurements for monitoring the mobility and physical function of older adults.1 Along with blood pressure, pulse rate, respiratory rate, body temperature, and pain, walking speed has been recommended as the “sixth vital sign” for assessing functional ability and overall health.2 Moreover, slowness in walking speed has been associated with adverse health outcomes, including disability, cognitive impairment, hospital admissions, falls, and all-cause mortality in those age 65 y and older.3,4 Recently, a systematic review has identified potentially modifiable risk factors for slow walking speed in community-dwelling adults age 45 y and older, including physical inactivity, low education, obesity, pain, and depression.5 Exposure to environmental factors such as lead,6 cadmium,7 and cobalt8 have been suggested to be associated with walking speed declines.
Phthalates are a group of chemicals used to make plastics more flexible and durable and are widely used in industrial materials, consumer products, and personal care products.9 The high molecular weight phthalates (ester side-chain lengths, five or more carbons) are used widely in polyvinyl chloride (PVC) polymers and plastisol applications, plastics, food packaging and processing materials, vinyl toys and floor coverings, and building products. The most important source of human exposure is diet, particularly foods packaged in plastic or PVC materials.10 The low molecular weight phthalates are used in non-PVC applications, such as personal care products, adhesives, and enteric-coated tablets, and their major sources of human exposure are reported as cosmetics and personal care products.9 Once in the body, phthalates are both endocrine disruptors11 and carcinogenic,12 and their accumulation has been associated with a wide range of health problems, including neurodevelopmental disorders,13 reproductive outcomes,14,15 respiratory16 and cardiovascular diseases,17 and metabolic outcomes (e.g., diabetes, insulin resistance, obesity, and kidney diseases).18 In addition, several epidemiological studies suggested that exposure to phthalates might be associated with neurocognitive and neurobehavioral disorders,19 frailty,20 and declines in both muscle strength21,22 and bone mineral density.23,24 All these outcomes were associated with poor physical performance and disability in adults age 50 y and older.25–28 In vivo experimental studies also showed that phthalate exposure could cause cognitive, neurosensory, and behavioral dysfunction29,30 and locomotor behavior defects through cellular and DNA damage in the brain.31,32 Other experimental studies reported that phthalates were associated with musculoskeletal impairment through glucose catabolic reactions in the muscle33 and developmental malformations in bones.34,35 Moreover, there is evidence that phthalates are associated with increases in oxidative stress and inflammation biomarkers such as C-reactive protein (CRP), interleukin 6 (IL-6), and tumor necrosis factor-α (TNFα),36,37 which are considered important mechanisms of age-related physical function decline.38–40 Despite this epidemiological and experimental evidence, the potential influence of phthalate exposure on walking speed among adults age 60 y and older remains unknown. Therefore, the present study aimed to evaluate whether exposure to phthalates estimated based on urinary mono-(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), mono-(2-ethyl-5-oxohexyl) phthalate (MEOHP), mono-n-butyl phthalate (MnBP), mono-(2-ethyl-5-carboxypentyl) phthalate (MECPP), and mono-benzyl phthalate (MBzP) concentrations, was associated with slowness of walking speed in adults age ≥60 y who participated in the Korean Elderly Environmental Panel II (KEEP II) study.
Methods
Study Participants
The KEEP II study was designed to examine relationships between environmental exposure and health outcomes in adults age ≥60 y. For this purpose, it conducted repeated interviews, physical examinations, and laboratory testing on participants age 60 y and over who visited two community welfare centers located in Seoul (urban area) and Asan (rural area), South Korea, between 2012 and 2014.21 At each examination, information on sociodemographic characteristics, medical and family history, and lifestyle behaviors was collected using structured questionnaires. Physical examinations and laboratory testing were conducted by certified health technicians who received intensive training on examination protocols. Detailed data collection structure is described in Table S1. Of 1,251 participants initially recruited without communication impairments, subjects who had no information on urinary phthalate metabolites concentrations (n=8) or walking speed (n=35) were excluded. Furthermore, we excluded 14 subjects with missing information on weight (n=10) or physical activity (n=4), as well as four subjects with unreliably high walking speed >1.8meters/second(m/s), leading to a final sample size of 1,190 participants. During the follow-up, subjects participated in each examination up to three times [450 (37.8%) subjects participated only once, 452 (38.0%) subjects participated twice, and 288 (24.2%) subjects participated three times] with 1-y interval [mean±standard deviation (SD): 1.10±0.38 y]. The number of participants was 740 at follow-up 1 and 288 at follow-up 2. Details for the enrollment and follow-up of study participants are presented in Figure 1. Additionally, the following subjects with extreme phthalate metabolite values were excluded from their specific analyses: MEHHP>500μg/L (n=2), MEOHP>380μg/L (n=2), MECPP>700μg/L (n=1), MnBP>1,000μg/L (n=2), and MBzP>100μg/L (n=1).
Figure 1. Classification of participants. Solid line box indicates the first visit for each participant (n=1,190 for all participants), and the dashed and dotted line boxes indicate the second (n=740 for follow-up 1 group) and third visits (n=288 for follow-up 2 group), respectively. Two metabolites (MECPP and MBzP) had one measurement at Exam II (Follow-up A1 and Enrollment B). Only cross-sectional analysis was possible. Thus, repeated measurements of MECPP and MBzP were not available.
Figure 1 is a flowchart with three steps, namely, Exam 1, Exam 2, and Exam 3. There are 1190 number of participants. Step 1: Under exam 1, there are three metabolites, namely, mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-oxohexyl) phthalate, and mono-n-butyl phthalate, and 729 cases under enrollment A. Step 2: Under Exam 2, there are 731 cases with five metabolites, namely, mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-oxohexyl) phthalate, mono-n-butyl phthalate, mono(2-ethyl-5-carboxypentyl), and monobenzyl phthalate, and 377 cases of A 1 follow-up, and 354 cases under Enrollment B. Step 3: Under Exam 3, there are three metabolites, namely, mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-oxohexyl) phthalate, and mono-n-butyl phthalate. Out of 377 cases of A 1 follow-up, 277 cases lead to A 2 follow-up. Out of 729 cases under enrollment A, 106 cases lead to A 1 follow-up. There are 354 cases of enrollment B, out of which 257 lead to B 1 follow-up. There are 107 cases under enrollment C.
The institutional review board of the Seoul National University Hospital (H-1209-006-424) approved and reviewed the study. All participants provided written informed consent before participation.
Walking Speed
At enrollment and each follow-up visit, participants were asked to walk at their usual pace a distance of 2.5 m (see Table S1). The walking time started when the participant’s foot crossed the starting line and fully touched the floor and ended when the participant’s foot completely passed the ending line and fully touched the floor. This process was repeated twice using a handheld stopwatch, and the faster value was used in the analyses. Walking speed (m/s) was computed by dividing the walked distance (in meters) by the recorded time of walk (in seconds). Walking speed has been proven to have a test-retest reliability.1 This measure has been used in previous studies of older adults including the Health and Retirement Study (HRS).41,42
For the main analyses, slowness was defined using the current 1.0m/s cutoff point suggested by the Asian Working Group for Sarcopenia (AWGS) in 2019.43 Additionally, in sensitivity analyses, slowness was defined using two alternative cutoff points: a) 0.8m/s44 and b) the lowest sex- and height-specific quintiles.45 The current AWGS definition of slowness (walking speed<1.0m/s) has been shown to predict mild physical disability and has been associated with a higher risk of hospitalization and lower extremity functional limitations,3 whereas the other two definitions have been associated with severe physical disability and an increased risk of falls and mortality.46,47
Urinary Phthalate Metabolites
Urine samples for all participants were collected at enrollment and at each follow-up visit. Phthalate exposure was estimated based on the following metabolites in the urine: MEHHP, MEOHP, MnBP, MECPP, and MBzP. Although MEHHP, MEOHP, and MnBP metabolites were available at enrollment and each follow-up visit (Exam I, Exam II, and Exam III), MECPP and MBzP metabolites were only measured at Exam II (see Table S1; Figure 1). Spot urine samples were collected from participants during their physical examination between 1000 hours to 1200 hours (10:00 A.M. to 12:00 P.M.) and frozen immediately at −20°C until laboratory analyses. Concentrations of urinary metabolites were analyzed using ultra performance liquid chromatography–tandem mass spectrometry (LC-MS/MS) (Xevo TQ-S; Waters) according to a previously reported method.21 The limits of detection (LODs) for MEHHP, MEOHP, MECPP, MnBP, and MBzP were 0.32μg/L, 0.20μg/L, 0.26μg/L, 0.35μg/L, and 0.19μg/L, respectively. Values below the LOD were replaced by the LOD divided by 2.
Other Covariates
Information on covariates was obtained at enrollment and at each follow-up visit (see Table S1). Based on a priori biological and epidemiological knowledge, the following were considered as potential confounders: sociodemographic factors, including age, sex, education level (≤elementary school, middle and high school, and >high school ), economic status (household income <USD $450, USD $450–2,659.9, or ≥USD $2,660), living arrangement (living alone, living with spouse, living with child, or others), and city of residence; anthropometric measurements: height and weight measured by certified health technicians; health-related behaviors, such as physical activity and smoking (no, yes); clinical factors, including self-reported depression and self-reported physician diagnosis of osteoarticular disease (osteoporosis and osteoarthritis), cardiovascular disease (angina, myocardial infarction, stroke, and cerebrovascular disease), respiratory disease (asthma and chronic obstructive pulmonary disease), cancer, hypertension, and diabetes; and urinary creatinine as measured using a Hitachi Automatic Analyzer (Hitachi 7600).
Because height and weight are known to affect walking speed,48 we controlled for these variables separately, instead of adjusting for body mass index. Physical activity [metabolic equivalents per week (METs-hours/week)] was computed as the sum of weekly METs of moderate and vigorous activity, as measured using the Korean version of the International Physical Activity Questionnaire (IPAQ).49 Individuals were classified into those who engaged in <7.5 (reference) and ≥7.5 METs-hours/week, which is recommended for Leisure Time Physical Activity by the World Health Organization.50 Current and former smokers were not considered separately because the number of current smokers was very low (n=57, 24, and 4 at enrollment and follow-up 1 and follow-up 2 visits). Depression was measured using the Korean version of the Short Form Geriatric Depression Scale (SGDS-K, range 0–15, with higher scores being indicative of more severe depression).51
Statistical Analyses
All statistical analyses were performed using SAS software (version 9.4; SAS Institute) and R (version 4.1.1; R Development Core Team). The statistical significance level was set as p<0.05. Characteristics of participants between subjects with and without slowness were compared using t-test for continuous variables and the chi-square test for categorical variables. Distributions of phthalate metabolite by participant characteristics were summarized using t-test for binomial variables and Wald F-test for categorical variables. We evaluated Spearman's correlations between all phthalate metabolites collected at a single point and Spearman correlations between enrollment and follow-up visits for each metabolite. Correlation between metabolites was analyzed using data at Exam II, the only time when all five metabolites were measured.
We performed a cross-sectional analysis for all five phthalate metabolites with the first measurement data and additionally examined longitudinal analysis for three phthalate metabolites with repeated measurement data. Urinary phthalate metabolite concentrations were modeled as continuous variables (log-transformed to normalize their distributions) and quartiles with p for trend. In addition, p-values for linear trend across quartiles of phthalate metabolites were computed by modeling categories of the metabolite as an ordinal variable coded using integer values (0–3). Furthermore, we computed estimates by comparing each of the upper three quartiles to the lowest quartile. The cutoff value of each quartile for phthalate metabolite concentration was applied using values obtained at enrollment. To estimate the odds ratios (ORs) of slowness associated with phthalate metabolite levels, logistic regression models (PROC LOGISTIC) for cross-sectional analyses and marginal logistic models based on generalized estimating equations (PROC GENMOD) for longitudinal analyses were used. To evaluate the change in walking speed (meters/second) associated with phthalate metabolite levels, linear regression models (PROC GLM) were used for cross-sectional analysis and linear mixed effect models (PROC MIXED) were used for longitudinal analysis. A random intercept and a random slope for the time elapsed from the first visit were used to account for the heterogeneity across subjects and subject-specific variability of walking speed over time. All models were adjusted for age, sex, education level, city of residence, height, weight, physical activity, smoking status, osteoarticular disease, cardiovascular disease, respiratory disease, cancer, hypertension, and diabetes; potentially time-varying covariates were collected at enrollment and every follow-up. To account for variations in urinary dilution, we fitted creatinine-corrected models dividing phthalate metabolite concentrations by the creatinine concentrations.
We evaluated the overall effect as a mixture of all five phthalate metabolites using Bayesian kernel machine regression (BKMR) while accounting for nonlinear exposure–response relationships and the relative importance of individual metabolites in the association between mixtures and walking speed.52,53 BKMR analysis was conducted using data at one time point (Exam II) when all five phthalate metabolites were measured to account for overall mixture effects and individual effects on walking speed. First, we evaluated overall mixture effects of phthalate metabolites on changes in walking speed (m/s) when all phthalate metabolites were at specific percentiles (i.e., first to 99th percentile) in comparison with their median values. Second, posterior inclusion probabilities (PIPs) were used to determine which phthalate metabolites were important contributors to the association between the overall mixture effects and walking speed.54 We also grouped phthalate exposure according to phthalate metabolite precursors and divided the five phthalate metabolites into three groups (group 1: MEHHP, MEOHP, and MECPP; group 2: MnBP; and group 3: MBzP) to fit hierarchical BKMR models.53
For all BKMR analyses, log-transformed creatinine-corrected phthalate metabolites levels were centered to a mean of 0 and scaled to an SD of 1. A PIP value of ≥0.5 was considered meaningful.54,55 All models were adjusted for age, sex, region, education level, smoking status, weight, height, physical activity, osteoarticular disease, cardiovascular disease, chronic respiratory diseases, cancer, diabetes, and hypertension.
Sensitivity Analyses
The following sensitivity analyses were conducted: First, we examined whether the results were different between the nonadjusted creatinine models and the creatinine-adjusted models, including creatinine concentrations as a covariate in the models.56 Second, we examined logistic regression models using 0.8m/s and the lowest quintile of sex- and height-adjusted walking speed as cutoff points for slowness. Third, because socioeconomic status (SES) affects walking speed,57 we examined whether the results differed after adjustment for additional SES factors, i.e., household income (<USD $450, USD $450–2,659.9, or ≥USD $2,660) and living arrangement (living alone, living with spouse, living with child, or others) in the subset sample with this information available (1,041 subjects with 1,721 observations). Fourth, because phthalate exposure was associated with the risk of depression in older adults58 and depression might affect walking speed in older adults,59 we examined whether results differed after adjusting for depression (1,189 subjects with 2,215 observations). Fifth, we conducted stratified models to assess potential effect modification by sex.
Results
Table 1 shows participants’ main characteristics by walking speed group. Among a total of 1,190 older adults (range of 60–98 y of age) included at enrollment, 30.67% were males, and their mean±SD age was 74.81±5.99 y. The participants with slowness (n=893, 75.04%) were significantly older and had lower height, weight, education, and physical activity levels than their counterparts.
Table 1 Enrollment and follow-up characteristics of 1,190 older adults from the Korean Elderly Environmental Panel II (2012–2014) by slowness status.
Total Slowness
<1 m/s Nonslowness
≥1 m/s p-Valuea
n Mean ±SD n Mean ±SD n Mean ±SD
Demographic characteristics at enrollment
Age (y) 1,190 74.81 ±5.99 893 75.75 ±5.91 297 71.99 ±5.31 <.001
BMI (kg/m2) 1,190 23.79 ±3.09 893 23.75 ±3.21 297 23.90 ±2.68 0.434
Height (cm) 1,190 155.96 ±8.59 893 155.61 ±8.75 297 157.02 ±8.01 0.014
Weight (kg) 1,190 57.96 ±9.36 893 57.62 ±9.68 297 58.98 ±8.25 0.019
Sex [n (%)] — — — — — — — — — 0.150
Male 365 30.67 264 29.56 — 101 34.01 — —
Female 825 69.33 629 70.44 — 196 65.99 — —
Education [n (%)] — — — — — — — — — <.001
≤Elementary school 797 66.97 — 668 74.80 — 129 43.13 — —
Middle and high school 310 26.05 — 183 20.49 — 127 42.76 — —
>High school 83 6.97 — 42 4.70 — 41 13.80 — —
Physical activity (METs-hours/week) [n (%)] — — — — — <.001
<7.5 988 83.03 — 785 87.91 — 203 68.35 — —
≥7.5 202 16.97 — 108 12.09 — 94 31.65 — —
Phthalate metabolitesb
MEHHP (μg/L)
Enrollment 1,190 21.25 ±2.42 893 22.79 ±2.39 297 17.23 ±2.45 <.001
Follow-up 1 740 17.94 ±2.19 583 18.84 ±2.18 157 14.98 ±2.16 0.001
Follow-up 2 288 15.99 ±2.27 154 17.46 ±2.34 134 14.45 ±2.16 0.050
MEOHP (μg/L)
Enrollment 1,190 15.44 ±2.45 893 16.37 ±2.39 297 12.95 ±2.56 <.001
Follow-up 1 740 12.38 ±2.39 583 12.97 ±2.39 157 10.41 ±2.30 0.005
Follow-up 2 288 11.51 ±2.23 154 12.58 ±2.30 134 10.39 ±2.12 0.043
MnBP (μg/L)
Enrollment 1,190 28.50 ±2.15 893 28.65 ±2.16 297 28.02 ±2.10 0.664
Follow-up 1 740 28.43 ±2.21 583 28.08 ±2.21 157 29.74 ±2.23 0.422
Follow-up 2 288 38.15 ±2.16 154 38.60 ±2.22 134 37.64 ±2.10 0.782
MECPP (μg/L)c
Exam II (A1+B) 731 25.00 ±2.15 629 25.62 ±2.15 102 21.45 ±2.17 0.030
MBzP (μg/L)c
Exam II (A1+B) 731 2.19 ±4.43 629 2.32 ±4.45 102 1.53 ±4.15 0.009
Note: All participants: Enrollment A+B+C in Figure 1. No covariate values were missing. —, no data; BMI, body mass index; GM, geometric mean; MBzP, mono-benzyl phthalate; MECPP, mono-(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono-(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono-(2-ethyl-5-oxohexyl) phthalate; MET, metabolic equivalent; MnBP, mono-n-butyl phthalate; m/s, meter/second; SD, standard deviation.
a p-Values were based on t-test for continuous variables and chi-square test for categorical variables.
b GMs and SDs are presented.
c MECPP and MBzP had one measurement (n=731) at Exam II (Follow-up A1 and Enrollment B). Only cross-sectional analysis was possible. Thus, repeated measurements of MECPP and MBzP were not available.
The geometric mean (GM) (SD) was 21.25 (2.42) μg/L for MEHHP, 15.44 (2.45) μg/L for MEOHP, 28.50 (2.15) μg/L for MnBP, 25.00 (2.15) μg/L for MECPP, and 2.19 (4.43) μg/L for MBzP. Except for MnBP, metabolite concentrations were higher in participants with slowness at enrollment and at each follow-up visit (Table 1). The overall distribution of phthalate metabolites is presented in Tables S2 and S3. Concentrations of oxidized metabolites of di(2-ethylhexyl) phthalate (DEHP; i.e., MEHHP, MEOHP, and MECPP) were also higher in older participants, as well as in those who lived in rural areas and had lower physical activity levels; whereas concentrations of MnBP were higher in participants with higher education level, those who lived in urban areas, and those with higher physical activity levels (Figure 2; the numeric values are presented in Table S4). All phthalate metabolites were significantly correlated with each other [all correlation coefficients (r) >0.5 with all p<0.001]; in particular, high molecular weight metabolites (MEHHP, MEOHP, and MECPP from the same parent compound) were strongly correlated with one another (r=0.894, 0.915, and 0.960; all p<0.001; Table 2). Correlations between enrollment and follow-up measures for each metabolite was relatively low (r between 0.296 and 0.463 with p<0.001; Table S5).
Figure 2. Geometric means (95% CIs) of urinary phthalate concentrations according to participant characteristics at enrollment in the KEEP II study. We used a survey t-test for binominal groups and Wald F-test for categorical groups. *Statistical significance at p<0.05. The dotted line includes overall geometric means of phthalate metabolites (numeric values are presented in Table S4). Note: CI, confidence interval.
Figure 2A to 2E are forest plots titled mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-oxohexyl) phthalate, mono-n-butyl phthalate, mono(2-ethyl-5-carboxypentyl), Monobenzyl phthalate, plotting Numbers, ranging as (bottom to top) 657 cases of hypertension, 533 cases of non-hypertension, 246 cases of diabetes, 944 cases of non-diabetes, 73 cases of cancer, 1117 cases of non-cancer, 44 cases of chronic respiratory disease, 1146 cases of non- chronic respiratory disease, 137 cases of cardiovascular disease, 1053 cases of non- cardiovascular disease, 432 cases of osteoarticular disease, 758 cases of non- osteoarticular disease, 202 cases of physical activity greater than or equal to 7.5 metabolic equivalents of task- hour per week, 988 cases of physical activity less than 7.5 metabolic equivalents of task- hour per week, 215 cases of smoker, 975 cases of non-smoker, 83 cases of greater than high school, 310 cases of middle and high school, 797 cases of less than or equal to elementary school, 642 cases of rural, 548 cases of urban, 825 cases of women, 365 cases of men, 237 cases of age greater than or equal to 80, 692 cases of age group between 70 and 79, 261 cases of age group between 60 to 69, and 1190 cases of total; 657 cases of hypertension, 533 cases of non-hypertension, 246 cases of diabetes, 944 cases of non-diabetes, 73 cases of cancer, 1117 cases of non-cancer, 44 cases of chronic respiratory disease, 1146 cases of non- chronic respiratory disease, 137 cases of cardiovascular disease, 1053 cases of non- cardiovascular disease, 432 cases of osteoarticular disease, 758 cases of non- osteoarticular disease, 202 cases of physical activity greater than or equal to 7.5 metabolic equivalents of task- hour per week, 988 cases of physical activity less than 7.5 metabolic equivalents of task- hour per week, 215 cases of smoker, 975 cases of non-smoker, 83 cases of greater than high school, 310 cases of middle and high school, 797 cases of less than or equal to elementary school, 642 cases of rural, 548 cases of urban, 825 cases of women, 365 cases of men, 237 cases of age greater than or equal to 80, 692 cases of age group between 70 and 79, 261 cases of age group between 60 to 69, and 1190 cases of total; 657 cases of hypertension, 533 cases of non-hypertension, 246 cases of diabetes, 944 cases of non-diabetes, 73 cases of cancer, 1117 cases of non-cancer, 44 cases of chronic respiratory disease, 1146 cases of non- chronic respiratory disease, 137 cases of cardiovascular disease, 1053 cases of non- cardiovascular disease, 432 cases of osteoarticular disease, 758 cases of non- osteoarticular disease, 202 cases of physical activity greater than or equal to 7.5 metabolic equivalents of task- hour per week, 988 cases of physical activity less than 7.5 metabolic equivalents of task- hour per week, 215 cases of smoker, 975 cases of non-smoker, 83 cases of greater than high school, 310 cases of middle and high school, 797 cases of less than or equal to elementary school, 642 cases of rural, 548 cases of urban, 825 cases of women, 365 cases of men, 237 cases of age greater than or equal to 80, 692 cases of age group between 70 and 79, 261 cases of age group between 60 to 69, and 1190 cases of total; 429 cases of hypertension, 302 cases of non-hypertension, 161 cases of diabetes, 570 cases of non-diabetes, 44 cases of cancer, 687 cases of non-cancer, 26 cases of chronic respiratory disease, 705 cases of non- chronic respiratory disease, 107 cases of cardiovascular disease, 624 cases of non- cardiovascular disease, 332 cases of osteoarticular disease, 399 cases of non- osteoarticular disease, 112 cases of physical activity greater than or equal to 7.5 metabolic equivalents of task- hour per week, 619 cases of physical activity less than 7.5 metabolic equivalents of task- hour per week, 112 cases of smoker, 619 cases of non-smoker, 54 cases of greater than high school, 187 cases of middle and high school, 490 cases of less than or equal to elementary school, 337 cases of rural, 394 cases of urban, 539 cases of women, 192 cases of men, 149 cases of age greater than or equal to 80, 450 cases of age group between 70 and 79, 132 cases of age group between 60 to 69, and 731 cases of total; and 429 cases of hypertension, 302 cases of non-hypertension, 161 cases of diabetes, 570 cases of non-diabetes, 44 cases of cancer, 687 cases of non-cancer, 26 cases of chronic respiratory disease, 705 cases of non- chronic respiratory disease, 107 cases of cardiovascular disease, 624 cases of non- cardiovascular disease, 332 cases of osteoarticular disease, 399 cases of non- osteoarticular disease, 112 cases of physical activity greater than or equal to 7.5 metabolic equivalents of task- hour per week, 619 cases of physical activity less than 7.5 metabolic equivalents of task- hour per week, 112 cases of smoker, 619 cases of non-smoker, 54 cases of greater than high school, 187 cases of middle and high school, 490 cases of less than or equal to elementary school, 337 cases of rural, 394 cases of urban, 539 cases of women, 192 cases of men, 149 cases of age greater than or equal to 80, 450 cases of age group between 70 and 79, 132 cases of age group between 60 to 69, and 731 cases of total (y-axis) across geomeans (microgram per liter), ranging from 10.0 to 40.0 in increments of 5.0, 10.0 to 40.0 in increments of 5.0, 10.0 to 40.0 in increments of 5.0, 10.0 to 40.0 in increments of 5.0, and 0.0 to 5.0 in increments of 1.0 (x-axis), respectively.
Table 2 Spearman’s coefficients of a single point correlation between phthalate metabolites at Exam II in the Korean Elderly Environmental Panel II study (n=731).
MEHHP (μg/L) MEOHP (μg/L) MnBP (μg/L) MECPPa
(μg/L) MBzPa (μg/L)
MEHHP (μg/L) 1 — — — —
MEOHP (μg/L) 0.960
(p<0.001) 1 — — —
MnBP (μg/L) 0.576
(p<0.001) 0.594
(p<0.001) 1 — —
MECPP (μg/L) 0.894
(p<0.001) 0.915
(p<0.001) 0.601
(p<0.001) 1 —
MBzP (μg/L) 0.523
(p<0.001) 0.528
(p<0.001) 0.505
(p<0.001) 0.537
(p<0.001) 1
Note: —, no data; MBzP, mono-benzyl phthalate; MECPP, mono-(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono-(2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono-(2-ethyl-5-oxohexyl) phthalate; MnBP, mono-n-butyl phthalate.
a MECPP and MBzP had one measurement (n=731) at Exam II (Follow-up A1 and Enrollment B). Only cross-sectional analysis was possible. Thus, repeated measurements of MECPP and MBzP were not available.
Table 3 shows logistic regression results for slowness according to each urinary phthalate metabolite (phthalate levels divided by creatinine and expressed as micrograms per gram creatinine) from cross-sectional analysis at enrollment and longitudinal analysis. In the cross-sectional analysis, we observed a linear trend between MBzP quartiles and slowness (p for trend=0.031). The OR per doubling of MBzP was 1.15 (95% CI: 1.02, 1.30), and the OR in the highest vs. lowest quartile was 2.20 (95% CI: 1.12, 4.35). We did not observe this trend with slowness in the other phthalate metabolites, although those in the second MECPP quartile and the highest MEHHP quartile (in comparison with those in the lowest) showed higher odds of slowness [OR=2.21 (95% CI: 1.17, 4.18) and 1.52 (95% CI: 0.96, 2.41), respectively]. In the longitudinal analysis, MEHHP levels were associated with an increased risk of slowness [OR per doubling: 1.15 (95% CI: 1.02, 1.29), OR in the highest quartile (vs. lowest): 1.47 (95% CI: 1.04, 2.06), and p for trend=0.035], whereas MnBP levels were associated with a decreased risk of slowness [OR per doubling: 0.84 (95% CI: 0.74, 0.96), OR in the highest quartile (vs. lowest): 0.64 (95% CI: 0.47, 0.87), and p for trend=0.006]. Although MEOHP did not show a linear exposure–response association with slowness, the participants in the highest quartile (vs. lowest) of this metabolite showed an increased risk of slowness [OR=1.52 (95% CI: 1.09, 2.13)].
Table 3 Odds ratios (95% CI) for slowness (<1.0m/s) according to urinary phthalate metabolite concentrations (micrograms per gram creatinine) from cross-sectional analysis at enrollment (n=1,190) and longitudinal analysis (n=2,218).
Phthalate metabolites Cross-sectional analysis at enrollment Longitudinal analysis
No. with slowness/no. of participants OR (95% CI) No. with slowness/no. of observations OR (95% CI)
MEHHP (μg/g cre)
Per doubling — 1.12 (0.96, 1.30) — 1.15 (1.02, 1.29)
Quartile 1 (1.07, 16.75) 191/299 1 Ref 421/660 1 Ref
Quartile 2 (16.78, 25.89) 228/297 1.14 (0.77, 1.73) 427/580 1.08 (0.83, 1.41)
Quartile 3 (25.98, 41.58) 220/297 0.97 (0.65, 1.44) 404/530 1.16 (0.87, 1.54)
Quartile 4 (41.74, 306.32) 252/296 1.52 (0.96, 2.41) 376/446 1.47 (1.04, 2.06)
p for trend — 0.198 — — 0.035 —
MEOHP (μg/g cre)
Per doubling — 1.06 (0.92, 1.23) — 1.05 (0.94, 1.17)
Quartile 1 (0.53, 11.98) 197/298 1 Ref 443/680 1 Ref
Quartile 2 (12.01, 18.66) 231/298 1.26 (0.84, 1.90) 423/573 1.04 (0.79, 1.36)
Quartile 3 (18.67, 30.96) 213/297 0.75 (0.50, 1.12) 387/519 0.91 (0.68, 1.22)
Quartile 4 (30.98, 263.23) 250/296 1.45 (0.92, 2.29) 375/444 1.52 (1.09, 2.13)
p for trend — 0.555 — — 0.105 —
MECPPa (μg/g cre)
Per doubling — 1.20 (0.92, 1.58) — NA NA
Quartile 1 (3.89, 18.65) 142/182 1 Ref NA NA NA
Quartile 2 (18.67, 27.29) 164/183 2.21 (1.17, 4.18) NA NA NA
Quartile 3 (27.29, 41.43) 162/183 1.65 (0.89, 3.06) NA NA NA
Quartile 4 (41.48, 216.94) 160/182 1.39 (0.74, 2.59) NA NA NA
p for trend — 0.326 — — NA —
MnBP (μg/g cre)
Per doubling — 1.03 (0.86, 1.25) — 0.84 (0.74, 0.96)
Quartile 1 (4.60, 24.63) 231/298 1 Ref 401/521 1 Ref
Quartile 2 (24.65, 33.97) 212/297 0.75 (0.49, 1.14) 380/517 0.72 (0.53, 0.99)
Quartile 3 (33.98, 49.07) 220/298 0.89 (0.58, 1.37) 403/565 0.67 (0.50, 0.91)
Quartile 4 (49.15, 764.57) 229/296 0.99 (0.64, 1.53) 145/614 0.64 (0.47, 0.87)
p for trend — 0.829 — — 0.006 —
MBzPa (μg/g cre)
Per doubling — 1.15 (1.02, 1.30) — NA NA
Quartile 1 (0.04, 1.13) 146/182 1 Ref NA NA NA
Quartile 2 (1.14, 2.76) 158/183 1.32 (0.72, 2.42) NA NA NA
Quartile 3 (2.76, 5.53) 158/183 1.32 (0.72, 2.43) NA NA NA
Quartile 4 (5.61, 66.47) 167/182 2.20 (1.12, 4.35) NA NA NA
p for trend — 0.031 — — NA —
Note: All participants: Enrollment A+B+C in Figure 1. All observations: Enrollment A+B+C and Follow-up A1+A2+B1 in Figure 1. Logistic regression models were used for cross-sectional analysis and generalized estimating equation models were used for longitudinal analysis. All models were adjusted for age, sex, region, education level, smoking status, weight, height, physical activity, osteoarticular disease, cardiovascular disease, chronic respiratory diseases, cancer, diabetes, and hypertension. —, no data; CI, confidence interval; Cre, creatinine; MBzP, mono-benzyl phthalate; MECPP, mono-(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono (2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono (2-ethyl-5-oxohexyl) phthalate; MnBP, mono-n-butyl phthalate; m/s, meter/second; NA, not available; Ref, reference.
a MECPP and MBzP had one measurement (n=731) at Exam II (Follow-up Al+Enrollment B). Only cross-sectional analysis was possible. Thus, repeated measurements of MECPP and MBzP were not available.
Table 4 presents linear regression results of changes in walking speed according to each urinary phthalate metabolite from cross-sectional analysis at enrollment and longitudinal analysis. In the cross-sectional analysis, MBzP quartiles showed a liner trend with slower walking speed (p for trend=0.048). The change in walking speed with an increased MBzP level was −0.01m/s (95% CI: −0.02, −0.00) per doubling and −0.05m/s (95% CI: −0.09, −0.01) in the highest quartile (vs. lowest). Although we did not observe this trend in other phthalate metabolites, MECPP levels were associated with slower walking speeds [change in walking speed: −0.02m/s (95% CI: −0.04, −0.00) per doubling, −0.06m/s (95% CI: −0.10, −0.02) and −0.05m/s (95% CI: −0.09, −0.01)] in the second and third quartiles (vs. lowest), whereas MnBP levels were associated with faster walking speeds [change in walking speed: 0.04m/s (95% CI: 0.00, 0.07) in the second quartile (vs. lowest)]. In longitudinal analysis, MEHHP quartiles showed a linear trend with slower walking speed, and MnBP quartiles showed a linear trend with faster walking speed (p for trend=0.026 and <0.001, respectively). The change in walking speed was −0.01m/s (95% CI: −0.02, −0.00) per doubling of MEHHP, and 0.03m/s (95% CI: 0.02, 0.04) per doubling of MnBP, and 0.04m/s (95% CI: 0.02, 0.07), 0.05m/s (95% CI: 0.03, 0.08), and 0.08m/s (95% CI: 0.05, 0.10) increases in the second, third, and fourth MnBP quartiles (vs. lowest), respectively.
Table 4 Change (95% CI) in walking speed (m/s) according to urinary phthalate metabolite concentrations (micrograms per gram cre) from cross-sectional analysis at enrollment (n=1,190) and longitudinal analysis (n=2,218).
Phthalate metabolites Cross-sectional analysis at enrollment Longitudinal analysis
No. of participants Estimate (m/s) (95% CI) No. of observations Estimate (m/s) (95% CI)
MEHHP (μg/g cre)
Per doubling — −0.01 (−0.02, 0.01) — −0.01 (−0.02, −0.00)
Quartile 1 (1.07, 16.75) 299 0 Ref 660 0 Ref
Quartile 2 (16.78, 25.89) 297 −0.01 (−0.05, 0.02) 580 0.00 (−0.03, 0.02)
Quartile 3 (25.98, 41.58) 297 −0.01 (−0.05, 0.02) 530 −0.02 (−0.05, 0.00)
Quartile 4 (41.74, 306.32) 296 −0.03 (−0.06, 0.01) 446 −0.03 (−0.05, 0.00)
p for trend — 0.174 — — 0.026 —
MEOHP (μg/g cre)
Per doubling — 0.00 (−0.01, 0.01) — 0.00 (−0.01, 0.01)
Quartile 1 (0.53, 11.98) 298 0 Ref 680 0 Ref
Quartile 2 (12.01, 18.66) 298 0.00 (−0.04, 0.03) 573 0.01 (−0.01, 0.04)
Quartile 3 (18.67, 30.96) 297 0.02 (−0.01, 0.06) 519 0.01 (−0.02, 0.03)
Quartile 4 (30.98, 263.23) 296 −0.01 (−0.04, 0.03) 444 −0.01 (−0.04, 0.02)
p for trend — 0.865 — — 0.493 —
MECPPa (μg/g cre)
Per doubling — −0.02 (−0.04, −0.00) — NA NA
Quartile 1 (3.89, 18.65) 182 0 Ref NA NA NA
Quartile 2 (18.67, 27.29) 183 −0.06 (−0.10, −0.02) NA NA NA
Quartile 3 (27.29, 41.43) 183 −0.05 (−0.09, −0.01) NA NA NA
Quartile 4 (41.48, 216.94) 182 −0.04 (−0.08, 0.00) NA NA NA
p for trend — 0.112 — — NA —
MnBP (μg/g cre)
Per doubling — 0.00 (−0.01, 0.02) — 0.03 (0.02, 0.04)
Quartile 1 (4.60, 24.63) 298 0 Ref 521 0 Ref
Quartile 2 (24.65, 33.97) 297 0.04 (0.00, 0.07) 517 0.04 (0.02, 0.07)
Quartile 3 (33.98, 49.07) 298 0.02 (−0.01, 0.05) 565 0.05 (0.03, 0.08)
Quartile 4 (49.15, 764.57) 296 0.03 (−0.01, 0.06) 614 0.08 (0.05, 0.10)
p for trend — 0.284 — — <.001 —
MBzPa (μg/g cre)
Per doubling — −0.01 (−0.02, −0.00) — NA NA
Quartile 1 (0.04, 1.13) 182 0 Ref NA NA NA
Quartile 2 (1.14, 2.76) 183 −0.04 (−0.08, 0.00) NA NA NA
Quartile 3 (2.76, 5.53) 183 −0.02 (−0.06, 0.02) NA NA NA
Quartile 4 (5.61, 66.47) 182 −0.05 (−0.09, −0.01) NA NA NA
p for trend — 0.048 — — NA —
Note: All participants: Enrollment A+B+C in Figure 1. All observations: Enrollment A+B+C and Follow-up A1+A1+B1 in Figure 1. Linear regression models were used for cross-sectional analysis and linear mixed-effects models were used for longitudinal analysis. All models were adjusted for age, sex, region, education level, smoking status, weight, height, physical activity, osteoarticular disease, cardiovascular disease, chronic respiratory diseases, cancer, diabetes, and hypertension. —, no data; CI, confidence interval; Cre, creatinine; MBzP, mono-benzyl phthalate; MECPP, mono-(2-ethyl-5-carboxypentyl) phthalate; MEHHP, mono (2-ethyl-5-hydroxyhexyl) phthalate; MEOHP, mono (2-ethyl-5-oxohexyl) phthalate; MnBP, mono-n-butyl phthalate; m/s, meters/second; NA, not available; Ref, reference.
a MECPP and MBzP had one measurement (n=731) at Exam II (Follow-up A1+Enrollment B). Only cross-sectional analysis was possible. Thus, repeated measurements of MECPP and MBzP were not available.
In the sensitivity analysis, results in the creatinine-unadjusted models, as well as in models adjusted for creatinine by including it as a covariate, were consistent, even though the statistical significances differed between the logistic (Table S6) and linear regression models (Table S7) depending on the metabolites. These results were consistent although when slowness was defined using the alternative proposed cutoff point of 0.8m/s, results were no longer statistically significant (∼49% of the population rated slow; Table S8), or the lowest quintile of walking speed was adjusted for sex and height (Table S9). Additionally, results were also consistent when models were further adjusted for SES factors and depression in subjects where this information was available (Table S10).
In stratified analyses, we observed an effect modification for MnBP with an inverse exposure–response association in women (p for interaction=0.018; Table S11).
After controlling for all covariates, the BKMR model showed that the overall mixture of five phthalate metabolites was inversely associated with walking speed when all phthalate metabolites were in the first to 99th percentiles in comparison with the median value (Figure 3; numeric values are presented in Table S12). For example, in comparison with the 50th percentile, the subjects with overall phthalate mixtures in the 75th and 95th percentiles had 0.021m/s [95% credible interval (Crl): −0.035, −0.006] and 0.047m/s (95% Crl: −0.093, −0.001) decreases in walking speed, respectively. Based on the estimated PIPs, none of the five phthalate metabolites (all PIPs<0.5) was identified as a significantly dominant contributor to the overall association (Table S13). In our hierarchical model, we observed that group 1 (MEHHP, MEOHP, and MECPP) had the highest group PIP, driving the main effect of the overall mixture (group PIP=0.51), in which MEHHP played the most important role (conditional PIP=0.72) (Table S13).
Figure 3. Overall effects of phthalate metabolite mixtures on changes in walking speed (m/s, meters/second) estimated by Bayesian kernel machine regression. Analysis was conducted with data at one time point (Exam II) when all five metabolites were measured (n=731). The model was adjusted for age, sex, region, education level, smoking status, weight, height, physical activity, osteoarticular disease, cardiovascular disease, chronic respiratory diseases, cancer, diabetes, and hypertension. The plot shows the estimated values when all log-transformed phthalates were at the respective percentiles (from the first to the 99th) in comparison with median values. Variation is presented using 95% credible intervals (numeric values are presented in Table S12).
Figure 3 is a ribbon plus line graph, plotting estimate changes in walking speed (meters per second), ranging from negative 0.1 to 0.1 in increments of 0.1 (y-axis) across percentile of phthalate concentration, ranging from 0.00 to 1.00 in increments of 0.25 (x-axis).
Discussion
In the current study, we evaluated the association between urinary phthalate metabolite levels and slowness in adults aged ≥60 years using data from the Korean Elderly Environmental Panel II (KEEP II) study. The cross-sectional data showed a positive exposure-response association between creatinine-corrected (micrograms per gram creatinine) MBzP levels and slowness, and higher concentrations of MEHHP were positively associated with the odds of slowness. The association for MEHHP became stronger in longitudinal analyses. MnBP showed a null result in the cross-sectional analysis but an inverse exposure–response association with slowness in the longitudinal analysis. Linear regression models using walking speed (meter/second) showed similar results as well as in MBzP, MEHHP, and MnBP. Furthermore, the current study found an overall joint effect of urinary phthalate metabolites in the association with decreased walking speed. The DEHP group (MEHHP, MEOHP, and MECPP) was observed to have the main effect of the overall mixture, in which MEHHP played the most important role.
Although there is growing evidence on the potential associations between environmental pollutants and declined physical function,20,60 and in particular on reduction in walking speed,6–8 this is the first epidemiological study, to our knowledge, that has evaluated the association between phthalate metabolites and slowness.
The associations observed may be explained through several mechanisms. First, phthalates could exert neurological effects on several cognitive, neurosensory, and behavioral abilities, which control walking function.61 In this respect, in vivo studies using Kunming mice, given oral exposure to DEHP and benzyl butyl phthalate (BBP) caused spatial learning and memory dysfunction;29,30 whereas phthalate exposure in zebrafish induced damage to the primary neurons and reduced the expression of genes associated with central nervous system development.32 There is also epidemiological evidence in humans suggesting that phthalate exposure may induce declines in memory62 and hearing.19 Second, phthalates have been associated with changes in the musculoskeletal system, manifested as reductions in bone mineral density,24 an increased risk of osteoporosis,23 and low grip strength.21,22 In other animal studies, exposure to phthalates has been associated with disruptions to skeletal formation and bone homeostasis.34,35 Third, a recent literature review has suggested that phthalates are associated with the cardiovascular system (e.g., hypertension, atherosclerosis, diabetes, and obesity),63 which might cause slower walking speed in adults age 45 y and older.64,65 However, adjusting for cardiovascular diseases and hypertension did not alter the observed associations. Fourth, there is evidence that phthalate exposure can induce oxidative stress and inflammation,36,37 both of which are important mechanisms associated with age-related physical functional decline.40 Indeed, several epidemiological studies have shown that inflammatory markers such as CRP, IL-6, and TNFα are associated with decreased physical function in the adults age 65 y and older through catabolic effects on muscle38–40 and through structural brain damage.66 In addition, Semba et al. have reported an inverse relationship between protein carbonyl level, as an indicator of oxidative damage to protein, and decreased walking speed among women age 65 y and older.67 Increased concentrations of superoxide anion production by nicotinamide adenine dinucleotide phosphate (NADPH) oxidase, another indicator of oxidative stress, have also been associated with slow walking speed in adults ages 76–84 y.68 Moreover, a growing body of evidence suggests that increased oxidative stress can induce mitochondrial DNA and microglia damage in the aging brain, which may lead to cognitive and neurodegenerative diseases.69 One in vivo study has reported that phthalate exposure can induce neurotoxicity through oxidative stress.31 Fifth, several in vitro studies have shown that phthalates can activate peroxisome proliferation–activated receptors (PPARs) in skeletal muscle, brain, and adipose tissues.70,71 Moreover, recent literature review and animal studies have suggested that some PPARs can not only be activated by phthalate metabolites but also can modify some adverse effects of phthalates on reproductive,72 hepatic,73 neurological,74 and cardiovascular outcomes.75 Last, an in vitro study using mice has found evidence that DEHP can induce PPARγ overexpression and result in apoptosis of undifferentiated neurons.76
However, we found inconsistent results in the association between low molecular weight phthalate metabolites (MnBP and MBzP) and walking speed. MBzP revealed an exposure–response association with slowness in the cross-sectional analysis, and MnBP showed a null result in the cross-sectional analysis but an inverse exposure–response association with slowness in the longitudinal analysis. The observed inverse association between MnBP and slowness might be affected by confounding of socioeconomic status [see directed acyclic graph (DAG)] in Figure S1, because the higher usage of personal care products, a major source of MnBP exposure, was correlated with higher SES (e.g., household income and education level),77 and individuals with higher SES were associated with better physical function.57 Therefore, we performed sensitivity analyses adjusting for household income, but the results were virtually the same (Table S10). An interesting finding was that urinary MnBP levels in the current study were marginally correlated with household income in women (p=0.054) but not in men (p=0.305) (Table S14). Silva et al. have reported that concentrations of MnBP used in personal care products, such as perfumes, lotions, cosmetics, and hair care products, are higher in women than in men.78 Thus, we performed a sensitivity analysis of the association between phthalate metabolites and slowness in subgroups according to sex (Table S11). An inverse exposure–response association was observed only in women but not in men. Taken together, a correlation between SES and personal care product use, one of major sources of MnBP in women, might lead to a change in the estimated association between MnBP and slowness in women (Figure S1B). Because the majority of the participants in the current study were women (69.3%), the associations in women may have driven the results in the overall study population.
Strengths of this study include the use of repeated measures for urinary phthalates and walking speed, the use of several phthalate metabolites, the fact that we adjusted for an important number of potential confounding factors, and the use of different approaches to account for urinary dilution. Another strength of the current study was the use of a sophisticated method, i.e., the BKMR approach, to evaluate the overall mixture effect of urinary phthalate metabolites on walking speed. It is challenging to detect an exposure–response relationship between chemical mixtures and health outcomes, in particular when the chemicals are highly correlated or from the same precursors. Among the potential limitations, we had a relatively low number of participants, which may have limited our ability to observe statistically significant associations. Unfortunately, we do not have information on potential biological mediators (i.e., inflammatory favors) and, due to a short biological half-life (24–48 h),79 our urinary phthalate metabolite levels could provide only an estimate of recent exposure. However, previous studies have demonstrated that single-spot urine samples of phthalates could be representative of long-term average levels of exposure. Thus, they could reflect an increased risk of cumulative exposure over time.80–82 Moreover, within-subject temporal variations are unlikely to be large because an individual’s lifestyle does not change significantly over time. Additionally, there might be an issue of healthy worker bias because phthalates exposure might differ in part depending on occupation. However, because most subjects in our study were retired or currently unemployed, their exposure sources were mainly from nonoccupational settings such as leisure or household activities. Thus, our study is unlikely to have a healthy worker bias.
Conclusion
In conclusion, urinary phthalate concentrations were associated with slowness in walking speed, which might reflect decreased functional ability and overall health in adults ages 60–98 y. Our findings add to the growing body of evidence demonstrating phthalate-mediated adverse health effects in the human body and support the need to further reduce the current exposure levels to prevent functional decline and promote healthy aging.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Korea Ministry of Education and the Korea Ministry of Science and Information and Communication Technology (grant numbers 2013R1A6A3A04059556; 2020R1A2C110170311). Also, this study was supported by the Susceptible Population Research Program (2008–2010) from the Korea Ministry of Environment (grant numbers 0411-20080013, 0411-20090007, 0411-20100016). E.G.-E. was supported by the Spanish Consortium for Research on Epidemiology and Public Health (CIBERESP) (ESP21PI04/2021).
The funders had no role in this study design, data collection, and analysis and prepared all results.
All authors participated in literature search and data interpretation. Y.C. supervised the study; Y.C. and Y.H. participated in designing the study; H.K., S.K., and K.K. acquired the data; J.Y., J. Kim, and J. Kwak analyzed data; J.Y. wrote the manuscript; E.G.-E., J. Kim, J. Kwak, H.K., Y.H., and Y.C. critically revised the manuscript.
Patient consent was obtained. This study was approved by the institutional review board (IRB) of Seoul National University Hospital/College of Medicine (IRB No. H-1209-004-424).
==== Refs
References
1. Rydwik E, Bergland A, Forsén L, Frändin K. 2012. Investigation into the reliability and validity of the measurement of elderly people’s clinical walking speed: a systematic review. Physiother Theory Pract 28 (3 ):238–256, PMID: , 10.3109/09593985.2011.601804.21929322
2. Middleton A, Fritz SL, Lusardi M. 2015. Walking speed: the functional vital sign. J Aging Phys Act 23 (2 ):314–322, PMID: , 10.1123/japa.2013-0236.24812254
3. Van Kan GA, Rolland Y, Andrieu S, Bauer J, Beauchet O, Bonnefoy M, et al. 2009. Gait speed at usual pace as a predictor of adverse outcomes in community-dwelling older people an International Academy on Nutrition and Aging (IANA) task force. J Nutr Health Aging 13 (10 ):881–889, PMID: , 10.1007/s12603-009-0246-z.19924348
4. Studenski S, Perera S, Patel K, Rosano C, Faulkner K, Inzitari M, et al. 2011. Gait speed and survival in older adults. JAMA 305 (1 ):50–58, PMID: , 10.1001/jama.2010.1923.21205966
5. Figgins E, Pieruccini-Faria F, Speechley M, Montero-Odasso M. 2021. Potentially modifiable risk factors for slow gait in community-dwelling older adults: a systematic review. Ageing Res Rev 66 :101253, PMID: , 10.1016/j.arr.2020.101253.33429086
6. Ji JS, Elbaz A, Weisskopf MG. 2013. Association between blood lead and walking speed in the National Health and Nutrition Examination Survey (NHANES 1999–2002). Environ Health Perspect 121 (6 ):711–716, PMID: , 10.1289/ehp.1205918.23603014
7. Kim J, Garcia-Esquinas E, Navas-Acien A, Choi Y-H. 2018. Blood and urine cadmium concentrations and walking speed in middle-aged and older U.S. adults. Environ Pollut 232 :97–104, PMID: , 10.1016/j.envpol.2017.09.022.28941716
8. Lang IA, Scarlett A, Guralnik JM, Depledge MH, Melzer D, Galloway TS, et al. 2009. Age-related impairments of mobility associated with cobalt and other heavy metals: data from NHANES 1999–2004. J Toxicol Environ Health A 72 (6 ):402–409, PMID: , 10.1080/15287390802647336.19199147
9. Wittassek M, Koch HM, Angerer J, Brüning T. 2011. Assessing exposure to phthalates–the human biomonitoring approach. Mol Nutr Food Res 55 (1 ):7–31, PMID: , 10.1002/mnfr.201000121.20564479
10. Wang Y, Zhu H, Kannan K. 2019. A review of biomonitoring of phthalate exposures. Toxics 7 (2 ):21, 10.3390/toxics7020021.30959800
11. Kabir ER, Rahman MS, Rahman I. 2015. A review on endocrine disruptors and their possible impacts on human health. Environ Toxicol Pharmacol 40 (1 ):241–258, PMID: , 10.1016/j.etap.2015.06.009.26164742
12. Caldwell JC. 2012. DEHP: genotoxicity and potential carcinogenic mechanisms—a review. Mutat Res/Rev Mutat Res 751 (2 ):82–157, PMID: , 10.1016/j.mrrev.2012.03.001.22484601
13. Radke EG, Braun JM, Nachman RM, Cooper GS. 2020. Phthalate exposure and neurodevelopment: a systematic review and meta-analysis of human epidemiological evidence. Environ Int 137 :105408, PMID: , 10.1016/j.envint.2019.105408.32045779
14. Radke EG, Braun JM, Meeker JD, Cooper GS. 2018. Phthalate exposure and male reproductive outcomes: a systematic review of the human epidemiological evidence. Environ Int 121 (pt 1 ):764–793, PMID: , 10.1016/j.envint.2018.07.029.30336412
15. Radke EG, Glenn BS, Braun JM, Cooper GS. 2019. Phthalate exposure and female reproductive and developmental outcomes: a systematic review of the human epidemiological evidence. Environ Int 130 :104580, PMID: , 10.1016/j.envint.2019.02.003.31351310
16. Hoppin JA, Ulmer R, London SJ. 2004. Phthalate exposure and pulmonary function. Environ Health Perspect 112 (5 ):571–574, PMID: , 10.1289/ehp.6564.15064163
17. Fu X, Xu J, Zhang R, Yu J. 2020. The association between environmental endocrine disruptors and cardiovascular diseases: a systematic review and meta-analysis. Environ Res 187 :109464, PMID: , 10.1016/j.envres.2020.109464.32438096
18. Radke EG, Galizia A, Thayer KA, Cooper GS. 2019. Phthalate exposure and metabolic effects: a systematic review of the human epidemiological evidence. Environ Int 132 :104768, PMID: , 10.1016/j.envint.2019.04.040.31196577
19. Shiue I. 2013. Urinary environmental chemical concentrations and vitamin D are associated with vision, hearing, and balance disorders in the elderly. Environ Int 53 :41–46, PMID: , 10.1016/j.envint.2012.12.006.23314200
20. Kim H, Lee S, Jung Y-I, Hong Y-C. 2021. Association between phthalate exposure and frailty among community-dwelling older adults: a repeated panel data study. Int J Environ Res Public Health 18 (4 ):1985, PMID: , 10.3390/ijerph18041985.33670787
21. Kim K-N, Lee M-R, Choi Y-H, Hwang H, Oh S-Y, Park Chee, et al. 2016. Association between phthalate exposure and lower handgrip strength in an elderly population: a repeated-measures study. Environ Health 15 (1 ):93, PMID: , 10.1186/s12940-016-0176-2.27581612
22. Sun L, Fan J, Song G, Cai S, Fan C, Zhong Y, et al. 2021. Exposure to phthalates is associated with grip strength in US adults. Ecotoxicol Environ Saf 209 :111787, PMID: , 10.1016/j.ecoenv.2020.111787.33333342
23. Min KB, Min JY. 2014. Urinary phthalate metabolites and the risk of low bone mineral density and osteoporosis in older women. J Clin Endocrinol Metab 99 (10 ):E1997–E2003, PMID: , 10.1210/jc.2014-2279.25050905
24. Reeves KW, Vieyra G, Grimes NP, Meliker J, Jackson RD, Wactawski-Wende J, et al. 2021. Urinary phthalate biomarkers and bone mineral density in postmenopausal women. J Clin Endocrinol Metab 106 (7 ):e2567–e2579, PMID: , 10.1210/clinem/dgab189.33754148
25. Kwon J, Suzuki T, Yoshida H, Kim H, Yoshida Y, Iwasa H, et al. 2007. Association between change in bone mineral density and decline in usual walking speed in elderly community‐dwelling Japanese women during 2 years of follow‐up. J Am Geriatr Soc 55 (2 ):240–244, PMID: , 10.1111/j.1532-5415.2007.01066.x.17302661
26. Vermeulen J, Neyens JCL, van Rossum E, Spreeuwenberg MD, de Witte LP. 2011. Predicting ADL disability in community-dwelling elderly people using physical frailty indicators: a systematic review. BMC Geriatr 11 :33, PMID: , 10.1186/1471-2318-11-33.21722355
27. Gray M, Gills JL, Glenn JM, Vincenzo JL, Walter CS, Madero EN, et al. 2021. Cognitive decline negatively impacts physical function. Exp Gerontol 143 :111164, PMID: , 10.1016/j.exger.2020.111164.33232795
28. McGrath R, Vincent BM, Jurivich DA, Hackney KJ, Tomkinson GR, Dahl LJ, et al. 2021. Handgrip strength asymmetry and weakness together are associated with functional disability in aging Americans. J Gerontol A Biol Sci Med Sci 76 (2 ):291–296, PMID: , 10.1093/gerona/glaa100.32319511
29. Min A, Liu F, Yang X, Chen M. 2014. Benzyl butyl phthalate exposure impairs learning and memory and attenuates neurotransmission and CREB phosphorylation in mice. Food Chem Toxicol 71 :81–89, PMID: , 10.1016/j.fct.2014.05.021.24937021
30. Tang J, Yuan Y, Wei C, Liao X, Yuan J, Nanberg E, et al. 2015. Neurobehavioral changes induced by di(2-ethylhexyl) phthalate and the protective effects of vitamin E in Kunming mice. Toxicol Res 4 (4 ):1006–1015, 10.1039/C4TX00250D.
31. Tseng I-L, Yang Y-F, Yu C-W, Li W-H, Liao VH-C. 2013. Phthalates induce neurotoxicity affecting locomotor and thermotactic behaviors and AFD neurons through oxidative stress in Caenorhabditis elegans. PLoS One 8 (12 ):e82657, PMID: , 10.1371/journal.pone.0082657.24349328
32. Tran CM, Do TN, Kim K-T. 2021. Comparative analysis of neurotoxicity of six phthalates in zebrafish embryos. Toxics 9 (1 ):5, PMID: , 10.3390/toxics9010005.33430197
33. Martinelli MI, Mocchiutti NO, Bernal CA. 2006. Dietary di(2-ethylhexyl) phthalate-impaired glucose metabolism in experimental animals. Hum Exp Toxicol 25 (9 ):531–538, PMID: , 10.1191/0960327106het651oa.17017006
34. McKee RH, Pavkov KL, Trimmer GW, Keller LH, Stump DG. 2006. An assessment of the potential developmental and reproductive toxicity of di-isoheptyl phthalate in rodents. Reprod Toxicol 21 (3 ):241–252, PMID: , 10.1016/j.reprotox.2005.09.002.16249068
35. Saillenfait AM, Gallissot F, Sabaté JP. 2009. Differential developmental toxicities of di‐n‐hexyl phthalate and dicyclohexyl phthalate administered orally to rats. J Appl Toxicol 29 (6 ):510–521, PMID: , 10.1002/jat.1436.19391110
36. Ferguson KK, Loch-Caruso R, Meeker JD. 2011. Urinary phthalate metabolites in relation to biomarkers of inflammation and oxidative stress: NHANES 1999–2006. Environ Res 111 (5 ):718–726, PMID: , 10.1016/j.envres.2011.02.002.21349512
37. Ferguson KK, Cantonwine DE, Rivera-González LO, Loch-Caruso R, Mukherjee B, Anzalota Del Toro LV, et al. 2014. Urinary phthalate metabolite associations with biomarkers of inflammation and oxidative stress across pregnancy in Puerto Rico. Environ Sci Technol 48 (12 ):7018–7025, PMID: , 10.1021/es502076j.24845688
38. Cesari M, Penninx BWJH, Pahor M, Lauretani F, Corsi AM, Williams GR, et al. 2004. Inflammatory markers and physical performance in older persons: the InCHIANTI study. J Gerontol A Biol Sci Med Sci 59 (3 ):M242–M248, PMID: , 10.1093/gerona/59.3.M242.24845688
39. Penninx BWJH, Kritchevsky SB, Newman AB, Nicklas BJ, Simonsick EM, Rubin S, et al. 2004. Inflammatory markers and incident mobility limitation in the elderly. J Am Geriatr Soc 52 (7 ):1105–1113, PMID: .15209648
40. Verghese J, Holtzer R, Oh-Park M, Derby CA, Lipton RB, Wang C. 2011. Inflammatory markers and gait speed decline in older adults. J Gerontol A Biol Sci Med Sci 66 (10 ):1083–1089, PMID: , 10.1093/gerona/glr099.21719612
41. Crimmins E, Guyer HM, Langa KM, Ofstedal MB, Wallace RB, Weir DR, et al. 2008. Documentation of Physical Measures, Anthropometrics and Blood Pressure in the Health and Retirement Study. HRS Documentation Report DR-011. Ann Arbor, MI: Survey Research Center University of Michigan.
42. Sonnega A, Faul JD, Ofstedal MB, Langa KM, Phillips JW, Weir DR. 2014. Cohort profile: the Health and Retirement Study (HRS). Int J Epidemiol 43 (2 ):576–585, PMID: , 10.1093/ije/dyu067.24671021
43. Chen LK, Woo J, Assantachai P, Auyeung TW, Chou MY, Iijima K, et al. 2020. Asian Working Group for Sarcopenia: 2019 consensus update on sarcopenia diagnosis and treatment. J Am Med Dir Assoc 21 (3 ):300–307, PMID: , 10.1016/j.jamda.2019.12.012.32033882
44. Chen L-K, Liu LK, Woo J, Assantachai P, Auyeung TW, Bahyah KS, et al. 2014. Sarcopenia in Asia: consensus report of the Asian Working Group for Sarcopenia. J Am Med Dir Assoc 15 (2 ):95–101, PMID: , 10.1016/j.jamda.2013.11.025.24461239
45. Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, et al. 2001. Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci 56 (3 ):M146–M157, PMID: , 10.1093/gerona/56.3.m146.11253156
46. Wang M-C, Li TC, Li CI, Liu CS, Lin WY, Lin CH, et al. 2019. Frailty, transition in frailty status and all-cause mortality in older adults of a Taichung community-based population. BMC Geriatr 19 (1 ):1–8, PMID: , 10.1186/s12877-019-1039-9.30606112
47. Meyer M-L, Fustinoni S, Henchoz Y, Hottinger AF, Santos-Eggimann B. 2021. Slowness predicts mortality: a comparative analysis of walking speed and Moberg picking-up tests. J Am Med Dir Assoc 22 (8 ):1652–1657.e2, PMID: , 10.1016/j.jamda.2021.02.028.33785308
48. Fiser WM, Hays NP, Rogers SC, Kajkenova O, Williams AE, Evans CM, et al. 2010. Energetics of walking in elderly people: factors related to gait speed. J Gerontol A Biol Sci Med Sci 65 (12 ):1332–1337, PMID: , 10.1093/gerona/glq137.20679072
49. Chun MY. 2012. Validity and reliability of Korean version of international physical activity questionnaire short form in the elderly. Korean J Fam Med 33 (3 ):144, PMID: , 10.4082/kjfm.2012.33.3.144.22787536
50. Bull FC, Al-Ansari SS, Biddle S, Borodulin K, Buman MP, Cardon G, et al. 2020. World Health Organization 2020 guidelines on physical activity and sedentary behaviour. Br J Sports Med 54 (24 ):1451–1462, PMID: , 10.1136/bjsports-2020-102955.33239350
51. Bae JN, Cho MJ. 2004. Development of the Korean version of the Geriatric Depression Scale and its short form among elderly psychiatric patients. J Psychosomatic Res 57 (3 ):297–305, PMID: , 10.1016/j.jpsychores.2004.01.004.15507257
52. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16 (3 ):493–508, PMID: , 10.1093/biostatistics/kxu058.25532525
53. Bobb JF, Claus Henn B, Valeri L, Coull BA. 2018. Statistical software for analyzing the health effects of multiple concurrent exposures via Bayesian kernel machine regression. Environ Health 17 (1 ):67, PMID: , 10.1186/s12940-018-0413-y.30126431
54. Barbieri MM, Berger JO. 2004. Optimal predictive model selection. Ann Stat 32 (3 ):870–897, 10.1214/009053604000000238.
55. Domingo-Relloso A, Grau-Perez M, Briongos-Figuero L, Gomez-Ariza JL, Garcia-Barrera T, Dueñas-Laita A, et al. 2019. The association of urine metals and metal mixtures with cardiovascular incidence in an adult population from Spain: the Hortega Follow-Up Study. Int J Epidemiol 48 (6 ):1839–1849, PMID: , 10.1093/ije/dyz061.31329884
56. Barr DB, Wilder LC, Caudill SP, Gonzalez AJ, Needham LL, Pirkle JL. 2005. Urinary creatinine concentrations in the US population: implications for urinary biologic monitoring measurements. Environ Health Perspect 113 (2 ):192–200, PMID: , 10.1289/ehp.7337.15687057
57. Wang J, Geng L. 2019. Effects of socioeconomic status on physical and psychological health: lifestyle as a mediator. Int J Environ Res Public Health 16 (2 ):281, PMID: , 10.3390/ijerph16020281.30669511
58. Lee K-S, Lim YH, Kim KN, Choi YH, Hong YC, Lee N. 2018. Urinary phthalate metabolites concentrations and symptoms of depression in an elderly population. Sci Total Environ 625 :1191–1197, PMID: , 10.1016/j.scitotenv.2017.12.219.29996415
59. Shimada H, Doi T, Lee S, Tsutsumimoto K, Bae S, Makino K, et al. 2021. Identification of disability risk in addition to slow walking speed in older adults. Gerontology 68 (6 ):625–634, PMID: , 10.1159/000516966.34261066
60. García-Esquinas E, Rodríguez-Artalejo F. 2017. Environmental pollutants, limitations in physical functioning, and frailty in older adults. Curr Environ Health Rep 4 (1 ):12–20, PMID: , 10.1007/s40572-017-0128-1.28091981
61. Ferrucci L, Bandinelli S, Benvenuti E, Di Iorio A, Macchi C, Harris TB, et al. 2000. Subsystems contributing to the decline in ability to walk: bridging the gap between epidemiology and geriatric practice in the InCHIANTI study. J Am Geriatr Soc 48 (12 ):1618–1625, PMID: , 10.1111/j.1532-5415.2000.tb03873.x.11129752
62. Shiue I. 2015. Arsenic, heavy metals, phthalates, pesticides, hydrocarbons and polyfluorinated compounds but not parabens or phenols are associated with adult remembering condition: US NHANES, 2011–2012. Environ Sci Pollut Res Int 22 (8 ):6381–6386, PMID: , 10.1007/s11356-015-4261-9.25744817
63. Mariana M, Cairrao E. 2020. Phthalates implications in the cardiovascular system. J Cardiovasc Develop Dis 7 (3 ):26, PMID: , 10.3390/jcdd7030026.32707888
64. Bouillon K, Batty GD, Hamer M, Sabia S, Shipley MJ, Britton A, et al. 2013. Cardiovascular disease risk scores in identifying future frailty: the Whitehall II prospective cohort study. Heart 99 (10 ):737–742, PMID: , 10.1136/heartjnl-2012-302922.23503403
65. Dumurgier J, Elbaz A, Dufouil C, Tavernier B, Tzourio C. 2010. Hypertension and lower walking speed in the elderly: the Three-City Study. J Hypertens 28 (7 ):1506–1514, PMID: , 10.1097/HJH.0b013e328338bbec.20404744
66. Windham BG, Wilkening SR, Lirette ST, Kullo IJ, Turner ST, Griswold ME, et al. 2016. Associations between inflammation and physical function in African Americans and European Americans with prevalent cardiovascular risk factors. J Am Geriatr Soc 64 (7 ):1448–1455, PMID: , 10.1111/jgs.14229.27310030
67. Semba RD, Ferrucci L, Sun K, Walston J, Varadhan R, Guralnik JM, et al. 2007. Oxidative stress and severe walking disability among older women. Am J Med 120 (12 ):1084–1089, PMID: , 10.1016/j.amjmed.2007.07.028.18060930
68. Baptista G, Dupuy A-M, Jaussent A, Durant R, Ventura E, Sauguet P, et al. 2012. Low-grade chronic inflammation and superoxide anion production by NADPH oxidase are the main determinants of physical frailty in older adults. Free Radic Res 46 (9 ):1108–1114, PMID: , 10.3109/10715762.2012.692784.22640231
69. Gemma C, Vila J, Bachstetter A, Bickford PC. 2007. Chapter 15: Oxidative stress and the aging brain: from theory to prevention. In: Brain Aging: Models, Methods, and Mechanisms. Riddle DR, ed. Boca Raton, FL: CRC Press/Taylor & Francis.
70. Bility MT, Thompson JT, McKee RH, David RM, Butala JH, Vanden Heuvel JP, et al. 2004. Activation of mouse and human peroxisome proliferator-activated receptors (PPARs) by phthalate monoesters. Toxicol Sci 82 (1 ):170–182, PMID: , 10.1093/toxsci/kfh253.15310864
71. Hurst CH, Waxman DJ. 2003. Activation of PPARα and PPARγ by environmental phthalate monoesters. Toxicol Sci 74 (2 ):297–308, PMID: , 10.1093/toxsci/kfg145.12805656
72. Latini G, Scoditti E, Verrotti A, De Felice C, Massaro M. 2008. Peroxisome proliferator-activated receptors as mediators of phthalate-induced effects in the male and female reproductive tract: epidemiological and experimental evidence. PPAR Res 2008 :359267, PMID: , 10.1155/2008/359267.18288285
73. Lapinskas PJ, Brown S, Leesnitzer LM, Blanchard S, Swanson C, Cattley RC, et al. 2005. Role of PPARα in mediating the effects of phthalates and metabolites in the liver. Toxicology 207 (1 ):149–163, PMID: , 10.1016/j.tox.2004.09.008.15590130
74. Miodovnik A, Edwards A, Bellinger DC, Hauser R. 2014. Developmental neurotoxicity of ortho-phthalate diesters: review of human and experimental evidence. Neurotoxicology 41 :112–122, PMID: , 10.1016/j.neuro.2014.01.007.24486776
75. Muscogiuri G, Colao A. 2017. Phthalates: new cardiovascular health disruptors? Arch Toxicol 91 (3 ):1513–1517, PMID: , 10.1007/s00204-016-1780-1.27358237
76. Lin C-H, Chen T-J, Chen S-S, Hsiao P-C, Yang R-C. 2011. Activation of Trim17 by PPARγ is involved in di(2-ethylhexyl) phthalate (DEHP)-induced apoptosis on neuro-2a cells. Toxicol Lett 206 (3 ):245–251, PMID: , 10.1016/j.toxlet.2011.08.002.21856391
77. Park G-H, Nam C, Hong S, Park B, Kim H, Lee T, et al. 2018. Socioeconomic factors influencing cosmetic usage patterns. J Expo Sci Environ Epidemiol 28 (3 ):242–250, PMID: , 10.1038/jes.2017.20.29666511
78. Silva MJ, Barr DB, Reidy JA, Malek NA, Hodge CC, Caudill SP, et al. 2004. Urinary levels of seven phthalate metabolites in the U.S. population from the National Health and Nutrition Examination Survey (NHANES) 1999–2000. Environ Health Perspect 112 (3 ):331–338, PMID: , 10.1289/ehp.6723.14998749
79. Benjamin S, Masai E, Kamimura N, Takahashi K, Anderson RC, Faisal PA, et al. 2017. Phthalates impact human health: epidemiological evidences and plausible mechanism of action. J Hazard Mater 340 :360–383, PMID: , 10.1016/j.jhazmat.2017.06.036.28800814
80. Hauser R, Meeker JD, Park S, Silva MJ, Calafat AM. 2004. Temporal variability of urinary phthalate metabolite levels in men of reproductive age. Environ Health Perspect 112 (17 ):1734–1740, PMID: , 10.1289/ehp.7212.15579421
81. Townsend MK, Franke AA, Li X, Hu FB, Eliassen AH. 2013. Within-person reproducibility of urinary bisphenol A and phthalate metabolites over a 1 to 3 year period among women in the Nurses’ Health Studies: a prospective cohort study. Environ Health 12 (1 ):80, PMID: , 10.1186/1476-069X-12-80.24034517
82. Dewalque L, Pirard C, Vandepaer S, Charlier C. 2015. Temporal variability of urinary concentrations of phthalate metabolites, parabens and benzophenone-3 in a Belgian adult population. Environ Res 142 :414–423, PMID: , 10.1016/j.envres.2015.07.015.26233661
|
PMC010xxxxxx/PMC10078803.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37022725
EHP12596
10.1289/EHP12596
Invited Perspective
Invited Perspective: Mixtures—Are They Worth the Risk (Assessment)?
Quiros-Alcala Lesliam 1
https://orcid.org/0000-0002-5566-2138
Barr Dana Boyd 2
1 Department of Environmental Health and Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA
2 Gangarosa Department of Environmental Health, Rollins School of Public Health, Emory University, Atlanta, Georgia, USA
Address correspondence to Dana Boyd Barr, 1518 Clifton Rd., CNR7007, Emory University, Rollins School of Public Health, Gangarosa Department of Environmental Health, Atlanta, GA 30322 USA. Email: [email protected]
6 4 2023
4 2023
131 4 04130114 12 2022
21 2 2023
28 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have no real or apparent conflicts of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP11899
==== Body
pmcHistorically, environmental health research has focused on studying exposure to a single chemical or chemical class and its relation to a single health outcome. In reality, exposures do not occur alone but involve a complex milieu of chemicals that may result in one or more health risks. With the advent of exposomic techniques for measuring a wide array of chemicals simultaneously1–4 and mixtures analysis techniques for evaluating the contributions of multiple chemicals to health outcomes,5–7 a more comprehensive and holistic evaluation of exposure and disease is now possible. Given this realization, it is not unreasonable to assume our current risk assessment techniques evaluating risk on a single chemical basis is outdated and likely an inaccurate representation of the true risk from complex chemical exposures.
In the commentary by Savitz and Hattersley,8 the authors address a complex issue in exposure science and epidemiology—mixtures—and how data on mixtures can be used to inform and maximize their usefulness in regulatory decision-making. As more epidemiologic studies integrate mixtures analyses using advanced statistical approaches, such as quantile g-computation,7 Bayesian kernel machine regression,5 or weighted quantile sum regression,6 data on individual chemical effects are often relegated to less critical findings of these studies. To update the risk assessment paradigm, Savitz and Hattersley present a framework for decision-making for evaluating chemical mixtures.8
The proposed framework suggests common mixture groupings based on product or exposure sources or common modes of action or effects. The authors examine the advantages and disadvantages of conducting studies on mixtures for advancing knowledge to inform policies, and they offer a strategy for epidemiologists and regulators to use in considering when and how to assess chemical mixtures. The authors conclude that conventional methods for assessing individual effects of chemicals remain preferable in certain situations. However, if the complexity and loss of generalizability that may occur when considering mixtures in a risk assessment are justified by dramatic improvements in the assessment, a mixtures approach may be warranted, especially if it is hypothesis-driven rather than data-driven exploration.
As the authors indicate, a critical need still exists to examine the individual effects of single chemicals to help discern the constituents of exposure mixtures that could be the “bad actors” (or main drivers associated with health end points) and to better inform further mixtures analyses. However, this approach may be inadequate given that the total risk of multiple chemicals may not be the sum of their individual risks. In fact, the few instances where risk has been evaluated with simple mixtures, the combination of chemicals appear to synergistically increase or attenuate associated end points.9 Individual chemicals may behave differently when present in a complex mixtures, which could enhance exposure, uptake, or intake or alter distribution, metabolism, elimination, or internal biological activity. These changes would not be captured by simply adding risk. In these situations, exposomics or data-driven techniques may be useful in identifying common metabolic pathways affected by observed chemical mixtures in individuals, especially when repeated temporal samples are measured. Many mediating pathways, such as oxidative stress, inflammation, and protein function, have been identified as common targets for many environmental chemicals that may ultimately drive potential adverse outcomes.
To illustrate, neurodegenerative diseases, such as Parkinson’s and Alzheimer’s diseases, are believed to have common pathogenic pathways, such as generation of reactive oxygen species (ROS), oxidative stress, or altered protein structures especially with misfolding, faulty degradation, DNA damage/mutations, mitochondrial dysfunctions, and neuroinflammatory processes.10 Different chemicals, such as rotenone and dieldrin, may cause epigenetic changes; manganese may alter protein folding; vanadium may produce excess ROS; and 1,1,1-trichloro-2,2-bis(p-chlorophenyl)ethane (DDT) may induce oxidative stress. However, they all synergistically work to alter mitochondrial performance that may lead to disease development or exacerbation.10 Exposomics may be able to identify chemicals that may contribute to these alterations and may further impact disease development. In ideal practice, all of the chemical contributors would be considered when evaluating risk.
Adding to the complexity of using mixtures analysis in risk assessment, Savitz and Hattersley highlight that significant spatial and temporal variation in exposures occur. Current exposure assessment methods may not adequately capture all exposures to all subpopulations, limiting the generalizability of these findings for risk assessment. For example, a mixtures approach could be effective in a worker population experiencing similar primary exposures, such as hairdressers who may come into contact with hair products containing similar classes of chemicals. Still, such results may not apply to the general population. Regardless, analyses of multiple serial samples collected across broad regions or populations may enable scientists to overcome this limitation.
Consideration of chemicals individually for risk assessments also has the advantage of potentially identifying “regrettable substitutions” that may otherwise go unnoticed. For example, di-2-ethylhexyl phthalate has been linked to adverse respiratory effects, including increased risk of asthma morbidity, as has its replacement product, di-2-ethylhexyl terephthalate.11 But realistically, these substitutions would be present in mixtures that may have synergistic biological effects as well.
Individual chemical risk assessment will always be necessary to fully characterize a given chemical’s harm. However, it is important to consider mixtures in risk assessment because they are more representative of real-world exposures and likely work collectively to induce a disease state. Application of advanced statistical methods and exposomic techniques for mixtures will continue to be useful and should be fully considered in the risk assessment process. Savitz and Hattersley have thrown down the gauntlet, and it is time for U.S. regulators to take up the challenge.
==== Refs
References
1. Barouki R, Audouze K, Becker C, Blaha L, Coumoul X, Karakitsios S, et al. 2022. The exposome and toxicology: a win–win collaboration. Toxicol Sci 186 (1 ):1–11, PMID: , 10.1093/toxsci/kfab149.34878125
2. Kalia V, Belsky DW, Baccarelli AA, Miller GW. 2022. An exposomic framework to uncover environmental drivers of aging. Exposome 2 (1 ):osac002, PMID: , 10.1093/exposome/osac002.35295547
3. Price EJ, Vitale CM, Miller GW, David A, Barouki R, Audouze K, et al. 2022. Merging the exposome into an integrated framework for “omics” sciences. iScience 25 (3 ):103976, PMID: , 10.1016/j.isci.2022.103976.35310334
4. Vermeulen R, Schymanski EL, Barabási AL, Miller GW. 2020. The exposome and health: where chemistry meets biology. Science 367 (6476 ):392–396, PMID: , 10.1126/science.aay3164.31974245
5. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16 (3 ):493–508, PMID: , 10.1093/biostatistics/kxu058.25532525
6. Carrico C, Gennings C, Wheeler DC, Factor-Litvak P. 2015. Characterization of weighted quantile sum regression for highly correlated data in a risk analysis setting. J Agric Biol Environ Stat 20 (1 ):100–120, PMID: , 10.1007/s13253-014-0180-3.30505142
7. Schmidt S. 2020. Quantile g-computation: a new method for analyzing mixtures of environmental exposures. Environ Health Perspect 128 (10 ):104004, PMID: , 10.1289/EHP7342.33074735
8. Savitz DA, Hattersley AM. 2023. Evaluating chemical mixtures in epidemiological studies to inform regulatory decisions. Environ Health Perspect 131 (4 ):045001, 10.1289/EHP11899.37022726
9. Martin O, Scholze M, Ermler S, McPhie J, Bopp SK, Kienzler A, et al. 2021. Ten years of research on synergisms and antagonisms in chemical mixtures: a systematic review and quantitative reappraisal of mixture studies. Environ Int 146 :106206, PMID: , 10.1016/j.envint.2020.106206.33120228
10. Huang M, Bargues-Carot A, Riaz Z, Wickham H, Zenitsky G, Jin H, et al. 2022. Impact of environmental risk factors on mitochondrial dysfunction, neuroinflammation, protein misfolding, and oxidative stress in the etiopathogenesis of Parkinson’s disease. Int J Mol Sci 23 (18 ):10808, PMID: , 10.3390/ijms231810808.36142718
11. Fandiño-Del-Rio M, Matsui EC, Peng RD, Meeker JD, Quirós-Alcalá L. 2022. Phthalate biomarkers and associations with respiratory symptoms and healthcare utilization among low-income urban children with asthma. Environ Res 212 (pt B ):113239, PMID: , 10.1016/j.envres.2022.113239.35405131
|
PMC010xxxxxx/PMC10078806.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37022726
EHP11899
10.1289/EHP11899
Commentary
Evaluating Chemical Mixtures in Epidemiological Studies to Inform Regulatory Decisions
Savitz David A. 1
Hattersley Anne M. 2
1 Department of Epidemiology, Brown University School of Public Health, Brown University, Providence, Rhode Island, USA
2 Global Safety Surveillance and Analysis, Procter & Gamble, Cincinnati, Ohio, USA
Address correspondence to Anne M. Hattersley, Procter & Gamble, MBC–South Building, 8700 Mason Montgomery Rd., Mason, OH 45040 USA. Email: [email protected]
6 4 2023
4 2023
131 4 04500122 7 2022
24 1 2023
28 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Epidemiological studies are increasingly going beyond the evaluation of health effects of individual chemicals to consider chemical mixtures. To our knowledge, the advantages and disadvantages of addressing chemical mixtures for informing regulatory decisions—as opposed to obtaining a more comprehensive understanding of etiology—has not been carefully considered.
Objectives:
We offer a framework for the study of chemical mixtures in epidemiological research intended to inform regulatory decisions. We identify a) the different ways mixtures originate (product source, pollution source, shared mode of action, or shared effect on health outcome), b) the use of indicator chemicals to address mixtures, and c) the requirements for epidemiological studies to be informative for regulatory purposes.
Discussion:
The principal advantage of considering mixtures is to obtain a more complete understanding of the role of the chemical environment as a determinant of health. Incorporating other exposures may improve the assessment of the net effect of the chemicals of interest. However, the increased complexity and potential loss of generalizability may limit the value of studies of mixtures, especially for mixtures based on mode of action or shared health outcomes. Our recommended strategy is to successively assess the marginal contribution of individual chemicals, joint effects with other specific chemicals, and hypothesis-driven evaluation of mixtures rather than applying hypothesis-free data exploration methods. Although more ambitious statistical approaches to mixtures may, in time, be helpful for guiding regulation, the authors believe conventional methods for assessing individual and combined effects of chemicals remain preferable. https://doi.org/10.1289/EHP11899
D.A.S. was supported as a paid consultant to Procter & Gamble, and A.M.H. is employed at Procter & Gamble.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Epidemiological research on health effects of environmental chemicals often seeks to inform regulatory decisions. When these studies focus on the health effects of single chemicals in isolation, it has become routine in our experience for epidemiologists to ask, “But what about mixtures?” The growing popularity of this approach is clear from a simple PubMed search on “chemical mixtures epidemiology” that identified <10 published papers per year in the 1990s, rising steadily to 20–30 per year in the 2000s, 30–75 per year in the 2010s, and >100 papers per year in 2020–2022. In our opinion, the question is not whether chemical mixtures are present (they are), whether chemicals may act jointly to influence health (they might), or whether interventions would affect multiple agents (they often do), but whether it is useful to conceptualize a set of chemicals as a mixture for the purposes of generating research that informs regulatory decisions. The mere existence of chemical mixtures is not a sufficient justification to incorporate them into studies.
Although recent literature highlights statistical methods for addressing chemical mixtures,1,2 studying chemical mixtures is not new in epidemiology.3,4 The effort to understand the net impact of the environment on health has resulted in innovative methods intended to incorporate more environmental complexity into epidemiological studies.5–7 However, in our view, these assessments have not considered fully the most effective approaches for ensuring these studies can inform regulation. A focus on applicability to regulation may lead to different methods than the more general goal of fully understanding disease etiology. Acknowledging the need for regulation to protect even the most susceptible segments of the population, it may be impractical to take susceptibility into account because it is determined in part by the presence of other chemicals. Each individual and each population is exposed to somewhat different exposure mixtures, but regulation requires research that is applicable to the entire population.
In this commentary, we offer our rationale for conceptualizing and studying sets of chemicals as a mixture in epidemiological studies. We also explore the advantages and disadvantages of addressing mixtures for advancing knowledge to inform policy and recommend a strategy for epidemiologists to consider regarding whether and how to evaluate mixtures. We do not review quantitative risk assessment or toxicologic literature—where the study of mixtures has generated a substantial body of research8,9—or provide detailed examples of specific mixtures that have been addressed in the literature. Instead, we focus on making epidemiological research more effective and impactful for informing regulatory decisions. Ultimate decisions will undoubtedly incorporate epidemiology, toxicology, exposure science, and chemical risk assessment, but the aim here is to maximize the utility of epidemiological research for this purpose.
Discussion
Defining Chemical Mixtures
At its most basic level, we define a chemical mixture as simply a set of individual chemicals that have been designated as a combined entity for some purpose, e.g., for assessing health effects, understanding biological processes, or mitigating exposure or health risks. The utility of studying chemical mixtures depends in part on the reasons that the combination of chemicals is conceptualized as a mixture. We consider four general types of chemical mixtures (Table 1).
Table 1 Four types of chemical mixtures with rationale, examples, and applications to epidemiological studies.
Chemical mixture type Rationale for defining as a mixture Examples Application to epidemiological studies of mixtures
Product source Multiple ingredients formulated for function Pesticides
Cosmetics
Cleaning products
Use of product; indicator chemical
Pollution source Generation of multiple by-products by industrial process Motor vehicle exhaust
Chlorinated drinking water
Exposure to source; indicator chemical
Shared mode of action Shared biological influence Endocrine disruptors
Chemicals resulting in oxidative stress, genotoxicity, or inflammation
Biomarkers
Shared effect on health outcome Impact on subclinical or clinical health effects Liver toxicity
Impaired immune function
Clinical health outcomes (e.g., adiposity, neurobehavioral symptoms, specific cancers)
Integrated impact of chemical mixture on health end point
Product source.
Some chemical mixtures are formulated in a laboratory or a manufacturing facility. Examples include pesticide compounds and consumer products such as cosmetics or cleaning agents. Use of the product automatically results in exposure to its constituent chemicals. Studies evaluating health effects of the product are thus inherently addressing a chemical mixture, even if the interest is in only one component of the mixture [e.g., per- and polyfluoroalkyl substances (PFAS) exposure due to use of firefighting foam]. The investigator may not explicitly state or intend the goal of the research to be studying mixtures but instead may present the study as research on the health impact of using the product. Exposure assessment is limited to the use of the product without the ability to isolate the individual components of the mixture. However, using biomarkers or toxicological evidence may allow identification of the likely contributors to adverse health effects.
Pollution source.
Many industrial processes generate a chemical mixture, not just a single chemical of health concern. For example, internal combustion engines produce a complex chemical mixture in the form of motor vehicle exhaust, with several of these “tailpipe emissions” subject to regulation.10 Chlorination of drinking water likewise generates hundreds of chemicals that are present in the finished water.11 Furthermore, some foods may result in exposure to a chemical mixture of health relevance, such as certain fish that contain mercury, polychlorinated biphenyls (PCBs), and fatty acids. For research and regulation, investigators may address pollution sources of chemical mixtures in the aggregate (e.g., “fatty fish consumption” or “proximity to roadways”) or on specific chemicals within the mixture (e.g., “dietary mercury” or “ambient levels of particulate matter”).
Shared mode of action.
A chemical mixture may be defined based on a shared mode of action, often informed by toxicological or mechanistic research. Concern with adverse health effects of specific chemicals or products may arise based on that mode of action and motivate epidemiological studies of potential health effects. Chemicals may interact in a variety of ways, including antagonism that mitigates individual chemical effects. However, the shared mode of action of regulatory concern generally involves joint effects such that the aggregate impact is greater than that of any one constituent chemical. In epidemiology, endocrine disruption is one of the most commonly considered modes of action, with the definition of chemical mixture based solely on shared effects on the endocrine system (i.e., a category of chemicals identified as a chemical mixture).12 Endocrine disruption can be an unintended effect of fire retardants, food packaging, food, or cosmetics. Other examples of shared mode of action are chemicals that result in oxidative stress, mutations, inflammation, or genotoxicity. Exposure to specific chemicals within the class may be assessed directly based on self-report or biomarkers. Assessment of the net biological effect (e.g., oxidative stress, inflammation) would not allow attribution to any one of the multiple agents that contribute. Because multiple chemicals from a wide range of sources contribute to the biological pathway of concern, it is difficult to link individual agents to the aggregate impact to guide regulatory decisions.
Shared effect on health outcome.
Related to the shared mode of action are chemicals that affect the same health end point. For instance, different chemicals may each result in liver toxicity or impaired immune function, possibly through different modes of action. This phenomenon can also apply to multiple chemicals that affect clinical health outcomes, such as adiposity (“obesogens”), neurobehavioral development, or specific cancers. These sets of chemicals are defined by their impact on subclinical or clinical health outcomes, without regard to source or mode of action. Comparable to categories of chemicals considered as chemical mixtures defined by a shared mode of action, mixtures of chemicals with similar health consequences may determine the aggregate impact of multiple agents. Although multiple chemicals contributing to the same pathway enhance the concern with adverse health effects, it is unclear how to incorporate shared contributions as members of a chemical class to regulate “endocrine disruptors” or “liver toxicants,” for example.
Addressing Mixtures with Indicator Chemicals
Frequently, research and regulation simplify complex mixtures through the selection of indicator chemicals (i.e., individual chemical constituents) from among the components of that mixture. This approach has been used to regulate drinking water disinfection by-products, a group of hundreds of chemicals formed when organic matter in source water is treated with chlorine to control microbial contamination.11 Rather than a complex summation or weighting of these chemicals, total trihalomethanes (four chemicals) and haloacetic acids (five chemicals) are monitored as indicators in the United States.11 Similarly, despite hundreds of different chemicals emitted by motor vehicles, regulation focuses on a subset of chemicals that are presumed to effectively represent health impact of the entire mixture.10 This approach is sometimes used with product mixtures in which the level of a particular component is studied and regulated on the presumption that it is the dominant threat to health (e.g., PFAS exposure due to use of firefighting foam).
The choice of which chemicals in a mixture are selected as indicators for research and regulation may be based on a variety of considerations. The focus can be on those found in the greatest concentration. There also can be a basis for zeroing in on the most toxic or persistent chemical in the mixture based on toxicology, exposure science, or other research, with the presumption that if the most concerning chemical is managed, the impact of the entire mixture will be effectively controlled as well. There is often uncertainty about whether monitoring and restricting the indicator chemical(s) are sufficient to reduce or eliminate health harm from the mixture. On the other hand, the simplicity and ease of monitoring make the use of indicator chemical(s) attractive for research and regulation.
Contributions of Epidemiology to Regulation
Epidemiologists conduct health studies for a variety of reasons, including understanding disease etiology and identifying risk factors. Much of the current interest in mixtures within epidemiology is motivated by the desire for a more complete understanding of the effect of the chemical environment on health outcomes, i.e., etiological research. For regulatory purposes, epidemiological studies not only estimate the health effects resulting from the chemical exposure of concern but also predict the consequences of reducing or eliminating that exposure while considering risks and benefits. It is essential that the observed effects be causal for changes in exposure to yield the desired health benefits (i.e., reducing exposure causes improved health). Research seeking to inform regulation may not, however, be fully aligned with the goal of attaining a comprehensive understanding of how environmental exposures affect human health. Quantifying the incremental health burden of the chemical agent to inform regulation does not necessarily need to contend with the many other influences on the health outcome, including other chemicals. As long as the identified association between chemical exposure and disease is causal, we believe that it will provide an estimate of the benefit from reducing exposure.
Studying biomarkers or subclinical effects may be useful in supporting or refining our understanding of disease etiology, but such indirect indicators of health effects are not the essential outcomes of greatest concern. Regulatory agencies do not generally regulate chemicals solely because they cause inflammation or oxidative stress without some basis for connecting these biological changes to health effects. From our perspective, clinically consequential health outcomes ultimately drive regulatory decisions.
Advantages and Disadvantages of Addressing Mixtures in Epidemiological Studies
When planning epidemiological studies to generate evidence to inform policy decisions, there are important considerations that argue in favor of evaluating chemical mixtures as aggregate entities rather than studying individual chemicals in isolation. Even if the research is intended to inform regulatory policy on single chemicals, failure to address the chemical mixture may produce misleading results by neglecting to identify adverse effects from coexposures within the mixture. When multiple chemicals act in a similar manner, as is thought to be the case for specific forms of PFAS,13 for example, ignoring associated chemicals and assessing the chemical in isolation may result in an inaccurate assessment of a given chemical’s incremental impact. For instance, if there is a threshold for adverse effects or the dose–response function is steeper in the higher dose range, then treating one chemical as though it were the only agent could fail to identify effects that result from that chemical in the presence of other similarly acting chemicals. When faced with the daunting task of regulating hundreds or thousands of chemicals that fall within the same class, we believe some means of considering the array of chemicals will be required on practical grounds.
There are also disadvantages to treating chemicals as a mixture in epidemiological studies. We believe that addressing multiple chemicals adds complexity to the study methods, results, and interpretation that makes it difficult to discern effects for any given agent and thus reduces the study’s value for guiding regulatory decisions. At the extreme, when investigators analyze a large set of chemicals in relation to one another (the “exposome”), the necessary statistical methods become increasingly challenging to conduct, replicate, and interpret. Results from studies of more comprehensive arrays of exposure generally do not provide dose–response information for any individual chemical. Furthermore, the constitution of the mixture will vary across settings depending on the sources and environmental conditions present, calling into question the generalizability of findings, which is essential for the results to be applicable to regulation. For instance, particulate air pollution in North America and western Europe may be similar enough to extrapolate findings. These same findings, however, may not be applicable to low-income regions where the particulate air pollution sources may be quite different. Although research that incorporates particular features of the specific study setting may more effectively reflect real-world complexity, it may be unable to distill and interpret that complexity in a manner that supports evidence-based decisions.
The balance of risks and benefits in studying mixtures may depend on the type of mixture (Table 1). In some cases, mixtures based on the product source might be regulated in the aggregate and thus be informed by epidemiological studies of the aggregate exposure. For example, a specific pesticide formulation or consumer product could be addressed as a unified entity and possibly removed from the market if found to be harmful. Even then, there would be an interest in identifying the harmful constituent chemical(s) to assess other sources of exposure to that agent and to consider modifying the product (e.g., by substituting other chemicals for the harmful chemical) to mitigate health risk. Similarly, aggregate pollutant mixtures are often regulated by limiting the most important contributor(s) with the intention that if the indicator chemical were controlled, other components of the pollution mixture would also be controlled.
Studies of chemical mixtures defined by mode of action or shared effects on the same health outcome are very difficult to connect to regulatory decisions. For example, we contend that it is infeasible to limit the level of aggregate exposure to endocrine-disrupting chemicals or oxidative stressors rather than targeting specific contributors to those pathways. Likewise, regulatory agencies cannot conceptualize regulating liver toxicants as an entity as opposed to individual contributors.
Recommended Strategy for Epidemiological Research on Mixtures
We suggest that specific requirements for informative studies of the health effects of chemical exposures should be tailored to the issue of concern. To maximize the value of epidemiological studies to inform regulation, consideration of the benefits and risks of studying mixtures offers some guidance. The range of options extends from examining a single chemical agent to assessing the exposome, with its hundreds or thousands of chemical exposures and other influences on disease risk, including nutrition, infection, and stress. For application to regulatory decisions, there is a trade-off between simplicity that results in ease of application and more comprehensive but increasingly complex chemical assessments. To quote Albert Einstein, “Everything should be made as simple as possible, but not simpler.”
In this commentary, we focus on whether to treat multiple chemicals as a single entity (i.e., a chemical mixture) for assessing health effects (Figure 1). The approach is driven by the question of interest.5 For regulation, the research questions are simply, “What is the net health effect from exposure to this chemical, and what health impact would result from modifying exposure?” To reconcile the priorities of policy relevance and an accurate reflection of etiological effects, we begin with the simplest approach: examining individual chemicals in isolation. For each chemical evaluated, we recommend starting by determining the association with health outcomes in conventional ways to assess dose–response relationships, bias (e.g., exposure measurement error), and confounding by other attributes, including exposure to other chemicals.
Figure 1. Proposed process to analyze chemical mixtures in epidemiological studies for regulatory decisions.
Figure 1 is a flowchart with eight steps. Step 1: Confirm research goal and chemical mixture type to inform regulatory decision-making. Step 2: Analyze individual chemicals. Step 3: Assess whether this is insufficient to inform regulatory decisions. Step 4: Analyze joint chemical exposures. Step 5: Do benefits of informing regulation exceed costs of added complexity? Step 6: Analyze multiple chemicals as a mixture. Step 7: Do benefits of informing regulation exceed costs of added complexity? Step 8: Translate findings from mixture analysis for regulatory application.
The more completely we understand the effect of individual chemical exposures, the more informative the assessment of mixtures will be. We believe that looking to the study of chemical mixtures to overcome shortcomings in understanding the effects of individual component chemicals is generally ineffective. Anchored in an understanding of constituent chemicals, it becomes more feasible to consider collective effects. Although addressing one chemical at a time is an impractical approach to assessing thousands of chemicals in a mixture that may be of concern, a clear understanding of at least some of the most important chemicals in a broader class will be helpful in making informed judgments about the full array of related chemicals. For example, addressing a PFAS mixture benefits from a relatively more complete understanding of the combined health effects of perfluorooctanoic acid and perfluorooctane sulfonate. If, on the other hand, the questions addressed in Figure 1 are answered “No”—i.e., benefits do not exceed costs—then it would not be of value to proceed with more complex approaches.
Working from the anchor of this one-at-a-time approach, we then consider joint exposures (i.e., coexposures) with guidance from the biological understanding of pathways by which the exposures may affect the health outcomes. As we move to address multiple chemical agents simultaneously, the exposure metrics may be as simple as dichotomies (i.e., chemical A alone, chemical B alone, chemicals A and B together) or ordinal categories (none, low, medium, high). If there is sufficient information from toxicological or prior epidemiological studies to develop more complicated chemical combinations based on quantitative weighting of the various exposures, then these can be considered.
The key point is that the investigators are imposing methods of combining individual agents as hypotheses about joint effects and are not deferring to a statistical algorithm to evaluate the impact of mixtures. This hypothesis-driven approach generates results that will indicate whether the evidence supports the hypothesis that the chemicals act jointly. The incremental benefit of considering mixtures can be identified relative to evaluating each component of the mixture in isolation. There may also be benefits of adjusting chemicals for one another to determine which appears to be the dominant exposure.
When the benefits of considering mixtures is evaluated with this methodical approach, we believe that the results are certain to be informative, whether addressing mixtures is found to be of value or not. At each stage—proceeding from considering individual chemicals to joint effects of two chemicals and then to a more complex mixture—investigators should balance the potential gain in understanding how multiple chemicals relate to health outcomes against the burden of added complexity and additional assumptions about the potential joint effects. Application of more complex statistical methods without a clear contribution to regulatory decision-making would favor the simpler approach.
Much of the current epidemiological investigation of chemical mixtures uses exploratory statistical approaches to discover profiles of chemicals that are predictive of health outcomes14,15 under the broad rubric of the exposome. Rather than imposing a method for aggregating chemicals into a single metric of exposure, these strategies search for harmful combinations of chemical agents, with the potential for discovering new groupings of chemical exposures (and chemical exposures acting jointly with other environmental, social, behavioral, or genetic attributes) that are predictive of health outcomes. This research may ultimately result in a more complete understanding of the association between the environment and health, at least for the population under investigation. The research, however, generally provides little or no value for informing regulatory decisions in the near term; it does not extract information that can be applied directly to informing the dose–response associations needed to guide regulation. Perhaps in time, the clues generated in such explorations will contribute to policy, but the potential for making that connection remains tenuous.
Given the inherent complexity of addressing chemical mixtures and the limitations of epidemiology, we contend that regulatory decisions should be based on a thoughtful, careful integration of epidemiology, toxicology, mechanistic research, and exposure science. Each line of investigation provides unique strengths (and limitations), and their collective value is certain to exceed each individual discipline’s contribution. Advancing methods for optimally integrating and extrapolating results across disciplines is an important goal for strengthening evidence-based regulation of mixtures.
Conclusions
Epidemiological research seeking to provide guidance to regulatory decision-making confronts the question of whether to incorporate chemical mixtures. Depending on the basis for the mixture being present and current scientific understanding, chemical mixtures may be amenable to regulation and feasible to study, encouraging epidemiologists to address them in their studies. However, the burden of proof falls on the argument that simple approaches (studying individual chemicals) are inadequate and more complex approaches (studying mixtures) are helpful for informing regulation. Studying mixtures based on mode of action may motivate research on specific chemicals or combinations of chemicals that, in time, result in findings of regulatory relevance, even if the mode-of-action studies themselves are not applicable. Considering both individual chemicals and chemical mixtures will provide a more complete understanding of the association between exposure and disease and better inform the strategy for managing health risks. Given the distinctive strengths and limitations of epidemiology in relation to toxicology, regulatory decisions should and likely will depend on their combination with a need to consider how best to maximize their complementarity.
Acknowledgments
The authors thank C. Mahony, M. Francis, M. Steinbuch, and V. Mehta for helpful comments on a previous version of this manuscript.
==== Refs
References
1. Joubert BR, Kioumourtzoglou MA, Chamberlain T, Chen HY, Gennings C, Turyk ME, et al. 2022. Powering Research through Innovative Methods for mixtures in Epidemiology (PRIME) program: novel and expanded statistical methods. Int J Environ Res Public Health 19 (3 ):1378, PMID: , 10.3390/ijerph19031378.35162394
2. Lyden GR, Vock DM, Barrett ES, Sathyanarayana S, Swan SH, Nguyen RH. 2022. A permutation-based approach to inference for weighted sum regression with correlated chemical mixtures. Stat Methods Med Res 31 (4 ):579–593, PMID: , 10.1177/09622802211013578.35128995
3. Samet JM, Speizer FE. 1993. Introduction and recommendations: working group on indoor air and other complex mixtures. Environ Health Perspect 101 (suppl 4 ):143–147. PMID: , 10.1289/ehp.93101s4143.8206023
4. Weisel CP. 2002. Assessing exposure to air toxics relative to asthma. Environ Health Perspect 110 (suppl 4 ):527–537, PMID: , 10.1289/ehp.02110s4527.12194882
5. Braun JM, Gennings C, Hauser R, Webster TF. 2016. What can epidemiological studies tell us about the impact of chemical mixtures on human health? Environ Health Perspect 124 (1 ):A6–9, PMID: , 10.1289/ehp.1510569.26720830
6. Braun JM, Sears CG. 2021. Invited perspective: how can studies of chemical mixtures and human health guide interventions and policy? Environ Health Perspect 129 (11 ):111304, PMID: , 10.1289/EHP10318.34817288
7. Taylor KW, Joubert BR, Braun JM, Dilworth C, Gennings C, Hauser R, et al. 2016. Statistical approaches for assessing health effects of environmental chemical mixtures in epidemiology: lessons from an innovative workshop. Environ Health Perspect 124 (12 ):A227–A229, PMID: , 10.1289/EHP547.27905274
8. EFSA Scientific Committee. More SJ, Bampidis V, Benford D, Bennekou SH, Bragard C, Halldorsson TI, et al. 2019. Guidance on harmonised methodologies for human health, animal health and ecological risk assessment of combined exposure to multiple chemicals. EFSA J 17 (3 ):e05634, PMID: , 10.2903/j.efsa.2019.5634.32626259
9. Martin O, Scholze M, Ermler S, McPhie J, Bopp SK, Kienzler A, et al. 2021. Ten years of research on synergisms and antagonisms in chemical mixtures: a systematic review and quantitative reappraisal of mixture studies. Environ Int 146 :106206, PMID: , 10.1016/j.envint.2020.106206.33120228
10. Jin T, Amini H, Kosheleva A, Danesh Yazdi M, Wei Y, Castro E, et al. 2022. Associations between long-term exposures to airborne PM2.5 components and mortality in Massachusetts: mixture analysis exploration. Environ Health 21 (1 ):96, PMID: , 10.1186/s12940-022-00907-2.36221093
11. Li XF, Mitch WA. 2018. Drinking water disinfection byproducts (DBPs) and human health effects: multidisciplinary challenges and opportunities. Environ Sci Technol 52 (4 ):1681–1689, PMID: , 10.1021/acs.est.7b05440.29283253
12. Samon SM, Rohlman D, Tidwell L, Hoffman PD, Oluyomi AO, Walker C, et al. 2023. Determinants of exposure to endocrine disruptors following hurricane Harvey. Environ Res 217 :114867, PMID: , 10.1016/j.envres.2022.114867.36423664
13. Barton KE, Zell-Baran LM, DeWitt JC, Brindley S, McDonough CA, Higgins CP, et al. 2022. Cross-sectional associations between serum PFASs and inflammatory biomarkers in a population exposed to AFFF-contaminated drinking water. Int J Hyg Environ Health 240 :113905, PMID: , 10.1016/j.ijheh.2021.113905.35065522
14. Cathey AL, Watkins DJ, Rosario ZY, Velez C, Mukherjee B, Alshawabkeh AN, et al. 2022. Biomarkers of exposure to phthalate mixtures and adverse birth outcomes in a Puerto Rico birth cohort. Environ Health Perspect 130 (3 ):37009, PMID: , 10.1289/EHP8990.35333099
15. Frenoy P, Perduca V, Cano-Sancho G, Antignac JP, Severi G, Mancini FR. 2022. Application of two statistical approaches (Bayesian Kernel Machine Regression and Principal Component Regression) to assess breast cancer risk in association to exposure to mixtures of brominated flame retardants and per- and polyfluorinated alkylated substances in the E3N cohort. Environ Health 21 :27, PMID: , 10.1186/s12940-022-00840-4.35216589
|
PMC010xxxxxx/PMC10081692.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37027338
EHP11958
10.1289/EHP11958
Research
Amniogenesis in Human Amniotic Sac Embryoids after Exposures to Organophosphate Flame Retardants
Xu Chenke 1
Zhang Chenhao 1
Liu Yanan 1
Ma Haojia 1
Wu Feifan 1
Jia Yingting 1
Hu Jianying 1
1 MOE Laboratory for Earth Surface Process, College of Urban and Environmental Sciences, Peking University, Beijing, China
Address correspondence to Jianying Hu, Laboratory for Earth Surface Processes, College of Urban and Environmental Sciences, Peking University, Beijing 10087 China. Telephone: 86-10-62765520. Email: [email protected]
7 4 2023
4 2023
131 4 04700702 8 2022
26 11 2022
10 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Amniogenesis is a key event in biochemical pregnancy, and its failure may result in human embryonic death. However, whether and how environmental chemicals affect amniogenesis remain largely unknown.
Objectives:
The objective of the present study was to screen chemicals that may disrupt amniogenesis in an amniotic sac embryoid model and to investigate the potential mechanism of amniogenesis failure, with a focus on organophosphate flame retardants (OPFRs).
Methods:
This study developed a high-throughput toxicity screening assay based on transcriptional activity of octamer-binding transcription factor 4 (Oct4). For the two positive OPFR hits with the strongest inhibitory activity, we used time-lapse and phase-contrast imaging to assess their effects on amniogenesis. Associated pathways were explored by RNA-sequencing and western blotting, and potential binding target protein was identified through a competitive binding experiment.
Results:
Eight positive hits exhibiting Oct4 expression were identified, with 2-ethylhexyl-diphenyl phosphate (EHDPP) and isodecyl diphenyl phosphate (IDDPP) showing the strongest inhibitory activity. EHDPP and IDDPP were found to disrupt the rosette-like structure of the amniotic sac or inhibit its development. Functional markers of squamous amniotic ectoderm and inner cell mass were also found disrupted in the EHDPP- and IDDPP-exposed embryoids. Mechanistically, embryoids exposed to each chemical exhibited abnormal accumulation of phosphorylated nonmuscle myosin (p-MLC-II) and were able to bind to integrin β1 (ITGβ1).
Conclusion:
The amniotic sac embryoid models suggested that OPFRs disrupted amniogenesis likely by inhibiting the ITGβ1 pathway, thus providing direct in vitro evidence associating OPFRs with biochemical miscarriage. https://doi.org/10.1289/EHP11958
Supplemental Material is available online (https://doi.org/10.1289/EHP11958).
All the authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Miscarriage has been recognized as the most noticeable complication during human pregnancy1 and has sharply increased globally over the past decades.2,3 About 23 million miscarriages occurred annually worldwide, translating to 0.73 pregnancy losses each second.4 Biochemical miscarriage was defined as very early fetal loss, in which the initial serum or urine biochemical test is positive, but it does not progress into a clinical pregnancy.5 Although 75% pregnancies end up as biochemical miscarriage,6 factors such as parental age and genetic diseases explain a small proportion of pathogenesis.4 Epidemiological studies have provided increasing evidence that environmental contaminants such as bisphenol A,7 polychlorinated biphenyls,8 and phthalates9 can stimulate the pathogenesis of biochemical miscarriage.
Organophosphate flame retardants (OPFRs) are of particular concern.10 OPFRs existed ubiquitously in building materials and various consumer products, such as mobile phones and baby products,11 and have therefore been detected in humans worldwide, including pregnant women and fetuses.12 Several epidemiological studies have associated OPFR exposure with biochemical miscarriage.13,14 Our previous study has proved that one of OPFRs, 2-ethylhexyl-diphenyl phosphate (EHDPP), can disrupt early placentation in trophoblast organoids and therefore may induce miscarriage in mice.15 Moreover, human early pregnancy is a stage at which the blastocyst contacts and invades the maternal endometrium and the amniotic sac emerges from the embryonic inner cell mass by differentiating into squamous amniotic ectoderm.16 The amniotic sac is an extraembryonic organ that forms a fluid-filled sac surrounding the embryo to provide mechanical protection and functions of embryonic respiration, nutrition, and excretion. Amniogenesis occurres during implantation and is one of the keystones for early human embryogenesis.17 Disrupted amniogenesis can lead to embryonic development failure and subsequent embryonic death at an early stage.18 The fundamental role of amniogenesis in early pregnancy led us to postulate that some OPFRs may also increase the risk of biochemical miscarriage by inhibiting this developmental process.
To test this hypothesis, we applied a high-throughput toxicity screening based on the transcriptional activity of octamer-binding transcription factor 4 (OCT4), one of the main pluripotency factors in early developmental stages, and then screened chemicals from 53 OPFRs that could potentially disrupt amniogenesis. Among eight positive hits, the two OPFRs with the highest inhibitory effects on Oct4 expression were further assessed for their effects on amniogenesis in an amniotic sac embryoid model using time-lapse phase-contrast imaging. Thus, our study contributes to elucidating the underlying process by which OPFRs induce biochemical miscarriage based on a key developmental node.
Materials and Methods
Chemicals and Reagents
Fifty-three OPFRs were used (Table S1). Forty-two OPFRs were listed in Toxicity Forecaster (ToxCast) list by U.S. Environmental Protection Agency (U.S. EPA) (https://comptox.epa.gov/dashboard/chemical-lists/FLAMERETARD). Eleven other OPFRs were detected from house dust, according to our previous study.19,20 Forty-seven have commercially available standards, which were purchased for the present study, and six OPFRs were synthesized in our laboratory. These OPFRs were used in toxicity screening for 48-h exposure, and EHDPP and IDDPP were also used in the following toxicity study for 72- or 96-h exposure.
Cell Culture
The human embryonic stem cell (hESC) line H1 was obtained from School of Aerospace Engineering in Tsinghua University and grown on Matrigel-coated plates with mTeSR1 medium (85850; STEMCELL Technologies Inc.). The Matrigel (354234; Corning) were added to the plate (40 μL/μm2) and smeared uniformly by pipette.
The OCT-tdTomato luciferase H1 cell line was established by the School of Basic Medical Sciences in Peking University.21 Briefly, OCT4 sgRNA was introduced into the modified Px330 (42230; Addgene), and the sequence of constructed plasmid is provided in Supplemental Material, “Sequence of OCT4-sgRNA Px330 Plasmid.” Next, OCT4-tdTomato plasmid was nucleofected together with OCT4-sgRNA Px330 plasmid in H1 cell line by 4D-Nucleofector System (Lonza) using a transfection kit (VAPH-5012; Lonza); the sequence of OCT4-tdTomato plasmid is provided in Supplemental Material, “Sequence of OCT4-tdTomato Plasmid.” The OCT-tdTomato-luciferase H1 cell medium included DMEM/F12 (C11330500BT; Coolaber) supplemented with 20% knockout serum replacement (KSR) (A3181502; Gibco), 1% GlutaMAX (35050-061; Thermo Fisher Scientific), 1% nonessential amino acids (11140-050; Thermo Fisher Scientific), 0.1 mM of β-mercaptoethanol (21985-023; Thermo Fisher Scientific), and 4 Octng/ml of fibroblast growth factor-basic (bFGF) (P5453; Beyotime). Both cultured H1 cells and OCT4-tdTomato luciferase cells were dissociated with 0.05% Trypsin-EDTA (T4090; Sigma) for 3 min at 37°C. All cells were then incubated in an atmosphere of 5% CO2 at 37°C. hESCs were passaged at a split ratio of 1:3 to 1:5 using Dispase II (CD4691; Coolaber) every 4–5 d.
OCT-Td luciferase H1 cells (3×104cells/cm2 in each passage) were grown on mouse embryonic fibroblast (Mef) feeder cells (1.5×105cells/cm2) using mTeSR1 medium (85850; STEMCELL Technologies Inc.). Mef feeder cells were only used for culturing, and OCT-Td luciferase H1 were then plated without Mef feeder cells for experiments. To prepare feeders, Mef cells (4×105cells/cm2) were prepared as described below and treated with 10μg/mL mitomycin C (HY-13316; MedChemExpress) for 2.5 h and washed with phosphate buffered saline (PBS; C14190500BT, Thermo Fisher Scientific).
Preparation of Mef Cells
For obtaining Mef cells, the 8-wk-old CD-1 pregnant mice with E13.5 fetuses from Charles River Laboratories were used in this study. Animal studies were approved by the Institutional Animal Care and Use Committee of Peking University (approval No. Urban-HuJY-1). The mice were anesthetized by 100mg/kg sodium pentobarbital via intramuscular injection and euthanized via cervical dislocation. The embryos were carefully stripped, and the bodies were fully cut into pieces, and the pieces were mixed with 5mL 0.05% Trypsin-EDTA (T4090; Sigma). After digestion for 5 min at 37°C and standing for 1 min, the supernatant was removed, and 2mL FBS (10091148; Thermo Fisher Scientific) was added to the supernatant. The digestion was repeated twice. After mixing the supernatants and filtration by 40μm cell strainer (251100; Abcam), the obtained cells were placed on 10 cm2 cell culture dishes at the density of 1×107/cm2 and cultured with DMEM (C11965500BT; Thermo Fisher Scientific) and 10% FBS (10091148; Thermo Fisher Scientific). When the cells reached ∼80%, 0.25% Trypsin-EDTA was used for digestion and expanded culture. Cells were then incubated in an atmosphere of 5% CO2 at 37°C and were passaged at a split ratio of 1:3 to 1:5 using 0.05% Trypsin-EDTA (T4090; Sigma).
Screening OPFRs That Down-Regulated OCT4 Expression
For chemical screening, the OCT4-tdTomato luciferase H1 cells were digested by Dispase II for 10 min and replated onto 96-well plates coated in 1% Matrigel (354230; Biodee) with Essential 8 medium (A1517001; Thermo Fisher Scientific) for 1 h before use. H1 cells were plated on the 96-well plates at the density of 1×104cells/cm2. After culturing for 2 d, the cells were treated with each OPFR in the list (10μM; Table S1). After the cells were exposed to OPFRs for a further 48 h, Sytox Green (R37168, Thermo Fisher Scientific; Dilution: 1:50) was added, followed by incubation at 37°C for 30 min, and then the H1 underwent high-throughput determination for fluorescence intensity using ImageXpress developed by Molecular Devices.
In Vitro Amniogenesis and Chemical Exposure
In vitro amniogenesis was performed according to recent methods.22 H1 cells were plated as single cells at 2×104cells/cm2 onto a gel bed (a nominal thickness ≥100μm) and then cultured in mTeSR1 containing 10μM of Y27632 (1254/10; Tocris), a specific inhibitor of rho-associated kinases, which can inhibit the apoptosis of stem cells after passage.23 After 24 h, the culture medium was replenished with fresh mTeSR1 and 4% (v:v) Geltrex supplement (A1413201; Thermo Fisher Scientific). Finally, the mTeSR1 medium with 4% (v:v) Geltrex supplement was replenished daily. During this 96-h process, H1 cells gradually gathered and the self-assembly embryoids were formed (Figure S1A,B), and columnar embryonic disc (OCT4+ and TFAP2A−) transformed into squamous amniotic ectoderm (TFAP2A+ and OCT4−),22 as shown in Figure S1C.
EHDPP has been reported to inhibit early embryonic development by disrupting extraembryonic development at 1 and 10μM of EHDPP.15 Thus, exposure concentration was set at 0.1, 1, or 10μM. Similar concentrations were selected for the IDDPP exposure experiment. Exposure concentration of Arg-Gly Asp (RGD) peptide was set at 1μM. The concentration of dimethylsulfoxide (DMSO; 0.1%) in the control groups (n=3) was the same as that in the exposure groups (n=3). For amniogenesis evaluation by time-lapse and phase-contrast imaging, cells were treated with either EHDPP or IDDPP for 96 h. For evaluation of functional markers in amniotic sac embryoids, cells were treated with either EHDPP or IDDPP for 72 and 96 h. For mechanism research, cells were treated with EHDPP or IDDPP for 72 h.
Time-Lapse and Phase-Contrast Imaging
For live-cell imaging, H1 were plated on four-chamber glass-bottom dishes (D35C4-20-1.5-N; Cellvis) at 2×104cells/cm2, using the culture and exposure procedure detailed in the preceding section. Geltrex (30μL) was used to generate the gel bed in each well. Live-cell imaging was performed by DeltaVision Elite (Applied Precision) with a 10× objective and at an interval of 15 min, and the Volocity Demo software was used to export images and videos.
Immunofluorescence of Amniotic Sac Embryoid
For confocal microscopy, H1 cells were plated on IbiTreat μ-plates (IB-80826, Ibidi GmbH) coated in 40μL of Geltrex per well, using the same culture and exposure procedure as in the previous two sections at 2×104cells/cm2. After H1 cells formed embryoids, 4% paraformaldehyde (C2055; Bioss) was used to fix the squamous cysts for 60 min, 0.25% Triton X-100 (CR00576; EBT Systems) was used to permeabilize for 30 min, 3% bovine serum albumin (BSA) (PM5130; Coolaber) was used to block for 60 min at 37°C. Primary antibody was used in staining overnight at 4°C, and secondary antibody was used in staining for 60 min at room temperature. The cells were then washed three times with PBS prior to counterstaining the cell nuclei with DAPI. Alexa-Fluor 488 phalloidin (A12379, Thermo Fisher Scientific; 1:200) was used as a pan-cell membrane marker. Information on the antibodies is listed in Table S2. Expression of amniotic markers at 72 h and 96 h in amniotic sac embryoids were demonstrated (Figure S1D). Isotype control was used as the negative technical control, which was shown in Figure S1D, and the proteins used in negative technical control are listed in Table S2. The images for the fixed samples were acquired by a High-Speed Spinning Disk Confocal Microscope (Andor Technology) with Leica 0.45 NA 10× air and Leica 1.3 NA 40× oil objectives. Fluorescence images were acquired by confocal microscopy and quantified with Fiji software (version 2.3.0).
RNA Sequencing (RNA-Seq)
H1 cells were plated on 12-well plates at 2×104cells/cm2 and exposed to EHDPP and IDDPP, respectively, for 72 h. After that, the embryoids were collected by 500μL TransZol Up (ET111-01, Transgen). Next, 100μL of chloroform (M1024442500; Merck) was used before vigorous shaking before an 12,000-rpm centrifugation at 20°C for 10 min. The top layer was aspirated, and 300μL of isopropyl alcohol (1.01040.4008; Merck) was added, and then the samples were gently shaken before an 12,000-rpm centrifugation at 20°C for 10 min. The supernatants were removed, and then 700μL of 75% ethanol (E801077; Macklin) was used per sample before gentle shaking and centrifugation at same centrifugation speed at 20°C for 5 min. After removing the supernatants, each sample was dissolved in 30μL of diethyl pyrocarbonate (DEPC) H2O (W274329; Aladdin). The quality and concentration of RNA were detected by a Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies). The RNA integrity number (RIN) was detected by Agilent 2100, and the RNA samples with a RIN >7 were considered to be qualified for RNA-seq. RNA-seq was achieved by the Beijing Genomics Institute. Poly A species were used to enrich RNA samples, and the libraries were constructed on the BGISEQ-500 platform (BGI). The sequencing depth of each library was ∼ 40 million reads. SOAPnuke (version 1.5.2), a software for FASTQ file filtering developed by the Beijing Genomics Institute, was used to filter the sequencing data, which treated all species in the same way.24 HISAT2 (version 2.0.4) and Bowtie2 (version 2.2.5) were used to map the reads against GRCh38.p12.25,26 Expectation-Maximization (version 1.2.8) was used to calculate the gene expression levels. This study performed gene set enrichment analysis (GSEA) based on all expressed genes between the control and each exposure group. Two human collections of the molecular signatures database, subcollection of C2 (curated gene sets) and canonical pathways (CP), were used for GSEA enrichment analysis. Additionally, differentially expressed genes [absolute value of log2 (fold change) >0.58 and adjusted p-value (Q-value) <0.05] were identified by DEGseq2 (version 1.34.0; Michael Love).
Quantitative Real-Time Polymerase Chain Reaction (RT-qPCR)
H1 cells were plated on 12-well plates at 2×104cells/cm2 and exposed to chemicals for 72 and 96 h. The total RNA was extracted following the protocol described in the “RNA Sequencing” section. Moloney Murine Leukemia Virus Reverse Transcriptase (639574; Takara), Recombinant RNase Inhibitor (2313Q; Takara), Oligo dT (3805; Takara), dNTP (4019; Takara), and Random Primers (3802; Takara) were used for reverse transcription on a S1000 Thermal Cycler (Bio-Rad) with the program: 40°C for 60 min, 70°C for 15 min, and 4°C for 30 min. We used SYBR Green (QPK-201; Toyobo) for real-time fluorescence detection on a StepOnePlus sequence detection system (Applied Biosystems), and the program was 95°C for 1 min, 40 cycles of 95°C for 15 s, and 60°C for 45 s. Relative gene expression was evaluated by the 2−△△Ct method, as suggested by Applied Biosystems. The primers were synthesized by Thermo Fisher Scientific, and the primer sequences are listed in Table S3.
Concentration Determination of EHDPP and IDDPP in Amniotic Sac Embryoid
H1 cells were plated on 12-well plates at 2×104cells/cm2 and exposed to EHDPP or IDDPP for 72 h. Embryoids were digested by 0.05% Trypsin-EDTA (T4090; Sigma) and then collected in 10mL glass bottles. After spiking 1 ng triphenyl phosphate (TPhP)-d15 (C/D/N Isotopes Inc.) as internal standard, embryoids were freeze-dried and weighed, and then 3mL ethyl acetate (E116131; Aladdin) was added to the extract. The mixtures were shaken for 20 min, and then were centrifugated at 4,000 rpm for 12 min. Extraction was repeated twice, and the combined supernatants were evaporated to dryness under high-purity nitrogen stream. The extracts were redissolved in 100μL methanol (A454; Thermo Fisher Scientific) and centrifugated at 12,000 rpm for 10 min. The supernatant was collected for analyzing EHDPP or IDDPP using an UltiMate 3000 UPLC system coupled with a TSQ Quantiva triple-quadrupole mass spectrometer (Thermo Fisher Scientific). The separation was performed on a BEH C18 (100×2.1mm, 1.7μm particle size; Waters) column. Instrument conditions are provided in Table S4. EHDPP and IDDPP were detected in blank with the concentrations of 0.58±0.26 ng/g dry weight (dw) and 0.71±0.63 ng/g dw, respectively. The final concentrations reported in this study were blank subtracted. The limit of quantitations (LOQs) were calculated as 10 times the standard deviations of procedural blanks, and the LOQs of EHDPP and IDDPP were 2.55 and 6.33 ng/g dw, respectively. The recoveries of EHDPP and IDDPP were 51.7%±7.3% and 48.1%±8.4%, respectively, and the recovery of internal standard was 56.1%±18.1%.
Western Blotting Analysis
Cell lysis buffer (9803S, CST) were used to collect the samples and concentrations of the protein were determined by BCA assay (P0012S; Beyotime). After a 12,000-rpm centrifugation at 4°C for 10 min, 5× loading buffer was added to each sample (n=3), and then each sample was heated to 100°C for 10 min. Each sample was loaded at 30μL per lane into polyacrylamide gel, in which protein concentration was 3μg/μL. After that step, nitrocellulose (NC) membrane (MH0322; Macklin) was used to transfer the protein, and 3% BSA in Tris-buffered saline (TBS) with 0.1% Tween 20 detergent (TBST) were used to block at room temperature for 1 h. The protein was then incubated with the primary antibody at 4°C overnight and then with the secondary antibody for 45 min at 37°C. After both primary and secondary antibody treatments, the strips were washed thrice with TBST. The membranes were washed twice with TBST and then shaken in stripping buffer (PMC1710125; Perfemiker) for 30 min. After stripping, the strips were washed thrice and then were treated from blocking to incubation of secondary antibody. The information on antibodies is listed in Table S2. The bands were developed by developing solution (YA0372; Solarbio) and fixing solution (YA0382; Solarbio) and were scanned by LiDE300 (Canon). The gray levels of the bands were analyzed by Fiji software (version 2.3.0).
Binding Affinity Assay
In vitro binding affinity tests were performed to assess whether OPFRs can bind to ITGβ1. Plates of 96 wells (3599; Corning) were coated overnight at 4°C with fibronectin (F0556; Sigma) diluted with carbonate buffer (0.5μg/mL, 15 mM of Na2CO3, 35 mM of NaHCO3, pH 9.6) and washed with PBST (0.1% Tween 20) in triplicate. The wells were blocked for 1 h at room temperature in BSA buffer (1% BSA; 20 mM of Tris-HCl, 150 mM of NaCl, 1 mM of CaCl2, 1 mM of MgCl2, 1 mM of MnCl2, pH 7.5) before adding 0.1, 1, 10, 100, 1,000, and 10,000 nM of EHDPP, IDDPP, or RGD peptide with a5b1 integrin protein (3230-A5-050; R&D; 2.0μg/mL) at room temperature for 1 h. After the plate was washed with PBST in triplicate, primary anti-integrin β1 antibody (ab150361, Abcam PLC; 1:2000) was added, and the plate was then washed again with PBST in triplicate. Rabbit antimouse antibody (ab150077; Abcam PLC) was added to the plate, which was then washed with PBST in triplicate before adding a 50-μL TMB solution (P0209; Beyotime) to each well. After 3 min, 50μL of H2SO4 (2M) was added, and the absorbance was detected at 450 nm using a microplate reader (SpectraMax i3X; Molecular Devices).
Statistical Analyses
All statistical analyses were performed by GraphPad Prism (version 9.0.0; GraphPad Software Inc.). Student’s t-tests with Welch’s correction (p<0.05 in F test) or without Welch’s correction (p>0.05 in F test) were used for multiple comparisons. A Pearson’s chi-square test was used to analyze differences in amniotic sac formation rate between the control and the EHDPP or IDDPP exposure groups. All experiments included n=3 independent biological replicates unless otherwise indicated. The screening, amniogenesis evaluation by time-lapse and phase-contrast imaging, and binding affinity experiment, and the evaluation of functional markers in amniotic sac embryoids all had three technical replicates. Western blotting, EHDPP and IDDPP concentration detection, and qPCR had two technical replicates. RNA-seq had no technical replicate. Each experiment was performed at least three times with the following significance levels: *p<0.05 and **p<0.01.
Data Availability
The data supporting the findings of this study are available from the corresponding author on request. Data of RNA-seq is available in Excel Table S1. The live-cell imaging videos are available in the Supplementary Movies 1–6.
Results
Screening OPFRs That Down-Regulated OCT4 Expression
This study used 53 OPFRs for screening (Table S1), as detailed in the “Methods” section, “Chemicals and Reagents.” This study used H1 stably transfected by OCT4-Td-luciferase plasmid for high-throughput screening of 53 OPFRs that might affect the pluripotency of early embryos (Figure 1A). After a 48-h exposure, OCT4 expression and cell survival were, respectively, detected using Td-luciferase and Sytox Green with any OPFRs that resulted in 20% lower OCT4 expression without significant effects on cell survival (compared to DMSO control) selected as positive hits (Figure 1B,C). Eight OPFRs among the 53 were identified as positive hits, including five aryl-OPFRs [EHDPP, IDDPP, tris(2,4-dimethylphenyl) phosphate (TXP), 4-tert-butylphenyl diphenyl phosphate (4tBPDPP), and isopropylphenyl diphenyl phosphate (2IPPDPP)], two alkyl-OPFRs [bis(2-ethylhexyl) phosphate (BEHP) and triamyl phosphate (TAP)], and one halogenate-OPFR [tris(2,3-dibromopropyl) phosphate (TDBPP)] (Figure 1D). Of the eight positive hits, EHDPP and IDDPP had the highest potency [the Td luciferase levels of H1 exposed to EHDPP and IDDPP were, respectively, 0.50±0.04 (p<0.01) and 0.53±0.15-fold (p<0.05) relative to the Td luciferase level in the control].
Figure 1. Screening of OPFRs that inhibited OCT4 expression. (A) Screening and toxicity evaluation of OPFRs; (B) Relative fluorescence intensity (means±SDs) of OCT4-luciferase (tdTomato) H1 cell line exposed to OPFRs (10,000 nM); (C) Relative fluorescence intensity (means±SDs) of Sytox Green (green) in hESC exposed to OPFRs (10μM); (D) Structure of identified chemicals, TXP, 2IPPDPP, 4tBPDPP, EHDPP, IDDPP, TAP, BEHP, TDBPP. Data in (C) and (D) are expressed relative to the levels in DMSO-treated hESC, which were set to 1. n=3. Data were analyzed using an unpaired two-tailed Student’s t-test. Indicated values are significantly different from the control value. Numeric data in (C) and (D) were listed in Table S5. Note: 2IPPDPP, 2-isopropylphenyl diphenyl phosphate; 4tBPDPP, 4-tert-butylphenyl diphenyl phosphate; BEHP, bis(2-ethylhexyl) phosphate; DMSO, dimethyl sulfoxide; EHDPP, 2-ethylhexyl-diphenyl phosphate; IDDPP, isodecyl diphenyl phosphate; OCT4, octamer-binding transcription factor 4; OPFRs, organophosphate flame retardants; SD, standard deviation; TAP, triamyl phosphate; TDBPP, tris(2,3-dibromopropyl) phosphate; TXP, tris(3,5-xylenyl) phosphate. *p<0.05. **p<0.01. Created with BioRender.com.
Figure 1A is an illustration with two steps, depicting the screening and toxicity evaluation of organophosphate flame retardants. Step 1: The human embryonic stem cell leads to the octamer-binding transcription factor 4- tdTomato luciferase assay. Step 2: The octamer-binding transcription factor 4- tdTomato luciferase assay leads to amniogenesis evaluation. Figures 1B and 1C are error bar graphs, plotting relative density of octamer-binding transcription factor 4, ranging from 0.0 to 1.4 in increments of 0.2 and relative intensity of Sytox green, ranging from 0.0 to 1.4 in increments of 0.2 (y-axis) across T P H P; C D P; T X P; T-3,5 X P; T Cr P; m-T Cr P; o-T Cr P; 2 I P P D P P; B 2 I P P P P; T 4 I P P P; T 2 I P P P; 4 t B P D P P; P D pt B P P; T 4 t B P P P; T Pe T Ph P; T 4 Dt B P P; T N P P; E H D P P; I D D P P; B D P P; B E H P P; D I D P P; D B P P; D P P; R B D P; B P A D P under aryl-organophosphate flame retardants; T M P; T E P; T Pr P; T I P P; T B P; T I B P; T E H P; T H P; T B O E P; T A P; D E E P; D M P P; D M P; D B P; B E H P under alkyl- organophosphate flame retardants; T C E P; T C I P P; T D C I P P; V 6; T 2,3 D C P P; T 2 C P P; T 3 C P P; T D B P P; T T B N P P under halogenated- organophosphate flame retardants; Fyrol 6; M D P A; and H P C T P under uppercase n-organophosphate flame retardants (x-axis). A heatmap depicts the following: T X P, 2 I P P D P P, 4 t B P D P P, E H D P P, I D D P P, T A P, B E H P, and T D B P P. Figure 1D displays structure of identified chemicals, T X P, 2 I P P D P P, 4 t B P D P P, E H D P P, I D D P P, T A P, B E H P, and T D B P P.
Effects of OPFRs on Amniogenesis
Given the relatively strong toxicity of the two aryl-OPFRs EHDPP and IDDPP in the inhibition of OCT4 transcription activity, this study applied an amniotic sac embryoid model to further investigate their toxicities in embryoids. No significant differences of cell survival were found in H1 cells exposed to EHDPP or IDDPP for 96 h (Figure S2A,B). We recorded the dynamics of amniogenesis for 2 d from 48 h to 96 h after culture in the control, EHDPP, and IDDPP groups, using the DeltaVision Elite (Figure 2A). In the control group, cells arranged around a shared point of apical constriction before 64 h, and then producing a rosette-like structure surrounding a central cavity between 72 and 96 h (“Normal” in Figure 2A and Supplementary Movie 1). Although significant differences in the arrangement of cells were not observed before 72 h between EHDPP-treated and the control groups, the rosette-like structure surrounding a central cavity either did not appear between 72 and 96 h (“I” in Figure 2A and Supplementary Movie 2) or was malformed (“II” in Figure 2A and Supplementary Movie 3). No significant difference in the amniotic sac formation rates was observed between the EHDPP-treated and control groups at 72 h but were 70% (35/50, normal amniotic sacs/total) and 59% (32/54) in 1μM and 10μM EHDPP-treated groups at 96 h, respectively, significantly lower than the control (88%, 43/49; p<0.05) (Figure 2B,C). Similarly, the formation rates (80% and 79%, respectively) in the IDDPP exposure groups (1μM and 10μM) were significantly lower than the control group at 96 h but showed no significant difference at 72 h (Figure 2D,E). After exposure to 1μM EHDPP or IDDPP for 72 h, the average concentrations of EHDPP and IDDPP in embryoids were 261.3±70.90 and 234.1±27.66 ng/g dw, respectively.
Figure 2. Amniogenesis evaluation of OPFRs by time-lapse and phase-contrast imaging. (A) Time-lapse and phase-contrast imaging in the control, EHDPP, and IDDPP exposure groups during amniogenesis of hESC; (B) 72 h cavity incidence rate of the control and EHDPP exposure groups; (C) 96-h cavity incidence rate of the control and EHDPP exposure groups; (E) 72-h cavity incidence rate of the control and IDDPP exposure groups; (D) 96-h cavity incidence rate of the control and IDDPP exposure groups. Embryoids treated with DMSO were used as the control. Scale bars, 20μm. Data were analyzed using a Pearson’s chi-square test. Indicated values are significantly different from the control value. Note: DMSO, dimethyl sulfoxide; EHDPP, 2-ethylhexyl-diphenyl phosphate; IDDPP, isodecyl diphenyl phosphate; OPFRs, organophosphate flame retardants. *p<0.05.
Figure 2A is a stained tissue displays seven columns, namely, 48 hours, 56 hours, 64 hours, 72 hours, 80 hours, 88 hours, and 96 hours, and three rows, namely, normal, One, and Two. Figures 2B and 2D are bar graphs, plotting 72 hours E H D P P cavity incidence rate, ranging from 0.0 to 1.0 in increments of 0.2 and 72 hours I H D P P cavity incidence rate, ranging from 0.0 to 1.0 in increments of 0.2 (y-axis) across D M S O, 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for cavity or total count. Figures 2C and 2E are bar graphs, plotting 96 hours E H D P P cavity incidence rate, ranging from 0 to 0.4 in increments of 0.1 and 0.5 to 1.0 in increments of 0.1, and 96 hours I D D P P cavity incidence rate, ranging from 0.0 to 0.5 in increments of 0.1 and 0.7 to 1.0 in increments of 0.1 (y-axis) across D M S O, 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for cavity or total count.
Evaluation of Functional Markers in Amniotic Sac Embryoid Exposed to OPFRs
The OCT4 expressions at 72 h in the 1μM and 10μM EHDPP exposure groups were 0.74±0.09 (p<0.05) and 0.68±0.16 (p<0.05) times as high, respectively, as those in the control group (Figure 3A–C). The CDX2 expressions at 96 h in the 1μM and 10μM EHDPP exposure groups were 0.61±0.03 and 0.55±0.10 times as high, respectively, as those in the control group, whereas the expressions of transcription factor AP-2α (TFAP2A) at 96 h in the 1μM and 10μM EHDPP exposure groups were 0.73±0.06 and 0.70±0.12 times as high, respectively, as those in the control group (Figure 3D–F). Although similar effects in the expression of these markers were also observed in the IDDPP exposure groups, the significant effect of IDDPP on OCT4 expression after 72 h exposure was observed in the 0.1μM exposure group (Figure 4A–F).
Figure 3. Evaluation of functional markers in amniotic sac embryoids exposed to EHDPP. (A) OCT4 (yellow) and DAPI (blue) in the control and EHDPP exposure groups at 72 h; (B) Relative intensity of OCT4 (means±SDs) in the control and EHDPP exposure groups at 72 h; (C) TFAP2A (red), CDX2 (green), and DAPI (blue) in the control and EHDPP exposure groups at 96 h; (D) Relative intensity of TFAP2A (means±SDs) in the control and EHDPP exposure groups at 96 h; (E) Relative intensity of CDX2 (means±SDs) in the control and EHDPP exposure groups at 96 h. Scale bars, 20μm. Data in (B), (D), and (E), were expressed relative to the levels in DMSO-treated embryoids, which were set to 1. n=3. Data were analyzed using Student’s t-test. Indicated values were significantly different from the control value. Numeric data in (B), (D) and (E) were listed in Table S6. Note: CDX2, caudal-type homeobox 2; DAPI, 4′,6-diamidino-2-phenylindole; DMSO, dimethyl sulfoxide; EHDPP, 2-ethylhexyl-diphenyl phosphate; OCT4, octamer-binding transcription factor 4; SD, standard deviation; TFAP2A, transcription factor AP-2α. *p<0.05. **p<0.01.
Figure 3A is a stained tissue displays three columns, namely, octamer-binding transcription factor 4, 4′,6-diamidino-2-phenylindole, and Merge, and four rows, namely, dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar. Figures 3B, 3D, and 3E are bar graphs, plotting Relative intensity of 72 hours octamer-binding transcription factor 4, ranging from 0.0 to 1.4 in increments of 0.2; Relative intensity of 96 hours caudal-type homeobox 2, 0.0 to 1.2 in increments of 0.3; Relative intensity of 96 hours transcription factor A P 2 lowercase alpha, ranging from 0.0 to 1.4 in increments of 0.2 (y-axis) across dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for 2-ethylhexyl-diphenyl phosphate. Figure 3C is a stained tissue with four columns, namely, transcription factor A P 2 lowercase alpha, caudal-type homeobox 2, 4′,6-diamidino-2-phenylindole, and merge, and four rows, namely, dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar.
Figure 4. Evaluation of functional markers in amniotic sac embryoids exposed to IDDPP. (A) OCT4 (yellow) and DAPI (blue) in the control and IDDPP exposure groups at 72 h; (B) Relative intensity of OCT4 (means±SDs) in the control and IDDPP exposure groups at 72 h; (C) TFAP2A (red), CDX2 (green), and DAPI (blue) in the control and IDDPP exposure groups at 96 h; (D) intensity of TFAP2A (means±SDs) in the control and IDDPP exposure groups at 96 h; (E) intensity of CDX2 (means±SDs) in control and IDDPP exposure groups at 96 h. Scale bars, 20μm. Data in (B), (D), and (E), were expressed relative to the levels in DMSO-treated embryoids, which were set to 1. n=3. Data were analyzed using Student’s t-test. Indicated values were significantly different from the control value. Numeric data in (B), (D) and (E) were listed in Table S6. Note: CDX2, caudal-type homeobox 2; DAPI, 4′,6-diamidino-2-phenylindole; DMSO, dimethyl sulfoxide; IDDPP, isodecyl diphenyl phosphate; OCT4, octamer-binding transcription factor 4; SD, standard deviation; TFAP2A, transcription factor AP-2α. *p<0.05. **p<0.01.
Figure 4A is a stained tissue displays three columns, namely, octamer-binding transcription factor 4, 4′,6-diamidino-2-phenylindole, and Merge, and four rows, namely, dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar. Figures 4B, 4D, and 4E are bar graphs, plotting Relative intensity of 72 hours octamer-binding transcription factor 4, ranging from 0.0 to 1.2 in increments of 0.3; Relative intensity of 96 hours caudal-type homeobox 2, 0.0 to 1.2 in increments of 0.3; Relative intensity of 96 hours transcription factor A P 2 A, ranging from 0.0 to 1.2 in increments of 0.3 (y-axis) across dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for isodecyl diphenyl phosphate. Figure 4C is a stained tissue with four columns, namely, transcription factor A P 2 A, caudal-type homeobox 2, 4′,6-diamidino-2-phenylindole, and merge, and four rows, namely, dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar.
Mechanism of Amniogenesis Failure
We then induced the H1 cells to embryoids in the presence of EHDPP or IDDPP for 72 h, isolated mRNA, and performed sequencing. Subsequent GSEA analysis showed that the pathways of integrin (ITG) 1 [Normalized enrichment score (NES)=−2.02, p<0.01; Figure 5A], ITG3 (NES=−1.66, p<0.01; Figure S3A), AVB3 ITG3 (NES=−1.81, p<0.01; Figure S3B), the extracellular matrix (NES=−1.61, p<0.01; Figure S3C), syndecan-1 (NES=−1.97, p<0.01; Figure S3D), focal adhesion kinase (FAK) (NES=−1.56, p<0.01; Figure 5B), phosphatidylinositol 3 kinase (PI3K) (NES=−1.81, p<0.05; Figure S5E), the Ras homolog gene family member A (RHOA) (NES=−1.56, p<0.05; Figure 5C), and Ras-related C3 botulinum toxin substrate 1 (RAC1) (NES=−1.31, p=0.11; Figure 5D) were the top enriched gene sets after EHDPP exposure. Similarly, ITG1 (NES=−1.97, p<0.01; Figure 5E), ITG3 (NES=−1.73, p<0.01; Figure S3F), AVB3 ITG3 (NES=−1.71, p<0.01; Figure S3G), the extracellular matrix (NES=−1.75, p<0.01; Figure S3H), syndecan-1 (NES=−1.26, p=0.15; Figure S3I), FAK (NES=−1.63, p<0.01; Figure 5F), PI3K (NES=−1.42, p<0.05; Figure S5J), RHOA (NES=−1.56, p<0.01; Figure 5G), and RAC1 (NES=−1.54, p<0.01; Figure 5H) were the top significantly enriched gene sets after IDDPP exposure.
Figure 5. GSEA analysis of amniotic sac embryoids exposed to EHDPP and IDDPP. (A) ITG1 pathway of GSEA analysis after EHDPP exposure (10μM); (B) FAK pathway of GSEA analysis after EHDPP exposure (10μM); (C) RAC1 of GSEA analysis after EHDPP exposure (10μM); (D) RHOA pathway of GSEA analysis after EHDPP exposure (10μM); (E) ITG1 pathway of GSEA analysis after IDDPP exposure (10μM); (F) FAK pathway of GSEA analysis after IDDPP exposure (10μM); (G) RHOA pathway of GSEA analysis after IDDPP exposure (10μM); (H) RAC1 pathway of GSEA analysis after IDDPP exposure (10μM); (I) Pathway summary. n=3. Indicated values were significantly different from the control value. Note: AVB3, α5β3; DMSO, dimethyl sulfoxide; EHDPP, 2-ethylhexyl-diphenyl phosphate; FAK, focal adhesion kinase pathway; GSEA, gene set enrichment analysis; IDDPP, isodecyl diphenyl phosphate; ITG, integrin; PI3K, phosphatidylinositol 3-kinase; RAC1, Ras-related C3 botulinum toxin substrate 1; RHOA, Ras homolog gene family, member A; SD, standard deviation. Created with BioRender.com.
Figures 5A to 5H are graphs titled integrin 1 pathway (2-ethylhexyl-diphenyl phosphate), focal adhesion kinase pathway (2-ethylhexyl-diphenyl phosphate), Ras-related C3 botulinum toxin substrate 1 pathway (2-ethylhexyl-diphenyl phosphate), Ras homolog gene family pathway (2-ethylhexyl-diphenyl phosphate), integrin 1 pathway (isodecyl diphenyl phosphate), focal adhesion kinase pathway (isodecyl diphenyl phosphate), Ras-related C3 botulinum toxin substrate 1 pathway (isodecyl diphenyl phosphate), Ras homolog gene family pathway (isodecyl diphenyl phosphate), plotting Enrichment Score, ranging from negative 0.6 to 0.0 in increments of 0.1; negative 0.5 to 0.1 in increments of 0.1; negative 0.5 to 0.1 in increments of 0.1; negative 0.5 to 0.1 in increments of 0.1; negative 0.6 to 0.0 in increments of 0.1; negative 0.5 to 0.1 in increments of 0.1; negative 0.5 to 0.1 in increments of 0.1; negative 0.5 to 0.1 in increments of 0.1; and Ranked list metric, ranging from negative 2 to 2 in increments of 2 (y-axis) across Rank in ordered dataset, ranging from 0 to 16000 in increments of 2000 (x-axis) for Signal 2 noise, ranging from negative 2 to 2 in unit increments; enrichment profile, hits, and ranking metric scores, respectively. Figure 5I is an illustration depicting the extracellular matrix, Syndecan-1. The integrins 1 and 3, alpha and beta 3 integrins, focal adhesion kinase pathway, and phosphatidylinositol 3-kinase lead to the Ras homolog gene family, member A, and the Ras-related C3 botulinum toxin in the cell.
By DEGseq-2 analysis, there were 354 differentially expressed genes (211 up-regulated genes and 143 down-regulated genes) between the EHDPP exposure group and the control (Figure 6A). And there were 1,349 differentially expressed genes (680 up-regulated genes and 669 down-regulated genes) between the IDDPP exposure group and the control (Figure 6B). There were 325 common differentially expressed genes in the EHDPP and IDDPP exposure groups in comparison with the control (Figure 6C). Among differentially expressed genes, there were several ITG-related genes such as dystonin (DST) (Figure 6A,B). By RT-qPCR, the gene expression of DST in the 10μM EHDPP exposure group was found to be 0.79±0.07 times as high (p<0.05) as those in the control, and gene expressions in the 1μM and 10μM IDDPP exposure groups were found to be 0.78±0.08 and 0.73±0.12 times (p<0.05) as high, respectively, as those in the control (Figure 6D,E).
Figure 6. Differential expressed gene analysis of amniotic sac embryoids exposed to EHDPP and IDDPP. (A) Differential expressed genes after EHDPP exposure by DEGseq2; (B) Differential expressed genes after IDDPP exposure by DEGseq2; (C) Number of differentially expressed genes in common after EHDPP and IDDPP exposure; (D) Relative expression (means±SDs) of DST in amniotic sac embryoids in 96 h EHDPP exposure; (E) Relative expression (means±SDs) of DST in amniotic sac embryoids in 96 h IDDPP exposure. n=3. Data in (D) and (E) were analyzed using Student’s t-test. Indicated values were significantly different from the control value. Numeric data in (D) and (E) were listed in Table S7. Note: ABCA1, ATP-binding cassette transporter A1; AKAP9, A-kinase anchoring protein 9; AR, androgen receptor; CD58, CD58 molecule; CENPK, centromere protein K; DICER1, ribonuclease type III; DMSO, dimethyl sulfoxide; DST, dystonin; EHDPP, 2-ethylhexyl-diphenyl phosphate; GLI2, GLI family zinc finger 2; GSEA, gene set enrichment analysis; HAX-1, HS1-associated protein X-1; IDDPP, isodecyl diphenyl phosphate; ITG, integrin; ITGAE, integrin subunit α E; ITGB3BP, integrin β3 binding protein; MMP16, matrix metallopeptidase 16; PCDH18, protocadherin 18; PRSS1, serine protease 1; ROCK1, Rho associated coiled-coil containing protein kinase 1; SD, standard deviation; THBS1, thrombospondin 1; VCAN, versican. *p<0.05.
Figures 6A and 6B are graphs, plotting I T G related genes, including C D 58, thrombospondin 1, dystonin, centromere protein K, Rho associated coiled-coil containing protein kinase 1, serine protease 1, protocadherin 18, matrix metallopeptidase 16, integrin subunit lowercase alpha E, androgen receptor, G L I family zinc finger 2, ITGB3BP, integrin lowercase beta 3 binding protein, ribonuclease type 3, A B C A 1, versican, H S 1-associated protein X-1, A-kinase anchoring protein 9 (y-axis) across dimethyl sulfoxide, including number 1, number 2, and number 3, and 2-ethylhexyl-diphenyl phosphate, including number 1, number 2, and number 3; dimethyl sulfoxide, including number 1, number 2, and number 3, and isodecyl-diphenyl phosphate, including number 1, number 2, and number 3 (x-axis). Figure 6C is a Ven diagram, depicting two circles. The following information is given: the circle on the left is labeled 21 cases of Count of Differential gene (2-ethylhexyl-diphenyl phosphate per dimethyl sulfoxide) and the circle on the right is labeled 817 cases of Count of Differential gene (isodecyl-diphenyl phosphate per dimethyl sulfoxide). The intersection area is labeled 325. Figures 6D and 6E are bar graphs, plotting Relative expression of 72 hours dystonin, ranging from 0.0 to 1.2 in increments of 0.3 (y-axis) across dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for 2-ethylhexyl-diphenyl phosphate and isodecyl-diphenyl phosphate.
Western blotting detected the activity of FAK, a cytoplasmic kinase downstream of ITG, and found that the relative p-FAK in the 1μM and 10μM EHDPP-treated groups were, respectively, 0.79±0.13 (p<0.05) and 0.71±0.11 (p<0.01) times as high, respectively, as those in the control group (Figure 7A,B).
Figure 7. Integrin activity evaluation of EHDPP. (A) Protein levels of FAK in the control and EHDPP exposure groups; (B) Relative protein level (means±SDs) of p-FAK/FAK; (C) p-MLC-II (green), ITGβ1 (red), and DAPI (blue) in the control and EHDPP exposure groups at 72 h; (D) Relative intensity levels (means±SDs) of p-MLC-II in the control and EHDPP exposure groups; (E) Relative intensity levels (means±SDs) of p-MLC-II (apical/basal) in the control and EHDPP exposure groups; (F) Relative intensity area (means±SDs) of p-MLC-II (apical/basal) in the control and EHDPP exposure groups; (G) Relative intensity levels (means±SDs) of ITGβ1 in the control and EHDPP exposure groups; (H) Relative binding affinity (means±SDs) of EHDPP, IDDPP, and RGD peptide. Scale bars, 20μm. Data in (B) and (D–H) were expressed relative to the levels in DMSO-treated embryoids, which were set to 1. n=3. Data were analyzed using Student’s t-test. Indicated values were significantly different from the control value. Numeric data in (B), (D), (E), (F) and (G) were listed in Table S8, numeric data in (H) was listed in Table S9. Note: DAPI, 4′,6-diamidino-2-phenylindole; EHDPP, 2-ethylhexyl-diphenyl phosphate; FAK, focal adhesion kinase pathway; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; IDDPP, isodecyl diphenyl phosphate; ITGβ1, integrin β1; p-MLC-II, phosphorylated nonmuscle myosin; RGD, Arg-Gly Asp; SD, standard deviation. *p<0.05. **p<0.01.
Figure 7A is a western blot with four columns, namely, 0, 0.1, 1, 10 under 2-ethylhexyl-diphenyl phosphate (micromolar) and three rows, namely, phosphorylated- focal adhesion kinase pathway, focal adhesion kinase pathway, and glyceraldehyde-3-phosphate dehydrogenase. Figures 7B, 7D, 7E, 7F, 7G are bar graphs, plotting Relative gray value of phosphorylated- focal adhesion kinase pathway or focal adhesion kinase pathway, ranging from 0.0 to 1.4 in increments of 0.2; Relative intensity of phosphorylated nonmuscle myosin, ranging from 0.0 to 1.5 in increments of 0.3, Relative intensity of phosphorylated nonmuscle myosin (apical or basal), ranging from 0.0 to 1.2 in increments of 0.2, Intensity area of phosphorylated nonmuscle myosin (apical or basal), ranging from 0.00 to 1.25 in increments of 0.25, Relative intensity of integrin lowercase beta 1, ranging from 0.0 to 1.5 in increments of 0.3 (y-axis) across dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for 2-ethylhexyl-diphenyl phosphate. Figure 7H is a line graph, plotting Relative integrin activity (percentage), ranging from 0 to 125 in increments of 25 (y-axis) across concentration (nanomolar), ranging as 10 begin superscript 0 end superscript, 10 begin superscript 1 end superscript, 10 begin superscript 2 end superscript, 10 begin superscript 3 end superscript, 10 begin superscript 4 end superscript, 10 begin superscript 5 end superscript (x-xis) for Arginylglycylaspartic acid peptide, 2-ethylhexyl-diphenyl phosphate and isodecyl-diphenyl phosphate. Figure 7C is a stained tissue with four columns, namely, phosphorylated nonmuscle myosin, integrin lowercase beta 1, 4′,6-diamidino-2-phenylindole, Merge and two rows, dimethyl sulfoxide and 2-ethylhexyl-diphenyl phosphate.
The phosphorylated nonmuscle myosin (p-MLC-II) expression levels after 72 h of exposure with 0.1μM and 10μM EHDPP were, respectively, 1.19±0.05 (p<0.01) and 1.32±0.08 (p<0.01) times higher than those in the control group (Figure 7C,D). The expression of p-MLC-II in the apical/basal domain of the cells in 0.1, 1, and 10μM EHDPP-treated groups were, 0.80±0.05 (p<0.05), 0.82±0.03 (p<0.05), and 0.84±0.03 (p<0.05) times as high, respectively, as those in the control group (Figure 7C,E), whereas the expression area of p-MLC-II in the apical/basal domain of the cells in 0.1, 1, and 10μM EHDPP exposure groups were, 0.59±0.05 (p<0.05), 0.48±0.02 (p<0.05), and 0.41±0.06 (p<0.05) times as high, respectively, as those in the control groups (Figure 7C,F). Similar results also occurred in the IDDPP exposure groups (Figure 8A–F). Because no significant difference in ITGβ1 expression in the EHDPP and IDDPP exposure groups were observed in comparison with the control (Figure 7C,G; Figure 8C,G), in vitro competitive binding experiments were performed to assess the affinities of EHDPP and IDDPP for the ITGβ1 protein. EHDPP and IDDPP were able to bind to ITGβ1 with half-maximal effective concentration (EC50) at concentrations of 753.2 nM and 887.5 nM, respectively, which were comparable with that of RGD peptide (370.2 nM) (Figure 7H), a positive ITGβ antagonist.
Figure 8. Integrin activity evaluation of IDDPP. (A) Protein levels of FAK in the control and IDDPP exposure groups; (B) Relative protein level (means±SDs) of FAK/β-actin; (C) p-MLC-II (green), ITGβ1 (red), and DAPI (blue) in the control and IDDPP exposure groups at 72 h; (D) Relative intensity levels (means±SDs) of p-MLC-II in the control and IDDPP exposure groups; (E) Relative intensity levels (means±SDs) of p-MLC-II (apical/basal) in the control and IDDPP exposure groups; (F) Relative intensity area (means±SDs) of pMLC-II (apical/basal) in the control and IDDPP exposure groups; (G) Relative intensity levels (means±SDs) of ITGβ1 in the control and IDDPP exposure groups. Scale bars, 20μm. Data in (B) and (D–G) were expressed relative to the levels in DMSO-treated embryoids, which were set to 1. n=3. Data were analyzed using Student’s t-test. Indicated values were significantly different from the control value. Numeric data in (B), (D), (E), (F) and (G) were listed in Table S10. Note: DAPI, 4′,6-diamidino-2-phenylindole; FAK, focal adhesion kinase pathway; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; IDDPP, isodecyl diphenyl phosphate; ITGβ1, integrin β1; p-MLC-II, phosphorylated nonmuscle myosin; SD, standard deviation. *p<0.05. **p<0.01.
Figure 8A is a western blot with four columns, namely, 0, 0.1, 1, 10 under isodecyl-diphenyl phosphate (micromolar) and three rows, namely, phosphorylated- focal adhesion kinase pathway, focal adhesion kinase pathway, and glyceraldehyde-3-phosphate dehydrogenase. Figures 8B, 8D, 8E, 8F, 8G are bar graphs, plotting Relative gray value of phosphorylated- focal adhesion kinase pathway or focal adhesion kinase pathway, ranging from 0.0 to 1.2 in increments of 0.3; Relative intensity of phosphorylated nonmuscle myosin, ranging from 0.0 to 1.5 in increments of 0.3, Relative intensity of phosphorylated nonmuscle myosin (apical or basal), ranging from 0.0 to 1.2 in increments of 0.2, Intensity area of phosphorylated nonmuscle myosin (apical or basal), ranging from 0.00 to 1.25 in increments of 0.25, Relative intensity of integrin lowercase beta 1, ranging from 0.0 to 1.5 in increments of 0.3 (y-axis) across dimethyl sulfoxide. 0.1 micromolar, 1 micromolar, and 10 micromolar (x-axis) for isodecyl-diphenyl phosphate. Figure 7C is a stained tissue with four columns, namely, phosphorylated nonmuscle myosin, integrin lowercase beta 1, 4′,6-diamidino-2-phenylindole, Merge and two rows, dimethyl sulfoxide and isodecyl-diphenyl phosphate.
After 96 h exposure of RGD peptide, the rosette-like structure did not appear or was destroyed (Figure 9A and Supplementary Movie 4–6). Although the amniotic sac formation rate did not significantly differ between the 72-h RGD peptide exposure group and the control group (Figure 9B), the rate after 96 h of exposure (38%, 14/37) was significantly lower than that of the control (86%, 32/37; p<0.05) (Figure 9C). OCT4 expression in the RGD peptide exposure group was 0.63±0.09 times as high as those in the control groups at 72 h (p<0.01; Figure 9D,E), whereas Tfap2a expression was 0.59±0.15 times as high as those in the control groups at 96 h (p<0.05; Figure 9D,F). The p-MLC-II expression level after 72 h of RGD peptide exposure was 1.31±0.02 (p<0.01) times higher than in the control (Figure 9G,H). The ratio of p-MLC-II expression in the apical/basal domain of the cells in the RGD peptide exposure group was 0.76±0.09 (p<0.05) times as high as those in the control groups (Figure 9G,I), and the ratio of the expression area for p-MLC-II in the apical/basal domain of cells in the RGD peptide exposure group was 0.40±0.14 (p<0.05) times as high as those in the control groups (Figure 9G,J), although no significant differences in ITGβ1 expression in the RGD peptide exposure groups were observed in comparison with the control (Figure 9K).
Figure 9. Amniogenesis of embryoids exposed to RGD peptide. (A) Time-lapse and phase-contrast imaging in the control and RGD peptide exposure groups during amniogenesis; (B) 72-h cavity incidence rate of the control and RGD peptide exposure groups; (C) 96-h cavity incidence rate of the control and RGD peptide exposure groups; (D) OCT4 (yellow), TFAP2A (red), and DAPI (blue) in the control and RGD peptide exposure groups; (E) Relative intensity of OCT4 (means±SDs) in the control and RGD peptide exposure groups at 72 h; (F) Intensity of TFAP2A (means±SDs) in the control and RGD peptide exposure groups at 96 h; (G) p-MLC-II (green), ITGβ1 (red), and DAPI (blue) in the control and RGD peptide exposure groups at 96 h; (H) Relative intensity levels (means±SDs) of p-MLC-II in the control and RGD peptide exposure groups; (I) Relative intensity levels (means±SDs) of p-MLC-II (apical/basal) in the control and RGD peptide exposure groups; (J) Relative intensity area (means±SDs) of pMLC-II (apical/basal) in the control and RGD peptide exposure groups; (K) Relative intensity levels (means±SDs) of ITGβ1 in the control and RGD peptide exposure groups. Scale bars, 20μm. Data in (B), (C), (E–F), and (H–K) were expressed relative to the levels in DMSO-treated embryoids, which were set to 1. n=3. Data in (B)–(C) were analyzed using a Pearson’s chi-square test. Data in (E), (F), and (H–K) were analyzed using Student’s t-test. Indicated values were significantly different from the control value. Numeric data in (E), (F), (H), (I), (J), and (K) were listed in Table S11. Note: DAPI, 4′,6-diamidino-2-phenylindole; ITGβ1, integrin β1; OCT4, octamer-binding transcription factor 4; p-MLC-II, phosphorylated nonmuscle myosin; RGD, Arg-Gly Asp; SD, standard deviation; TFAP2A, transcription factor AP-2α. *p<0.05. **p<0.01.
Figure 9A is a stained tissue with seven columns, namely, 48 hours, 56 hours, 64 hours, 72 hours, 80 hours, 88 hours, 96 hours and two rows, namely, dimethyl sulfoxide and Arginylglycylaspartic acid peptide. Figures 9B, 9C, 9E, 9F, 9H, 9I, 9J, 9K are bar graphs, plotting 72 hours cavity incidence rate, ranging from 0.0 to 1.0 in increments of 0.2; 96 hours cavity incidence rate, ranging from 0.0 to 1.0 in increments of 0.2; Relative intensity of octamer-binding transcription factor 4, ranging from 0.0 to 1.2 in increments of 0.3; Relative intensity of transcription factor A P-2 lowercase alpha, ranging from 0.0 to 1.2 in increments of 0.3; Relative Intensity of phosphorylated nonmuscle myosin, ranging from 0.0 to 1.5 in increments of 0.3; Relative Intensity of phosphorylated nonmuscle myosin (apical/basal), ranging from 0.00 to 1.25 in increments of 0.25; Intensity area of phosphorylated nonmuscle myosin (apical/basal), ranging from 0.00 to 1.25 in increments of 0.25; and Relative intensity of integrin lowercase beta 1, ranging from 0.0 to 1.5 in increments of 0.3 (y-axis) across dimethyl sulfoxide and Arginylglycylaspartic acid peptide (x-axis). Figure 9D is stained tissue with six columns, namely, octamer-binding transcription factor 4, 4′,6-diamidino-2-phenylindole, Merge, transcription factor A P-2 lowercase alpha, 4′,6-diamidino-2-phenylindole, Merge and two rows, namely, dimethyl sulfoxide and Arginylglycylaspartic acid peptide. Figure 9G is a stained tissue with four columns, namely, phosphorylated nonmuscle myosin, integrin lowercase beta 1, 4′,6-diamidino-2-phenylindole, Merge and two rows, namely, dimethyl sulfoxide and Arginylglycylaspartic acid peptide.
Discussion
A number of environmental chemicals have been linked with biochemical miscarriage by epidemiological studies,27 yet why and how these pernicious effects occur in human pregnancy remains unknown. By evaluating amniogenesis in a human amniotic sac embryoid model, we found that two environmental chemicals, EHDPP and IDDPP, impeded the cells developing to amniotic sac embryoids or broke the rosette-like structure of the amniotic sac by inhibiting the ITGβ1 pathway.
Production of OPFRs as substitutes for conventional flame retardants has rapidly increased, but their health risks remain central to the debate over the safety of their use.10,28 OPFRs have been detected in pregnant women,11,12 and have been associated with adverse pregnancy outcomes, particularly for biochemical miscarriage, in both epidemiological14 and animal studies.15 Because various OPFRs have been produced to meet industry needs, preliminary developmental toxicity screening is desirable. Our study applied a high-throughput toxicity screening based on the transcriptional activity of OCT4, a well-known pluripotency marker of preimplantation embryos,29,30 and identified five aryl-OPFRs, two alkyl-OPFRs, and one halogenated-OPFR that down-regulated its transcription, with EHDPP and IDDPP displaying the strongest inhibition of OCT4 expression. EHDPP was widely used in manufacturing, from electronic equipment to furniture and textiles, and was even permitted in food packaging.31,32 Although the use and environmental occurrence of IDDPP have not been as well reported, it has been detected in indoor house dust, in which the maximum concentration of IDDPP (37,649 ng/g) was 5-fold higher than that of EHDPP (7,388 ng/g).20
During amniogenesis, the primary lineage converts from a cluster of disordered cells into a structured epithelial amniotic sac, in which cells arrange around a shared point of apical constriction.33 Amniogenesis specification follows strikingly dissimilar mechanisms in different species of mammals. The amniotic epithelial cells are produced by peri-implantation epiblast originally in humans, whereas the amniotic membrane is formed by the folding of embryonic tissues during late gastrulation in mice, rabbits, and lower primates.34 Thus, an in vitro amniogenesis model of humans is needed. Human amniotic sac embryoids can be derived from human embryonic stem cells,22 which provided a feasible approach to assess the toxicity of chemicals in human amniogenesis. This study found that in the OPFR exposure groups, the structured epithelium did not form, because the cells failed to successfully arrange around a shared point, resulting in a rosette-like structure with a broken central cavity. In other cases, the cavity failed to expand and the embryoid remained a small pellet. The amniotic sac formation rates were lower in the OPFR-treated groups than the rates in the control, demonstrating that OPFRs disrupted amniogenesis. The concentration of EHDPP in amniotic sac embryoids exposed to EHDPP for 72 h at 1μM was comparable with that (<LOQ-873 ng/g dw; mean±SD: 44.4±129.3 ng/g dw) in human villi,12 showing our results have human relevance.
OCT4 is highly expressed in the embryonic disc during amniogenesis,16 whereas CDX2 and TFAP2A were highly expressed in squamous amniotic ectoderm.20,35 Therefore, the lower OCT4, CDX2, and TFAP2A expressions in amniotic sac embryoids exposed to OPFRs suggests that these chemicals disrupted both the embryonic disc and the squamous amniotic ectoderm in amniotic sac embryoids. Because amniogenesis plays a fundamental role in early pregnancy,16 the results of the present research implied a causal relationship between OPFRs exposure and biochemical miscarriage. These results illustrated that amniogenesis failure induced by OPFRs may contribute to severe adverse pregnancy outcomes and provide evidence of how OPFRs impel early pregnancy termination. Our previous study also observed OPFRs-induced implantation failure in mice, which was attributed to placentation-disrupting effects.15 Epidemiological studies have associated two OPFR metabolites, diphenyl phosphate (DPHP) and isopropylphenyl phenyl phosphate (ip-PPP), with the risk of biochemical miscarriage.13 Because DPHP is a common metabolic product of EHDPP, triphenyl phosphate (TPhP), and resorcinol bis(diphenyl phosphate) (RDP)36 and can be directly transferred to humans due to its use as a plasticizer in consumer products,37 it remains unclear whether biochemical miscarriage can be attributed to OPFR or DPHP itself. Nonetheless, this study offers evidence that EHDPP may be a major contributor to amniogenesis failure.
Amniogenesis is a complicated and transient process in mammals. Although multiple developmental pathways regulate this dynamic embryonic expansion, few have been thoroughly explored.38 GSEA analysis showed that several ITG-related gene sets including ITG1, ITG3, the extracellular matrix, and syndecan-139,40 were enriched after OPFRs exposure, indicating that EHDPP and IDDPP may inhibit the ITG1 pathway in an amniotic sac embryoid. ITG can induce kinase-mediated signaling via FAK and therefore activate downstream pathways, such as RHOA and RAC1.41 The downregulation of the FAK, RHOA, and RAC1 pathway gene sets and the inhibition of p-FAK in the OPFRs exposure groups further demonstrated that OPFRs inhibited ITG pathways in amniotic sac embryoids. Non-muscle myosin (MLC) was the major molecular driver in the intracellular actomyosin-dependent motility, and its activity was determined by p-MLC-II.42 It has been reported that ITG pathways induced amniogenesis failure by driving the abnormal location and accumulation of p-MLC-II.43 In the OPFRs exposure groups, p-MLC-II was highly expressed in embryoids, ectopically clustered at the basal domain of the rosette but suppressed to localize at the apical domain. Thus, OPFRs may induce amniogenesis failure by accumulating p-MLC-II protein in the basal amniotic sac. Because ITGβ1 is one of the major subtypes of ITG,41 we used a positive ITGβ1 antagonist, RGD peptide,44 and found an abnormal accumulation of p-MLC-II and amniogenesis failure similar to those observed in the OPFR exposure groups, suggesting that OPFRs may inhibit amniogenesis by disrupting ITGβ1-mediated p-MLC-II accumulation. Although no observable difference in ITGβ1 expression was found between the OPFRs exposure and control groups, competitive binding of OPFRs to ITGβ1 suggests that OPFRs may directly bind to ITGβ1 and therefore inhibit its activity. Although the potency of OPFR binding activity was weaker than that of the positive ITGβ1 antagonist, RGD peptide, the strong ITG-antagonistic activity of OPFRs was nonetheless unexpected given its design as an industrial chemical rather than a drug.
In summary, this study illustrated the adverse effects of OPFRs on human amniogenesis by inducing the ITGβ1-mediated abnormal accumulation of p-MLC-II in amniotic sac embryoids. Although the determinants of biochemical miscarriage remain unclear, our study provides evidence for the association between early-life exposure to OPFRs and biochemical miscarriage. However, this study is limited by its use of a symmetrical amniotic sac embryoid, whereas human amniogenesis is asymmetrical, with one side developing into an embryonic disc and the other forming a squamous amniotic ectoderm.22 The evolutionary formation of an ectodermal cavity was a key biochemical process during amniogenesis, allowing fetuses to live independently of an aquatic environment.34 This study nonetheless simulated a dynamic process of cavitation, which transformed embryonic discs into squamous amniotic ectoderms and demonstrated the impacts of OPFRs on amniogenesis failure in human amniotic sac embryoids. This hESC-based in vitro platform may facilitate the precise and rapid evaluation of industrial and consumer products that may induce biochemical miscarriage due to amniogenesis failure.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank S. Qin at the Core Facilities of the School of Life Sciences and thank the National Center for Protein Sciences at Peking University in Beijing, China.
This study received financial support from the National Natural Science Foundation of China (21737001, 41821005, and 22076005).
C.X. and J.H. designed the study. C.X., C.Z., Y.L., H.M., and F.W. performed the experiments and collected data. C.X., Y.W., and J.H. analyzed the data. C.X. and J.H. wrote the manuscript.
==== Refs
References
1. Practice Committee of the American Society for Reproductive Medicine. 2012. Evaluation and treatment of recurrent pregnancy loss: a committee opinion. Fertil Steril 98 (5 ):1103–1111, PMID: , 10.1016/j.fertnstert.2012.06.048.22835448
2. Rossen LM, Ahrens KA, Branum AM. 2018. Trends in risk of pregnancy loss among us women, 1990–2011. Paediatr Perinat Epidemiol 32 (1 ):19–29, PMID: , 10.1111/ppe.12417.29053188
3. Li X-L, Du D-F, Chen S-J, Zheng S-H, Lee AC, Chen Q. 2016. Trends in ectopic pregnancy, hydatidiform mole and miscarriage in the largest obstetrics and gynaecology hospital in China from 2003 to 2013. Reprod Health 13 (1 ):58, PMID: , 10.1186/s12978-016-0181-5.27206481
4. Quenby S, Gallos ID, Dhillon-Smith RK, Podesek M, Stephenson MD, Fisher J, et al. 2021. Miscarriage matters: the epidemiological, physical, psychological, and economic costs of early pregnancy loss. Lancet 397 (10285 ):1658–1667, PMID: , 10.1016/S0140-6736(21)00682-6.33915094
5. Bottomley C, Bourne T. 2009. Diagnosing miscarriage. Best Pract Res Clin Obstet Gynaecol. 23 (4 ):463–477, PMID: , 10.1016/j.bpobgyn.2009.02.004.19502113
6. Norwitz ER, Schust DJ, Fisher SJ. 2001. Mechanisms of disease – implantation and the survival of early pregnancy. N Engl J Med 345 (19 ):1400–1408, PMID: , 10.1056/NEJMra000763.11794174
7. Krieg SA, Shahine LK, Lathi RB. 2016. Environmental exposure to endocrine-disrupting chemicals and miscarriage. Fertil Steril 106 (4 ):941–947, PMID: , 10.1016/j.fertnstert.2016.06.043.27473347
8. Kumar S. 2011. Occupational, environmental and lifestyle factors associated with spontaneous abortion. Reprod Sci 18 (10 ):915–930, PMID: , 10.1177/1933719111413298.21960507
9. Guimarães MT, Cunha MG, Carvalho DP, Ribeiro TS, Martins LC, Braga ALF, et al. 2015. Influence of environmental contamination on pregnancy outcomes. Environ Sci Pollut Res Int 22 (19 ):14950–14959, PMID: , 10.1007/s11356-015-4704-3.26002366
10. Boer J, Stapleton H. 2019. Toward fire safety without chemical risk. Science 364 (6437 ):231–232, PMID: , 10.1126/science.aax2054.31000649
11. Chupeau Z, Bonvallot N, Mercier F, Le Bot B, Chevrier C, Glorennec P. 2020. Organophosphorus flame retardants: a global review of indoor contamination and human exposure in Europe and epidemiological evidence. Int J Environ Res Public Health 17 (18 ):6713, PMID: , 10.3390/ijerph17186713.32942622
12. Zhao F, Chen M, Gao F, Shen H, Hu J. 2017. Organophosphorus flame retardants in pregnant women and their transfer to chorionic villi. Environ Sci Technol 51 (11 ):6489–6497, PMID: , 10.1021/acs.est.7b01122.28516762
13. Carignan CC, Mínguez-Alarcón L, Butt CM, Williams PL, Meeker JD, Stapleton HM, et al. 2017. Urinary concentrations of organophosphate flame retardant metabolites and pregnancy outcomes among women undergoing in vitro fertilization. Environ Health Perspect 125 (8 ):087018, PMID: , 10.1289/EHP1021.28858831
14. Messerlian C, Williams PL, Mínguez-Alarcón L, Carignan CC, Ford JB, Butt CM, et al. 2018. Organophosphate flame-retardant metabolite concentrations and pregnancy loss among women conceiving with assisted reproductive technology. Fertil Steril 110 (6 ):1137–1144.e1131, PMID: , 10.1016/j.fertnstert.2018.06.045.30396558
15. Xu C, Ma H, Gao F, Zhang C, Hu W, Jia Y, et al. 2022. Screening of organophosphate flame retardants with placentation-disrupting effects in human trophoblast organoid model and characterization of adverse pregnancy outcomes in mice. Environ Health Perspect 130 (5 ):57002, PMID: , 10.1289/EHP10273.35503735
16. Gilbert SF, Barresi MJF. 2020. Developmental Biology. 11th ed. Sunderland, MA: Sinauer Associates, Inc.
17. Deglincerti A, Croft GF, Pietila LN, Zernicka-Goetz M, Siggia ED, Brivanlou AH. 2016. Self-organization of the in vitro attached human embryo. Nature 533 (7602 ):251–254, PMID: , 10.1038/nature17948.27144363
18. Luckett WP. 1975. The development of primordial and definitive amniotic cavities in early rhesus monkey and human embryos. Am J Anat 144 (2 ):149–167, PMID: , 10.1002/aja.1001440204.810017
19. Wang L, Kang Q, Jia Y, Li X, Hu J. 2021. Identification of three novel chloroalkyl organophosphate triesters in house dust using halogenation-guided nontarget screening combined with suspect screening. Environ Sci Technol 55 (4 ):2482–2490, PMID: , 10.1021/acs.est.0c07278.33502167
20. Wang L, Jia Y, Kang Q, Song W, Hu J. 2020. Nontarget discovery of 11 aryl organophosphate triesters in house dust using high-resolution mass spectrometry. Environ Sci Technol 54 (18 ):11376–11385, PMID: , 10.1021/acs.est.0c01970.32830962
21. Liu B, Chen S, Xu Y, Lyu Y, Wang J, Du Y, et al. 2021. Chemically defined and xeno-free culture condition for human extended pluripotent stem cells. Nat Commun 12 (1 ):3017, PMID: , 10.1038/s41467-021-23320-8.34021145
22. Shao Y, Taniguchi K, Townshend RF, Miki T, Gumucio DL, Fu J. 2017. A pluripotent stem cell-based model for post-implantation human amniotic sac development. Nat Commun 8 (1 ):208, PMID: , 10.1038/s41467-017-00236-w.28785084
23. Watanabe K, Ueno M, Kamiya D, Nishiyama A, Matsumura M, Wataya T, et al. 2007. A ROCK inhibitor permits survival of dissociated human embryonic stem cells. Nat Biotechnol 25 (6 ):681–686, PMID: , 10.1038/nbt1310.17529971
24. Cock PJA, Fields CJ, Goto N, Heuer ML, Rice PM. 2010. The sanger fastq file format for sequences with quality scores, and the solexa/illumina fastq variants. Nucleic Acids Res 38 (6 ):1767–1771, PMID: , 10.1093/nar/gkp1137.20015970
25. Kim D, Langmead B, Salzberg SL. 2015. Hisat: a fast spliced aligner with low memory requirements. Nat Methods 12 (4 ):357–360, PMID: , 10.1038/nmeth.3317.25751142
26. Langmead B, Trapnell C, Pop M, Salzberg SL. 2009. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol 10 (3 ):R25, PMID: , 10.1186/gb-2009-10-3-r25.19261174
27. Radke EG, Glenn BS, Braun JM, Cooper GS. 2019. Phthalate exposure and female reproductive and developmental outcomes: a systematic review of the human epidemiological evidence. Environ Int 130 :104580, PMID: , 10.1016/j.envint.2019.02.003.31351310
28. Blum A, Behl M, Birnbaum LS, Diamond ML, Phillips A, Singla V, et al. 2019. Organophosphate ester flame retardants: are they a regrettable substitution for polybrominated diphenyl ethers? Environ Sci Technol Lett 6 (11 ):638–649, PMID: , 10.1021/acs.estlett.9b00582.32494578
29. Wu G, Scholer HR. 2014. Role of oct4 in the early embryo development. Cell Regen 3 (1 ):7, PMID: , 10.1186/2045-9769-3-7.25408886
30. Nichols J, Zevnik B, Anastassiadis K, Niwa H, Klewe-Nebenius D, Chambers I, et al. 1998. Formation of pluripotent stem cells in the mammalian embryo depends on the POU transcription factor Oct4. Cell 95 (3 ):379–391, PMID: , 10.1016/S0092-8674(00)81769-9.9814708
31. Dueñas-Mas MJ, Ballesteros-Gómez A, Rubio S. 2020. Supramolecular solvent-based microextraction of aryl-phosphate flame retardants in indoor dust from houses and education buildings in Spain. Sci Total Environ 733 :139291, PMID: , 10.1016/j.scitotenv.2020.139291.32450379
32. U.S. Food and Drug Administration. Inventory of Food Contact Substances Listed in 21 CFR: diphenyl-2-Ethylhexyl Phosphate. https://www.cfsanappsexternal.fda.gov/scripts/fdcc/index.cfm?set=IndirectAdditives&id=DIPHENYLETHYLHEXYLPHOSPHATE [accessed 20 April 2022].
33. Bedzhov I, Zernicka-Goetz M. 2014. Self-organizing properties of mouse pluripotent cells initiate morphogenesis upon implantation. Cell 156 (5 ):1032–1044, PMID: , 10.1016/j.cell.2014.01.023.24529478
34. Rostovskaya M, Andrews S, Reik W, Rugg-Gunn PJ. 2022. Amniogenesis occurs in two independent waves in primates. Cell Stem Cell 178 (5 ):39–50, PMID: , 10.1016/j.stem.2022.03.014.35439430
35. Mendjan S, Mascetti VL, Ortmann D, Ortiz M, Karjosukarso DW, Ng Y, et al. 2014. NANOG and CDX2 pattern distinct subtypes of human mesoderm during exit from pluripotency. Cell Stem Cell 15 (3 ):310–325, PMID: , 10.1016/j.stem.2014.06.006.25042702
36. Van den Eede N, Maho W, Erratico C, Neels H, Covaci A. 2013. First insights in the metabolism of phosphate flame retardants and plasticizers using human liver fractions. Toxicol Lett 223 (1 ):9–15, PMID: , 10.1016/j.toxlet.2013.08.012.23994729
37. Klein D, Kessler W, Semder B, Pütz C, Lichtmannegger J, Otter R, et al. 2016. Di-(2-propylheptyl) phthalate (DPHP) and its metabolites in blood of rats upon single oral administration of DPHP. Toxicol Lett 259 :80–86, PMID: , 10.1016/j.toxlet.2016.07.025.27452279
38. Rostovskaya M, Andrews S, Reik W, Rugg-Gunn PJ. 2022. Amniogenesis occurs in two independent waves in primates. Cell Stem Cell 29 (5 ):744–759.e746, PMID: , 10.1016/j.stem.2022.03.014.35439430
39. Ley K, Rivera-Nieves J, Sandborn WJ, Shattil S. 2016. Integrin-based therapeutics: biological basis, clinical use and new drugs. Nat Rev Drug Discov 15 (3 ):173–183, PMID: , 10.1038/nrd.2015.10.26822833
40. Stepp MA, Pal-Ghosh S, Tadvalkar G, Pajoohesh-Ganji A. 2015. Syndecan-1 and its expanding list of contacts. Adv Wound Care 4 (4 ):235–249, PMID: , 10.1089/wound.2014.0555.25945286
41. Cooper J, Giancotti FG. 2019. Integrin signaling in cancer: mechanotransduction, stemness, epithelial plasticity, and therapeutic resistance. Cancer Cell 35 (3 ):347–367, PMID: , 10.1016/j.ccell.2019.01.007.30889378
42. Kazakova OA, Khapchaev AY, Ragimov AA, Salimov EL, Shirinsky VP. 2019. Western blotting-based quantitative measurement of myosin ii regulatory light chain phosphorylation in small amounts of non-muscle cells. Biochemistry (Mosc) 84 (1 ):11–19, PMID: , 10.1134/S0006297919010024.30927521
43. Molè MA, Weberling A, Fässler R, Campbell A, Fishel S, Zernicka-Goetz M. 2021. Integrin β1 coordinates survival and morphogenesis of the embryonic lineage upon implantation and pluripotency transition. Cell Rep 34 (10 ):108834, PMID: , 10.1016/j.celrep.2021.108834.33691117
44. Kapp TG, Rechenmacher F, Neubauer S, Maltsev OV, Cavalcanti-Adam EA, Zarka R, et al. 2017. A comprehensive evaluation of the activity and selectivity profile of ligands for RGD-binding integrins. Sci Rep 7 (1 ):39805, PMID: , 10.1038/srep39805.28074920
|
PMC010xxxxxx/PMC10081693.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37027337
EHP11600
10.1289/EHP11600
Research
Assessment of Nonalcoholic Fatty Liver Disease Symptoms and Gut–Liver Axis Status in Zebrafish after Exposure to Polystyrene Microplastics and Oxytetracycline, Alone and in Combination
Zhou Weishang 1 *
Shi Wei 1 *
Du Xueying 1
Han Yu 1
Tang Yu 1
Ri Sanghyok 1 2
Ju Kwangjin 1 3
Kim Tongchol 1 2
Huang Lin 1
Zhang Weixia 1
Yu Yihan 1
Tian Dandan 1
Yu Yingying 1
Chen Liangbiao 4
Wu Zhichao 4
https://orcid.org/0000-0002-1039-7801
Liu Guangxu 1
1 College of Animal Sciences, Zhejiang University, Hangzhou, P.R. China
2 College of Life Science, Kim Hyong Jik University of Education, Pyongyang, DPR Korea
3 College of Aquaculture, Wonsan Fisheries University, Wonsan, DPR Korea
4 Key Laboratory of Exploration and Utilization of Aquatic Genetic Resources, Ministry of Education, Shanghai Ocean University, Shanghai, P.R. China
Address correspondence to Guangxu Liu, College of Animal Sciences, Zhejiang University, 866 Yuhangtang Rd., Hangzhou, P.R. China 310058. Email: [email protected]
7 4 2023
4 2023
131 4 04700623 5 2022
31 10 2022
23 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Environmental pollution may give rise to the incidence and progression of nonalcoholic fatty liver disease (NAFLD), the most common cause for chronic severe liver lesions. Although knowledge of NAFLD pathogenesis is particularly important for the development of effective prevention, the relationship between NAFLD occurrence and exposure to emerging pollutants, such as microplastics (MPs) and antibiotic residues, awaits assessment.
Objectives:
This study aimed to evaluate the toxicity of MPs and antibiotic residues related to NAFLD occurrence using the zebrafish model species.
Methods:
Taking common polystyrene MPs and oxytetracycline (OTC) as representatives, typical NAFLD symptoms, including lipid accumulation, liver inflammation, and hepatic oxidative stress, were screened after 28-d exposure to environmentally realistic concentrations of MPs (0.69mg/L) and antibiotic residue (3.00μg/L). The impacts of MPs and OTC on gut health, the gut–liver axis, and hepatic lipid metabolism were also investigated to reveal potential affecting mechanisms underpinning the NAFLD symptoms observed.
Results:
Compared with the control fish, zebrafish exposed to MPs and OTC exhibited significantly higher levels of lipid accumulation, triglycerides, and cholesterol contents, as well as inflammation, in conjunction with oxidative stress in their livers. In addition, a markedly smaller proportion of Proteobacteria and higher ratios of Firmicutes/Bacteroidetes were detected by microbiome analysis of gut contents in treated samples. After the exposures, the zebrafish also experienced intestinal oxidative injury and yielded significantly fewer numbers of goblet cells. Markedly higher levels of the intestinal bacteria-sourced endotoxin lipopolysaccharide (LPS) were also detected in serum. Animals treated with MPs and OTC exhibited higher expression levels of LPS binding receptor (LBP) and downstream inflammation-related genes while also exhibiting lower activity and gene expression of lipase. Furthermore, MP-OTC coexposure generally exerted more severe effects compared with single MP or OTC exposure.
Discussion:
Our results suggested that exposure to MPs and OTC may disrupt the gut–liver axis and be associated with NAFLD occurrence. https://doi.org/10.1289/EHP11600
Supplemental Material is available online (https://doi.org/10.1289/EHP11600).
* These authors contributed equally to this work and are joint first authors of this manuscript.
The authors declare no competing interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Nonalcoholic fatty liver disease (NAFLD), characterized by excessive fat accumulation in hepatocytes, was suggested to be the most common cause of chronic liver lesions.1 Recent surveys have demonstrated that NAFLD is prevalent worldwide, specifically, ∼31.79%,2 30.45%,2 and 27.37%2 of the population in the Middle East, South America, and Asia, respectively, suffered from different degrees of NAFLD.2–3 In addition, a significantly higher prevalence rate of NAFLD (i.e., up to 68.2% in obese children compared with 2.1% in normal children in China) was detected in obese individuals.4–5 Considering the growing global epidemic of metabolic disorders, such as obesity, one group has predicted a 178% increase in liver-related deaths among individuals with NAFLD by 2030.6–7 Therefore, with no effective and specific medication currently available for NAFLD, knowledge of NAFLD pathogenesis is particularly important for the development of effective prevention.
In recent years, accumulating data have suggested that exposure to environmental pollutants could be a tangible risk factor for NAFLD incidence and progression.8–9 For example, it has been shown that exposure of mice to particulate matter ≤2.5μm in aerodynamic diameter (PM2.5) resulted in typical NAFLD symptoms, such as hepatic lipidosis, elevation of plasma triglycerides (TGs) and low-/very-low-density lipoproteins, and liver inflammation.10 Similar NAFLD-inducing impacts have also been reported for a series of other environmental pollutants, such as thiamethoxam (TMX),11 bisphenol A,12 and 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD).13 However, the potential toxicity of many emerging pollutants known to be associated with the development of NAFLD, such as microplastics (MPs) and antibiotics, remains unknown.
Owing to the massive production14 and use15 of plastics in both industry and daily life (∼350–400 million tons of plastics are produced globally every year,16 and the annual plastics consumption is predicted to surpass 1 billion tons by 205017), plastic wastes are ubiquitously present in various environments, forming an emerging pollution phenomenon—plastic pollution.18–19 Plastic pollution can be formed from the breakdown of larger pieces of plastic through weathering and biodegradation20–21 or through the manufacture of small plastics de novo22; a large proportion of environmental plastics have a diameter of <5mm and are collectively termed as MPs.23–24 Humans may be exposed to environmental MPs through multiple routes,25 for example, dermal contact,26 inhalation,27 and ingestion.28 Recently, various types of MPs have been detected in different human tissues.29–30 For instance, particles with sizes ≥700 nm of different types of plastics, including polystyrene (PS), polymethyl methacrylate (PMMA), polypropylene (PP), polyethylene (PE), and polyethylene terephthalate (PET), were detected in whole blood samples collected from 22 healthy volunteers.29 Similarly, a case study reported the detection of MPs (including PP and others that remained unidentified) with sizes ranging from 5 to 10μm in the placenta of a pregnant woman.30 Although drawing increasing public concern, the potential impact of MP exposure on individual health remains largely understudied.
In addition, characterized by their large specific surface area and hydrophobicity, MPs may adsorb other environmental pollutants, such as various antibiotics, through Van der Waals forces,31 pore filling,32 and electrostatic interaction.33 For example, the Freundlich Kf value was estimated to be 425±26 (mg/kg) (mg/L)1/n for the absorption of oxytetracycline (OTC) on the surface of PS MPs.32 In addition, indirect evidence (indicated by aggravated bioaccumulation of antibiotics in aquatic organisms in the co-presence of MPs) supporting antibiotics absorption on MPs surface was also increasingly documented.34–35 Although adsorption capacities varied among pollutants, plastic types, and environmental conditions,36 MPs may act as vectors for secondary pollutant internalization.37 This highlights the need to investigate the health risks of MPs along with other common pollutants, such as antibiotics.
Since their discovery, antibiotics have been widely used (∼34.8 billion daily defined doses in 201538) for the treatment of diseases caused by pathogenic bacteria; however, massive application has led to the ubiquitous presence of antibiotic residues in various environments worldwide.39–40 For example, it has been suggested that ∼30%–90% of OTC prescribed to patients failed to be degraded in vivo and ended up in the environment.41 According to reports, the concentrations of OTC in soil, groundwater, and surface water can reach as high as 287mg/kg (Southern China),42 237.0 ng/L (Southwestern China),43 and 361.1μg/L (Northern China),44 respectively. Currently, in addition to consuming antibiotic-contaminated water and taking medically prescribed antibiotics, we are facing a severe risk of extra-dietary antibiotic (i.e., OTC) exposure. For instance, a significant amount of OTC residues has been found in milk (196.6–206.9μg/L, India),45 eggs (421.0–568.0μg/kg, Nigeria),46 honey (89.0–92.0μg/kg, Turkey),47 and seafood (2.7–8.6μg/kg, USA).48 With potential deleterious effects, such as gut microbiota dysbiosis, environmental antibiotic residues are widely regarded as a great threat to human health.49
Owing to the gut–liver axis, gut health has been shown to be closely related to that of the liver.50–51 Previous case studies carried out in animal models (i.e., zebrafish52 and mice53) have demonstrated that intestinal exposure to MPs and antibiotics may lead to microbiota dysbiosis and physiological damages. Although it has been stated in previous studies,50,54 comprehensive and systematic investigations are urgently needed to assess the corresponding risk and detail the mechanisms of action where these effects have the potential to induce NAFLD through the gut–liver axis.
With the merits of high sequence homologies to humans and ease of manipulation, the zebrafish (Danio rerio) has been widely used as a model species to study NAFLD.55–56 In the present study, the impacts of MPs and antibiotics (taking the commonly found environmental PS MPs and OTC as representatives) on the predisposition to NAFLD were assessed in zebrafish by screening typical NAFLD symptoms, including lipid accumulation (the size and abundance of lipid droplets, and lipid composition profiling of hepatocytes), liver inflammation (the content of pro-inflammatory cytokine and expressions of inflammation-related genes), and hepatic oxidative stress [in vivo content of reactive oxygen species (ROS), degree of lipid peroxidation, and activities of antioxidant enzymes]. In addition, gut health (the microbiome of gut contents, histomorphological characteristics, and oxidative injury), the level of the intestinal bacteria-sourced endotoxin lipopolysaccharide (LPS) in the circulatory system, as well as the gene expression of the endotoxin binding receptor in hepatocytes and the lipid catabolic activity of liver (activity of hepatic lipase and its gene expression) upon MP and OTC exposure were also evaluated to elucidate the potential mechanisms underlying NAFLD symptoms.
Materials and Methods
Experimental Animals, Materials, and Chemicals
Adult zebrafish (wild type, TU strain, 4 months old), commercial PS MPs (monosphere, diameter at 490±25 nm; micrograph and physiochemical properties are provided in Figure S1 and Table S1), and standards of OTC (analytical grade, purity >95%) were purchased from FishBio Co., Ltd., Regal Nano-plastic Engineering Research Institute, and Solarbio Life Sciences (YZ130305), respectively. All experiments were approved by the animal care committee of Zhejiang University, and all methods were performed in accordance with the Guidelines for the Care and Use of Animals for Research and Teaching at Zhejiang University (ETHICS CODE permit no. ZJU20220031).
Exposure Experiments
After acclimation in dechlorinated tap water (aeration for 72 h before use) for a week, zebrafish (480 individuals in total) were randomly assigned to four experimental groups (three replicates for each experimental group and 40 zebrafish for each replicate), namely, a control group, an MP-exposure group, an OTC-exposure group, and an MP-OTC coexposure group. According to previous studies, to simulate environmentally realistic pollution scenarios for fish species, 0.69mg/L (equivalent to the MP level reported in the Miri River57) and 3.00μg/L (equivalent to the average level of OTC reported in the Yangtze River and Tai Lakes58) were adopted as the exposure concentrations for MPs and OTC, respectively.
Exposure was conducted in tanks filled with 20L of dechlorinated tap water containing the corresponding designated concentrations. Water was filtered through a 0.45-μm membrane filter before use to minimize potential waterborne MP contamination. During the 28-d exposure, the temperature and pH of the experimental water were maintained at 28.0±0.5°C and 7.1±0.2, respectively; in addition, a light cycle of 14-h light/10-h dark was adopted. Commercial food pellets (FishBio Co., Ltd.) were provided and the experimental water was renewed daily after feeding. After corresponding exposure, the zebrafish were anesthetized in 0.02% tricaine (E10521; Sigma-Aldrich) and sacrificed in ice water before obtaining the tissue specimens.
Following reported methods, the background and working concentrations of MPs and OTC in each exposure group (Table S2) were determined using high-performance liquid chromatography–mass spectrometry (HPLC/MS)59 and light microscopy,60 respectively. Briefly, 1L of water sampled from each experimental tank was filtered through a 0.7-μm glass fiber filter (GF/F; Whatman), acidified to pH 3.0 with 40% sulfuric acid (vol/vol), and then transferred into an activated solid-phase extraction column (HLB; 6mL, 500mg; Waters). After methanol elution and evaporation, pure water was added to the sample to a final volume of 1.0mL. The OTC concentration in the sample was then determined by HPLC/MS (TSQ Quantum; Thermo Scientific) at the detection limit of 11.43 ng/L. To quantify the concentration of MPs in water, 10μL of water was sampled from each experimental tank and the number of MPs in the sample was determined under a microscope (BX53; Olympus) with a hemocytometer at the magnification of 1,000×.
To verify the interactions between MPs and OTC in the experimental medium, the absorption of OTC on surface of the MPs and the effect of OTC on MP dispersing characteristics were determined with Fourier infrared spectroscopic (FT-IR) analysis61 and dynamic light scattering measurement,62 respectively (Figures S2–S4). In brief, MP water samples with and without OTC were injected into a closed liquid pool with a thickness of 1mm, and the transmittance of samples were measured with a FT-IR spectrometer (NICOLET iS50FT-IR; Thermo Scientific) in the wavenumber range of 4,000–400/cm. The particle size and zeta potential of MPs in water with and without OTC were determined with a Zetasizer (Nano ZS90; Malvern) at 25°C at the detection angle of 173° with a red laser (633 nm, 4 mW) and an avalanche photodiode detector (quantum efficiency >50% at 633 nm). In this measurement, the attenuator position was set at 4.65mm and the attenuation speed was set to be automatically based on the size and concentration of the MPs.
Histological Observation and Biochemical Profiling of Lipids in the Liver
Following methods previously reported,63 lipid accumulation in zebrafish livers was assessed microscopically using Oil Red O staining. Briefly, after the corresponding exposure, hepatic tissues dissected from six individuals in each experimental group (n=6) were fixed with 4% paraformaldehyde, washed with phosphate-buffered saline (PBS), dehydrated with sucrose solution for 6 h, and then preserved at −80°C individually. Tissue samples were embedded in optimal cutting temperature compound (OCT; 4583; SAKURA), cryosectioned into 16-μm sections using a cryostat microtome (CM 1950; Leica), and subsequently stained with Oil Red O (0.5mg Oil Red in 100mL anhydrous isopropyl alcohol and then diluted with 40% distilled water) according to standard procedures. In brief, frozen sections were washed with distilled water to remove redundant OCT and steeped with 60% isopropyl alcohol for 2 min, followed by incubation in Oil Red O solution for 5 min. After immediate rinsing with running water, samples were stained with hematoxylin for 2 min, differentiated with 1% alcohol hydrochloride for 2 s, and then washed again with distilled water. Once sealed with glycerin gelatin, the size and abundance of lipid droplets accumulated in the hepatic tissue were examined under a light microscope (Eclipse Ci-L; Nikon).
The triglyceride (TG), total cholesterol (TCHO), free fatty acid (FFA), and total bile acid (TBA) levels in zebrafish livers were determined using the corresponding commercial kits [BC0625, BC1985, and BC0595 (Solarbio) and E003-2-1 (Njjcbio), respectively].64 Briefly, livers dissected from three individuals from the same replicate of an experimental group were pooled as one sample [18 individuals and six samples (n=6) in total for each experimental group] with each parameter investigated to meet the quantity requirements of analysis. After homogenization and the addition of the extraction solution provided in kit, the sample was immediately centrifuged at 4°C for 10 min. The color development reaction was then conducted by mixing the supernatant collected with the corresponding chromogenic reagent, following the manufacturer’s instructions. The absorption values of the supernatants (wavelengths at 420, 500, 550, and 405 nm for TGs, TCHO, FFAs, and TBAs, respectively) were determined using a microplate reader (Multiskan GO; Thermo Scientific). After protein content quantification using the Bradford method (P0006; Beyotime),65 the contents of TG, TCHO, FFA, and TBA in zebrafish livers were calculated using the absorption values obtained and expressed as micromoles per milligram of protein.
Assessment of Hepatic Inflammation
The expression levels of four classic inflammation-related genes, including myeloid differentiation primary response 88 (MyD88), tumor necrosis factor (TNF) receptor–associated factor 6 (TRAF6), nuclear factor kappa-light-chain-enhancer of activated B cells p105 subunit (NFκB), and TNF-α, in zebrafish livers after the corresponding exposures were assessed using real-time polymerase chain reaction (PCR). Nine zebrafish were selected from each experimental group, and the livers of three individuals from the same replicate were pooled as one sample (n=3 for each experimental group). After RNA extraction with the EASYspin Plus RNA extraction kit (RN2802; Aidlab) and complementary DNA (cDNA) generation with PrimeScript RT Reagent (RR037A; TaKaRa), real-time PCR was performed with a Bio-Rad real-time system (CFX96; Bio-Rad) using the following program: initial denaturation at 95°C for 5 min, followed by 40 cycles of 95°C for 10 s, 60°C for 30 s, and 72°C for 30 s, and a final step of 72°C for 5 min. β-actin was used as the internal reference, and all primer information (including both sequence and accession number of corresponding gene) are provided in Table S3. All PCR primers were synthesized by Tsingke Biotechnology Co., Ltd., and relative expression levels were analyzed using the 2−ΔΔCT method.66
Hepatic inflammation status of zebrafish after the corresponding 28-d treatments was further evaluated by quantifying TNF-α, a pro-inflammatory cytokine, using western blotting.67 Three individuals were selected from each experimental group (one individual from each replicate tank and n=3 for each experimental group) and dissected on ice. Fresh individual liver tissue was homogenized and mixed with commercial radioimmunoprecipitation assay buffer (P0013B; Beyotime) supplemented with the protease inhibitor phenylmethylsulfonyl fluoride (ST506; Beyotime) and 1×complete (11697498001; Roche), followed by centrifugation at 13,800×g at 4°C for 15 min. After protein content determination as described above, the collected supernatant was mixed with sodium dodecyl sulfate (SDS) buffer (total protein/SDS=1:2), heated at 100°C for 5 min, separated on a 10% SDS–polyacrylamide gel electrophoresis system, and then electrotransferred onto a polyvinylidene fluoride membrane. The membrane was blocked with QuickBlock Western Occluder (P0252; Beyotime) at room temperature for 4 h and then immunoblotted overnight at 4°C with the primary antibodies TNF-α (1:500) and β-actin (1:1,000) (R1203-1 and R1207-1, respectively; HuaBio). After incubation with horseradish peroxide (HRP)–conjugated goat anti-rabbit immunoglobulin G (1:2,000) (HA1001; HuaBio) and exposure with a gel imager (ClinxChemiScope 3400; Clinx), TNF-α expression levels were determined using ImageJ software.68
Evaluation of Oxidative Stress in the Livers
After corresponding exposures, the in vivo ROS content in hepatic tissue was determined in situ using ROS-specific fluorescent staining.69 Following the method described above, cryosections (20-μm thick) were prepared from fresh hepatic tissues collected individually from six zebrafish in each experimental group (2 individuals from each replicate and n=6 for each experimental group). After staining with ROS-specific fluorescent dihydroethidium [DHE; dissolved in dimethyl sulfoxide (DMSO) to 2mg/mL and diluted with PBS at a ratio of 1:500] dye (D7008; Sigma) at 37°C for 30 min and three rounds of PBS washes on a decolorizing shaker (5 min each), samples were incubated with 4′,6-diamidino-2-phenylindole (DAPI; D9542; Sigma) in the dark at room temperature for 10 min. Images were then captured using a fluorescence microscope (Eclipse E100; Nikon) at excitation and emission wavelengths of 535 and 610 nm, respectively. The ROS-specific fluorescence intensity of each sample was subsequently determined using ImageJ software, according to the method reported.70
The degree of lipid peroxidation indicated by the content of its terminal product, malondialdehyde (MDA), and the activities of two antioxidant enzymes, superoxide dismutase (SOD) and catalase (CAT), in zebrafish livers were measured using the corresponding commercial kits (BC0025, BC0175, and BC0205, respectively; Solarbio). For each parameter investigated, 18 individuals were selected from each experimental group (6 of each replicate), and the livers from 3 zebrafish were pooled as one sample (n=6 in total for each experimental group). Following the protocols reported in our previous study,71 pooled samples were homogenized on ice with the extraction solution provided. The supernatant was collected by centrifugation at 8,000×g and 4°C for 10 min and used to determine the MDA content and SOD and CAT activities. For MDA content estimation, the supernatant was mixed with thiobarbituric acid at 100°C for an hour followed by centrifugation at 10,000×g for 10 min. The absorption values at three wavelengths (450, 532, and 600 nm) were then recorded with a microplate reader (Multiskan GO; Thermo Scientific). For SOD and CAT enzymatic activities, supernatant was mixed with the corresponding reaction solution (nitro-blue tetrazolium and hydrogen peroxide for SOD and CAT, respectively) for 30 min at 37°C. The absorption value was then determined at 560 and 240 nm for SOD and CAT, respectively. After protein content determination as described above, the MDA content and the SOD and CAT activities in the sample were calculated with the absorption values obtained following the manufacturer’s instructions and standardized with the protein content of the sample.
Microbiome Analysis of Gut Contents
After a 28-d exposure and a 24-h fasting period, the intestines were dissected individually from 18 zebrafish for each experimental group (6 individuals from each replicate), and the gut contents squeezed out from the 6 individuals from the same replicate were pooled as one sample (n=3 for each experimental group) to meet the minimum amount required for analysis. Total genomic DNA of each sample was then extracted using the cetyltrimethylammonium bromide (CTAB) method according to the reported protocol.72 Briefly, samples were incubated with CTAB lysis buffer [100 mM Tris-hydrochloride, 100 mM ethylenediaminetetraacetic acid (EDTA), 100 mM sodium phosphate, 1.5M sodium chloride, 1% CTAB] and protease K (10mg/mL) for 30 min at 37°C. After protease treatment and adding phenol/chloroform/isoamyl alcohol (vol:vol:vol, 25:24:1) solution, the sample was centrifuged at 12,000×g for 10 min. Chloroform/isoamyl alcohol (vol:vol, 24:1) was then added to the obtained supernatant, followed by centrifugation at 12,000×g for 10 min. Isoamyl alcohol was subsequently added to the obtained supernatant to precipitate DNA. After purification with 75% ethyl alcohol and RNA removal using RNase solution, the DNA sample was collected by centrifugation (12,000×g for 10 min) and dissolved in double-distilled water, followed by quality verification and quantity determination using electrophoresis (1% agarose gel) and a NanoDrop spectrophotometer (Thermo Scientific), respectively. The hypervariable region V4 of the bacterial 16S rRNA was amplified using specific primers 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) obtained from the Novogene Co., Ltd. Amplification was carried out in 15μL of Phusion High-Fidelity PCR Master Mix (New England Biolabs) containing 2μM of forward and reverse primers and 10 ng of template DNA using the following thermal cycles: initial denaturation at 98°C for 1 min, followed by 30 cycles of denaturation at 98°C for 10 s, annealing at 50°C for 30 s, and elongation at 72°C for 30 s, and a final step at 72°C for 5 min. The purified PCR products were then subjected to Illumina-based high-throughput sequencing (Novogene Co., Ltd.). The obtained sequencing data were submitted to the National Center for Biotechnology Information (NCBI) database under the accession number of PRJNA890774.
According to methods reported previously,73 after removing chimeric sequences, Uparse software (version 7.0.1001) was used to analyze the sequences obtained. Sequences with ≥97% similarity were assigned to the same operational taxonomic units (OTUs), and a representative sequence for each OTU was screened for further annotation using the Silva database. OTU abundance information was normalized using a standard sequence number corresponding to the sample with the least sequences. The Firmicutes/Bacteroidetes ratio (F/B ratio), a key indicator for obesity in both human74 and fish75 according to previous studies, was then calculated based on the corresponding abundance obtained.
Histopathological Examination and Oxidative Injury of the Intestines
Following the methods reported,76 intestines dissected individually from six zebrafish in each experimental group (2 individuals from each replicate and n=6 for each experimental group) were fixed with 4% paraformaldehyde for 24 h, dehydrated with ethanol (75%, 85%, 95%, and 100% ethanol, sequentially), transparentized with xylene solution, embedded in paraffin wax, and then transversely sliced into 5-μm sections with a rotary microtome (RM2016; Leica). After dewaxing and rehydration, the samples were stained with hematoxylin for 5 min and then differentiated with 0.1% hydrochloric acid ethanol. After removing extra hematoxylin with distilled water, the sample was stained with eosin for 3 min. The stained sample was dehydrated with ethanol as described above and then sealed with glycerin gelatin. After the abovementioned hematoxylin and eosin (H&E) staining, the sample was examined microscopically (Eclipse Ci-L; Nikon). The number of intestinal goblet cells, an important indicator of the intestinal barrier in zebrafish according to previous studies,76–77 was determined and used to calculate their density in the observed area.
After the corresponding 28-d exposure, intestines were dissected individually from six zebrafish in each experimental group (2 individuals from each replicate and n=6 for each experimental group) to estimate the intestinal accumulation of OTC and degree of oxidative damage. Following the methods reported in our previous study,34 the OTC concentrations in zebrafish intestines (Figure S5) were determined using HPLC/MS. Briefly, after EDTA–McIlvaine (0.1M) ultrasonic extraction at 4°C for 30 min and centrifugation at 900×g for 5 min, the supernatant collected from the intestine sample was transferred into an HLB solid-phase extraction column, followed by elution with methanol and ethyl acetate. The eluate was collected, dried with nitrogen, and then redissolved in 1mL of methanol/trifluoroacetic acid mixture (vol:vol, 1:19). After filtration through a 0.45-μm membrane, the amount of OTC in the sample was determined by HPLC-MS (ACQUITY I-Class/Xevo TQ-S; Waters). According to the methods described in the section “Evaluation of Oxidative Stress in the Livers,” the level of oxidative injury was estimated by measuring the MDA content in the intestines. In brief, after incubating the supernatant collected from the homogenized sample with thiobarbituric acid at 100°C for an hour, the absorption values (at 450, 532, and 600 nm) of the sample were determined with a microplate reader (Multiskan GO; Thermo Scientific). The MDA content in the sample was subsequently calculated with the absorption values obtained following the manufacturer’s instructions and standardized with the protein content of the sample.
Quantification of LPS Content in Serum and Expression Analysis of Its Receptor in Hepatocytes
Following the reported methods,78 the content of bacteria-sourced endotoxin LPS in zebrafish serum was measured using a commercial LPS enzyme-linked immunosorbent assay kit (JL13861; Jonln Biotechnology). To obtain sufficient serum for analysis, 54 zebrafish were used from each experimental group (18 individuals from each replicate) and blood extracts from 9 individuals from the same replicate were pooled as one sample (n=6 for each experimental group in total). After clotting at 4°C overnight and centrifugation at 1,000×g for 20 min, the obtained supernatant (serum) was incubated with HRP-labeled antibodies at 37°C for an hour, followed by 15 min of incubation (37°C) with the chromogenic reagent provided. Upon adding the stop solution, the absorption value of each sample was measured at a wavelength of 450 nm using a microplate reader (Multiskan GO; Thermo Scientific). After quantifying the protein content of each sample as described above, the LPS content in the serum was calculated using the absorption value obtained and standardized with the protein content following the manufacturer’s instructions.
Following methods described in the section “Assessment of Hepatic Inflammation,” the gene expression of LPS binding protein (LBP) in zebrafish liver (n=3 for each experimental group) after corresponding exposures was analyzed by real-time PCR using β-actin as an internal reference. Briefly, after RNA extraction with the EASYspin Plus RNA extraction kit (RN2802; Aidlab) and cDNA generation with PrimeScript RT Reagent (RR037A; TaKaRa), real-time PCR was performed with a Bio-Rad real-time system (CFX96; Bio-Rad) using the amplification program described in section “Assessment of Hepatic Inflammation” and primers described in Table S3.
Determination of Lipase Activity and Its Gene Expression in the Liver
After the corresponding 28-d treatments, the lipase activity in the zebrafish livers was determined using a commercial lipase activity kit (BC2345; Solarbio) following the methods previously reported.79 Eighteen zebrafish were selected from each experimental group (6 individuals from each replicate) and the livers of 3 individuals from the sample replicate were pooled as one sample (n=6 for each experimental group). After homogenization on ice and centrifugation at 4°C (12,000×g), the obtained supernatant was fully mixed with the reaction substrate (olive oil) at 37°C for 10 min. Once the decomposition reaction was terminated by the provided stop solution, the sample was fully mixed with copper sulfate solution for color development. The absorbance value of each supernatant obtained was then determined at 710 nm with a microplate reader (Multiskan GO; Thermo Scientific) and used to calculate the lipase activity of the sample, following the manufacturer’s instructions. One unit of lipase activity was defined as the amount of lipase that catalyzes the release of 1μmol of FFAs from olive oil per milligram of protein per minute at 37°C.
In addition, expression level of the gene encoding lipase (LIP) in the zebrafish liver (1 individual collected from each replicate and n=3 for each experimental group) after the corresponding 28-d treatments was analyzed by real-time PCR following the same method described above. Primers used for LIP expression analysis are listed in Table S3.
Statistical Analysis
All parameters were compared among the different experimental groups using one-way analysis of variance and Tukey’s post hoc tests after verification of data normality and variance homogeneity with the Shapiro–Wilk test and Levene’s test, respectively. All analyses were conducted using OriginPro (version 8.0), and statistical significance was set at p<0.05.
Results
Hepatic Lipid Accumulation in Zebrafish
Compared with control, lipid-specific Oil Red O staining results (Figure 1) demonstrated that zebrafish exposed to MPs, OTC, and MP-OTC exhibited more (Figure 1B; F3,8=41.28, p<0.05) and larger (Figure 1C; F3,8=60.11, p<0.05) lipid droplets (Lds) accumulated in their livers. In addition, markedly higher TG (Figure 2A; F3,20=38.69, p<0.05) and TCHO (Figure 2B; F3,20=18.24, p<0.05) contents, whereas lower FFA (Figure 2C; F3,20=34.40, p<0.05) and TBA (Figure 2D; F3,20=95.57, p<0.05) contents were observed in the liver samples treated with MPs, OTC, and MP-OTC. Among all experimental groups, the most and largest Lds, the highest TG and TCHO contents, and the least FFA and TBA contents were detected in MP-OTC coexposed samples (Figures 1 and 2).
Figure 1. (A) Lipid droplet staining using Oil Red O (n=6 and randomly selected images presented as representatives), (B) numbers of lipid droplets per millimeter squared (n=3), and (C) quantified area of lipid droplets to that of the whole tissue observed (n=3) in zebrafish livers after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively. The black arrows in (A) indicate lipid droplets (Lds) (magnification at 200× and scale bar: 100μm). Numbers of lipid droplets were counted manually and the percentages of lipid droplet area to that of the whole tissue observed were estimated with ImageJ in (B) and (C), respectively. The corresponding numeric data of (B) and (C) are provided in Table S4. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; SE, standard error.
Figure 1A is a stained tissue displaying two columns, namely, without oxytetracycline and oxytetracycline and two rows, namely, without microplastics, and microplastics. Figures 1B and 1C are bar graphs plotting number of lipid droplets (10 begin superscript 3 end superscript per millimeter squared), ranging from 0 to 12 in increments of 2 and percentage of lipid droplets (percentage), ranging from 0 to 15 in increments of 3 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Figure 2. The contents of (A) triglycerides (TGs), (B) total cholesterol (TCHO), (C) free fatty acids (FFAs), and (D) total bile acids (TBAs) in zebrafish livers after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively (n=6 for each experimental group). All parameters were determined using corresponding commercial kits [BC0625, BC1985, and BC0595 (Solarbio) and E003-2-1 (Njjcbio), respectively] with a microplate reader (Multiskan GO; Thermo Scientific) and standardized with the protein content of the sample. The corresponding numeric data are provided in Table S5. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; prot, protein; SE, standard error.
Figures 2A to 2D are bar graphs, plotting the content of triglycerides in liver (micromoles per milligram protein), ranging from 0.0 to 0.4 in increments of 0.1; the content of total cholesterol in liver (micromoles per milligram protein), ranging from 0.00 to 0.30 in increments 0.05; the content of free fatty acids in liver (micromoles per milligram protein), ranging from 0.00 to 1.50 in increments of 0.25; and the content of total bile acids in liver (micromoles per milligram protein), ranging from 0.00 to 0.10 in increments of 0.02 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Status of Hepatic Inflammation
Compared with the control fish, the zebrafish treated with MPs, OTC, and MP-OTC yielded significantly higher expression levels of the four classic inflammation-related genes MyD88, TRAF6, NFκB, and TNF-α in their livers (Figure 3A). Similarly, markedly higher levels of pro-inflammatory cytokine (TNF-α) were detected by western blotting in samples exposed to MPs, OTC, and MP-OTC and which were ∼1.20, 1.42, and 1.65 times that of the control, respectively (Figure 3B; F3,8=27.19, p<0.05). In addition, compared with those treated with MPs or OTC alone, animals coexposed to MPs and OTC showed significantly higher expression levels of the inflammation-related genes tested and TNF-α (Figure 3).
Figure 3. (A) Expression levels of inflammation-related genes and (B) western blot of TNF-α in zebrafish livers after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively (n=3 for each experimental group for both gene expression and western blot analyses). Gene expressions were determined by real-time PCR with a Bio-Rad real-time system (CFX96; Bio-Rad) and the relative TNF-α expression levels were quantified using ImageJ. The corresponding numeric data are provided in Table S6. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: MDA, malondialdehyde; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; PCR, polymerase chain reaction; ROS, reactive oxygen species; SE, standard error; TNF-α, tumor necrosis factor-α.
Figure 3A is a heatmap, plotting Tumor necrosis factor lowercase alpha, Nuclear factor kappa-light-chain-enhancer of activated B cells, Tumor necrosis factor receptor-associated factor 6, and Myeloid differentiation primary response 88 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis). A color scale ranges from 2 to 12 in increments of 2. Figure 3B is a set of one western blot and one bar graph. The western blot, plotting control, oxytetracycline, microplastics, and microplastics plus oxytetracycline (columns) and beta-actin with 42 K Da and Tumor necrosis factor alpha with 17 K Da (rows). The bar graph, plotting Tumor necrosis factor lowercase alpha relative expression, ranging from 0.0 to 2.0 in increments of 0.5 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Status of Oxidative Stress in Liver
DHE staining (Figure 4) demonstrated that zebrafish exposed to MPs, OTC, and MP-OTC for 28 d had significantly higher ROS-specific fluorescent intensity in their livers, ∼2.40, 3.29, and 4.23 times that of the control, respectively (Figure 4B; F3,20=133.52, p<0.01). Similarly, compared with the control, higher levels of MDA content were detected in MPs, OTC, and MP-OTC groups, ∼1.77, 2.12, and 2.46 times that of the control, respectively (Figure 4C; F3,20=39.61, p<0.05). In addition, significantly lower SOD activity was observed in MPs, OTC, and MP-OTC samples, ∼65.50%, 63.55%, and 31.96% of that of the control, respectively (Figure 5A; F3,20=17.14, p<0.05). Zebrafish of the OTC- and the MP-OTC–exposure groups also exhibited lower CAT activity in their livers, ∼96.61% and 84.40% of that of the control, respectively (Figure 5B; F3,20=108.34, p<0.05). Finally, zebrafish coexposed to MPs and OTC were shown to have markedly higher ROS-specific fluorescent intensity and MDA content, but lower activities of SOD and CAT compared with those singly treated with MPs or OTC (Figures 4 and 5).
Figure 4. (A) ROS-specific fluorescent staining, (B) quantified ROS fluorescent intensities, and (C) MDA contents in zebrafish livers after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively [n=6 for each experimental group, and randomly selected images are presented as representatives for (A)]. The hepatocyte and ROS were stained in blue (DAPI) and red (DHE) in (A), respectively (magnification at 400× and scale bar: 50μm). The ROS-specific fluorescence intensity was determined using ImageJ and the MDA content was measured with a commercial kit (BC0025; Solarbio) using a microplate reader (Multiskan GO; Thermo Scientific). The corresponding numeric data are provided in Table S7. Data (means±SEs) with different superscripts above in (B) and (C) were significantly different between groups at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: DAPI, 4′,6-diamidino-2-phenylindole; DHE, dihydroethidium; MDA, malondialdehyde; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; prot, protein; ROS, reactive oxygen species; SE, standard error.
Figure 4A is a stained tissue displaying Hepatocyte, Reactive oxygen species, and Merge (columns) and control, Microplastics, Oxytetracycline, and Microplastics plus Oxytetracycline (rows). Figures 4B and 4C are bar graphs, plotting fluorescence of intensity of reactive oxygen species, ranging from 0 to 45 in increments of 5 and malondialdehyde content in liver (micromoles per milligram protein), ranging from 0.0 to 0.6 in increments of 0.1 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Figure 5. The activities of (A) SOD and (B) CAT in zebrafish livers after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively (n=6 for each experimental group). The enzymatic activities of SOD and CAT were measured with corresponding commercial kits (BC0175 and BC0205, respectively; Solarbio) using a microplate reader (Multiskan GO; Thermo Scientific). The corresponding numeric data are provided in Table S8. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: CAT, catalase; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; prot, protein; SE, standard error; SOD, superoxide dismutase.
Figures 5A and 5B are bar graphs, plotting Superoxide dismutase activity (units per milligram protein), ranging from 0 to 100 in increments of 25 and Catalase activity (units per milligram protein), ranging from 0 to 35 in increments of 35, 35 to 50 in increments of 5 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Microbiome of Gut Contents
Significant differences in microbiome composition among the experimental groups were detected in the gut contents by sequencing analysis (Figure 6). At the phylum level, Proteobacteria was the most abundant microbial found in the gut contents of zebrafish and accounted for ∼81.45%, 78.73%, 68.91%, and 76.48% for the control, MP, OTC, and MP-OTC groups, respectively (Figure 6A). In addition, zebrafish exposed to MPs, OTC, and MP-OTC gave higher F/B ratios, ∼6.85%, 8.82, and 9.68 times that of the control, respectively (Figure 6B). In this study, the highest F/B ratio was observed in zebrafish coexposed to MPs and OTC (Figure 6B).
Figure 6. The (A) top 10 abundant microbial phyla and the (B) Firmicutes/Bacteroidetes ratios of the gut content microbiome of zebrafish after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively (n=3 for each experimental group). Microbiome of gut contents (under accession number of PRJNA890774 in NCBI database) were obtained by Illumina-based high-throughput sequencing (Novogene Co., Ltd.). Different microbial phyla are labeled with different colors and corresponding phyla names are listed on the right side in (A). Blue and orange colors in (B) indicate the relative abundance of Firmicutes and Bacteroidetes, respectively. The corresponding numeric data for (A) are provided in Table S9. The number above each data column in (B) indicates the Firmicutes/Bacteroidetes ratio for the corresponding experimental group. Note: F/B, Firmicutes/Bacteroidetes (ratio); MP-OTC, microplastics and oxytetracycline; MPs, microplastics; NCBI, National Center for Biotechnology Information; OTC, oxytetracycline.
Figure 6A is a stacked bar graph, plotting relative abundance, ranging from 0 to 1 in increments of 0.25 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis) for others, Planctomycetes, Verrucomicrobiota, Spirochaetota, unidentified underscore Bacteria, Actinobacteiota, Fusobacteriota, Bacteroidota, Cyanobacteria, Firmicutes, and Proteobacteria. Figure 6B is a stacked bar graph, plotting relative abundance (Firmicutes and Bacteroidetes ratio), ranging from 0.00 to 0.12 in increments of 0.02 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis) for Firmicutes and Bacteroidetes.
Physiological Fitness of Zebrafish Intestines
Compared with control, H&E staining of the intestine showed that zebrafish exposed to MPs, OTC, and MP-OTC harbored significantly fewer goblet cells (Figure 7A). Statistically, after 28-d treatment with MPs, OTC, and MP-OTC, the density of goblet cells was only ∼74.06%, 61.59%, and 40.72% of that of the control, respectively (Figure 7B; F3,20=20.36, p<0.05). In addition, zebrafish from MPs-, OTC-, and MP-OTC–exposure groups exhibited MDA contents in their intestines that were ∼2.12, 1.22, and 3.41 times that of the control, respectively (Figure 8; F3,20=118.15, p<0.01). Compared with those treated with MPs or OTC alone, zebrafish coexposed to MPs and OTC possessed significantly fewer goblet cells and a higher level of MDA content in their intestines (Figures 7 and 8).
Figure 7. (A) Histological images of intestines and (B) quantified intestinal goblet cell densities in zebrafish exposed to control, MPs, OTC, and MP-OTC, respectively [n=6 for both analysis for each experimental group, and randomly selected images are presented as representatives for (A)]. The images of intestines presented in (A) were stained with H&E. The magnification and scale bar were 200× and 100μm for (A), respectively. Goblet cells are indicated by black arrows in (A). The corresponding numeric data for (B) are provided in Table S10. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: H&E, hematoxylin and eosin; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; SE, standard error.
Figure 7A is a stained tissue displaying two columns, namely, without oxytetracycline and oxytetracycline and two rows, namely, without microplastics, and microplastics. Figure 7B is a bar graph, plotting intestinal goblet cell density (cells per 10 begin superscript 3 end superscript micrometer squared), ranging from 0.0 to 2.5 in increments of 0.5 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Figure 8. MDA contents in the intestines of zebrafish after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively (n=6 for each experimental group). The MDA content was measured with a commercial kit (BC0025; Solarbio) using a microplate reader (Multiskan GO; Thermo Scientific) and standardized with the protein content of the sample. The corresponding numeric data are provided in Table S11. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: MDA, malondialdehyde; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; prot, protein; SE, standard error.
Figure 8 is bar graph, plotting Malondialdehyde content in intestine (micromoles per milligram protein), ranging 0.0 to 3.0 in increments of 0.5 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Content of LPS in Serum and Expression of LBP in Liver
The LPS levels in serum of MPs-, OTC-, and MP-OTC–treated zebrafish were ∼1.15, 1.24, and 1.33 times that of the control, respectively (Figure 9A; F3,20=35.62, p<0.05). After 28-d exposure of zebrafish to MPs, OTC, and MP-OTC, the expression levels of LBP in the livers were ∼5.15-, 4.87-, and 10.64-fold that of the control, respectively (Figure 9B; p<0.05). In addition, zebrafish coexposed to MPs and OTC yielded significantly higher levels of LPS in serum and LBP expression in liver than those treated with MPs or OTC alone (Figure 9).
Figure 9. (A) LPS contents in serum (n=6 for each experimental group) and (B) expression levels of LBP in the livers (n=3 for each experimental group of zebrafish after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively). The LPS content in the sample was measured with the commercial LPS ELISA kit (JL13861; Jonln Biotechnology) using a microplate reader (Multiskan GO; Thermo Scientific), and LBP expression was determined by real-time PCR with a Bio-Rad real-time system (CFX96; Bio-Rad). The corresponding numeric data are provided in Table S12. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: ELISA, enzyme-linked immunosorbent assay; LBP, gene encoding lipopolysaccharide binding receptor; LPS, lipopolysaccharide; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; PCR, polymerase chain reaction; prot, protein; SE, standard error.
Figures 9A and 9B are bar graphs, plotting Lipopolysaccharide concentration in serum (micrograms per milligram protein), ranging from 0.000 to 0.100 in increments of 0.100, 0.100 to 0.225 in increments of 0.025 and the relative expression of gene encoding lipopolysaccharide binding receptor, ranging from 0 to 16 in increments of 4 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Hepatic Lipid Catabolic Activity
The activity of hepatic lipase in zebrafish exposed to MPs, OTC, and MP-OTC was ∼6.67%, 9.71%, and 11.87% lower than that of control, respectively (Figure 10A; F3,20=36.68, p<0.05). Although the MP-exposure group showed an LIP expression level that was similar to that of the control, significantly lower LIP expression was observed in the livers of zebrafish from the OTC- and the MP-OTC–exposure groups, ∼74.98% and 42.48% of that of the control, respectively (Figure 10B; F3,8=27.17, p<0.05). Compared with those treated with MPs or OTC alone, zebrafish coexposed to MPs and OTC revealed significantly lower lipase activity and LIP expression in their livers (Figure 10).
Figure 10. (A) Activities of lipase (n=6 for each experimental group) and (B) expressions of LIP (n=3 for each experimental group) in zebrafish livers after 28-d exposure to control, MPs, OTC, and MP-OTC, respectively. The activity of lipase was determined with the commercial lipase activity kit (BC2345; Solarbio) using a microplate reader (Multiskan GO; Thermo Scientific) and LIP expression was determined by real-time PCR with a Bio-Rad real-time system (CFX96; Bio-Rad). The corresponding numeric data are provided in Table S13. Data (means±SEs) with different superscripts above were significantly different at p<0.05 (one-way analysis of variance and Tukey’s post hoc tests). Note: LIP, gene encoding lipase; MP-OTC, microplastics and oxytetracycline; MPs, microplastics; OTC, oxytetracycline; PCR, polymerase chain reaction; prot, protein; SE, standard error.
Figures 10A and 10B are bar graphs, plotting lipase activity (units per milligram protein), ranging from 0 to 16 in increments of 16, 16 to 19 in unit increments and the relative expression of gene encoding lipase, ranging from 0.0 to 1.2 in increments of 0.2 (y-axis) across control, microplastics, oxytetracycline, and microplastics plus oxytetracycline (x-axis).
Discussion
Regardless of the great threat posed by NAFLD prevalence on human health, the potential impacts of some emerging pollutants, such as MPs and OTC, to the development of NAFLD remains largely unknown. Data obtained in the present study demonstrate that zebrafish exposed to MPs and OTC at the levels equivalent to those reported by previous surveys57–58 for 28 d exhibited a series of significant differences from control fish. This suggested the presence of NAFLD and was consistent with progression of the disease.
In addition to the most common NAFLD symptom of hepatic lipidosis [more and larger lipid droplets accumulated in hepatocytes (Figure 1), in conjunction with higher levels of TG and TCHO contents (Figure 2)], zebrafish exposed to MPs and OTC also displayed markedly higher expression levels of immune-related genes (MyD88, TRAF6, NFκB, and TNF-α), as well as TNF-α protein, in their livers (Figure 3). This supported the presence of hepatic inflammation in these animals, a typical more severe symptom that was normally detected in the severe form of NAFLD, nonalcoholic steatohepatitis (NASH).80–81 In addition, zebrafish exposed to MPs and OTC for 28 d revealed significantly higher levels of ROS and MDA (Figure 4), whereas lower activities of SOD and CAT (Figure 5) were found in their livers, suggesting the occurrence of oxidative injury (indicated by higher level of lipid peroxidation) probably due to an induction of in vivo ROS and a disruption of antioxidant enzymes. Given that high degrees of lipidosis, extensive inflammation, and pathological lesions often progress to severe liver diseases, such as hepatocyte necrosis,82 hepatocyte fibrosis,83 and liver cirrhosis,84 our results give an indication where exposure to environmental MPs and OTC could pose a considerable threat to liver health.
Based on the data obtained in this study, we hypothesize that the gut–liver axis might be the target of MPs and OTC in zebrafish with NAFLD. First, microbiome analysis of gut contents showed that zebrafish from the MPs-, OTC-, and MP-OTC–exposure groups yielded a smaller relative abundance of Proteobacteria but higher F/B ratios compared with that of the control fish (Figure 6), which the authors postulate to be a sign of gut microbiota dysbiosis. According to previous studies,74–75 a high F/B ratio might be an important indicator of obesity, a common complication of NAFLD, thus the higher F/B ratios observed in zebrafish after 28-d exposure to MPs and OTC may suggest a higher risk of obesity and NAFLD. In addition, it has been suggested that change in microbial composition of the gut may increase the release of intestinal endotoxins, such as LPS.85–86 Based on the relationship between intestinal microbial composition and LPS release reported in zebrafish,87 we reason that zebrafish exposed to MPs and OTC might have higher levels of LPS in their intestines due to the reduced abundance of the dominant microbial Proteobacteria in their gut contents.
In accordance with previous studies,76–77 our data showed that zebrafish exposed to MPs and OTC for 28 d exhibited significantly higher MDA contents and fewer goblet cells in their intestines than the control fish (Figures 7 and 8). On one hand, the higher MDA contents observed suggested the occurrence of intestinal oxidative injury in MPs- and OTC-treated zebrafish, which is believed by the authors to be the potential cause for the fewer goblet cells detected. On the other hand, given that intestinal mucus (the mucosal barrier on the intestinal surface that prevents harmful substances from entering the circulatory system) was primarily secreted by goblet cells,88–89 we hypothesize that MPs- and OTC-treated zebrafish with fewer goblet cells in their intestines may have a disrupted intestinal mucosal barrier and thus higher intestinal permeability to harmful substances, such as endotoxins.
In this study, significantly higher levels of LPS were detected in the serum of zebrafish after 28-d exposure (Figure 9A). Meanwhile, MPs- and OTC-treated zebrafish exhibited higher levels of LBP compared with the control (Figure 9B). One interpretation of this finding might be the activation of the LBP by high levels of LPS. According to the literature, once activated by the potent endotoxin LPS, a series of immune responses will be triggered,78,90 offering a probable explanation for the inflammatory symptoms detected. Specifically, we found higher levels of MyD88, NFκB, TRAF6, and TNF-α, as well as TNF-α in the livers of zebrafish exposed to MPs and OTC, suggesting that LPS binding to its receptor, LBP, may activate MyD88, triggering the downstream NFκB-signaling pathway and the release of pro-inflammatory TNF-α.
According to previous studies,63,91 both inflammation and oxidative injury of the liver induced by exposure of zebrafish to MPs and OTC could lead to hepatic lipid metabolism disorders, which may give rise to hepatic lipidosis. For instance, it has been demonstrated that lipid peroxidation caused by oxidative stress could damage hepatocytes and inhibit normal hepatic lipid metabolism in zebrafish.92 In addition, it has been shown that zebrafish lipase was sensitive to oxidative stress and inflammation.92–93 In accordance with these previous studies, our data illustrated that zebrafish exposed to MPs and OTC for 28 d produced significantly lower activity of hepatic lipase and LIP expression than that of the control fish (Figure 10). In addition, MPs- and OTC-treated zebrafish harbored markedly less FFA and TBA (catabolites for TG and TCHO, respectively) in their livers (Figure 2C,D). We believe that the observation of low levels of FFAs and TBA under high levels of TG and TCHO in the livers of MPs- and OTC-treated zebrafish indicates a significant reduction in lipid catabolism, which may partially account for the excessive accumulation of lipids in the liver. Furthermore, it has been suggested that the pro-inflammatory factor TNF-α could interfere with insulin signaling and lead to insulin resistance,94 which might be not only the primary cause of diabetes but also closely related to obesity progression95–96 and NAFLD.97–98 Thus, in addition to disrupting hepatic lipid metabolism, we believe that interfering with insulin signaling could be another route for the incidence and progression of NAFLD observed in MPs- and OTC-treated zebrafish.
Complementing previous studies carried out in other species (i.e., the thick-shell mussel99 and the blood clam71), our study found coexposure of zebrafish to MPs and OTC generally exerted more severe effects on the parameters investigated compared with single treatments. On one hand, this may be due to the summation of adverse impacts on shared common targets (i.e., almost all the physiological parameters investigated in this study) of MPs and OTC. On the other hand, according to the intestinal OTC accumulation (Figure S2) and the FT-IR results (Figure S3) of the present study and those reported previously,32,36 MPs could also absorb OTC residues from the environment (the aqueous environment in the present study). Therefore, we hypothesized that MPs may condense environmental OTC and act as vectors to facilitate the internalization of OTC, which could be another explanation for the more severe impacts detected.
Due to the frequent detection of MPs and OTC in food materials,100–101 sauces,19 and drinking water,28,58 ingestion through contaminated food and drinking water has been suggested to be the primary exposure route in humans.102–103 For example, it has been demonstrated that the average MPs in five frequently consumed fruits and vegetables (apples, pears, broccoli, lettuce, and carrots) in Catania, Italy, reached as high as 132,740 particles/g.104 In addition, based on the survey conducted in Catania, Italy,28 the estimated daily intakes of MPs through drinking bottled water were estimated to be ∼1.53×106 particles/kg-body weight (BW) per day (40.1mg/kg-BW per day) and 3.35×106 particles/kg-BW per day (87.8mg/kg-BW per day) for adults and children, respectively. Furthermore, given the exposure risk caused by consuming formula prepared in infant feeding bottles, the daily intake of MPs by infants was estimated to be ∼1.62×107 particles/day based on data collected from 48 regions globally.105 Similarly, the estimated daily intakes of OTC merely through consuming contaminated seafood was estimated to be ∼0.141–0.447μg/d per capita based on the survey results (∼2.7–8.6μg/kg OTC) of seafood in the U.S. market48 and the seafood consumption per capita (52.05g/d) of the Food and Agriculture Organization of the United Nations.106
Although both MPs and OTC potentially pose a great threat to human health, the realistic overall exposure concentrations of these two pollutants remain unclear. Therefore, it is hard to directly extrapolate the results obtained in this study to humans (the zebrafish were exposed to the environmental realistic concentrations of MPs and OTC for fish species). However, given the high genome sequence homologies between human and zebrafish,107 our results contribute significant implications for the environmental induction of human NAFLD. Considering the long-term amount of MPs and OTC that can accumulate in our body via daily ingestion of contaminated food and water, the potential NAFLD risk of these pollutants should not be overlooked.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
W.S.Z., W.S., and G.X.L. contributed to experimental design, statistical analysis and manuscript preparation. W.S.Z., W.S., and X.Y.D. carried out the experiments; Y.H., Y.T., S.H.R., K.J.J., T.C.K., L.H., W.X.Z., Y.H.Y., D.D.T., L.B.C., W.Z.C., and Y.Y.Y. contributed to experiment preparation and data analysis. G.X.L. contributed to substantive discussion of the results and revision of the manuscript.
This work was funded by the National Key R&D Program of China (no. 2018YFD0900601), the National Natural Science Foundation of China (no. 32172944), the Natural Science Foundation of Zhejiang Province (LQ21C190003), and the Open Fund of Zhejiang Key Laboratory of Exploration and Preservation of Costal Bio-resources (J2021001).
==== Refs
References
1. Pais R, Maurel T. 2021. Natural history of NAFLD. J Clin Med 10 (6 ):1161, PMID: , 10.3390/jcm10061161.33802047
2. Younossi ZM, Koenig AB, Abdelatif D, Fazel Y, Henry L, Wymer M. 2016. Global epidemiology of nonalcoholic fatty liver disease—meta-analytic assessment of prevalence, incidence, and outcomes. Hepatology 64 (1 ):73–84, PMID: , 10.1002/hep.28431.26707365
3. Younossi Z, Tacke F, Arrese M, Chander Sharma B, Mostafa I, Bugianesi E, et al. 2019. Global perspectives on nonalcoholic fatty liver disease and nonalcoholic steatohepatitis. Hepatology 69 (6 ):2672–2682, PMID: , 10.1002/hep.30251.30179269
4. Fu JF, Shi HB, Liu LR, Jiang P, Liang L, Wang CL, et al. 2011. Non-alcoholic fatty liver disease: an early mediator predicting metabolic syndrome in obese children? World J Gastroenterol 17 (6 ):735–742, PMID: , 10.3748/wjg.v17.i6.735.21390143
5. Fan JG. 2013. Epidemiology of alcoholic and nonalcoholic fatty liver disease in China. J Gastroenterol Hepatol 28 (suppl 1 ):11–17, PMID: , 10.1111/jgh.12036.23855290
6. Estes C, Razavi H, Loomba R, Younossi Z, Sanyal AJ. 2018. Modeling the epidemic of nonalcoholic fatty liver disease demonstrates an exponential increase in burden of disease. Hepatology 67 (1 ):123–133, PMID: , 10.1002/hep.29466.28802062
7. O’Neill S, O’Driscoll L. 2015. Metabolic syndrome: a closer look at the growing epidemic and its associated pathologies. Obes Rev 16 (1 ):1–12, PMID: , 10.1111/obr.12229.25407540
8. Arciello M, Gori M, Maggio R, Barbaro B, Tarocchi M, Galli A, et al. 2013. Environmental pollution: a tangible risk for NAFLD pathogenesis. Int J Mol Sci 14 (11 ):22052–22066, PMID: , 10.3390/ijms141122052.24213605
9. Rajak S, Raza S, Tewari A, Sinha RA. 2022. Environmental toxicants and NAFLD: a neglected yet significant relationship. Dig Dis Sci 67 (8 ):3497–3507, PMID: , 10.1007/s10620-021-07203-y.34383198
10. Zheng Z, Xu X, Zhang X, Wang A, Zhang C, Hüttemann M, et al. 2013. Exposure to ambient particulate matter induces a NASH-like phenotype and impairs hepatic glucose metabolism in an animal model. J Hepatol 58 (1 ):148–154, PMID: , 10.1016/j.jhep.2012.08.009.22902548
11. Yang D, Zhang X, Yue L, Hu H, Wei X, Guo Q, et al. 2021. Thiamethoxam induces nonalcoholic fatty liver disease in mice via methionine metabolism disturb via nicotinamide N-methyltransferase overexpression. Chemosphere 273 :129727, PMID: , 10.1016/j.chemosphere.2021.129727.33524747
12. Verstraete SG, Wojcicki JM, Perito ER, Rosenthal P. 2018. Bisphenol A increases risk for presumed non-alcoholic fatty liver disease in Hispanic adolescents in NHANES 2003–2010. Environ Health 17 (1 ):12, PMID: , 10.1186/s12940-018-0356-3.29391015
13. Angrish MM, Dominici CY, Zacharewski TR. 2013. TCDD-elicited effects on liver, serum, and adipose lipid composition in C57BL/6 mice. Toxicol Sci 131 (1 ):108–115, PMID: , 10.1093/toxsci/kfs277.22977169
14. Geyer R, Jambeck JR, Law KL. 2017. Production, use, and fate of all plastics ever made. Sci Adv 3 (7 ):e1700782, PMID: , 10.1126/sciadv.1700782.28776036
15. Okoffo ED, Donner E, McGrath SP, Tscharke BJ, O’Brien JW, O’Brien S, et al. 2021. Plastics in biosolids from 1950 to 2016: a function of global plastic production and consumption. Water Res 201 :117367, PMID: , 10.1016/j.watres.2021.117367.34182349
16. PlasticsEurope. 2021. Plastics—the facts 2021: an analysis of European plastics production, demand and waste data. https://plasticseurope.org/knowledge-hub/plastics-the-facts-2021/ [accessed 5 March 2022].
17. Geyer R. 2020. Chapter 2—production, use, and fate of synthetic polymers. In: Plastic Waste and Recycling: Environmental Impact, Societal Issues, Prevention, and Solutions. Letcher T, ed. New York, NY: Academic Press, 13–32.
18. Danso D, Chow J, Streit WR. 2019. Plastics: environmental and biotechnological perspectives on microbial degradation. Appl Environ Microbiol 85 (19 ):e01095-19, PMID: , 10.1128/AEM.01095-19.31324632
19. Peixoto D, Pinheiro C, Amorim J, Oliva-Teles L, Guilhermino L, Vieira MN. 2019. Microplastic pollution in commercial salt for human consumption: a review. Estuar Coast Shelf Sci 219 :161–168, 10.1016/j.ecss.2019.02.018.
20. Fan P, Yu H, Xi B, Tan W. 2022. A review on the occurrence and influence of biodegradable microplastics in soil ecosystems: are biodegradable plastics substitute or threat? Environ Int 163 :107244, PMID: , 10.1016/j.envint.2022.107244.35436719
21. Arp HPH, Kühnel D, Rummel C, MacLeod M, Potthoff A, Reichelt S, et al. 2021. Weathering plastics as a planetary boundary threat: exposure, fate, and hazards. Environ Sci Technol 55 (11 ):7246–7255, PMID: , 10.1021/acs.est.1c01512.33973471
22. Summers S, Henry T, Gutierrez T. 2018. Agglomeration of nano- and microplastic particles in seawater by autochthonous and de novo-produced sources of exopolymeric substances. Mar Pollut Bull 130 :258–267, PMID: , 10.1016/j.marpolbul.2018.03.039.29866555
23. Gouin T, Roche N, Lohmann R, Hodges G. 2011. A thermodynamic approach for assessing the environmental exposure of chemicals absorbed to microplastic. Environ Sci Technol 45 (4 ):1466–1472, PMID: , 10.1021/es1032025.21268630
24. Thompson RC, Olsen Y, Mitchell RP, Davis A, Rowland SJ, John AWG, et al. 2004. Lost at sea: where is all the plastic? Science 304 (5672 ):838, PMID: , 10.1126/science.1094559.15131299
25. Domenech J, Marcos R. 2021. Pathways of human exposure to microplastics, and estimation of the total burden. Curr Opin Food Sci 39 :144–151, 10.1016/j.cofs.2021.01.004.
26. De Falco F, Gullo MP, Gentile G, Di Pace E, Cocca M, Gelabert L, et al. 2018. Evaluation of microplastic release caused by textile washing processes of synthetic fabrics. Environ Pollut 236 :916–925, PMID: , 10.1016/j.envpol.2017.10.057.29107418
27. Liu K, Wang X, Wei N, Song Z, Li D. 2019. Accurate quantification and transport estimation of suspended atmospheric microplastics in megacities: implications for human health. Environ Int 132 :105127, PMID: , 10.1016/j.envint.2019.105127.31487610
28. Zuccarello P, Ferrante M, Cristaldi A, Copat C, Grasso A, Sangregorio D, et al. 2019. Exposure to microplastics (<10 μm) associated to plastic bottles mineral water consumption: the first quantitative study. Water Res 157 :365–371, PMID: , 10.1016/j.watres.2019.03.091.30974285
29. Leslie HA, van Velzen MJM, Brandsma SH, Vethaak AD, Garcia-Vallejo JJ, Lamoree MH. 2022. Discovery and quantification of plastic particle pollution in human blood. Environ Int 163 :107199, PMID: , 10.1016/j.envint.2022.107199.35367073
30. Ragusa A, Svelato A, Santacroce C, Catalano P, Notarstefano V, Carnevali O, et al. 2021. Plasticenta: first evidence of microplastics in human placenta. Environ Int 146 :106274, PMID: , 10.1016/j.envint.2020.106274.33395930
31. Xu B, Liu F, Brookes PC, Xu J. 2018. The sorption kinetics and isotherms of sulfamethoxazole with polyethylene microplastics. Mar Pollut Bull 131 (pt A ):191–196, PMID: , 10.1016/j.marpolbul.2018.04.027.29886936
32. Zhang H, Wang J, Zhou B, Zhou Y, Dai Z, Zhou Q, et al. 2018. Enhanced adsorption of oxytetracycline to weathered microplastic polystyrene: kinetics, isotherms and influencing factors. Environ Pollut 243 (pt B ):1550–1557, PMID: , 10.1016/j.envpol.2018.09.122.30296753
33. Yu F, Li Y, Huang G, Yang C, Chen C, Zhou T, et al. 2020. Adsorption behavior of the antibiotic levofloxacin on microplastics in the presence of different heavy metals in an aqueous solution. Chemosphere 260 :127650, PMID: , 10.1016/j.chemosphere.2020.127650.32693263
34. Zhou W, Han Y, Tang Y, Shi W, Du X, Sun S, et al. 2020. Microplastics aggravate the bioaccumulation of two waterborne veterinary antibiotics in an edible bivalve species: potential mechanisms and implications for human health. Environ Sci Technol 54 (13 ):8115–8122, PMID: , 10.1021/acs.est.0c01575.32531169
35. Guilhermino L, Vieira LR, Ribeiro D, Tavares AS, Cardoso V, Alves A, et al. 2018. Uptake and effects of the antimicrobial florfenicol, microplastics and their mixtures on freshwater exotic invasive bivalve Corbicula fluminea. Sci Total Environ 622–623 :1131–1142, PMID: , 10.1016/j.scitotenv.2017.12.020.29890582
36. Li J, Zhang K, Zhang H. 2018. Adsorption of antibiotics on microplastics. Environ Pollut 237 :460–467, PMID: , 10.1016/j.envpol.2018.02.050.29510365
37. Verdú I, González-Pleiter M, Leganés F, Rosal R, Fernández-Piñas F. 2021. Microplastics can act as vector of the biocide triclosan exerting damage to freshwater microalgae. Chemosphere 266 :129193, PMID: , 10.1016/j.chemosphere.2020.129193.33310522
38. Klein EY, Van Boeckel TP, Martinez EM, Pant S, Gandra S, Levin SA, et al. 2018. Global increase and geographic convergence in antibiotic consumption between 2000 and 2015. Proc Natl Acad Sci USA 115 (15 ):E3463–E3470, PMID: , 10.1073/pnas.1717295115.29581252
39. Robles-Jimenez LE, Aranda-Aguirre E, Castelan-Ortega OA, Shettino-Bermudez BS, Ortiz-Salinas R, Miranda M, et al. 2021. Worldwide traceability of antibiotic residues from livestock in wastewater and soil: a systematic review. Animals (Basel) 12 (1 ):60, PMID: , 10.3390/ani12010060.35011166
40. Booth A, Aga DS, Wester AL. 2020. Retrospective analysis of the global antibiotic residues that exceed the predicted no effect concentration for antimicrobial resistance in various environmental matrices. Environ Int 141 :105796, PMID: , 10.1016/j.envint.2020.105796.32422499
41. Alcock RE, Sweetman A, Jones KC. 1999. Assessment of organic contaminant fate in waste water treatment plants. I: selected compounds and physicochemical properties. Chemosphere 38 (10 ):2247–2262, PMID: , 10.1016/S0045-6535(98)00444-5.10101865
42. Li YW, Wu XL, Mo CH, Tai YP, Huang XP, Xiang L. 2011. Investigation of sulfonamide, tetracycline, and quinolone antibiotics in vegetable farmland soil in the Pearl River Delta area, southern China. J Agric Food Chem 59 (13 ):7268–7276, PMID: , 10.1021/jf1047578.21623636
43. Chen L, Lang H, Liu F, Jin S, Yan T. 2018. Presence of antibiotics in shallow ground water in the northern and southwestern regions of China. Ground Water 56 (3 ):451–457, PMID: , 10.1111/gwat.12596.28940442
44. Jiang Y, Li M, Guo C, An D, Xu J, Zhang Y, et al. 2014. Distribution and ecological risk of antibiotics in a typical effluent-receiving river (Wangyang River) in north China. Chemosphere 112 :267–274, PMID: , 10.1016/j.chemosphere.2014.04.075.25048915
45. Hebbal MA, Latha C, Menon KV, Deepa J. 2020. Occurrence of oxytetracycline residues in milk samples from Palakkad, Kerala, India. Vet World 13 (6 ):1056–1064, PMID: , 10.14202/vetworld.2020.1056-1064.32801555
46. Olatoye O, Kayode ST. 2012. Oxytetracycline residues in retail chicken eggs in Ibadan, Nigeria. Food Addit Contam Part B Surveill 5 (4 ):255–259, PMID: , 10.1080/19393210.2012.702791.24786405
47. Gunes ME, Gunes N, Cibik R. 2009. Oxytetracycline and sulphonamide residues analysis of honey samples from Southern Marmara region in Turkey. Bulg J Agric Sci 15 :163–167.
48. Done HY, Halden RU. 2015. Reconnaissance of 47 antibiotics and associated microbial risks in seafood sold in the United States. J Hazard Mater 282 :10–17, PMID: , 10.1016/j.jhazmat.2014.08.075.25449970
49. Ribeiro CFA, Silveira GGOS, Cândido ES, Cardoso MH, Espínola Carvalho CM, Franco OL. 2020. Effects of antibiotic treatment on gut microbiota and how to overcome its negative impacts on human health. ACS Infect Dis 6 (10 ):2544–2559, PMID: , 10.1021/acsinfecdis.0c00036.32786282
50. Compare D, Coccoli P, Rocco A, Nardone OM, De Maria S, Cartenì M, et al. 2012. Gut–liver axis: the impact of gut microbiota on non alcoholic fatty liver disease. Nutr Metab Cardiovasc Dis 22 (6 ):471–476, PMID: , 10.1016/j.numecd.2012.02.007.22546554
51. Ohtani N, Kawada N. 2019. Role of the gut–liver axis in liver inflammation, fibrosis, and cancer: a special focus on the gut microbiota relationship. Hepatol Commun 3 (4 ):456–470, PMID: , 10.1002/hep4.1331.30976737
52. Dong CD, Chen CW, Chen YC, Chen HH, Lee JS, Lin CH. 2020. Polystyrene microplastic particles: in vitro pulmonary toxicity assessment. J Hazard Mater 385 :121575, PMID: , 10.1016/j.jhazmat.2019.121575.31727530
53. Jiang J, Chen L, Wu S, Lv L, Liu X, Wang Q, et al. 2020. Effects of difenoconazole on hepatotoxicity, lipid metabolism and gut microbiota in zebrafish (Danio rerio). Environ Pollut 265 (pt A ):114844, PMID: , 10.1016/j.envpol.2020.114844.32480235
54. Kirpich IA, Marsano LS, McClain CJ. 2015. Gut–liver axis, nutrition, and non-alcoholic fatty liver disease. Clin Biochem 48 (13–14 ):923–930, PMID: , 10.1016/j.clinbiochem.2015.06.023.26151226
55. Goessling W, Sadler KC. 2015. Zebrafish: an important tool for liver disease research. Gastroenterology 149 (6 ):1361–1377, PMID: , 10.1053/j.gastro.2015.08.034.26319012
56. Pham DH, Zhang C, Yin C. 2017. Using zebrafish to model liver diseases—where do we stand? Curr Pathobiol Rep 5 (2 ):207–221, PMID: , 10.1007/s40139-017-0141-y.29098121
57. Liong RMY, Hadibarata T, Yuniarto A, Tang KHD, Khamidun MH. 2021. Microplastic occurrence in the water and sediment of Miri River Estuary, Borneo Island. Water Air Soil Pollut 232 (8 ):342, 10.1007/s11270-021-05297-8.
58. Qin Y, Wen Q, Ma Y, Yang C, Liu Z. 2018. Antibiotics pollution in Gonghu Bay in the period of water diversion from Yangtze River to Taihu Lake. Environ Earth Sci 77 (11 ):419, 10.1007/s12665-018-7558-4.
59. Jiang L, Hu X, Yin D, Zhang H, Yu Z. 2011. Occurrence, distribution and seasonal variation of antibiotics in the Huangpu River, Shanghai, China. Chemosphere 82 (6 ):822–828, PMID: , 10.1016/j.chemosphere.2010.11.028.21131021
60. Song YK, Hong SH, Jang M, Han GM, Rani M, Lee J, et al. 2015. A comparison of microscopic and spectroscopic identification methods for analysis of microplastics in environmental samples. Mar Pollut Bull 93 (1–2 ):202–209, PMID: , 10.1016/j.marpolbul.2015.01.015.25682567
61. Chen Y, Li J, Wang F, Yang H, Liu L. 2021. Adsorption of tetracyclines onto polyethylene microplastics: a combined study of experiment and molecular dynamics simulation. Chemosphere 265 :129133, PMID: , 10.1016/j.chemosphere.2020.129133.33276997
62. Philips LA, Ruffner DB, Cheong FC, Blusewicz JM, Kasimbeg P, Waisi B, et al. 2017. Holographic characterization of contaminants in water: differentiation of suspended particles in heterogeneous dispersions. Water Res 122 :431–439, PMID: , 10.1016/j.watres.2017.06.006.28624726
63. Guo W, Lei L, Shi X, Li R, Wang Q, Han J, et al. 2021. Nonalcoholic fatty liver disease development in zebrafish upon exposure to bis(2-ethylhexyl)-2,3,4,5-tetrabromophthalate, a novel brominated flame retardant. Environ Sci Technol 55 (10 ):6926–6935, PMID: , 10.1021/acs.est.1c01476.33938212
64. Chen L, Lam JCW, Tang L, Hu C, Liu M, Lam PKS, et al. 2020. Probiotic modulation of lipid metabolism disorders caused by perfluorobutanesulfonate pollution in zebrafish. Environ Sci Technol 54 (12 ):7494–7503, PMID: , 10.1021/acs.est.0c02345.32459962
65. Kruger NJ. 1994. The Bradford method for protein quantitation. Methods Mol Biol 32 :9–15, PMID: , 10.1385/0-89603-268-X:9.7951753
66. Livak KJ, Schmittgen TD. 2001. Analysis of relative gene expression data using real-time quantitative PCR and the 2−ΔΔCT method. Methods 25 (4 ):402–408, PMID: , 10.1006/meth.2001.1262.11846609
67. Chen J, Ng SM, Chang C, Zhang Z, Bourdon JC, Lane DP, et al. 2009. p53 isoform Δ113p53 is a p53 target gene that antagonizes p53 apoptotic activity via BclxL activation in zebrafish. Genes Dev 23 (3 ):278–290, PMID: , 10.1101/gad.1761609.19204115
68. Gallo-Oller G, Ordoñez R, Dotor J. 2018. A new background subtraction method for Western blot densitometry band quantification through image analysis software. J Immunol Methods 457 :1–5, PMID: , 10.1016/j.jim.2018.03.004.29522776
69. Carter WO, Narayanan PK, Robinson JP. 1994. Intracellular hydrogen peroxide and superoxide anion detection in endothelial cells. J Leukoc Biol 55 (2 ):253–258, PMID: , 10.1002/jlb.55.2.253.8301222
70. Qiang L, Cheng J. 2021. Exposure to polystyrene microplastics impairs gonads of zebrafish (Danio rerio). Chemosphere 263 :128161, PMID: , 10.1016/j.chemosphere.2020.128161.33297137
71. Zhou W, Tang Y, Du X, Han Y, Shi W, Sun S, et al. 2021. Fine polystyrene microplastics render immune responses more vulnerable to two veterinary antibiotics in a bivalve species. Mar Pollut Bull 164 :111995, PMID: , 10.1016/j.marpolbul.2021.111995.33493858
72. Yin Z, Gong Y, Liu Y, He Y, Yao C, Huang W, et al. 2022. Fucoidan improves growth, digestive tract maturation, and gut microbiota in large yellow croaker (Larimichthys crocea) larvae. Nutrients 14 (21 ):4504, PMID: , 10.3390/nu14214504.36364770
73. Edgar RC. 2013. UPARSE: highly accurate OTU sequences from microbial amplicon reads. Nat Methods 10 (10 ):996–998, PMID: , 10.1038/nmeth.2604.23955772
74. Magne F, Gotteland M, Gauthier L, Zazueta A, Pesoa S, Navarrete P, et al. 2020. The Firmicutes/Bacteroidetes ratio: a relevant marker of gut dysbiosis in obese patients? Nutrients 12 (5 ):1474, PMID: , 10.3390/nu12051474.32438689
75. Kan H, Zhao F, Zhang XX, Ren H, Gao S. 2015. Correlations of gut microbial community shift with hepatic damage and growth inhibition of Carassius auratus induced by pentachlorophenol exposure. Environ Sci Technol 49 (19 ):11894–11902, PMID: , 10.1021/acs.est.5b02990.26378342
76. Zhou L, Limbu SM, Shen M, Zhai W, Qiao F, He A, et al. 2018. Environmental concentrations of antibiotics impair zebrafish gut health. Environ Pollut 235 :245–254, PMID: , 10.1016/j.envpol.2017.12.073.29291524
77. Qiao R, Deng Y, Zhang S, Wolosker MB, Zhu Q, Ren H, et al. 2019. Accumulation of different shapes of microplastics initiates intestinal injury and gut microbiota dysbiosis in the gut of zebrafish. Chemosphere 236 :124334, PMID: , 10.1016/j.chemosphere.2019.07.065.31310986
78. Yang LL, Wang GQ, Yang LM, Huang ZB, Zhang WQ, Yu LZ. 2014. Endotoxin molecule lipopolysaccharide-induced zebrafish inflammation model: a novel screening method for anti-inflammatory drugs. Molecules 19 (2 ):2390–2409, PMID: , 10.3390/molecules19022390.24566310
79. Gao Y, Hou L, Wang Y, Guo S, Yuan D, Jiang Y, et al. 2020. Octreotide alleviates pancreatic damage caused by paraquat in rats by reducing inflammatory responses and oxidative stress. Environ Toxicol Pharmacol 80 :103456, PMID: , 10.1016/j.etap.2020.103456.32673753
80. Wree A, Broderick L, Canbay A, Hoffman HM, Feldstein AE. 2013. From NAFLD to NASH to cirrhosis—new insights into disease mechanisms. Nat Rev Gastroenterol Hepatol 10 (11 ):627–636, PMID: , 10.1038/nrgastro.2013.149.23958599
81. Sumida Y, Yoneda M. 2018. Current and future pharmacological therapies for NAFLD/NASH. J Gastroenterol 53 (3 ):362–376, PMID: , 10.1007/s00535-017-1415-1.29247356
82. Arrese M, Cabrera D, Kalergis AM, Feldstein AE. 2016. Innate immunity and inflammation in NAFLD/NASH. Dig Dis Sci 61 (5 ):1294–1303, PMID: , 10.1007/s10620-016-4049-x.26841783
83. Kumar S, Duan Q, Wu R, Harris EN, Su Q. 2021. Pathophysiological communication between hepatocytes and non-parenchymal cells in liver injury from NAFLD to liver fibrosis. Adv Drug Deliv Rev 176 :113869, PMID: , 10.1016/j.addr.2021.113869.34280515
84. Traussnigg S, Kienbacher C, Halilbasic E, Rechling C, Kazemi-Shirazi L, Hofer H, et al. 2015. Challenges and management of liver cirrhosis: practical issues in the therapy of patients with cirrhosis due to NAFLD and NASH. Dig Dis 33 (4 ):598–607, PMID: , 10.1159/000375353.26159280
85. Hersoug LG, Møller P, Loft S. 2016. Gut microbiota-derived lipopolysaccharide uptake and trafficking to adipose tissue: implications for inflammation and obesity. Obes Rev 17 (4 ):297–312, PMID: , 10.1111/obr.12370.26712364
86. Di Lorenzo F, De Castro C, Silipo A, Molinaro A. 2019. Lipopolysaccharide structures of Gram-negative populations in the gut microbiota and effects on host interactions. FEMS Microbiol Rev 43 (3 ):257–272, PMID: , 10.1093/femsre/fuz002.30649292
87. Luo T, Wang X, Jin Y. 2021. Low concentrations of imidacloprid exposure induced gut toxicity in adult zebrafish (Danio rerio). Comp Biochem Physiol C Toxicol Pharmacol 241 :108972, PMID: , 10.1016/j.cbpc.2020.108972.33418081
88. Bhatia S, Prabhu PN, Benefiel AC, Miller MJ, Chow J, Davis SR, et al. 2015. Galacto-oligosaccharides may directly enhance intestinal barrier function through the modulation of goblet cells. Mol Nutr Food Res 59 (3 ):566–573, PMID: , 10.1002/mnfr.201400639.25421108
89. McCauley HA, Guasch G. 2015. Three cheers for the goblet cell: maintaining homeostasis in mucosal epithelia. Trends Mol Med 21 (8 ):492–503, PMID: , 10.1016/j.molmed.2015.06.003.26144290
90. Jirillo E, Caccavo D, Magrone T, Piccigallo E, Amati L, Lembo A, et al. 2002. The role of the liver in the response to LPS: experimental and clinical findings. J Endotoxin Res 8 (5 ):319–327, PMID: , 10.1177/09680519020080050501.12537690
91. Lai Y, Zhou C, Huang P, Dong Z, Mo C, Xie L, et al. 2018. Polydatin alleviated alcoholic liver injury in zebrafish larvae through ameliorating lipid metabolism and oxidative stress. J Pharmacol Sci 138 (1 ):46–53, PMID: , 10.1016/j.jphs.2018.08.007.30245287
92. Pan YX, Luo Z, Zhuo MQ, Wei CC, Chen GH, Song YF. 2018. Oxidative stress and mitochondrial dysfunction mediated Cd-induced hepatic lipid accumulation in zebrafish Danio rerio. Aquat Toxicol 199 :12–20, PMID: , 10.1016/j.aquatox.2018.03.017.29604498
93. Zhao F, Guo M, Zhang M, Duan M, Zheng J, Liu Y, et al. 2022. Sub-lethal concentration of metamifop exposure impair gut health of zebrafish (Danio rerio). Chemosphere 303 (pt 2 ):135081, PMID: , 10.1016/j.chemosphere.2022.135081.35636611
94. Hotamisligil GS, Arner P, Caro JF, Atkinson RL, Spiegelman BM. 1995. Increased adipose tissue expression of tumor necrosis factor-alpha in human obesity and insulin resistance. J Clin Invest 95 (5 ):2409–2415, PMID: , 10.1172/JCI117936.7738205
95. Esser N, Legrand-Poels S, Piette J, Scheen AJ, Paquot N. 2014. Inflammation as a link between obesity, metabolic syndrome and type 2 diabetes. Diabetes Res Clin Pract 105 (2 ):141–150, PMID: , 10.1016/j.diabres.2014.04.006.24798950
96. Kahn SE, Hull RL, Utzschneider KM. 2006. Mechanisms linking obesity to insulin resistance and type 2 diabetes. Nature 444 (7121 ):840–846, PMID: , 10.1038/nature05482.17167471
97. Tanase DM, Gosav EM, Costea CF, Ciocoiu M, Lacatusu CM, Maranduca MA, et al. 2020. The intricate relationship between type 2 diabetes mellitus (T2DM), insulin resistance (IR), and nonalcoholic fatty liver disease (NAFLD). J Diabetes Res 2020 :3920196, PMID: , 10.1155/2020/3920196.32832560
98. Khan RS, Bril F, Cusi K, Newsome PN. 2019. Modulation of insulin resistance in nonalcoholic fatty liver disease. Hepatology 70 (2 ):711–724, PMID: , 10.1002/hep.30429.30556145
99. Han Y, Zhou W, Tang Y, Shi W, Shao Y, Ren P, et al. 2021. Microplastics aggravate the bioaccumulation of three veterinary antibiotics in the thick shell mussel Mytilus coruscus and induce synergistic immunotoxic effects. Sci Total Environ 770 :145273, PMID: , 10.1016/j.scitotenv.2021.145273.33513513
100. Smith M, Love DC, Rochman CM, Neff RA. 2018. Microplastics in seafood and the implications for human health. Curr Environ Health Rep 5 (3 ):375–386, PMID: , 10.1007/s40572-018-0206-z.30116998
101. Boonsaner M, Hawker DW. 2013. Evaluation of food chain transfer of the antibiotic oxytetracycline and human risk assessment. Chemosphere 93 (6 ):1009–1014, PMID: , 10.1016/j.chemosphere.2013.05.070.23790827
102. De-la-Torre GE. 2020. Microplastics: an emerging threat to food security and human health. J Food Sci Technol 57 (5 ):1601–1608, PMID: , 10.1007/s13197-019-04138-1.32327770
103. Wang Y, Zhao X, Zang J, Li Y, Dong X, Jiang F, et al. 2022. Estimates of dietary exposure to antibiotics among a community population in East China. Antibiotics (Basel) 11 (3 ):407, PMID: , 10.3390/antibiotics11030407.35326870
104. Oliveri Conti G, Ferrante M, Banni M, Favara C, Nicolosi I, Cristaldi A, et al. 2020. Micro- and nano-plastics in edible fruit and vegetables. The first diet risks assessment for the general population. Environ Res 187 :109677, PMID: , 10.1016/j.envres.2020.109677.32454310
105. Li D, Shi Y, Yang L, Xiao L, Kehoe DK, Gun’ko YK, et al. 2020. Microplastic release from the degradation of polypropylene feeding bottles during infant formula preparation. Nat Food 1 (11 ):746–754, 10.1038/s43016-020-00171-y.37128027
106. Food and Agriculture Organization of the United Nations. 2013. FAO Statistical Yearbook. http://www.fao.org/docrep/018/i3107e/i3107e00.htm [accessed 24 December 2021].
107. Barbazuk WB, Korf I, Kadavi C, Heyen J, Tate S, Wun E, et al. 2000. The syntenic relationship of the zebrafish and human genomes. Genome Res 10 (9 ):1351–1358, PMID: , 10.1101/gr.144700.10984453
|
PMC010xxxxxx/PMC10084927.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37036791
EHP12860
10.1289/EHP12860
Invited Perspective
Invited Perspective: Still Hazy? Air Pollution and Acute Kidney Injury
Hsu Simon 1
Bi Jianzhao 2
https://orcid.org/0000-0003-1571-7592
de Boer Ian H. 1
1 Division of Nephrology and Kidney Research Institute, Department of Medicine, University of Washington, Seattle, Washington, USA
2 Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, USA
Address correspondence to Ian H. de Boer, 325 9th Ave., Box 359606, Seattle, WA 98104 USA. Telephone: (206) 616-5403. Email: [email protected]
10 4 2023
4 2023
131 4 04130207 2 2023
22 2 2023
23 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have no conflicts of interest to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP10729
==== Body
pmcAir pollution poses prevailing and increasing threats to human health.1 Beyond its well-recognized role in respiratory and cardiovascular diseases, studies have now linked air pollution to numerous other illnesses, including chronic kidney disease (CKD).2 Defined as kidney damage or abnormal function persisting for at least 3 months,3 CKD afflicts >800 million people worldwide and is most commonly detected by routine blood testing showing a decline in glomerular filtration rate (GFR) estimated from serum creatinine concentration (eGFR) or elevated urinary albumin excretion.4 Epidemiological evidence over the last few years has associated long-term exposure to air pollutants—including fine [≤2.5μm in aerodynamic diameter (PM2.5)] and coarse particulate matter [≤10μm in aerodynamic diameter (PM10)], nitrogen dioxide (NO2), and carbon monoxide (CO)—with increased risk of incident CKD, eGFR decline, and progression to kidney failure,5–7 which requires dialysis or kidney transplantation for survival. Given that CKD is a leading cause of worldwide mortality and is usually asymptomatic until its advanced stages,4 elucidating the adverse kidney effects of air pollution and their pathogenesis may inform vital preventive strategies.
In this issue of Environmental Health Perspectives, Lee et al. extend the potential risks of long-term air pollution exposure on kidney health to include acute kidney injury (AKI).8 They report the results of a longitudinal study of >60 million Medicare beneficiaries, examining associations of long-term exposures to major air pollutants with risk of first hospitalization for AKI. They ascertained AKI status through hospital discharge diagnosis codes over 1-y periods of follow-up, which started from the participant’s entry into the cohort through the first admission with diagnosis codes for AKI, death, or the end of the study period, whichever came first. After adjustment for demographics and neighborhood-level socioeconomic status indicators, exposure to PM2.5, NO2, and summertime ozone (O3) were each associated with increased risk of first hospitalization with AKI. Notably, these associations were attenuated in analyses restricted to participants presumed to have CKD; after restriction, only exposure to PM2.5 still indicated an increased risk with AKI hospitalization, although smaller than in the overall study population.
The authors’ work draws welcome attention to the potential impact of air pollution on kidney disease, which had been relatively overlooked until the past few years. Most published studies have examined CKD,2,5–7 whereas AKI is a related but distinct condition (Figure 1), defined by an abrupt (within 48 h) and often reversible decline in GFR.9 Although prior work has associated short-term exposure to air pollution with increased risk of AKI,10 the broader causal and temporal relationships between air pollution and kidney disease remain to be defined.
Figure 1. Relationship of air pollution, kidney disease, and select risk factors. Note: AKI, acute kidney injury; CKD, chronic kidney disease.
Figure 1 is a flowchart with five steps. Step 1: Low socioeconomic status leads to air pollution, chronic kidney disease, acute kidney injury, hypertension, and diabetes. Step 2: Hypertension and diabetes lead to chronic kidney disease and acute kidney injury. Step 3: Air pollution leads to hypertension, diabetes, and chronic kidney disease. Step 4: Chronic kidney disease leads to acute kidney injury. Step 5: Acute kidney injury leads to chronic kidney disease. Does air pollution lead to acute kidney injury?
The underlying biological mechanisms by which chronic exposure to air pollution could result in AKI are not fully clear, although a few possibilities exist. First, the proposed mechanisms that link air pollution to a gradual reduction in eGFR to result in CKD2—oxidative stress, systemic inflammation, DNA damage, and atherosclerosis—may increase susceptibility for AKI. The causes of AKI itself are vast and heterogenous, from hypotension and dehydration (accounting for up to 40% of hospital-acquired AKI globally11) to drugs, infection, ischemia, inflammation, vascular injury, and outflow obstruction.9 The pathogenic effects of air pollution may thus contribute to a “multi-hit” process that overwhelms the kidneys’ functional reserve to increase the risk of AKI. Second, the effects of air pollution on AKI may be mediated by comorbid conditions (Figure 1). These include hypertension, type 1 and type 2 diabetes,12,13 and even CKD, which markedly increases the risk for hospital-acquired AKI,14 and could explain why effect estimates in the present study were attenuated among participants with CKD. Last, air pollution could have acute, direct cytotoxic effects on the kidney akin to what has been proposed for iodinated contrast.15 This mechanism of kidney injury is plausible given studies in humans and mice showing that inhaled gold nanoparticles translocated from the lungs into systemic circulation and became detectable in the urine, indicating direct contact with the vasculature and tubules of the kidneys.16
As recognized by the authors, the use of hospital discharge diagnosis codes to define AKI comes with important limitations. Although reasonably specific for adjudicated AKI events and convenient for health care research, these codes have poor sensitivity and undercount AKI incidence, including cases of AKI that occur without hospitalization.17 Moreover, diagnosis codes do not provide granularity on the cause of AKI or its timing relative to hospital admission. For instance, a participant’s AKI may a) be acquired in the community and be among the primary reasons for hospital admission, b) be absent on admission but develop quickly as a complication of the admitting diagnosis, or c) be absent on admission, but develop iatrogenically as a complication of the hospitalization, diagnostic interventions, or treatments. A direct causal role for air pollution in AKI, or increased susceptibility to AKI caused by exposure to air pollution, would be expected to have stronger relationships with AKI acquired in the community or early in the course of hospitalization, as well as with AKI caused by toxic or ischemic processes. Restriction to more specific AKI outcomes may help infer causal relationships and better quantify magnitudes of relevant associations.
Accurate assessment of exposures and potential confounders are also critical for epidemiological inferences. Exposure assessments of ambient air pollution have historically relied on measurements from regulatory air monitoring stations operated by government agencies,18 which have been sparsely deployed in urban and populated regions to determine compliance with ambient air quality standards.19 Thus, improving air pollution assessment requires data sources other than the regulatory monitors, whether they are from satellite remote sensing data (a strength of the current study), investigator-led reference monitoring campaigns, or broadly deployed low-cost sensors. On confounding, air pollution, CKD, and AKI all share a common risk factor in socioeconomic status20–22—itself a complex construct with multiple domains—and interact with other characteristics and each other (Figure 1). Disentangling the true effects of air pollution on kidney disease is thus challenging but important. For example, it may well be that unmeasured behavioral or lifestyle factors contribute to hypertension and diabetes, two leading risk factors for CKD and AKI, as well as an increase in air pollution exposure that is then interpreted to “cause” the onset of AKI.
Lee et al.8 contribute to the important and growing literature on the adverse kidney effects of air pollutants. Future studies that incorporate multiplatform data sources to improve exposure estimates, include more granular covariate data, and assess CKD and AKI concurrently and with increased granularity would help address some of the knowledge gaps posed above. Along with experimental studies aimed at identifying mechanisms of AKI, we may move closer toward a more holistic view of how air pollution affects the kidney.
==== Refs
References
1. Keswani A, Akselrod H, Anenberg SC. 2022. Health and clinical impacts of air pollution and linkages with climate change. NEJM Evidence 1 (7 ):EVIDra2200068, 10.1056/EVIDra2200068.
2. Shubham S, Kumar M, Sarma DK, Kumawat M, Verma V, Samartha RM, et al. 2022. Role of air pollution in chronic kidney disease: an update on evidence, mechanisms and mitigation strategies. Int Arch Occup Environ Health 95 (5 ):897–908, PMID: , 10.1007/s00420-021-01808-6.34716808
3. Levey AS, Eckardt KU, Tsukamoto Y, Levin A, Coresh J, Rossert J, et al. 2005. Definition and classification of chronic kidney disease: a position statement from Kidney Disease: Improving Global Outcomes (KDIGO). Kidney Int 67 (6 ):2089–2100, PMID: , 10.1111/j.1523-1755.2005.00365.x.15882252
4. Kovesdy CP. 2022. Epidemiology of chronic kidney disease: an update 2022. Kidney Int Suppl (2011) 12 (1 ):7–11, PMID: , 10.1016/j.kisu.2021.11.003.35529086
5. Bowe B, Xie Y, Li T, Yan Y, Xian H, Al-Aly Z. 2018. Particulate matter air pollution and the risk of incident CKD and progression to ESRD. J Am Soc Nephrol 29 (1 ):218–230, PMID: , 10.1681/ASN.2017030253.28935655
6. Bowe B, Xie Y, Li T, Yan Y, Xian H, Al-Aly Z. 2017. Associations of ambient coarse particulate matter, nitrogen dioxide, and carbon monoxide with the risk of kidney disease: a cohort study. Lancet Planet Health 1 (7 ):e267–e276, PMID: , 10.1016/S2542-5196(17)30117-1.29851625
7. Blum MF, Surapaneni A, Stewart JD, Liao D, Yanosky JD, Whitsel EA, et al. 2020. Particulate matter and albuminuria, glomerular filtration rate, and incident CKD. Clin J Am Soc Nephrol 15 (3 ):311–319, PMID: , 10.2215/CJN.08350719.32108020
8. Lee W, Wu X, Heo S, Kim JM, Fong KC, Son JY, et al. 2023. Air pollution and acute kidney injury in the U.S. medicare population: a longitudinal cohort study. Environ Health Perspect 131 (4 ):047008, 10.1289/EHP10729.37036790
9. Kellum JA, Romagnani P, Ashuntantang G, Ronco C, Zarbock A, Anders HJ. 2021. Acute kidney injury. Nat Rev Dis Primers 7 (1 ):52, PMID: , 10.1038/s41572-021-00284-z.34267223
10. Bi J, Barry V, Weil EJ, Chang HH, Ebelt S. 2021. Short-term exposure to fine particulate air pollution and emergency department visits for kidney diseases in the Atlanta metropolitan area. Environ Epidemiol 5 (4 ):e164, PMID: , 10.1097/EE9.0000000000000164.34414347
11. Mehta RL, Burdmann EA, Cerdá J, Feehally J, Finkelstein F, García-García G, et al. 2016. Recognition and management of acute kidney injury in the International Society of Nephrology 0by25 Global Snapshot: a multinational cross-sectional study. Lancet 387 (10032 ):2017–2025, PMID: , 10.1016/S0140-6736(16)30240-9.27086173
12. Yan YH, Chou CCK, Wang JS, Tung CL, Li YR, Lo K, et al. 2014. Subchronic effects of inhaled ambient particulate matter on glucose homeostasis and target organ damage in a type 1 diabetic rat model. Toxicol Appl Pharmacol 281 (2 ):211–220, PMID: , 10.1016/j.taap.2014.10.005.25454026
13. Afsar B, Elsurer Afsar R, Kanbay A, Covic A, Ortiz A, Kanbay M. 2019. Air pollution and kidney disease: review of current evidence. Clin Kidney J 12 (1 ):19–32, PMID: , 10.1093/ckj/sfy111.30746128
14. Hsu CY, Ordoñez JD, Chertow GM, Fan D, McCulloch CE, Go AS. 2008. The risk of acute renal failure in patients with chronic kidney disease. Kidney Int 74 (1 ):101–107, PMID: , 10.1038/ki.2008.107.18385668
15. Persson PB, Hansell P, Liss P. 2005. Pathophysiology of contrast medium-induced nephropathy. Kidney Int 68 (1 ):14–22, PMID: , 10.1111/j.1523-1755.2005.00377.x.15954892
16. Miller MR, Raftis JB, Langrish JP, McLean SG, Samutrtai P, Connell SP, et al. 2017. Inhaled nanoparticles accumulate at sites of vascular disease. ACS Nano 11 (5 ):4542–4552, PMID: , 10.1021/acsnano.6b08551.28443337
17. Vlasschaert MEO, Bejaimal SAD, Hackam DG, Quinn R, Cuerden MS, Oliver MJ, et al. 2011. Validity of administrative database coding for kidney disease: a systematic review. Am J Kidney Dis 57 (1 ):29–43, PMID: , 10.1053/j.ajkd.2010.08.031.21184918
18. Sorek-Hamer M, Chatfield R, Liu Y. 2020. Review: strategies for using satellite-based products in modeling PM2.5 and short-term pollution episodes. Environ Int 144 :106057, PMID: , 10.1016/j.envint.2020.106057.32889481
19. Hall ES, Kaushik SM, Vanderpool RW, Duvall R, Beaver M, Long R, et al. 2014. Integrating sensor monitoring technology into the current air pollution regulatory support paradigm: practical considerations. Am J Environ Eng 4 (6 ):147–154, 10.5923/j.ajee.20140406.02.
20. Bowe B, Xie Y, Yan Y, Al-Aly Z. 2019. Burden of cause-specific mortality associated with PM2.5 air pollution in the United States. JAMA Netw Open 2 (11 ):e1915834, PMID: , 10.1001/jamanetworkopen.2019.15834.31747037
21. Bello AK, Peters J, Rigby J, Rahman AA, El Nahas M. 2008. Socioeconomic status and chronic kidney disease at presentation to a renal service in the United Kingdom. Clin J Am Soc Nephrol 3 (5 ):1316–1323, PMID: , 10.2215/CJN.00680208.18579673
22. Hounkpatin HO, Fraser SDS, Johnson MJ, Harris S, Uniacke M, Roderick PJ. 2020. The association of socioeconomic status with incidence and outcomes of acute kidney injury. Clin Kidney J 13 (2 ):245–252, PMID: , 10.1093/ckj/sfz113.32297881
|
PMC010xxxxxx/PMC10084931.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37036790
EHP10729
10.1289/EHP10729
Research
Air Pollution and Acute Kidney Injury in the U.S. Medicare Population: A Longitudinal Cohort Study
https://orcid.org/0000-0001-5723-9061
Lee Whanhee 1
Wu Xiao 2
Heo Seulkee 3
Kim Joyce Mary 4
Fong Kelvin C. 3
Son Ji-Young 3
Sabath Matthew Benjamin 5
Trisovic Ana 6 7
Braun Danielle 6 7
Park Jae Yoon 8 9
Kim Yong Chul 10
Lee Jung Pyo 10 11
Schwartz Joel 12
Kim Ho 13 14
Dominici Francesca 6 7
Al-Aly Ziyad 15 16 17 18
Bell Michelle L. 3
1 School of Biomedical Convergence Engineering, Pusan National University, Yangsan, Republic of Korea
2 Department of Biostatistics, Mailman School of Public Health, Columbia University, New York, New York, USA
3 Yale School of the Environment, Yale University, New Haven, Connecticut, USA
4 Department of Environmental Medicine, College of Medicine, Ewha Womans University, Seoul, Republic of Korea
5 Faculty of Arts and Sciences Research Computing Department, Harvard University, Boston, Massachusetts, USA
6 Department of Data Science, Dana-Farber Cancer Institute, Boston, Massachusetts, USA
7 Department of Biostatistics, Harvard TH Chan School of Public Health, Boston, Massachusetts, USA
8 Department of Internal Medicine, Dongguk University Ilsan Hospital, Republic of Korea
9 Department of Internal Medicine, Dongguk University College of Medicine, Republic of Korea
10 Department of Internal Medicine, Seoul National University Hospital, Seoul, Republic of Korea
11 Department of Internal Medicine, Seoul National University Boramae Medical Center, Republic of Korea
12 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
13 Department of Public Health Science, Graduate School of Public Health, Seoul National University, Seoul, Republic of Korea
14 Institute for Sustainable Development, Graduate School of Public Health, Seoul National University, Republic of Korea
15 Nephrology Section, Medicine Service, Veterans Affairs Saint Louis Health Care System, Saint Louis, Missouri, USA
16 Department of Medicine, Washington University School of Medicine, Saint Louis, Missouri, USA
17 Division of Public Health Sciences, Department of Surgery, Washington University School of Medicine, Saint Louis, Missouri, USA
18 Institute for Public Health, Washington University School of Medicine, Saint Louis, Missouri, USA
Address correspondence to Whanhee Lee, School of Biomedical Convergence Engineering, Pusan National University, 49 Busandaehak-ro, Yangsan-si, Gyeongsangnam-do, 50612, South Korea. Telephone: (82) 51-510-8599. Email: [email protected]
10 4 2023
4 2023
131 4 04700804 12 2021
14 2 2023
23 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Recent studies have reported the association between air pollution exposure and reduced kidney function. However, it is unclear whether air pollution is associated with an increased risk of acute kidney injury (AKI).
Objectives:
To address this gap in knowledge, we investigated the effect estimates of long-term exposures to fine particulate matter [PM ≤2.5μm in aerodynamic diameter (PM2.5)], nitrogen dioxide (NO2), and ozone (O3) on the risk of first hospital admission for AKI using nationwide Medicare data.
Methods:
This nationwide population-based longitudinal cohort study included 61,300,754 beneficiaries enrolled in Medicare Part A fee-for-service (FFS) who were ≥65 years of age and resided in the continental United States from the years 2000 through 2016. We applied Cox-equivalent Poisson models to estimate the association between air pollution and first hospital admission for AKI.
Results:
Exposure to PM2.5, NO2, and O3 was associated with increased risk for first hospital admission for AKI, with hazard ratios (HRs) of 1.17 (95% CI: 1.16, 1.19) for a 5-μg/m3 increase in PM2.5, 1.12 (95% CI: 1.11, 1.13) for a 10-ppb increase in NO2, and 1.03 (95% CI: 1.02, 1.04) for a 10-ppb increase in summer-period O3 (June to September). The associations persisted at annual exposures lower than the current National Ambient Air Quality Standard.
Discussion:
This study found an association between exposures to air pollution and the risk of the first hospital admission with AKI, and this association persisted even at low concentrations of air pollution. Our findings provide beneficial implications for public health policies and air pollution guidelines to alleviate health care expenditures and the disease burden attributable to AKI. https://doi.org/10.1289/EHP10729
Supplemental Material is available online (https://doi.org/10.1289/EHP10729).
M.L.B. has grant support and membership in review panels for policy and proposal review (e.g., NIH, EPA, HEI), but not directly related to this research. All other authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Acute kidney injury (AKI), formerly known as acute renal failure, is a clinical syndrome characterized by a sudden decrease in renal excretory function.1 AKI is common (50.8 admissions per 1,000 persons in 2018) in the United State Medicare population2 and is more common among intensive care unit admissions. A previous review study reported that the incidence of AKI ranges from 20% to 50% during intensive care unit admissions, based on 51 individual studies published between 2006 and 2012.3 Furthermore, AKI is closely related to the incidence and progression of chronic kidney disease (CKD) and end-stage renal disease2,4 and is also associated with greater likelihood of long-term care, higher health care costs, and increased mortality.2,5 The incidence of dialysis-treated AKI has increased during the last decades.2,6 The cumulative 1-y incidence of death after initiation of outpatient hemodialysis for AKI treatment was >31.6% in Medicare beneficiaries (2017–2018).2
Despite its importance, studies on the effect estimates of environmental stressors on AKI are scarce. There are several biological mechanisms that can link air pollution exposure and kidney disease. Inhaled air pollution can directly lead to a decrease in renal function, oxidative stress, DNA damage in renal tissue, and AKI exacerbation.7,8 In addition, long-term exposure to fine particulate matter [PM ≤2.5μm in aerodynamic diameter (PM2.5)] and nitrogen oxide (NOx) can prematurely age blood vessels and gradually restrict blood flow to the heart and other major blood vessels over time. The long-term exposure to air pollution also gradually increases the likelihood of incident cardiovascular events, such as stroke and cardiac infarction,9 which are the major triggers of AKI.1 Long-term exposure to air pollution is also closely related to gradual deteriorations in respiratory, urinary tract, and pulmonary function that can develop into sepsis,10–12 and a previous study has shown that ∼60% of patients with septic shock developed AKI.13
The findings from the previous studies provide biological plausibility supporting the hypothesis of an association between air pollution and AKI. Therefore, in this study, we aimed to estimate the association between air pollution and the first hospital admissions associated with AKI, using a nationwide longitudinal cohort study covering >61 million Medicare Part A FFS beneficiaries from 2000 through 2016, with 451.3 million person-years of follow-up.
Materials and Methods
Ethics Considerations
This study was conducted under a protocol approved by the Yale Institutional Review Board. The need to obtain informed consent was waived because this study used existing anonymous data sources. This study followed the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guideline (see the Supplemental Material, “A. STROBE Statement”).
Data Sharing Information
Data were collected by the Yale–Harvard Medicare collaboration under a data user agreement with Centers for Medicare and Medicaid Services (CMS). The data cannot be made publicly available.
Study Design and Participants
We constructed a longitudinal Medicare cohort that included beneficiaries who were enrolled in Medicare Part A FFS Medicare (≥65 years of age) in the contiguous United States from 1 January 2000 to 31 December 2016. The Medicare inpatient hospital claims were obtained from the Medicare Provider and Analysis Review (MedPAR) files that contain one record per hospital admission. For each beneficiary, we extracted the first admission date for AKI, age, sex, race, ZIP code of residence, and Medicaid eligibility (as a proxy for low socioeconomic status; hereafter “dual” refers to Medicaid beneficiaries who were also eligible for Medicaid, and “non-dual,” to Medicare beneficiaries who were not eligible for Medicaid)14 in each follow-up year. Beneficiaries become eligible to enter Medicare when they turn 65 years of age. For our study, the follow-up for each beneficiary started on 1 January 2000, or 1 January of the year following their entry into the cohort, and individuals were followed-up through the first admission with diagnosis codes for AKI, death, or the end of the study period, whichever came first. We used primary or secondary discharge diagnosis codes categorized according to the International Statistical Classification of Diseases, Ninth Revision15 (ICD 9; WHO 1978) or the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision16 (ICD-10; WHO 2016), namely, ICD-9 code 584; ICD-10 code N17 to address any inpatient hospital admissions with any AKI onsets. We used the first 10 diagnosis fields to define the primary (the first diagnosis field) and secondary (as the diseases from the 2nd to the 10th diagnosis fields) diagnosis codes for AKI. The directed acyclic graph (DAG) for this study is displayed in the Supplemental Material, “B. DAG for this study.”
Air Pollution Data
A nationwide air pollution data set including high-resolution PM2.5, NO2, and ozone (O3) estimates obtained from prediction models was used. Specifically, predicted daily concentrations of ambient levels of PM2.5, NO2, and O3 (daily maximum 8-h O3) at 1-km2 spatial resolution across the contiguous United States were obtained from well-validated published models.17–19 These predictions were estimated using hybrid ensemble models incorporating random forest, gradient boosting, and neural networks. Multiple predictor variables from monitoring data, satellite data, meteorological conditions, land-use variables, and chemical transport models were used. The technical details of the prediction models have been previously published17–20 with excellent performance: cross-validated R2s of 0.89 for annual PM2.5, 0.84 for annual NO2, and 0.88 for summer-season O3 across the continental United States. Daily concentration predictions at 1 km2 were aggregated to each ZIP code by averaging the predictions at grid cells with centroid points inside the boundary of that ZIP code.14,20,21 For each calendar year, we assigned the annual (PM2.5 and NO2) and summer-season (June to September: O3) ZIP code-level average concentrations to Medicare Part A enrollees according to their residential ZIP code as the main exposures.14,20–22 Ground-level O3 is formed by chemical reactions from precursor pollutants, such as fuel combustion, road transport, and vegetation, and the reactions are catalyzed by heat and sunlight. Thus, the summer-season O3 was used in this study to assess the long-term effect estimates of O3.22,23 To examine the correlation among pollutants, we calculated Pearson’s correlation coefficients.
Confounders
We considered confounding variables that could affect the associations between long-term exposure to air pollution and hospital admissions for AKI. First, we collected four individual-level variables from Medicare files: age at cohort entry in 2-y categories (65–66, 67–68, …, ≥85 y old), sex, race [self-reported; White, Black, and other (Asian, Hispanic, American Indian or Alaskan Native, and unknown)],14,21 and Medicaid eligibility. This racial categorization was for stable statistical estimation, and further divisions of race were not possible owing to the structure of the data provided from the CMS.
In addition, we collected 10 neighborhood-level socioeconomic status indicators that have been associated with both kidney disease and air pollution exposures,24–26 that is, eight ZIP code-level indicators and two county-level indicators, as well as indicator variables indicating geographical regions. The eight indicators available at ZIP Code Tabulation Areas (ZCTA) level were derived from the 2000 U.S. Census, the 2010 U.S. Census (https://www.census.gov/), and the American Community Survey (https://www.census.gov/programs-surveys/acs) from 2005–2016. If indicators were missing for a year, we linearly interpolated or extrapolated their values using available data. The ZCTA indicators included the percentage of the population below the poverty level, population density (persons per kilometer squared), median home value (in USD), percentage of the population that is Black, percentage of the population that is Hispanic, median household income (in USD), percentage of homes with owner-occupied housing, and percentage of the population without a high school education. These ZCTA data were matched to ZIP code. In addition, two county-level indicators [average body mass index (BMI) and percentage of the population that had ever smoked] were collected from the Behavioral Risk Factor Surveillance System (BRFSS; https://www.cdc.gov/brfss/) for the period of 2000–2016. These county-level indicators were matched to ZIP code if the ZIP code centroids fell within the country boundary. Finally, we included indicator variables for the region (Northeast, Southeast, Midwest, Southwest, and West) and calendar year in the main model to adjust for potential residual confounding by spatial and temporal trends.
Statistical Analysis
We used a Cox-equivalent reparameterized Poisson model to address the computational challenges of the conventional Cox proportional hazard model.21 This Poisson model is mathematically identical to a time-varying Cox hazard model under an Anderson–Gill representation.21,27 Specifically, for each pollutant (i.e., single-pollutant model), a stratified Poisson model was fit to estimate the association between time-varying (current year) annual air pollution and the first hospital admissions with AKI diagnosis code. The dependent variable was the count of the first hospital admissions in each follow-up year, (time-varying) calendar year, and ZIP code within strata specified by individual-level variables, namely: age at study entry in 2-y categories (65–66, 67–68, …, ≥85 y old), sex, race, and Medicaid eligibility. The total person-time of Medicare Part A FFS beneficiaries within each stratum was used as the offset. We adjusted for neighborhood-level indicators as covariates and indicator variables for region in the model. This study performed a complete-case analysis and did not consider missing-data imputation. The m-out-of-n bootstrap method with ZIP code units was applied to calculate empirical confidence intervals (CIs).28 The mathematical equation on the equivalence between the Cox proportional hazard model and the stratified Poisson model that we used is as follows: (1) hc,z(a,t)=h0c(a)exp(β1Wz,t+β2Cz,t),
where hc,z(a,t) indicates the hazard of hospitalization at follow-up year a, calendar year t, and ZIP code z for individual-level strata c (age group, sex, race, and Medicaid eligibility), and h0c(a) is a baseline hazard function. Wz,t denotes the annual average air pollutants in ZIP code z in year t. Cz,t indicates time-varying covariates. Model 1 can be written as follows: (2) E(Ya,tc,z)Ta,tc,z=h0c(a)exp(β1Wz,t+β2Cz,t),
where E(Ya,tc,z) denotes the expected number of events for each stratum c, and Ta,tc,z is the corresponding total person-time in that stratum. Taking the log of both sides, model 2 can be written as (3) log(E(Ya,tc,z))=log (Ta,tc,z)+ log(h0c(a))+β1Wz,t+β2Cz,t.
Model 3 is the equation for the stratified Poisson model, which is equivalent to model 1. We used R software (version 4.0.3; R Development Core Team) to perform statistical analyses with the package gnm.29
In addition, to examine whether the effect estimate of air pollution on AKI exists at low concentrations (hereafter referred to as the low-pollution cohort), we repeated the main analysis but restricted it to the subset of the cohort with annual exposures lower than <12 μg/m3 PM2.5; based on the current National Ambient Air Quality Standard (NAAQS), 20 ppb NO2, and 50 ppb warm-season O3 during the entire study period given that such analysis is highly important to inform future policy decisions. Finally, to assess any potential deviations from linearity in the concentration–response curves, we replaced the linear term of air pollution in the main model with a B-spine function with three equally distributed internal knots (at the 25th, 50th, and 75th percentiles of the air pollution concentrations) for each pollutant.30 To draw the maps in Figure 1, we used a shapefile provided from the Environmental Systems Research Institute, Inc. (ESRI), with used R software (version 4.0.3). The attributes of this file included the five-digit ZIP code, the two-letter abbreviation for the state in which the ZIP code point was located, the area of the ZIP code area based on in square miles Albers Equal Area Projection, and the Federal Information Processing Standard publication (FIPS) code for the county in which the ZIP code was located.
Figure 1. Nationwide first hospital admissions for the acute kidney injury (AKI) and concentrations of air pollution across the contiguous United States (2000–2016). (A) Occurrence of first hospital admissions per 100,000 Medicare Part A fee-for-service beneficiaries (≥65 years of age), (B) 17-y average concentration of annual fine particulate matter (PM2.5), (C) concentration of nitrogen dioxide (NO2), and (D) concentration of warm-season ozone (O3) (June to September O3).
Figure 1 is a set of four maps of the United States of America, depicting the first hospital admissions for the acute kidney injury and concentrations of air pollution across the contiguous United States from the year 2000 to 2016. Figure 1A is a map of the United States of America, depicting the occurrence of first hospital admissions. A scale depicts per 100,000 ranging from 0 to 40,000 in increments of 10,000. Figure 1B is a map of the United States of America, depicting the concentration of particulate matter begin subscript 2.5 end subscript. A scale depicts the particulate matter begin subscript 2.5 end subscript (micrograms per meter cubed) ranging from 5 to 15 in increments of 5. Figure 1C is a map of the United States of America, depicting the concentration of nitrogen dioxide. A scale depicts the nitrogen dioxide (parts per billion), ranging from 10 to 50 in increments of 10. Figure 1D is a map of the United States of America, depicting the concentration of warm-season ozone. A scale depicts the ozone (parts per billion), ranging from 30 to 50 in increments of 10.
Subpopulation Analysis
To identify subpopulations who showed higher or lower vulnerability, we repeated the same analyses stratified by race (White, Black, and other), age group (65–74 and ≥75 y old), sex, and Medicaid eligibility (as a proxy of low socioeconomic status). In addition, we repeated the analysis by region to consider potential heterogeneity among regions caused by differing chemical compositions, environmental factors, climatic conditions, and population characteristics.
CKD Status and Primary Diseases Accompanied by AKI
We examined the potential difference in the association between air pollution and AKI by CKD status prior to the first hospital admission for AKI and primary diseases accompanied by AKI. The main analysis was repeated with data stratified by a) the hospital admission for CKD (inpatient hospital admissions) with primary or secondary ICD discharge diagnosis codes (ICD-9 code 585; ICD-10 code N18) prior to the first hospital admission for AKI, b) the primary diseases of the first hospital admission with a secondary AKI diagnosis code: circulatory system disease (ICD-9 codes 390–458; ICD-10 codes I00–I99), ischemic heart disease (ICD-9 codes 410–414; ICD-10 codes I20–I25), heart failure (ICD-9 code 428; ICD-10 code I50), acute myocardial infarction (ICD-9 code 410; ICD-10 code I21), cerebrovascular disease (ICD-9 codes 430–438; ICD-10 codes I60–I69), pneumonia (ICD-9 codes 480–486; ICD-10 codes J12–J18), diabetes mellitus (ICD-9 code 250; ICD-10 codes E08–E14), and urinary tract infection (ICD-9 code 599.0; ICD-10 code N39.0).
Sensitivity Analysis
We performed sensitivity analyses to examine whether our main results were robust to the selection of confounders (we examined the risk estimates with or without neighborhood-level indicator or region indicators adjustments in the analytic procedures). We also applied 1-y lag period exposures as an alternative exposure window and conducted a sensitivity analysis restricting kidney outcomes to only those with primary diagnoses codes. Moreover, to exclude potentially prevented cases, we repeated the analyses using data excluding beneficiaries who had the first hospital admission for these outcomes in their first 2 y of follow-up. Finally, we fit two-pollutant models for all combinations of air pollutants to examine the effect estimate of each pollutant by including the other pollutant as a potential confounder.
Results
The full cohort data set included 61,390,754 beneficiaries living in 34,918 ZIP codes. Descriptive information on the beneficiaries is displayed in Table 1. There were 451.3 million person-years of follow-up for AKI, and the total number of first admissions with AKI primary or secondary diagnosis code was ∼9.3 million (Table 2). Of those first AKI admissions, 23.5% (2,180,045 cases) of hospitalizations were identified as the primary discharge diagnosis code, and the median follow-up was 6 y. The low-pollution cohort included 126.9–145.3 million person-years of follow-up depending on pollutant. The correlations among air pollutants were as follows: PM2.5 and NO2, 0.4; NO2 and summer-season O3, 0.3; and PM2.5 and summer O3, 0.2.
Table 1 Descriptive cohort characteristics [n (%) or mean±SD] for U.S. Medicare Part A beneficiaries (≥65 years of age), 2000–2016.
Characteristics Full cohort Low-pollution cohort
(N=61,390,754) PM2.5 <12μg/m3 (n=19,456,404) NO2 <20 ppb (n=16,671,751) O3 <50 ppb (n=12,982,688)
Age at entry (y)
65–74 47,086,254 (76.7) 14,992,132 (77.1) 12,976,559 (77.8) 11,754,531 (76.6)
75–84 10,494,944 (17.1) 3,273,073 (16.8) 2,717,030 (16.3) 2,634,673 (17.2)
≥85 3,809,556 (6.2) 1,191,199 (6.1) 978,162 (5.9) 964,484 (6.3)
Sex
Men 27,545,251 (44.9) 8,984,178 (46.2) 7,612,285 (45.7) 7,027,957 (45.8)
Women 33,845,503 (55.1) 10,472,226 (53.8) 9,059,466 (54.3) 8,325,731 (54.2)
Race
White 51,731,626 (84.3) 17,475,710 (89.8) 14,976,507 (89.8) 12,939,265 (84.3)
Black 5,391,156 (8.8) 576,461 (3) 1,088,577 (6.5) 942,711 (6.1)
Othera 3,598,102 (5.9) 1,182,646 (6.1) 480,830 (2.9) 1,288,161 (8.4)
Medicaid eligibility
Not eligible 53,793,102 (87.6) 17,388,739 (89.4) 14,691,420 (88.1) 13,318,146 (86.7)
Eligible 7,597,652 (12.4) 2,067,665 (10.6) 1,980,331 (11.9) 2,035,542 (13.3)
Air pollution concentration
Annual PM2.5 (μg/m3) 9.8±3.1 7.2±2.2 9.1±2.8 8.5±2.7
Annual NO2 (ppb) 18.9±10.1 15.4±8.4 9.8±3.4 16.1±8
Summer-period O3 (ppb) 45.2±8.5 42.7±9.9 43.3±7.1 36±6.3
Potential confounder (neighborhood-level characteristic)
Below poverty level (%) 10.5±8.1 9.4±7.5 11.1±8 10.4±7.8
Population density (persons per km2) 3,269.8±8,519 1,411.4±2,376.4 256.5±662.1 2,775±4,562.1
Median home value ($1,000) 203,912.7±167,755.5 201,364±157,472.9 134,910.8±106,900.5 232,379.7±194,675.2
Black (%) 11.9±18.4 5±9.9 9.5±16.1 9.4±14.9
Hispanic (%) 7.2±8.8 6.8±7.6 4.2±6.3 7.9±8.6
Median household income ($1,000) 52,689±23,079 52,418.1±20,871.4 44,573.1±15,620.5 53,696.4±22,725.5
Owner-occupied housing (%) 67.4±17.7 70.7±14.7 75.1±11.4 66.6±17.3
Below high school education (%) 26.9±16.1 22.6±14.9 29.4±16.2 24.8±15.8
Ever smoked (%) 46±7.3 47.4±7.9 48.3±8.3 46±8.1
BMI (kg/m2) 27.5±1 27.3±1 27.8±1.1 27.3±1
Note: Table 1 consists of complete data without missing variables. BMI, body mass index; NO2, nitrogen dioxide; O3, ozone; PM2.5, fine particulate matter; SD, standard deviation.
a Race of other included Asian, Hispanic, American Indian or Alaskan Native, and unknown. Further divisions of race were not possible owing to the structure of the data.
Table 2 Admissions for acute kidney injury (AKI) and association between air pollution and the first hospital admission for AKI for the United States.
Cohort/category PM2.5 NO2 O3
Full cohort
Admissions (n) 9,272,274 9,272,274 9,272,274
Total person-years 451,305,627 451,305,627 451,305,627
Median follow-up (y) 6 6 6
Hazard ratio (95% CI) 1.17 (1.16, 1.19) 1.12 (1.11, 1.13) 1.03 (1.02, 1.04)
Low-pollution cohort
Admissions (n) 2,582,170 2,379,401 2,757,486
Total person-years 145,327,939 126,903,255 135,797,051
Median follow-up (y) 6 6 6
Hazard ratio (95% CI) 1.20 (1.17, 1.22) 1.07 (1.05, 1.09) 1.03 (1.01, 1.04)
Note: The low-pollution cohort includes the subset of the cohort with annual average levels below NAAQS levels for the entire follow-up duration. Hazard ratio: PM2.5 (per 5 μg/m3, annual), NO2 (per 10 ppb, annual), and O3 (per 10 ppb, summer-season). Individual-level confounders (age, sex, race, Medicaid eligibility), neighborhood-level indicators [percentage of the population below the poverty level, population density (persons per kilometer squared), median home value (USD), percentage of the population that is Black, percentage of the population that is Hispanic, median household income (USD), percentage of homes with owner-occupied housing, percentage of the population without a high school education, average BMI, percentage of the population that had ever smoked], calendar year, and indicator variables for the region (Northeast, Southeast, Midwest, Southwest, and West) were adjusted in the results. Study population: Medicare Part A fee-for-service beneficiaries (≥65 years of age) from 2000 to 2016. A Cox-equivalent reparameterized Poisson model was used to estimate the hazard ratios. BMI, body mass index; CI, confidence interval; NAAQS, National Ambient Air Quality Standard; NO2, nitrogen dioxide; O3, ozone; PM2.5, fine particulate matter.
The geographical distributions of first hospital admission occurrences with AKI and air pollution concentrations are displayed in Figure 1. The Southeast Region showed the most frequent occurrence of first AKI hospital admissions. The average annual concentrations of air pollution over the study period were 9.7 μg/m3 for annual PM2.5, 23 ppb for annual NO2, and 25 ppb for summer-period O3. PM2.5 concentrations were highest in California and in the Eastern and Southeastern Regions of the United States. The highest NO2 concentrations were generally in metropolitan areas (New York, Los Angeles, and Chicago), and the highest O3 concentrations were observed in California.
The estimated concentration–response curves for all pollutants (Figure 2; see Table S1 for corresponding numeric data) suggest an approximately linear association between air pollution concentrations and the first hospital admission for AKI, and results from the linear model are shown in Table 2. In the full cohort (i.e., considering all levels of air pollution), air pollution was positively associated with AKI for all pollutants, with hazard ratios (HRs) of 1.17 (95% CI: 1.16, 1.19) for a 5-μg/m3 increase in PM2.5, 1.12 (95% CI: 1.11, 1.13) for a 10-ppb increase in NO2, and 1.03 (95% CI: 1.02, 1.04) for a 10-ppb increase in warm-season O3. In the low-pollution cohort (i.e., considering air pollution levels below the NAAQS levels), positive associations were still observed with HRs of 1.20 (95% CI: 1.17, 1.22) for a 5-μg/m3 increase in PM2.5, 1.07 (95% CI: 1.05, 1.09) for a 10-ppb increase in NO2, and 1.03 (95% CI: 1.01, 1.04) for a 10-ppb increase in summer-season O3.
Figure 2. Concentration–response curves for the association between long-term air pollution exposure and kidney diseases: (A) PM2.5, (B) NO2, and (C) O3. Dotted vertical lines: 10th and 90th percentiles of each air pollution concentration. Shaded areas: 95% CIs. Reference exposure points: 0 μg/m3 for PM2.5 and 0 ppb for NO2 and O3. Individual-level confounders (age, sex, race, Medicaid eligibility) and neighborhood-level socioeconomic status indicators were adjusted in the results. Study population: Medicare Part A fee-for-service beneficiaries (≥65 years of age) from 2000 to 2016. See Table S1 for corresponding numeric data. A Cox-equivalent reparameterized Poisson model was used to estimate the hazard ratios. Note: CI: confidence interval; NO2, nitrogen dioxide; O3, ozone; PM2.5, fine particulate matter.
Figures 2A to 2C are ribbon plus line graphs titled particulate matter begin subscript 2.5 end subscript, nitrogen dioxide, and ozone, plotting hazard ratio, ranging from 1.0 to 1.8 in increments of 0.2; 1.0 to 1.6 in increments of 0.1; and 0.98 to 1.10 in increments of 0.02 (y-axis) across micrograms per meter cubed, ranging from 5 to 15 in increments of 5; parts per billion, ranging from 10 to 40 in increments of 10; and parts per billion, ranging from 30 to 60 in increments of 10 (x-axis), respectively.
In general, a positive association between air pollution and AKI was observed across all subpopulations (Figure 3; see Table S2 for corresponding numeric data). Those who were White or not eligible for Medicaid generally showed higher air pollution effect estimates compared with those who were Black and other populations and people who were eligible for Medicaid. These patterns were also observed in race–Medicaid eligibility stratified analysis (Table S3). Furthermore, for PM2.5 and NO2, people ≥75 years of age showed more pronounced impacts of air pollution than people 65–74 years of age. In addition, the positive associations with air pollution were observed across all regions (except for the association with O3 in the Southwest Region), although the effect sizes varied by region; the highest effect estimates were observed in the Midwest and Northeast Regions (Table S4).
Figure 3. Subpopulation-specific association between air pollution and first hospital admission for acute kidney disease (AKI). Hazard ratio: (A) PM2.5 (per 5 μg/m3), (B) NO2 (per 10 ppb), and (C) O3 (per 10 ppb). Dual: Eligible for Medicaid, Non-dual: Noneligible for Medicaid. Individual-level confounders (age, sex, race, Medicaid eligibility), neighborhood-level indicators (percentage of the population below the poverty level, population density (persons per kilometers squared), median home value (USD), percentage of the population that is Black, percentage of the population that is Hispanic, median household income (USD), percentage of homes with owner-occupied housing, percentage of the population without a high school education, average BMI, percentage of the population that had ever smoked), calendar year, and indicator variables for the region (Northeast, Southeast, Midwest, Southwest, and West) were adjusted in the results. Study population: Medicare Part A fee-for-service beneficiaries (≥65 years of age) from 2000 to 2016. See Table S2 for corresponding numeric data. A Cox-equivalent reparameterized Poisson model was used to estimate the hazard ratios. Note: BMI, body mass index; NO2, nitrogen dioxide; O3, ozone; PM2.5, fine particulate matter.
Figures 3A to 3C are error bar graphs titled particulate matter begin subscript 2.5 end subscript, nitrogen dioxide, and ozone, plotting hazard ratio, ranging from 1 to 1.3 in increments of 0.1 (y-axis) across White, Black, Other, Male, Female, 65 to 74 years, 75 years or older, Dual, and Non-dual (x-axis), respectively.
Table 3 displays the association between air pollution and AKI differ by CKD hospitalization prior to the hospital admission for AKI and primary diseases of the first hospital admission with AKI secondary diagnosis code. Beneficiaries who had the hospital admission for CKD prior to the first AKI hospital admission showed lower air pollution effect estimates compared with the total population. Meanwhile, for PM2.5 and NO2, the association between air pollution and AKI was more pronounced in hospitalized beneficiaries for AKI with heart failure, cerebrovascular disease, pneumonia, and urinary tract infection as a primary disease compared with the total population. For O3, the association between air pollution and AKI was more pronounced in hospitalized beneficiaries for AKI with cerebrovascular disease and pneumonia as a primary disease.
Table 3 Association between air pollution and the first hospital admission for acute kidney injury (AKI) by chronic kidney disease admission record prior to AKI and primary disease of the first hospital admission with a secondary AKI diagnosis code (full cohort, N=61,390,754).
Hospital admission/disease N (%) PM2.5 NO2 O3
HR (95% CI) HR (95% CI) HR (95% CI)
Prior to the hospital admission for AKI
Chronic kidney disease 3,189,290 (34.4) 1.07 (1.05, 1.08) 0.98 (0.97, 0.99) 0.98 (0.98, 0.99)
Primary disease of the first hospital admission with a secondary AKI diagnostic code
Circulatory system disease 2,880,526 (31.1) 1.18 (1.16, 1.20) 1.13 (1.12, 1.14) 1.02 (1.01, 1.03)
Ischemic heart disease 782,960 (8.4) 1.17 (1.15, 1.19) 1.12 (1.11, 1.14) 1.02 (1.01, 1.03)
Heart failure 935,738 (10.1) 1.23 (1.21, 1.26) 1.15 (1.13, 1.17) 1.01 (1.00, 1.02)
Acute myocardial infarction 574,103 (6.2) 1.16 (1.14, 1.18) 1.01 (0.99, 1.02) 1.01 (1.00, 1.02)
Cerebrovascular disease 321,819 (3.5) 1.27 (1.24, 1.30) 1.16 (1.14, 1.19) 1.06 (1.05, 1.07)
Pneumonia 618,543 (6.7) 1.25 (1.22, 1.27) 1.12 (1.10, 1.13) 1.06 (1.05, 1.07)
Diabetes mellitus 194,903 (2.1) 1.17 (1.13, 1.20) 1.16 (1.14, 1.18) 1.03 (1.01, 1.04)
Urinary tract infection 335,871 (3.6) 1.28 (1.24, 1.32) 1.16 (1.13, 1.18) 1.02 (1.01, 1.04)
Note: Hazard ratio: PM2.5 (per 5 μg/m3, annual), NO2 (per 10 ppb, annual), and O3 (per 10 ppb, summer-season). Individual-level confounders (age, sex, race, Medicaid eligibility), neighborhood-level indicators [percentage of the population below the poverty level, population density (persons per kilometer squared), median home value (USD), percentage of the population that is Black, percentage of the population that is Hispanic, median household income (USD), percentage of homes with owner-occupied housing, percentage of the population without a high school education, average BMI, percentage of the population that had ever smoked], calendar year, and indicator variables for the region (Northeast, Southeast, Midwest, Southwest, and West) were adjusted in the results. Study population: Medicare Part A fee-for-service beneficiaries (≥65 years of age) from 2000 to 2016. A Cox-equivalent reparameterized Poisson model was used to estimate the hazard ratios. BMI, body mass index; CI, confidence interval; HR, hazard ratio; NO2, nitrogen dioxide; O3, ozone; PM2.5, fine particulate matter.
Finally, the results of our sensitivity analysis were generally consistent with the main results, except for several of the O3 results. Our results were robust to confounder adjustments, removal of prevalent cases, and use of a different lag period. In addition, the exclusion of cases identified by secondary diagnostic codes did not change the main results (Table S5). Two-pollutant models showed that estimates were consistent with those from the single-pollutant models, although effect sizes slightly decreased (Table S6).
Discussion
This study investigated the association between air pollution and the first hospital admission for AKI using a nationwide large prospective cohort covering >61 million Medicare Part A FFS beneficiaries from 2000 to 2016. Annual exposures to PM2.5, NO2, and summer-period O3 were associated with an increased risk of the first hospital admission for AKI, and the associations existed even at levels below the current annual NAAQS levels for PM2.5 (12 μg/m3) and NO2 (53 ppb). These findings suggest that improving air quality may lead to public health benefits and reduce the risk of AKI. To the best of our knowledge, this is the first and largest epidemiological study to investigate the association between long-term exposure to air pollution and AKI development using nationwide Medicare cohort data.
The findings of this study are consistent with previous studies examining the association between exposure to long-term air pollution and a decrease in renal function. Three U.S. military veteran cohort studies reported that long-term (annual average) exposure to higher concentrations of air pollution [PM2.5,31–33 PM10, NO2, and carbon monoxide (CO)34] is associated with a reduced estimated glomerular filtration rate (eGFR). Another cohort study conducted in four U.S. counties reported that higher annual average PM2.5 was associated with a higher urinary albumin–creatinine ratio.35 A Taiwanese cohort study also showed that an increased concentration of PM2.5 was associated with an increased risk of CKD development, which was defined by an eGRF of <60mL/min per 1.73 m2.36 A recent cohort study in South Korea revealed a positive long-term effect estimate of PM2.5 on the mortality of CKD patients.37 In addition, a recent time-stratified case-crossover study reported a positive association between short-term exposure to PM2.5 (lag 0–1 d) and risk of urgent or emergent hospital admissions for AKI in the U.S. Medicare population,11 and another population-based time-series study in South Korea also showed a positive short-term effect estimate of air pollution on emergency department visits due to AKI.38
This study provides scientific evidence that the public health benefits of stricter air pollution standards may alleviate the potential risk of AKI. The concentration–response curves (Figure 2) indicate no evidence of a threshold value for pollution for the development of AKI, although the O3 curve that showed fluctuations should be interpreted carefully because of its uncertain estimates at very low and high values. Furthermore, the results from low-pollution cohorts showed that the effect estimate of exposure to air pollution on the first hospital admission for AKI persisted at the low concentrations across all air pollutants. The results indicate that the health-benefit-per-unit decrease in the concentration of these air pollutants are consistent across concentrations that are below the current NAAQS levels.
Nevertheless, this study had several limitations. First, although we performed stratified analyses, the Medicare Part A data we used in this study is an administrative database for Medicare FFS claims, thus we were limited in assessing the confounding and interactive effects of medication and underlying medical conditions, such as sepsis, CKD, heart diseases, diabetes mellitus, and hypertension, which are identified risk factors associated with AKI.1,39 In addition, like previous studies based on health insurance claim data,38,40,41 we had a limitation in identifying the complex pathways and mediating diseases of AKI that were affected by exposure to air pollution. Further, we were not able to collect detailed information on mortality (e.g., causes of death) that were sufficiently informative to consider the competing risk models between mortality and the first hospital admission for AKI; thus, our estimates on the association between air pollution and hospital admissions for AKI should be interpreted as the overall effect of air pollution on any inpatient hospitalizations associated with AKI irrespective of causative medical conditions. This limitation is important and should be addressed carefully in future studies.
Second, because the Medicare cohort includes only individuals who were ≥65 y old, our results are limited in their ability to represent the entire U.S. population. Our results represent only the Medicare Part A FFS population, which does not include all Medicare beneficiaries (the Medicare FFS population covers up to ∼65.8% of the Medicare population in 2016) or persons ≥65 years of age.21 Because we had no information on Medicare–HMO (managed care) claims and younger Medicare-eligible population with disabilities or end-stage renal diseases, we were unable to cover the entire Medicare population. Third, the first hospital admission with an ICD-diagnosis code for AKI has limitations when interpreting it as the onset of AKI. AKI is generally diagnosed through laboratory tests (e.g., the accumulation of end products of nitrogen metabolism or decrease urine outputs).1 A previous study in Scotland and a systematic review including 25 studies in four counties (United States, Canada, Australia, and Spain) reported that the incidence of AKI was substantially underestimated when ICD diagnostic codes were used to define AKI42,43: A median of positive predictive values (PPVs) was ∼67%–70%. (A study by Logan et al.42 was performed at acute hospitals in two Scottish Health Boards, and a review study by Vlasschaert43 included 8 U.S. studies based on the California Hospital Discharge Abstract Database, the Partners Health Care System Research Patient Data Registry, and the Veterans Affairs Patient Treatment File). In addition, hospital admissions can occur at more advanced stages of the disease or for treating complications attributed to AKI, such as volume overload, electrolyte, and acid–base disturbances.44 Thus, hospital admission records cannot adequately represent the incidence of AKI, and our cohort could undercount AKI onset. In addition, although this study includes multiple neighborhood-level indicators as the potential confounders, Medicare claims do not include extensive individual-level data on behavioral and socioeconomic risk factors, which could be crucial confounders. Therefore, the potential effects of unmeasured confounders should be considered in depth in future studies.
Together with the aforementioned limitations, several points should be addressed further. With the large study size of the Medicare data set, we were able to estimate differences in the air pollution–AKI risks among subpopulations: the older (≥75 y old), White, and Medicaid-noneligible beneficiaries generally showed higher effect estimates of air pollution on the first hospital admission for AKI. Nevertheless, the results corresponding to those of White and Medicaid-noneligible persons should be investigated further because they are seemingly different from results of previous U.S. studies that revealed a higher air pollution–mortality risk in non-White and Medicaid-eligible populations, albeit for different health outcomes.14,45 There are several plausible explanations. First, there may have been underdiagnosis. In general, despite Medicare, low-income individuals and racial minorities can have lower accessibility to medical facilities,46 which could result in underdiagnosis. Particularly, based on a recent U.S. national study, the hospitalization for AKI was associated with an increase in excess hospitalization costs of $1,800–$7,900 compared with patients without AKI, and the excess costs were generally higher than those of other acute medical conditions.47 Thus, we postulate that the high economic burden of AKI might be considerably associated with the underdiagnosis problem, especially in socially marginalized populations. Although we considered Medicaid eligibility and neighborhood-level socioeconomic variables, additional variables that reflect individual-level disparities in accessibility to medical resources (e.g., income level, medical expenditure, and accessibility to nephrologists) should be addressed in future studies because these variables do not fully capture factors related to disparities.
Second, there may be confounding effects. A cohort study in Korea reported that CKD patients with healthy lifestyles (normal weight, nonsmokers, and nondrinkers) showed a higher air pollution–mortality risk than CKD patients with less-healthy lifestyles.37 A study in Taiwan also reported that cohort participants with comorbidities showed a lower risk of PM2.5 on CKD incidence.36 These results imply the possibility of confounding effects between air pollution and behavior or biological factors on the development of kidney disease. Especially, in the United States, high-income White individuals generally have lower prevalence of hypertension, obesity, and diabetes, as well as better health behaviors46 than the general population. Therefore, the results of this study could be influenced by the potential confounding effects of underlying health behaviors and other biological conditions.
Finally, future research could also address the complex air pollution mixture and should disentangle the association with short- and long-term exposure to air pollution. We focused on annual average air pollutant levels of PM2.5, NO2, and O3 individually although the actual air pollution mixture is complex, with simultaneous exposure to these and other pollutants, as well as relationships among these pollutants (e.g., NO2 and a precursor of O3) and different chemical structures of PM2.5. In addition, based on these complex chemical compositions and seasonality of air pollutants, seemingly inconsistent sensitivity analysis results of O3 should be interpreted carefully and examined in depth in future studies using data with higher temporal resolution.
In summary, we found an association between exposures to air pollution and the risk of the first hospital admission for AKI, and this association persisted even at low concentrations of air pollution. Our findings suggest beneficial implications for public health policies to alleviate health care expenditures and disease burden attributable to AKI and also provide epidemiological evidence on the value of air pollution guidelines for potential AKI patients.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
W.L. designed the study, coordinated the work, conducted the statistical analysis, and took the lead in drafting the manuscript and interpreting the results. M.L.B. supported the whole procedures of this study as a senior author. X.W. and F.D. supported the statistical analysis and result interpretation. Z.A., S.H., K.C.F., J.Y.S., H.K., D.B., F.D., and J.S. contributed to the interpretation of the results and reviewed the manuscript. M.B.S., D.B., and J.S. provided the data. Z.A., J.M.K., J.Y.P., Y.C.K., and J.P.L provided medical input in interpreting the results and writing the manuscript. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Data were collected by the Yale–Harvard Medicare collaboration (directed by F.D. and M.L.B.) under a data user agreement with the Centers for Medicare and Medicaid Services and cannot be made publicly available.
This paper was developed under Assistance Agreement RD835871 awarded by the U.S. Environmental Protection Agency (EPA) to Yale University (M.L.B.). It has not been formally reviewed by the U.S. EPA. The views expressed in this document are solely those of M.L.B. and other coauthors and do not necessarily reflect those of the agency. The U.S. EPA does not endorse any products or commercial services mentioned in this publication. W.L. was supported by the 2020 Science and Technology Subsequent Generation Support Project (NRF-2021R1A6A3A03038675), implemented by the National Research Foundation of Korea. This work also was supported by BK21 Four, Korean Southeast Center for the 4th Industrial Revolution Leader Education (W.L.) and Korea Institute for Advancement of Technology (KIAT) grant funded by the Korea Government (Ministry of Education-Ministry of Trade, Industry and Energy) (W.L.).
==== Refs
References
1. Bellomo R, Kellum JA, Ronco C. 2012. Acute kidney injury. Lancet 380 (9843 ):756–766, PMID: , 10.1016/S0140-6736(11)61454-2.22617274
2. United States Renal Data System. 2020. USRDS Annual Data Report: Epidemiology of Kidney Disease in the United States. Bethesda, MD: National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases. https://usrds-adr.niddk.nih.gov/2020 [accessed 1 June 2022].
3. Case J, Khan S, Khalid R, Khan A. 2013. Epidemiology of acute kidney injury in the intensive care unit. Crit Care Res Pract 2013 :479730, PMID: , 10.1155/2013/479730.23573420
4. Waikar SS, Liu KD, Chertow GM. 2008. Diagnosis, epidemiology and outcomes of acute kidney injury. Clin J Am Soc Nephrol 3 (3 ):844–861, PMID: , 10.2215/CJN.05191107.18337550
5. Pavkov ME, Harding JL, Burrows NR. 2018. Trends in hospitalizations for acute kidney injury—United States, 2000–2014. MMWR Morb Mortal Wkly Rep 67 (10 ):289–293, PMID: , 10.15585/mmwr.mm6710a2.29543788
6. Sawhney S, Fraser SD. 2017. Epidemiology of AKI: utilizing large databases to determine the burden of AKI. Adv Chronic Kidney Dis 24 (4 ):194–204, PMID: , 10.1053/j.ackd.2017.05.001.28778358
7. Nemmar A, Karaca T, Beegam S, Yuvaraju P, Yasin J, Hamadi NK, et al. 2016. Prolonged pulmonary exposure to diesel exhaust particles exacerbates renal oxidative stress, inflammation and DNA damage in mice with adenine-induced chronic renal failure. Cell Physiol Biochem 38 (5 ):1703–1713, PMID: , 10.1159/000443109.27160713
8. Nemmar A, Al-Salam S, Zia S, Yasin J, Al Husseni I, Ali BH. 2010. Diesel exhaust particles in the lung aggravate experimental acute renal failure. Toxicol Sci 113 (1 ):267–277, PMID: , 10.1093/toxsci/kfp222.19797351
9. Kaufman JD, Adar SD, Barr RG, Budoff M, Burke GL, Curl CL, et al. 2016. Association between air pollution and coronary artery calcification within six metropolitan areas in the USA (the Multi-Ethnic Study of Atherosclerosis and Air Pollution): a longitudinal cohort study. Lancet 388 (10045 ):696–704, PMID: , 10.1016/S0140-6736(16)00378-0.27233746
10. Chuang KJ, Chan CC, Su TC, Lee CT, Tang CS. 2007. The effect of urban air pollution on inflammation, oxidative stress, coagulation, and autonomic dysfunction in young adults. Am J Respir Crit Care Med 176 (4 ):370–376, PMID: , 10.1164/rccm.200611-1627OC.17463411
11. Wei Y, Wang Y, Di Q, Choirat C, Wang Y, Koutrakis P, et al. 2019. Short term exposure to fine particulate matter and hospital admission risks and costs in the Medicare population: time stratified, case crossover study. BMJ 367 :l6258, PMID: , 10.1136/bmj.l6258.31776122
12. Ciencewicki J, Jaspers I. 2007. Air pollution and respiratory viral infection. Inhal Toxicol 19 (14 ):1135–1146, PMID: , 10.1080/08958370701665434.17987465
13. Plataki M, Kashani K, Cabello-Garza J, Maldonado F, Kashyap R, Kor DJ, et al. 2011. Predictors of acute kidney injury in septic shock patients: an observational cohort study. Clin J Am Soc Nephrol 6 (7 ):1744–1751, PMID: , 10.2215/CJN.05480610.21734090
14. Di Q, Wang Y, Zanobetti A, Wang Y, Koutrakis P, Choirat C, et al. 2017. Air pollution and mortality in the Medicare population. New Engl J Med 376 (26 ):2513–2522, PMID: , 10.1056/NEJMoa1702747.28657878
15. WHO (World Health Organization). 1978. International Statistical Classification of Diseases, Ninth Revision, Basic Tabulation List with Alphabetic Index. http://apps.who.int/classifications/icd10/browse/2016/en [accessed 1 June 2022].
16. WHO. 1993. ICD Classification of Mental and Behavioural Disorders, Diagnostic Criteria for Research, 10th ed. Geneva, Switzerland: WHO.
17. Di Q, Rowland S, Koutrakis P, Schwartz J. 2017. A hybrid model for spatially and temporally resolved ozone exposures in the continental United States. J Air Waste Manag Assoc 67 (1 ):39–52, PMID: , 10.1080/10962247.2016.1200159.27332675
18. Di Q, Kloog I, Koutrakis P, Lyapustin A, Wang Y, Schwartz J. 2016. Assessing PM2.5 exposures with high spatiotemporal resolution across the continental United States. Environ Sci Tech 50 (9 ):4712–4721, PMID: , 10.1021/acs.est.5b06121.27023334
19. Di Q, Amini H, Shi L, Kloog I, Silvern R, Kelly J, et al. 2020. Assessing NO2 concentration and model uncertainty with high spatiotemporal resolution across the contiguous United States using ensemble model averaging. Environ Sci Technol 54 (3 ):1372–1384, PMID: , 10.1021/acs.est.9b03358.31851499
20. Wei Y, Yazdi MD, Di Q, Requia WJ, Dominici F, Zanobetti A, et al. 2021. Emulating causal dose-response relations between air pollutants and mortality in the Medicare population. Environ Health 20 (1 ):53, PMID: , 10.1186/s12940-021-00742-x.33957920
21. Shi L, Wu X, Yazdi MD, Braun D, Abu Awad Y, Wei Y, et al. 2020. Long-term effects of PM2·5 on neurological disorders in the American Medicare population: a longitudinal cohort study. Lancet Planet Health 4 (12 ):e557–e565, PMID: , 10.1016/S2542-5196(20)30227-8.33091388
22. Jerrett M, Burnett RT, Pope CA III, Ito K, Thurston G, Krewski D, et al. 2009. Long-term ozone exposure and mortality. New Engl J Med 360 (11 ):1085–1095, PMID: , 10.1056/NEJMoa0803894.19279340
23. Di Q, Wang Y, Zanobetti A, Wang Y, Koutrakis P, Choirat C, et al. 2017. Air pollution and mortality in the Medicare population. New Engl J Med 376 (26 ):2513–2522, PMID: , 10.1056/NEJMoa1702747.28657878
24. Nicholas SB, Kalantar-Zadeh K, Norris KC. 2015. Socioeconomic disparities in chronic kidney disease. Adv Chronic Kidney Dis 22 (1 ):6–15, PMID: , 10.1053/j.ackd.2014.07.002.25573507
25. Holmes J, Phillips D, Donovan K, Geen J, Williams JD, Phillips AO, et al. 2019. Acute kidney injury, age, and socioeconomic deprivation: evaluation of a national data set. Kidney Int Rep 4 (6 ):824–832, PMID: , 10.1016/j.ekir.2019.03.009.31194105
26. Hajat A, Hsia C, O’Neill MS. 2015. Socioeconomic disparities and air pollution exposure: a global review. Current Environ Health Rep 2 (4 ):440–450, PMID: , 10.1007/s40572-015-0069-5.26381684
27. Andersen PK, Gill RD. 1982. Cox’s regression model for counting processes: a large sample study. Ann Stat 10 (4 ):1100–1120, 10.1214/aos/1176345976.
28. Bickel PJ, Götze F, van Zwet WR. 2012. Resampling fewer than n observations: gains, losses, and remedies for losses. In: Selected works of Willem van Zwet. New York, NY: Springer, 267–297.
29. Turner H, Firth D. 2007. Generalized nonlinear models in R: an overview of the gnm package. Report no. NCRM Working Paper Series 06/07.
30. Liu C, Chen R, Sera F, Vicedo-Cabrera AM, Guo Y, Tong S, et al. 2019. Ambient particulate air pollution and daily mortality in 652 cities. New Engl J Med 381 (8 ):705–715, PMID: , 10.1056/NEJMoa1817364.31433918
31. Bowe B, Xie Y, Li T, Yan Y, Xian H, Al-Aly Z. 2018. Particulate matter air pollution and the risk of incident CKD and progression to ESRD. J Am Soc Nephrol 29 (1 ):218–230, PMID: , 10.1681/ASN.2017030253.28935655
32. Kuźma Ł, Małyszko J, Bachórzewska-Gajewska H, Kralisz P, Dobrzycki S. 2021. Exposure to air pollution and renal function. Sci Rep 11 (1 ):11419, PMID: , 10.1038/s41598-021-91000-0.34075149
33. Mehta AJ, Zanobetti A, Bind MAC, Kloog I, Koutrakis P, Sparrow D. 2016. Long-term exposure to ambient fine particulate matter and renal function in older men: the veterans Administration Normative Aging Study. Environ Health Perspect 124 (9 ):1353–1360, PMID: , 10.1289/ehp.1510269.26955062
34. Bowe B, Xie Y, Li T, Yan Y, Xian H, Al-Aly Z. 2017. Associations of ambient coarse particulate matter, nitrogen dioxide, and carbon monoxide with the risk of kidney disease: a cohort study. Lancet Planet Health 1 (7 ):e267–e276, PMID: , 10.1016/S2542-5196(17)30117-1.29851625
35. Blum MF, Surapaneni A, Stewart JD, Liao D, Yanosky JD, Whitsel EA, et al. 2020. Particulate matter and albuminuria, glomerular filtration rate, and incident CKD. Clin J Am Soc Nephrol 15 (3 ):311–319, PMID: , 10.2215/CJN.08350719.32108020
36. Chan TC, Zhang Z, Lin BC, Lin C, Deng HB, Chuang YC, et al. 2018. Long-term exposure to ambient fine particulate matter and chronic kidney disease: a cohort study. Environ Health Perspect 126 (10 ):107002, PMID: , 10.1289/EHP3304.30392394
37. Jung J, Park JY, Kim YC, Lee H, Kim E, Kim YS, et al. 2021. Effects of air pollution on mortality of patients with chronic kidney disease: a large observational cohort study. Sci Total Environ 786 :147471, PMID: , 10.1016/j.scitotenv.2021.147471.33971609
38. Lee W, Prifti K, Kim H, Kim E, Yang J, Min J, et al. 2022. Short-term exposure to air pollution and attributable risk of kidney diseases: a nationwide time-series study. Epidemiology 33 (1 ):17–24, PMID: , 10.1097/EDE.0000000000001430.34711735
39. Finlay S, Bray B, Lewington AJ, Hunter-Rowe CT, Banerjee A, Atkinson JM, et al. 2013. Identification of risk factors associated with acute kidney injury in patients admitted to acute medical units. Clin Med (Lond) 13 (3 ):233–238, PMID: , 10.7861/clinmedicine.13-3-233.23760694
40. Bragg-Gresham J, Morgenstern H, McClellan W, Saydah S, Pavkov M, Williams D, et al. 2018. County-level air quality and the prevalence of diagnosed chronic kidney disease in the US Medicare population. PLoS One 13 (7 ):e0200612, PMID: , 10.1371/journal.pone.0200612.30063741
41. Lee W, Wu X, Heo S, Fong KC, Son JY, Sabath MB, et al. 2022. Associations between long term air pollution exposure and first hospital admission for kidney and total urinary system diseases in the US Medicare population: nationwide longitudinal cohort study. BMJ Med 1 (1 ):e000009, PMID: , 10.1136/bmjmed-2021-000009.36936557
42. Logan R, Davey P, De Souza N, Baird D, Guthrie B, Bell S. 2019. Assessing the accuracy of ICD-10 coding for measuring rates of and mortality from acute kidney injury and the impact of electronic alerts: an observational cohort study. Clin Kidney J 13 (6 ):1083–1090, PMID: , 10.1093/ckj/sfz117.33391753
43. Vlasschaert ME, Bejaimal SAD, Hackam DG, Quinn R, Cuerden MS, Oliver MJ, et al. 2011. Validity of administrative database coding for kidney disease: a systematic review. Am J Kidney Dis 57 (1 ):29–43, PMID: , 10.1053/j.ajkd.2010.08.031.21184918
44. Prowle JR, Echeverri JE, Ligabo EV, Ronco C, Bellomo R. 2010. Fluid balance and acute kidney injury. Nat Rev Nephrol 6 (2 ):107–115, PMID: , 10.1038/nrneph.2009.213.20027192
45. Pope CA III, Lefler JS, Ezzati M, Higbee JD, Marshall JD, Kim SY, et al. 2019. Mortality risk and fine particulate air pollution in a large, representative cohort of U.S. adults. Environ Health Perspect 127 (7 ):077007, PMID: , 10.1289/EHP4438.31339350
46. Dubay LC, Lebrun LA. 2012. Health, behavior, and health care disparities: disentangling the effects of income and race in the United States. Int J Health Serv 42 (4 ):607–625, PMID: , 10.2190/HS.42.4.c.23367796
47. Silver SA, Long J, Zheng Y, Chertow GM. 2017. Cost of acute kidney injury in hospitalized patients. J Hosp Med 12 (2 ):70–76, PMID: , 10.12788/jhm.2683.28182800
|
PMC010xxxxxx/PMC10094192.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37042841
EHP11896
10.1289/EHP11896
Research
Association between Organophosphate Ester Exposure and Insulin Resistance with Glycometabolic Disorders among Older Chinese Adults 60–69 Years of Age: Evidence from the China BAPE Study
Ding Enmin 1 2
Deng Fuchang 1
Fang Jianlong 1
Li Tiantian 1 3
Hou Minmin 4
Liu Juan 1
Miao Ke 1
Yan Wenyan 3
Fang Ke 1
Shi Wanying 1
Fu Yuanzheng 1
Liu Yuanyuan 1
Dong Haoran 1
Dong Li 1
Ding Changming 1
Liu Xiaohui 5
Pollitt Krystal J. Godri 6
Ji John S. 7
Shi Yali 4
Cai Yaqi 4
https://orcid.org/0000-0003-3219-1422
Tang Song 1 3
https://orcid.org/0000-0002-7071-571X
Shi Xiaoming 1 3
1 China CDC Key Laboratory of Environment and Population Health, National Institute of Environmental Health, Chinese Center for Disease Control and Prevention, Beijing, China
2 Institute of Occupational Disease Prevention, Jiangsu Provincial Center for Disease Control and Prevention, Nanjing, Jiangsu, China
3 Center for Global Health, School of Public Health, Nanjing Medical University, Nanjing, Jiangsu, China
4 State Key Laboratory of Environmental Chemistry and Ecotoxicology, Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, China
5 National Protein Science Technology Center and School of Life Sciences, Tsinghua University, Beijing, China
6 Department of Environmental Health Sciences, Yale School of Public Health, New Haven, Connecticut, USA
7 Vanke School of Public Health, Tsinghua University, Beijing, China
Address correspondence to Xiaoming Shi, No. 7 Panjiayuan Nanli, Chaoyang District, Beijing 100021, China. Telephone: 86-10-5093-0101. E-mail: [email protected]. And, Song Tang, No. 7 Panjiayuan Nanli, Chaoyang District, Beijing, 100021 China. Telephone: 86-10-5093-0184. Email: [email protected]
12 4 2023
4 2023
131 4 04700922 7 2022
10 2 2023
16 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Organophosphate esters (OPEs) are common endocrine-disrupting chemicals, and OPE exposure may be associated with type 2 diabetes (T2D). However, greater knowledge regarding the biomolecular intermediators underlying the impact of OPEs on T2D in humans are needed to understand biological etiology.
Objectives:
We explored the associations between OPE exposure and glycometabolic markers among older Chinese adults 60–69 years of age to elucidate the underlying mechanisms using a multi-omics approach.
Methods:
This was a longitudinal panel study comprising 76 healthy participants 60–69 years of age who lived in Jinan city of northern China. The study was conducted once every month for 5 months, from September 2018 to January 2019. We measured a total of 17 OPEs in the blood, 11 OPE metabolites in urine, and 4 glycometabolic markers (fasting plasma glucose, glycated serum protein, fasting insulin, and homeostatic model assessment for insulin resistance). The blood transcriptome and serum/urine metabolome were also evaluated. The associations between individual OPEs and glycometabolic markers were explored. An adverse outcome pathway (AOP) was established to determine the biomolecules mediating the associations.
Results:
Exposure to five OPEs and OPE metabolites (trimethylolpropane phosphate, triphenyl phosphate, tri-iso-butyl phosphate, dibutyl phosphate, and diphenyl phosphate) was associated with increased levels of glycometabolic markers. The mixture effect analysis further indicated the adverse effect of OPE mixtures. Multi-omics analyses revealed that the endogenous changes in the transcriptional and metabolic levels were associated with OPE exposure. The putative AOPs model suggested that triggers of molecular initiation events (e.g., insulin receptor and glucose transporter type 4) with subsequent key events, including disruptions in signal transduction pathways (e.g., phosphatidylinositol 3-kinase/protein kinase B and insulin secretion signaling) and biological functions (glucose uptake and insulin secretion), may constitute the diabetogenic effects of OPEs.
Discussion:
OPEs are associated with the elevated risk of T2D among older Chinese adults 60–69 years of age. Implementing OPE exposure reduction strategies may help reduce the T2D burden among these individuals, if the relationship is causal. https://doi.org/10.1289/EHP11896
Supplemental Material is available online (https://doi.org/10.1289/EHP11896).
None of the authors declare any conflict of interest.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Diabetes mellitus, mainly type 2 diabetes (T2D), is one of the four most prevalent noncommunicable diseases.1,2 In 2021, 536.6 (10.5%) million adults 20–79 years of age worldwide were estimated to have diabetes, which resulted in 6.7 million deaths and health expenditures totaling USD $966 billion.3 Diabetes impacts aging societies, affecting an estimated 19.3% of adults 65–99 years of age worldwide.4 Most of the diabetes burden falls on low- and medium-income countries, such as China and India.5 In the past decade, the number of diabetic patients in China increased from 90 (age-adjusted prevalence: 8.8%) to 140 million (10.6%),3 and the number of older individuals (>65 years of age) with diabetes was 35.5 million, ranking first globally and accounting for 25% of the elderly diabetes patients worldwide.4
The risk of T2D is multifactorial and determined by an interplay of genetic and environmental factors.6 Family history of diabetes, older age, unhealthy diet, physical inactivity, smoking, and overweight and obesity are associated with an increased risk of T2D; of these, overweight and obesity was found to be the strongest risk factor.7 Of note, the age-standardized prevalence of general obesity among Chinese adults, is 14.0%,8 considerably lower than that among American adults (43.3%).9 However, the estimated age-standardized prevalence of diabetes is almost the same among Chinese and American adults (12.4% vs. 14.3%).10,11 Genetic factors, a thrifty phenotype,12–14 and unique dietary patterns (e.g., low intake of whole grains but high intake of refined grains)15 are suspected to be responsible for the increased risk of T2D in the Chinese population. In addition, since industrialization advancements and immense growth took place after the 1980s, chemical pollution has been the leading environmental cause of diseases and deaths in China.16 Emerging pollutants are ever-evolving and are widely used in the industry and for daily necessities worldwide, and their production is gradually shifting to China owing to the increasingly strict regulations in Europe and America. Extensive evidence suggests that disproportionate exposure to environmental pollutants may be an underappreciated contributor to disparities in the incidence of T2D.17
Long-term exposure to environmental pollutants combined with frailty and a decline of immune function makes the elderly population particularly susceptible to T2D.18 Cohort studies have revealed that exposure to endocrine-disrupting chemicals (EDCs) disturbed the onset and progression of T2D among older Swedish and Korean adults.19–21 As a substitute for polybrominated diphenyl ethers, organophosphate esters (OPEs) are commonly used as flame retardants and plasticizers,22 and the global annual consumption of OPEs exceeded 680,000 tons in 2015, with 30% of the usage was reported in China.23 As a result, OPEs are ubiquitously detected in Chinese individuals in the microgram to milliliter range.24 Such high OPE exposure levels adversely impact the reproductive, endocrine, cardiovascular, and nervous systems.25,26 In vivo and in vitro studies have reported that OPEs may cause diabetogenic effects27–29; however, evidence confirming these findings at the population level is relatively scarce,30,31 and the mechanisms underlying the indicated association remain largely unknown. Further research and intervention are required to reduce the individual and societal burden of diabetes among the Chinese elderly population. Such studies should incorporate a comprehensive clarification of modifiable diabetogenic OPEs and the mechanistic standpoints in terms of the etiology of T2D. Furthermore, high-throughput multi-omics (e.g., transcriptomics and metabolomics) approaches can be used to fill in the knowledge gaps between exposure to pollutants and the pathophysiology of T2D.32–34
The present study was part of a well-characterized longitudinal multi-omics study, the China Biomarkers of Air Pollutant Exposure (BAPE) Study, conducted among Chinese individuals 60–69 years of age, with monthly longitudinal monitoring conducted over 5 months.35 The study aimed to a) explore the associations between internal OPE exposure and glycometabolic markers, b) identify the key OPEs associated with glycometabolic markers using mixture effect analysis, and c) elucidate the potential biological perturbations and mechanisms underlying the associations between key OPEs and glucose homeostasis using multi-omics approaches and an adverse outcome pathways (AOPs) model.
Methods
Study Design and Population
This investigation draws on the China BAPE longitudinal panel study that included a visit each month for 5 months to healthy older Chinese adults 60–69 years of age. In brief, 76 healthy adults 60–69 years of age (50% males, 100% Chinese Han nationality) were recruited from within the Jinan megacity, China. Each participant was assessed five times over a 5-month period from 10 September 2018 to 19 January 2019, with a 1-month interval between each assessment. To minimize the potential impact of different dietary patterns, all participants were freely provided three standardized and nutritionally balanced meals (e.g., rice, meat, vegetable, and fruit; in line with local dietary habits) per day for 5 continuous days before bio-sample collection and physical examination. All of the participants were required to finish a questionnaire regarding family information, personal information, and time–activity patterns prior to the physical examination. During the physical examination, fasting venous blood samples and midstream urine samples were collected from the participants at 0700 hours and immediately stored at −80°C until further processing. Of the 76 participants, 58 (76%) finished 5 visits; 12 (16%) finished 4 visits; 3 (4%) finished 3 visits; and 3 (4%) finished 2 visits. A total of 353 person-visits were included in our analyses. The study protocol was approved by the ethics committee of the National Institute of Environmental Health (NIEH, Chinese Center for Disease Control and Prevention, No. 201816). Explicit written informed consent was acquired from all the participants. The detailed protocol of the China BAPE Study has been described in a previous publication.35
Characterization of OPE Exposure
Seventeen OPEs were measured in the blood samples (n=352), and 11 OPE metabolites (m-OPEs) were evaluated in the urine samples (n=353) (Table S1). Analytical procedures for the measurement of OPEs and the internal concentrations of OPEs were previously reported.36 Urine creatinine (Cr) was measured using a Flex reagent cartridge in a modified kinetic Jaffe assay (model RxL; Dade Behring). Cr-adjusted concentrations of urine m-OPEs were used in further analyses. Concentrations less than the limit of detection (LOD) were imputed using the LOD divided by 2 for each OPE biomarker. The pairwise correlations of the OPE exposures were calculated using the Spearman coefficient. All participants were self-reported nonsmoking participants; thus, plasma cotinine levels were measured using Hypersil GOLD C18 Selectivity HPLC Columns (ThermoFisher Scientific) interfaced with an LC-Q-Exactive Orbitrap Mass Spectrometer (ThermoFisher Scientific) to assess whether the participants were either actively or passively exposed to tobacco.
Assessment of Glycometabolic Markers
Glycometabolic marker levels, including fasting plasma glucose (FPG) and glycated serum protein [GSP, an indicator of short-to-medium–term (latest 2–3 wk) average blood glucose levels], were measured in all blood samples (n=353) at Calibra Diagnostics Co. Ltd. using the Cobas 8000 c702 and Cobas 6000 c501 (Roche) modular analyzer series, respectively. Fasting insulin (FINS) levels were determined using a Millipex Human Metabolic Hormone Panel V3 (HMHEMAG-34K-07; Merck) with a fluorescence detection system (Magpix; Luminex Corporation) and the xPonent 4.2 (Luminex Corporation) and Bio-Plex Manager (version 6.1; Bio-Rad) software, according to the manufacturer’s protocol. Homeostatic model assessment for insulin resistance (HOMA-IR) was performed using the following formula: [FPG (in millimoles/liter)×FINS (in micro units/milliliter)]/22.5.37
Transcriptome Analysis
Total RNA was extracted from the leukocytes of venous blood samples (n=346). In brief, leukocytes were isolated from 5mL of whole blood. After centrifugation at 3,000 rpm for 5 minutes, the leukocytes aggregated in the medium were extracted and washed two times with phosphate-buffered saline. The leukocytes were then lysed in 1mL TRIzol Reagent (Invitrogen Corp.). Next, the total RNA was assessed for quality and quantified using a NanoDrop ND-2000 (ThermoFisher Scientific) and Agilent 2100 Bioanalyzer. Library preparation with 100 ng of the extracted RNA was conducted through the TruSeq Stranded Total RNA Library Prep Kit (Illumina, Inc.). Last, the RNA was sequenced using the Illumina Hiseq X Ten System (Novogene). Genes were identified by using HISAT2 software and comparing the data with the human database.38 Gene expression quantification was conducted using the featureCounts read summarization program in the Subread software package.39
Nontargeted Metabolome Assessment
Nontargeted metabolomics analyses were performed in serum (n=353) and urine (n=346) samples. Sample processing, quality control and data extraction, metabolite identification, data curation, data quantification, and data normalization were conducted according to previously published protocols.40 In brief, the samples were precipitated with methanol for 2 min under severe vibration (GenoGrinder 2000), and then the metabolites were extracted by centrifugation. The supernatant was then separated into five fractions: Two fractions were analyzed by two different reverse phase/ultra-performance liquid chromatography–mass spectrum (RP/UPLC-MS/MS) methods using positive ion mode electrospray ionization (ESI). The remaining three fractions were analyzed by hydrophilic interaction liquid chromatography (HILIC)/UPLC-MS/MS with negative ion mode ESI, RP/UPLC-MS/MS with negative ion mode ESI, or kept for backup, respectively. The samples were put in TurboVap (Zymark) for a short time to remove the organic solvent and then kept at –80°C overnight until subsequent analysis.
The metabolomics analysis was conducted using Waters ACQUITY UPLC and Q-Exactive high resolution/accurate mass spectrometer (Thermo Fisher), which were connected to an Orbitrap mass analyzer running at 35,000 mass resolution and a heated ESI source. The first and second aliquots of sample extracts were analyzed following acidic positive ion conditions that were chromatographically optimized for more hydrophilic and hydrophobic compounds, respectively. The third aliquot of sample extracts was gradient eluted from a separate dedicated C18 column and analyzed using basic negative ion–optimized conditions. The fourth aliquot of sample extracts was eluted from a HILIC column and analyzed via negative ionization. The mass spectrum analysis used dynamic exclusion that alternated between MS or data-dependent multistage mass spectrometry (MSn) scans with scan range covering 70–1,000 m/z.
After the raw mass spectrum data extraction and peak recognition, the metabolites were identified by comparison with an internal library. The library was based on certified standards that included a) m/z, b) retention time, and c) chromatographic data of all molecules. In addition, metabolite identifications were based on the following three criteria: a) retention time within 100 ms tolerance, b) accurate mass match to the internal library (10-ppm mass tolerance), and b) MS/MS fractions based on the ions present in the experimental spectrum compared with the ions present in the library spectrum. To ensure data quality, all peaks were manually checked. Then the area under the curve was used to quantify and all data were normalized before further analysis.
For quality control (QC) purposes, three types of controls were analyzed in concert with the experimental samples: a) process blanks, b) mixed matrix samples (10μL of each sample), and c) a cocktail of QC standards spiked into samples. All the experimental samples were randomly distributed on the platform and QC samples were also evenly distributed among the injections.
Statistical Analyses
All data were analyzed according to the following pipeline: a) a linear mixed-effects model (LMM) was conducted to explore the associations between OPE exposures and the glycometabolic markers; b) quantile g-computation (qgcomp)41 was used to assess the effects of OPE mixtures on the glycometabolic markers and to identify the most important OPEs with relative positive weights >10% within the OPE mixtures; c) the associations between each key OPE and each biomolecule (transcripts and metabolites obtained using multi-omics profiling) were examined using LMM, and the biomolecules mediating the impact of each key OPE on a specific glycometabolic marker were determined using the causal inference test (CIT)42; and d) Integrative Pathway Analysis (IPA) of the biomolecular intermediators was conducted to investigate the underlying biological mechanisms43 (Figure 1). The detailed processes are described in the following sections.
Figure 1. Overview of the study design. Diagram of the present study design. Internal OPE exposure of 76 healthy older Chinese adults 60–69 years of age with five monthly longitudinal sample (blood and urine) collections was characterized previously. Glycometabolic markers (FPG, GSP, FINS, and HOMA-IR) were measured, and multi-omics profiling (peripheral blood transcriptome, serum metabolome, and urine metabolome) was conducted. Exposure–health outcome associations and multi-omics integrative analyses were further used to identify the key OPEs and to interpret the biological mechanisms underlying the perturbations of glycometabolic markers, respectively. Note: CIT, causal inference test; FINS, fasting insulin; FPG, fasting plasma glucose; GSP, glycated serum protein; HOMA-IR, homeostatic model assessment for insulin resistance; IPA, Ingenuity Pathway Analysis; LMM, linear mixed-effects model; OPE, organophosphate ester; qgcomp, quantile g-computation.
Figure 1 is an illustration with three parts. On the top, the illustration is titled Population and has four steps. Step 1: Under Survey, there are 72 older adults aged 60 to 69 years with 353 measurements. From visits 1 to 5, blood and urine samples were collected. Step 2: organophosphate esters Step 3: Glycometabolism markers, including fasting plasma glucose, glycated serum protein, fasting insulin, and homeostatic model assessment for insulin resistance. Step 4: Transcriptome and metabolome, including genes and metabolites. In the middle, the illustration is titled Exposure–Outcome Association Analysis and has two steps. Step 1: Linear mixed-effects model: organophosphate esters include fasting plasma glucose, glycated serum protein, fasting insulin, and homeostatic model assessment for insulin resistance. Step 2: Quantile g-computation At the bottom, the illustration titled Multi-omics Integration has three steps. Step 1: Key organophosphate esters Step 2: The blood transcriptome, serum metabolome, and urine metabolome lead to causal inference tests plus Ingenuity Pathway Analysis. Step 3: Glycometabolism markers.
LMM
The measurement values of certain OPE exposures and glycometabolic markers were logarithm (base 10)-, square root-, power one-third-, square-, or cubic-transformed to approach Gaussian distribution (Table S1). The maximum missing rate for certain exposure variables (blood OPEs) was 0.28% owing to the runout of one blood sample. After imputing the missing data for exposures using a chained equation (mice package with the predictive mean matching method),44 the exposures were standardized through z-score normalization to denote a change of 1 standard deviation (SD) in the glycometabolic marker values to facilitate their comparisons. A LMM with participant-specific intercepts and “unstructured” covariance structure was used to independently assess the associations between exposure to each OPE and each glycometabolic marker. Each main model was adjusted for the set of predefined adjustment factors according to previous studies45,46: age (continuous), sex (female/male), body mass index [BMI; continuous; calculated as weight divided by height squared (in kilograms per meter squared)], education level (below primary school, primary school, junior school, senior high school, and university), financial income [continuous; annual household income (in 10,000 CNY)], plasma cotinine concentration (continuous), and other diet (the total frequency of extra food consumption in addition to the provided standardized meals during the 3-d investigation). Among them, information on age, sex, education level, and financial income was assessed at the first visit for all of the participants, whereas other time-varying factors (BMI, plasma cotinine concentration, and other diet) were assessed at all five visits. Other diet information was recorded using a daily time–activity questionnaire at each visit, and the summarized results are shown in Table S2. Stratification analysis of the associations between OPE exposures and glycometabolic markers were conducted by sex. Multiple hypothesis testing-corrected p-values were obtained by calculating the false discovery rate (FDR) and the estimated proportion of false discoveries made vs. the number of total discoveries made at a given significance level (α).47 In multiple testing corrections, FDR was statistically significant at <5%. β estimates and standard errors from the models were converted to percentage change values with the 95% confidence intervals (95% CIs) associated with 1-SD increases in individual OPE concentrations.
qgcomp
The OPEs that were found to be significantly associated with each glycometabolic marker after multiple testing using LMM were packed as a chemical mixture. The qgcomp method was performed to assess the effects of the OPE mixtures on the glycometabolic markers based on parameter inference via the qgcomp package in R. This approach combined the inferential simplicity of weighted quantile sum regression (WQS) with the flexibility of g-computation.41 The advantage of qgcomp was that exposures could interact with outcomes in any directions. Gaussian distributions were specified as link functions, and parameter q was set to four in the linear model. Five hundred bootstrap iterations were performed to calculate the 95% CI for each mixture. With the exception of predefined adjustment factors, visit time was included as an adjusting covariate for qgcomp analysis. OPEs with positive weights of >10% were identified as key components.
CIT
To identify the biomolecular intermediates, the causal relationship inference of OPEs–biomolecular intermediators–glycometabolic markers was assessed using the CIT.42 In brief, each OPE–metabolite/transcript–glycometabolic marker relationship was individually analyzed to classify the components as consequential to/mediated by/independent of expression of gene or metabolite. In this study, causal inference was defined by the following four criteria: a) the OPEs and glycometabolic markers were significantly correlated, b) the OPEs were significantly associated with the expression of biomolecules after adjusting for the glycometabolic markers, c) the expression of biomolecules was significantly associated with the glycometabolic markers after adjusting for OPEs, and d) the OPEs were independent of the glycometabolic markers after adjusting for the expression of biomolecules. In addition to predefined adjustment factors, visit time (month of sample collection) was included as an adjustment covariate for the CIT model. To summarize the p-values for the whole CIT, the intersection–union test framework was used as the maximum p-value of the four test components.42
IPA
IPA canonical pathway analysis of all the significant biomolecular intermediators (both genes and metabolites) of the causal relationship between each OPE and glycometabolic markers was performed using IPA software (version 68752261; Qiagen; https://digitalinsights.qiagen.com/IPA). The corresponding biological pathways of these transcripts and metabolites were identified for each OPE independently. Canonical pathways with a p<0.05 (Fischer’s exact test) were regarded as statistically significant. IPA helps to uncover new biological insights and interpretations by combining an enormous schemata of existing knowledge from the literature with a massive collection of gene and metabolite expression measurements.
Sensitivity Analysis
Sensitivity analysis was performed to assess various models by controlling for one of the following covariates in the main model: month of sample collection, tea consumption (number of cups over 3 d), and frequency of alcohol consumption (over 3 d). A two-sided p<0.05 in sensitivity analysis indicated statistical significance. All statistical evaluations were conducted using R (R Development Core Team) with the lme4 and qgcomp packages.
Results
Overview of OPE Exposure and Glycometabolic Markers
Participant characteristics are presented in Table 1. Of all participants, 38 (50%) were women, and the mean±SD participant age was 64.5±4.5 y. The concentrations of the 17 blood OPEs and 11 urinary OPE metabolites over five longitudinal visits are summarized in Figure 2A and Table S1 (see also Excel Table S1). The concentrations of tributyl phosphate (TBP), triphenyl phosphate (TPHP), and tri(1-chloro-2-propyl) phosphate (TCPP) were relatively high in the blood samples, whereas those of bis(2-chloroethyl) phosphate (BCEP), di(2-ethylhexyl) phosphate (DEHP), and bis(1,3-dichloro-2-propyl) phosphate (BDCPP) were relatively high in the urine samples. The pairwise Spearman correlations of the OPE exposures are shown in Figure S1. Meanwhile, four glycometabolic markers (FPG, GSP, FINS, and HOMA-IR) reflecting the status of insulin resistance and glycometabolic homeostasis are shown in Figure 2B (see also Excel Table S2).
Table 1 Demographics of the study participants in the China BAPE Study 2018–2019 (n=76 with 353 measurements).
Variables n (%) or mean±SD
Age (y) 65.1±2.8
Sex (male) 38 (50.0)
BMI (kg/m2) 25.0±2.4
Highest education level
Primary school or below 8 (10.5)
Junior middle or high school 54 (71.1)
College graduate or beyond 14 (18.4)
Annual financial income (×10,000 CNY)
≤7 25 (32.9)
7–10 26 (34.2)
>10 25 (32.9)
Tea consumption (number of cups/3 d) 8.0±10.4
Alcohol consumption (frequency/3 d) 0.02±0.2
Plasma cotinine concertation (ng/mL) 1.0±5.7
Other diet (frequency/3 d) 2.2±2.6
Month of sampling (n samples)
September 66 (18.7)
October 74 (21.0)
November 71 (20.1)
December 71 (20.1)
January 71 (20.1)
Note: BAPE, Biomarkers of Air Pollutant Exposure; BMI, body mass index; SD, standard deviation.
Figure 2. Characterization of the internal exposure of OPEs and glycometabolic markers as part of this study (n=76 with 353 measurements). (A) Profiling of the internal exposures (blood and urine) of OPEs among healthy older adults over the five monthly longitudinal visits (see also Excel Table S1). (B) Distribution of glycometabolic markers among healthy older adults over the five monthly longitudinal visits (see also Excel Table S2). Note: Cr, creatinine; Dec, December; FINS, fasting insulin; FPG, fasting plasma glucose; GSP, glycated serum protein; HOMA-IR, homeostatic model assessment for insulin resistance; Jan, January; Nov, November; Oct, October; OPE, organophosphate ester; Sep, September.
Figure 2A is a set of two graphs titled Blood and Urine, plotting chemical rank by concentration, ranging from 0 to 10 in increments of 10 (y-axis) across concentration in blood (micrograms per liter), ranging as 10 begin superscript negative 2 end subscript and 10 begin superscript negative 1 end subscript and concentration in urine (micrograms per gram creatinine), ranging as 10 begin superscript negative 2 end subscript and 10 begin superscript negative 1 end subscript (x-axis) for T B P, T P H P, T C P P, T I B P, T E P, T C E P, T N B P, T E H P, T M P, T B O E P, T D C P P, E H D P P, C D P P, R D P, T M P P, T P R P, B A B P; and B C E P, D E H P, B D C P P, D P H P, B B O E P, D B P, B C I P P, D P H P-O H, B M P P, B B O E H E P, E H D P P-O H in the months of September to January, respectively. Figure 2B is a graph, plotting a homeostatic model assessment for insulin resistance, fasting insulin (micro units per milliliter), fasting plasma glucose (millimoles per liter), and glycated serum protein (micromoles per liter) (y-axis) across concentration in blood, ranging as 10 begin superscript negative 1 end subscript, 10 begin superscript 0 end subscript, 10 begin superscript 1 end subscript. 10 begin superscript 2 end subscript, and 10 begin superscript 3 end subscript (x-axis).
Association between OPE Exposure and Glycometabolic Markers
The percentage change in estimates and 95% CIs of the glycometabolic markers for a 1-SD increase in the concentration of each OPE are shown in Figure 3A and Excel Table S3. After multiple testing corrections, 6, 7, 3, and 9 OPEs were found to be significantly associated with FPG [TPHP, trimethylolpropane phosphate (TMPP), 2-ethylhexyl diphenyl phosphate (EHDPP), diphenyl phosphate (DPHP), dibutyl phosphate (DBP), and BCEP], GSP [TPHP, TMPP, EHDPP, hydroxyphenyl 2-ethylhexyl diphenyl phosphate (EHDPP-OH), DPHP, DBP, and BCEP], FINS (TPHP, DPHP, and DBP), and HOMA-IR [TPHP, tri-n-butyl phosphate (TnBP), TMPP, tri-iso-butyl phosphate (TiBP), TBP, EHDPP, DPHP, DBP, and BCEP], respectively. Of these OPEs, blood TPHP and urinary DBP, and DPHP showed an undesirable positive influence on all four glycometabolic markers. The results of the sensitivity analysis were consistent with the findings of the main model (Figure S2 and Excel Table S4). A stratification analysis of effect modification of sex was conducted (Figure S3 and Excel Table S5), and the correlation directions and significance between OPE exposure and glycometabolic marker changes in different sex were generally consistent.
Figure 3. OPE exposure–glycometabolic marker association and qgcomp analyses–determined exposure to the key OPEs among healthy older Chinese adults 60–69 years of age. (A) Forest plots of the results of the LMM between OPE exposure and glycometabolic markers (FPG, GSP, FINS, and HOMA-IR). The FDR-adjusted p-values of each predictor are given as *FDR <0.05 (see also Excel Table S3). (B) Effect diagrams of the changes in the z-scores of glycometabolic markers with a quantile increase in the mixture concentration (see also Table S3). (C) Bar plots of the relative weight of each pollutant within four chemical mixtures constructed to assess their effects on glycometabolic markers (see also Table S4). Note: FDR, false discovery rate; FINS, fasting insulin; FPG, fasting plasma glucose; GSP, glycated serum protein; HOMA-IR, homeostatic model assessment for insulin resistance; LMM, linear mixed-effects model; OPE, organophosphate ester; qgcomp, quantile g-computation.
Figure 3A is a set of two error bar graphs titled Blood organophosphate esters and Urine organophosphate esters, plotting percentage change (percent), ranging from negative 10 to 20, in increments of 10, negative 5 to 15 in increments of 5; negative 5 to 5 in increments of 5; and negative 5 to 10 in increments of 5 (left y-axis) and homeostatic model assessment for insulin resistance, fasting insulin, glycated serum protein, and fasting plasma glucose (right y-axis) across B A B P, C D P P, E H D P P, R D P, T B O E P, T B P, T C E P, T C I P P, T D C P P, T E H P, T E P, T I B P, T M P, T M P P, T n B P, T P H P, T P r P, B B O E H E P, B B O E P, B C E P, B C I P P, B D C P P, B M P P, D B P, D E H P, D P H P, D P H P−O H, and E H D P P−O H (x-axis) for trend, positive, negative, and not significant. Figure 3B is a set of four graphs titled fasting plasma glucose, glycated serum protein, fasting insulin, and homeostatic model assessment for insulin resistance, plotting Uppercase y, ranging from 0.7 to 1.1 in increments of 0.1; 2.4 to 2.7 in increments of 0.1; 0.3 to 0.7 in increments of 0.1; and negative 0.2 to 0.4 in increments of 0.2 (y-axis) across joint exposure quantile, ranging from 0.00 to 1.00 in increments of 0.25 (x-axis), respectively. Figure 3C is a set of four bar graphs titled fasting plasma glucose, glycated serum protein, fasting insulin, and homeostatic model assessment for insulin resistance, plotting B C E P, D B P, D P H P, E H D P P, T M P P, T P H P; B C E P, D B P, D P H P, E H D P P−O H, T M P P, T P H P; D B P, D P H P, T P H P; and B C E P, D B P, D P H P, E H D P P, T B P, T I B P, T M P P, T n B P, and T P H P (y-axis) across negative weights, ranging from 1 to 0 in increments of 0.5 and positive weights, ranging from 0 to 1 in increments of 0.5 (x-axis), respectively.
Identification of the Key OPEs
After multiple testing corrections, OPEs that were associated with the four glycometabolic markers were grouped as four different mixtures for subsequent qgcomp analyses. Figure 3B shows the log-transformed concentrations of four OPE mixtures that were positively associated with the z-scores of FPG (p<0.001), GSP (p=0.001), FINS (p=0.002), and HOMA-IR (p=0.005), respectively (see also Table S3). Each quartile increment in the mixture concentration was associated with elevated z-scores of 0.06 (95% CI: 0.03, 0.08), 0.04 (95% CI: 0.02, 0.05), 0.06 (95% CI: 0.02, 0.09), and 0.12 (95% CI: 0.04, 0.20) for FPG, GSP, FINS, and HOMA-IR, respectively. In Figure 3C and Table S4, the weight of each OPE reflects the contribution of the correlated components to the overall mixture effect. Within the mixtures, TMPP had the highest weight (54%) for elevated FPG, followed by DBP (24%) and TPHP (14%), whereas exposure to BCEP and DPHP contributed minimally (<10%) to the overall mixture positive effect. TMPP contributed to 46% of the overall positive mixture effect of elevated GSP, followed by DBP and TPHP (26% and 19%, respectively). TPHP and TiBP had the highest weights of contribution to FINS and HOMA-IR for positive weight, at 44% and 29%, respectively. For negative weight, EHDPP had the highest weights for FPG, GSP, and HOMA-IR with 100%, 97%, and 57%, respectively. Ultimately, five OPEs (TMPP, TPHP, TiBP, DBP, and DPHP) with positive weights of >10% were considered the key OPEs and were suspected to be responsible for the increased risks of glycometabolic disorder and therefore were included in further analyses.
Longitudinal Correlations of OPEs–Metabolites
The numbers of endogenous metabolites, based on metabolomics classifications in the serum and urine samples, are shown in Figure 4A. The numbers and directions of the associations between the five key OPEs and the respective serum and urine metabolomes are presented in Figure 4B,C. A total of 1,354 associations were found (710 and 644 for the serum and urine metabolomes, respectively; FDR <0.05; Excel Tables S6 and S7). The network diagram shows the exposure–metabolite pairwise associations with FDR <0.001 for the serum and urine metabolomes (Figure 4D,E; Excel Tables S6 and S7; raw serum and urine metabolome data in Excel Tables S8 and S9). Specifically, for the serum metabolome, the top three OPEs of high connective degree included TPHP, TMPP, and DBP. The top three metabolite classes with positive associations were peptides (20%), cofactors and vitamins (12%), and energy (12%), whereas those with negative associations were peptides (18%), carbohydrates (15%), and nucleotides (12%) (Figure 4F; Table S5). For the urine metabolome, DBP, DPHP, and TPHP were the three OPEs with the highest connective degree and were found to have mostly negative correlations. The top three metabolite classes with positive associations were carbohydrates (10%), secondary metabolites (10%), and cofactors and vitamins (8%), whereas those with negative associations were nucleotides (17%), global metabolites (12%), and cofactors and vitamins (11%) (Figure 4G; Table S6).
Figure 4. Serum and urine metabolome profiling of exposure to the key OPEs among healthy older Chinese adults 60–69 years of age. (A) The number of endogenous metabolites based on metabolic classifications in the serum and urine. (B,C) Volcano plots of the coefficient estimates of the key OPEs vs. the FDR values in the associations of the exposure–serum metabolome (B; see also Excel Table S6) and exposure–urine metabolome (C; see also Excel Table S7). Coefficient estimates are expressed as percentage changes (%) in FPG, GSP, FINS, and HOMA-IR per 1-SD change in each exposure, which was previously transformed to approach normality. The dashed horizontal line shows where the FDR value equals 0.05. (D,E) Network diagram of the association analysis between exposure to the key OPEs and the serum/urine metabolome (only metabolites with associations of FDR <0.001 are shown; see also Excel Tables S6 and S7). The size of the node represents the degree of the exposure–metabolite connection, and the color of the edge represents the coefficient estimate of the exposure–metabolite association. (F,G) Stacking histograms of the percentages (%) of positively and negatively associated metabolites within each class, as well as the overall average for the serum and urine metabolomes with an FDR value of <0.05 (see also Tables S5 and S6). Note: FDR, false discovery rate; FINS, fasting insulin; FPG, fasting plasma glucose; GSP, glycated serum protein; HOMA-IR, homeostatic model assessment for insulin resistance; OPE, organophosphate ester; SD, standard deviation.
Figure 4A is set of two pie charts. The first pie chart is titled Serum metabolome and displays the following information: Lipid count is 429, amino acid count is 210, xenobiotics count is 153, peptide count is 44, nucleotide count is 39, cofactors and vitamins count is 38, carbohydrate count is 22, partially characterized molecules count is 12, and energy count is 10. The second pie chart is titled Urine metabolome and displays the following information: amino acid count is 239, carbohydrate count is 73, nucleotide count is 59, lipid count is 46, cofactors and vitamins count is 40, global count is 29, other secondary metabolites count is 10, energy count is 5, xenobiotics count is 3, terpenoids and polyketides count is 1, and other count is 444. Figure 4B is a set of five volcano plots titled T I B P, T M P P, T P H P, D B P, and D P H P, plotting negative log 10 of false discovery rate, ranging from 0 to 6 in increments of 2; 0.0 to 10.0 in increments of 2.5; 0 to 20 in increments of 10; 0 to 9 in increments of 3; and 0 to 6 in increments of 2 (y-axis) across percentage changes (percent), ranging from negative 50 to 50 in increments of 25 (x-axis) for positive, negative, and not significant. Figure 4C is a set of five volcano plots titled T I B P, T M P P, T P H P, D B P, and D P H P, plotting negative log 10 of false discovery rate, ranging from 0 to 5 in unit increments; 0 to 5 in unit increments; 0 to 6 in increments of 2; 0 to 9 in increments of 3; and 0.0 to 7.5 in increments of 2.5 (y-axis) across percentage changes (percent), ranging from negative 50 to 50 in increments of 25 (x-axis) for positive, negative, and not significant. Figure 4D is a network diagram depicting the association between exposure to the key organophosphate esters, including T P H P, T M P P, D P H P, T I B P, and D B P, and the serum or urine metabolome. The size of the node represents the degree, including 40, 80, 120, and 160, of the exposure–metabolite connection, and the shade of the edge represents the coefficient estimate of the exposure–metabolite association. The scale depicts effects ranging from negative 0.3 to 0.3 in increments of 0.3. The class includes organophosphate esters, xenobiotics, carbohydrate, nucleotide, peptide, other secondary metabolites, global, lipid, amino acid, cofactors and vitamins, energy, partially characterized molecules, terpenoids and polyketides, and other. Figure 4E is a network diagram depicting the association between exposure to the key organophosphate esters, including T P H P, D P H P, and D B P, and the serum or urine metabolome. The size of the node represents the degree, including 30, 60, 90, of the exposure–metabolite connection, and the shade of the edge represents the coefficient estimate of the exposure–metabolite association. The scale depicts effects ranging from negative 0.3 to 0.3 in increments of 0.3. The class includes organophosphate esters, xenobiotics, carbohydrate, nucleotide, peptide, other secondary metabolites, global, lipid, amino acid, cofactors and vitamins, energy, partially characterized molecules, terpenoids and polyketides, and other. Figures 4F and 4G, each are a set of two horizontal stacked bar graphs, Overall, T i B P, T M P P, T P H P, D B P, D P H P (y-axis) across percentage of positive metabolite, ranging from 200 to 0 in decrements of 50 and percentage of negative metabolite, ranging from 0 to 200 in increments of 50 (x-axis) for class, including Amino Acid, Cofactors and Vitamins, Lipid, Partially Characterized Molecules, Xenobiotics, Carbohydrate, Energy, Nucleotide, Peptide, Global, Other, and Other secondary metabolites.
Longitudinal Correlations of OPEs–Transcripts
The associations between five key OPEs and the blood transcriptome were also explored, and the numbers and directions of the associations are displayed using volcano plots (Figure 5A; Excel Table S10). A total of 16,585 associations (8,673 positive and 7,912 negative) were significant. The network diagram shows the OPE–transcript pairwise associations of the top 100 transcripts selected for absolute effect value, with an FDR of <0.001 for each OPE (Figure 5B; Excel Table S10). Specifically, the top three OPEs correlated with gene expressions of high connective degree were TPHP (3,875 positive and 3,476 negative associations), DBP (3,056 positive and 2,653 negative associations), and TMPP (1,741 positive and 1,783 negative associations) (Figure 5C).
Figure 5. Transcriptome profiling of exposure to the key OPEs among healthy older Chinese adults 60–69 years of age. (A) Volcano plots of the coefficient estimates for exposure to the key OPEs vs. the FDR values among the exposure–transcriptome associations (see also Excel Table S10). Coefficient estimates are expressed as percentage changes (%) of FPG, GSP, FINS, and HOMA-IR per 1-SD change in each exposure, which was previously transformed to approach normality. The dashed horizontal line shows where the FDR value equals 0.05. (B) Network diagram of the association analysis between exposure to the key OPEs and the blood transcriptome (only the top 100 transcripts of each exposure with associations of FDR <0.001 are shown). The size of the node represents the degree of exposure–transcript connection, and the color of the edge represents the coefficient estimates of exposure–transcript association (see also Excel Table S10). (C) Stacking histograms of the counts of the positively and negatively associated transcripts for each exposure with an FDR value of <0.05. (D) Tripartite network of the inferred causal relationships (top 150 genes selected for the absolute effect value for each OPE with an FDR value of <0.1) of exposure to the key OPEs to the outcome (FPG and GSP) through transcriptome mediators (see also Excel Table S12). The size of the node represents the degree of exposure–transcript–outcome connection. (B,D) share a common legend. Note: FDR, false discovery rate; FINS, fasting insulin; FPG, fasting plasma glucose; GSP, glycated serum protein; HOMA-IR, homeostatic model assessment for insulin resistance; OPE, organophosphate ester; SD, standard deviation.
Figure 5A is a set of five volcano plots titled T i B P, T M P P, T P H P, D B P, D P H P, plotting negative log 10 of (false discovery rate), ranging from 0 to 3, in unit increments; 0 to 5 in unit increments; 0 to 15 in increments of 5; 0.0 to 12.5 in increments of 2.5; and 0 to 3 in unit increments (y-axis) across percentage change (percent), ranging from negative 50 to 50 in increments of 25 (x-axis) for positive, negative, and not significant. Figure 5B is a network diagram depicting the association between exposure to the key organophosphate esters, including T P H P, T M P P, and D B P, and the blood transcriptome. The size of the node represents the degree, including 25, 50, 75, 100, of the exposure–transcript connection, and the shade of the edge represents the coefficient estimate of the exposure–transcript association. The scale depicts effects ranging from negative 0.3 to 0.3 in increments of 0.3. Figure 5C is a set of two horizontal bar graphs, plotting organophosphate esters, including T M P P, T P H P, and D B P (y-axis) across Count of positive transcripts, ranging from 4,000 to 0 in decrements of 1,000 and count of negative transcripts, ranging from 0 to 4,000 in increments of 1,000 (x-axis). Figure 5D is a tripartite network, depicting the inferred causal relationships, including the top 150 genes selected for the absolute effect value for each organophosphate ester, including T M P P, D B P, T P H P, G S P, and F P G, with a false discovery rate of exposure to the key organophosphate esters to the outcome, including fasting plasma glucose and glycated serum protein through transcriptome mediators. The size of the node represents the degree, including 30, 60, and 90 of exposure—transcription—outcome connection. The class includes organophosphate esters, fasting plasma glucose, glycated serum protein, fasting insulin, homeostatic model assessment for insulin resistance, and chromosomes, including from 1 to 22, uppercase x, and uppercase m t.
Biomolecules Mediating the Effects of OPEs on Glycometabolic Markers
A tripartite network plot of OPEs–metabolites–glycometabolic markers constructed using all of the significant metabolites and transcripts as potential biomolecular intermediators is presented in Figure 6A (see also Excel Table S11). A total of 106 serum and 29 urinary metabolites were inferred as biomolecular intermediators, with an FDR of <0.2. Specifically, 87, 60, 31, and 10 metabolites were inferred as biomolecular intermediators of the effects of TMPP, TPHP, DPHP, and DBP on glycometabolic markers, respectively (the number of common and specific metabolites of each OPE are shown in Figure S4A). In addition, 1,651 transcripts were inferred as biomolecular intermediators, with an FDR of <0.1, and a network plot of OPEs–transcript–glycometabolic markers is presented in Figure 5D (top 150 genes selected for the absolute effect value for each OPE; see also Excel Table S12). Specifically, 1,248, 649, and 116 transcripts were inferred as biomolecular intermediators of the effects of TMPP, TPHP, and DBP on glycometabolic markers, respectively (the number of common and specific transcripts of each OPE are shown in Figure S4B).
Figure 6. Inferred biomolecular intermediators of insulin resistance and glycometabolic disorders and the results of Integrative Pathway Analysis among healthy older Chinese adults 60–69 years of age. (A) A tripartite network of the inferred causal relationships (causal inference analysis with an FDR value of <0.1) between exposure to the key OPEs and the outcomes (FPG, GSP, FINS, and HOMA-IR) through serum and urine metabolite mediators. The size of the node represents the degree of the exposure–metabolite–outcome connection (see also Excel Table S11). (B) Bar plots of the representative canonical pathways of the statistically significant biomolecular intermediators (genes and serum/urine metabolites) for TMPP, TPHP, DBP, and DPHP (p<0.05, see also Excel Table S13). The negative logarithm of the p-value is displayed on the x-axis, and the color of the bar represents the gene/metabolite ratio. The dashed vertical line shows where p=0.05. The red and blue texts represent the pathways previously reported in experimental studies and reported for the first time, respectively. Note: DBP, dibutyl phosphate; DPHP, diphenyl phosphate; FINS, fasting insulin; FPG, fasting plasma glucose; GSP, glycated serum protein; HOMA-IR, homeostatic model assessment for insulin resistance; OPE, organophosphate ester; TMPP, trimethylolpropane phosphate; TPHP, triphenyl phosphate.
Figure 6A is a tripartite network, depicting the inferred causal relationships between exposure to the key organophosphate esters and the outcomes, including fasting plasma glucose, glycated serum protein, fasting insulin, homeostatic model assessment for insulin resistance through serum and urine metabolite mediators. The size of the node represents the degree, including 20, 40, and 60 of the exposure–metabolite–outcome connection. The class includes Organophosphate esters, Lipid, Xenobiotics, Amino Acid, Carbohydrate, Cofactors and Vitamins, Nucleotide, Energy, Peptide, Partially Characterized Molecules, Global Other secondary metabolites, Terpenoids and Polyketides, Other, Fasting plasma glucose, Glycated serum protein, Fasting insulin, Homeostatic model assessment for insulin resistance. Figure 6B is a set four horizontal bar graphs titled T M P P, T P H P, D B P, and D P H P, plotting R A R Activation, Induction of Apoptosis by H I V 1, G lowercase alpha s Signaling, E R K or M A P K Signaling, Pyroptosis Signaling Pathway, E R K 5 Signaling, N A D Signaling Pathway, Purine Ribonucleosides Degradation, Production of N O and R O S in Macrophages, Sumoylation Pathway, Death Receptor Signaling, Inflammasome pathway, M I F−mediated Glucocorticoid Regulation, Relaxin Signaling, Leptin Signaling in Obesity, G lowercase alpha 12 of 13 Signaling, Insulin Secretion Signaling Pathway, Insulin Receptor Signaling, P P A R lowercase alpha and R X R lowercase alpha Activation, N R F 2−mediated Oxidative Stress Response, Autophagy, M Y C Mediated Apoptosis Signaling, Apoptosis Signaling, and P I3 K and A K T Signaling (y-axis) across negative log to the base 10 of (lowercase italic p), including 1 to 3 in increments of 2 and 3 to 10 in increments of 7 (x-axis) for gene to metabolite ratio ranges from 0.25 to 1.00 in increments of 0.25.
IPA for Biomolecular Intermediators
Biomolecular intermediators of the diabetogenic effects of TPHP, TMPP, DBP, and DPHP were included in the IPA analysis. Representative canonical pathways (p<0.05) shared by at least two OPEs are presented in Figure 6B and Excel Table S13. Among these, several pathways [e.g., peroxisome proliferator-activated receptor alpha (PPARα)/retinoid X receptor alpha (RXRα) activation pathway, leptin signaling in obesity, G-protein alpha-s (Gαs) signaling, myelocytomatosis viral oncogene (MYC)-mediated apoptosis signaling, and pyroptosis signaling] simultaneously appeared in the analysis results of at least three key OPEs. Core perturbed genes and metabolites within the abovementioned representative pathways were integrated to propose putative AOPs for impaired glucose homeostasis after OPE exposure (Figure 7A). These AOPs began with possible molecular initiation events (MIEs): activation of first apoptosis signal ligand (FasL) and protease-activated receptor (PAR), as well as inhibition of tumor necrosis factor (TNF), TNF receptor (TNFR), insulin receptor (IR), lysophosphatidic acid receptor (LPAR), and glucose transporter type 4 (GLUT4). Alterations of these ligands and membrane receptors caused a series of key events (KEs) at the molecular/cellular levels, such as KE-1, those involved in aberrant expression of kinases [phosphatidylinositol 3-kinase (PI3K), protein kinase B (AKT), and glycogen synthase kinase-3 (GSK3)], the apoptosis/autophagy regulator Bcl2-antagonist of cell death protein (BAD), the transcription factor nuclear factor kappa B (NF-κB), metabolic enzymes (e.g., citrate synthase and succinate dehydrogenase), and tricarboxylic acid (TCA) cycle–related metabolites (e.g., citrate and succinate). The downstream KEs at the organ/system levels (KE-2; e.g., those involved in glucose uptake, glycogen synthesis, insulin secretion, energy metabolism, oxidative stress, and inflammation) were subsequently altered and were responsible for perturbations in glucose homeostasis eventually leading to T2D. In general, the up-regulation or down-regulation trends of most MIEs and KEs were consistent among the key OPEs, except for certain molecules (e.g., PAR, PPARα, BAD, and GSK3) for urinary DBP and DPHP.
Figure 7. AOP linking OPE exposure and adverse outcome and schematic of the putative biological mechanisms of OPEs exposure–induced type 2 diabetes outcome. (A) An AOP diagram depicting the MIEs in response to exposure to the key OPEs and the subsequent series of KEs, for example, multiple signal transduction and metabolic pathway perturbations (KE-1 at the molecular/cellular level) and impaired biological functions (KE-2 at the organ/system level), which ultimately induced glycometabolic disorder-related adverse outcomes. Alterations of genes (blue) and metabolites (purple) are represented with colored boxes. (B) Schematic of the putative biological mechanisms that mediate the linkages between OPEs exposure and apical type 2 diabetes outcome. Note: AOP, adverse outcome pathway; KEs, key events; MIEs, molecular initiating events; OPE, organophosphate ester; T2DM, type 2 diabetes mellitus.
Figure 7A is an illustration depicting the molecular initiating events in response to exposure to the key organophosphate esters, including T M P P, T P H P, D B P, D P H P and the subsequent series of key events, for example, multiple signal transduction and metabolic pathway perturbations. The key events -1 are at the molecular or cellular level and result in impaired biological functions, and the key events -2 are at the organ or system level, including inflammation, insulin secretion, fatty acid oxidation, oxidative stress, apoptosis or autophagy, glycogen synthesis, glucose uptake, and energy metabolism, which ultimately induce glycometabolic disorder-related adverse outcomes. The genes and metabolites include up with significance, down with significance, up, and down Figure 7B is an illustration, depicting the putative biological mechanisms that mediate the linkages between organophosphate esters exposure and apical type 2 diabetes outcome. The illustration is divided into three parts, namely, Exposure, Pathway, and Outcome. Exposure includes T M P P, T P H P, D B P, and D P H P. The pathway depicts the reactions on P 13 K-A K T signaling, insulin receptor signaling, apoptosis, autophagy, and oxidative stress; P P R A lowercase alpha and R X R lowercase alpha activation; leptin signaling in obesity; glucocorticoids regulation; insulin secretion signaling, and G lowercase alpha 12 of 13 signaling. In the outcome, insulin resistance plus hyperglycemia leads to type 2 diabetes mellitus.
Discussion
OPEs are widely used as a substitute for polybrominated diphenyl ethers and are ubiquitously detectable in environmental matrices and human bodies, which has caused considerable concern in terms of their adverse effects worldwide. The present study, for the first time, comprehensively profiled the associations between OPE exposure and insulin resistance and glycometabolic homeostasis among healthy older Chinese adults 60–69 years of age and systematically explored the underlying biological mechanisms using multi-omics profiling.
Identification of the Key OPEs
In this study, exposure to five key OPEs and m-OPEs (TMPP, TPHP, TiBP, DBP, and DPHP) was found to be associated with elevated levels of glycometabolic markers, independent of the traditional risk factors for T2D (e.g., age, sex, BMI, and diet). To the best of our knowledge, the observed associations were first reported at the population level; moreover, some associations had, to varying extents, a priori credibility based on the previous experimental literature (Table S7).28,29,48–62 Specifically, although TMPP, TPHP, and DPHP were widely detected in the Chinese population at concentration ranges of 0.24–1.68 ng/mL,24,63,64 to the best of our knowledge, our study is the first to reveal the adverse correlations between blood TPHP and its metabolite DPHP in urine and glucose homeostasis within humans. These findings are in line with those of previous experimental studies reporting the diabetogenic effects of TPHP and DPHP in vivo and in vitro. For instance, exposure to TPHP results in a higher glucose level and HOMA-IR index and inhibits the level of adiponectin, an insulin-sensitizing hormone, in adult male mice49 and pubertal mice.27 Likewise, numerous in vitro studies have confirmed the diabetogenic effects of TPHP and DPHP on 3T3-L1 adipose and HepG2 cells.59–61 However, no previous studies have yet reported the relationships between exposure to TMPP, TiBP, and DBP and abnormal glucose metabolism. Thus, further health hazard assessments through epidemiological studies or in vivo and in vitro experiments are warranted. In addition, it is noteworthy that the dominant pollutants differed between specific glycometabolic markers. For instance, TMPP and TPHP were the leading contributors of elevated FPG and FINS levels, respectively, possibly because of their different diabetogenic potential in triggering MIEs or downstream KEs. Therefore, subsequent CIT and multi-omics analyses were conducted to determine the biomolecular intermediators and obtain mechanistic insights into specific exposure–outcome associations.
Determination of Biomolecular Intermediators and Relevant Pathways
In this study, we comprehensively characterized the biomolecular changes in genes and metabolites induced by the key OPEs in healthy older adults for the first time. Some endogenous changes were first unveiled at the population level, including PI3K/AKT signaling [PI3K, AKT, and mitogen-activated protein kinase kinase kinase 8 (MAP3K8) inhibition], IR signaling [IR and growth factor receptor-bound protein 2 (GRB2) inhibition and IR substrate activation], PPARα/RXRα signaling (protein kinase A and PPARα activation, and RXRα and NF-κB inhibition), insulin secretion signaling [eukaryotic initiation factor 2 (EIF2) activation], and TCA cycle processes (citrate synthase, succinate dehydrogenase, and citrate inhibition). Our results were consistent with those of a previous in vitro study that reported that PI3K, AKT, and PPARα mRNAs were significantly down-regulated after TPHP treatment of L02 cells.65 Dysregulation in PI3K/AKT signaling, especially AKT activity inhibition, increases insulin resistance among patients with T2D.66 Besides, an in vivo study conducted in mice, showed that TPHP reduces the abundance of citrate and succinate within the TCA cycle52; moreover, the down-regulation of succinate dehydrogenase can impair glucose metabolism in islet cells in T2D.67 Similar to our findings, previous toxicological studies have reported that TPHP could induce cell apoptosis,68,69 autophagy,70 oxidative stress,71 and glucocorticoid regulation disruption72–74 (Table S8).48,60,68,75–93 Alterations of some other pathways (e.g., leptin signaling,94 relaxin signaling,95 and Gα12/13 signaling96) are also closely related to insulin resistance and glucose metabolism, although they have not yet been reported to induce diabetogenic effects upon OPE exposure. In addition, to a certain extent, blood TMPP and TPHP share the most common biomolecular intermediators and pathways. In contrast, urinary DBP and DPHP have relatively unique molecular characteristics, suggesting differential modes of action and diabetogenic potentials, which are perhaps caused by their different chemical structures.
Putative AOPs Linking OPE Exposure and the Outcomes of T2D
AOPs can address the existing gaps in biological knowledge and provide evidence-based mechanistic insights into the associations between environmental exposure and health outcomes. To the best of our knowledge, our study is the first to present a putative network of AOPs linking OPE exposure with possible MIEs to downstream KEs that ultimately manifest as outcomes of T2D. The AOPs presented herein revealed that OPE exposure could trigger various MIEs, for example, aberrant alterations of ligands and membrane receptors (TNF/FasL, IR, TNFR, and GLUT4) that can subsequently disturb the downstream signal transduction and metabolic pathways [e.g., PI3K-AKT signaling, IR signaling, PPARα/RXRα activation, and TCA cycle (as KE-1)] and biological processes [e.g., glucose metabolism (as KE-2)]. These pathways primarily communicate through the PI3K-AKT signaling hub, which is a well-recognized major effector of metabolic insulin action. Specifically, IRs stimulate a signaling cascade, leading to the phosphorylation/activation of IR substrates and the subsequent activation and phosphorylation of the PI3K cascade and AKT, respectively.97 This further regulates GSK3 activation98 and GLUT4 expression on the cell membrane,99 contributing to glycogen synthesis and glucose transportation, respectively. Moreover, we observed broad perturbations in the downstream events of AKT phosphorylation, with abnormal pathophysiological functions (e.g., apoptosis, autophagy, oxidative stress, and inflammation), resulting in the outcome of T2D.100 Our findings are also in line with those of previous in silico and in vitro analyses showing that TPHP exposure can trigger the PI3K/AKT pathway as a KE,65 highlighting its potential role as a promising preventive intervention or pharmacological target for OPE-associated T2D.
Strengths and Limitations
The main strength of this study is the multidimensional integration of the transcriptome and metabolome using two different biological markers (i.e., blood and urine), providing a reliable assessment of OPE exposure and endogenous biomolecular perturbations while considering the critical aging window.101 Besides, the five longitudinal repeated measurements helped avoid biases introduced by exposure measurement errors of one-time sampling and presented the molecular scenarios of the effects of differential exposure at individual levels, with the potential to explore causal associations. In addition, the availability of detailed and longitudinal data pertaining to participant demographics and time–activity helped us adjust for confounding factors in the statistical analyses and minimize the effects of other risk factors on our findings.
This study also has some limitations. First, the complete exclusion of potential uncontrolled confounders was impossible because reliance on uncontrolled longitudinal associations limits the ability to make causal determinations. Second, the population representation and sample size of our study could be inadequate, warranting further evaluations among different ethnic groups and larger population sizes with a nested case–control design or validations in toxicological research in vivo. Third, as potential EDCs, OPEs may have sexually dimorphic effects on glucose metabolism. Given the relatively small sample size and complex mixture of OPE exposures, it was underpowered to assess sex-stratified analyses in our exploratory study statistically. Future studies with a larger sample size may evaluate the sex-specific effects of these OPEs on glucose metabolism. Last, owing to the observational design of this study and the high dimensional feature of multi-omics data, caution is warranted with regard to whether these associations are causal.
Conclusions and Implications
The present exploratory study suggests that several OPEs may be independently and jointly associated with abnormal glycometabolism. High production volumes and the broad applications of OPEs in China, combined with the persistence and mobility characteristics, have led to the ever-increasing bioaccumulation of OPEs, possibly contributing to the high risk of T2D among the Chinese population. The putative AOPs linking OPE exposure to T2D, including the activation of a series of MIEs (e.g., IR, TNFR, and PAR) and the disruption of multiple downstream KEs (e.g., PI3K/AKT, PPARα/RXRα activation, and insulin secretion), were also established (Figure 7B). Our findings advance the existing knowledge on the environmental bases of the etiology of T2D among older Chinese adults 60–69 years of age and shed light on the evidence linking chemical safety and noncommunicable disease in public health.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank all the participants of the China Biomarkers of Air Pollutant Exposure (BAPE) Study, the Dianliu Community, the Ankang Community Hospital, the Shandong Center for Disease Control and Prevention (CDC), the Jinan CDC, the China BAPE Study team, and the Metabolomics Facilities in Tsinghua University Protein Research Center, as well as Hangzhou Calibra Diagnostics Ltd., Dian Diagnostics, and Metabolon.
This study was financially supported by the National Research Program for Key Issues in Air Pollution Control of China (DQGG0401), the National Natural Science Foundation of China (82025030, 81941023, and 92043301), and the National Key Research and Development Program of China (2016YFC0206500 and 2022YFC3702700) to X.S.
==== Refs
References
1. Weil AR. 2015. The growing burden of noncommunicable diseases. Health Aff (Millwood) 34 (9 ):1439, PMID: , 10.1377/hlthaff.2015.0974.26355043
2. WHO (World Health Organization). 2017. Noncommunicable Diseases Progress Monitor 2017. https://www.who.int/publications/i/item/9789241513029 [accessed May 14, 2022].
3. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB, et al. 2022. IDF Diabetes Atlas: global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract 183 :109119, PMID: , 10.1016/j.diabres.2021.109119.34879977
4. Sinclair A, Saeedi P, Kaundal A, Karuranga S, Malanda B, Williams R. 2020. Diabetes and global ageing among 65–99-year-old adults: findings from the International Diabetes Federation Diabetes Atlas, 9th edition. Diabetes Res Clin Pract 162 :108078, PMID: , 10.1016/j.diabres.2020.108078.32068097
5. Zhang P, Gregg E. 2017. Global economic burden of diabetes and its implications. Lancet Diabetes Endocrinol 5 (6 ):404–405, PMID: , 10.1016/S2213-8587(17)30100-6.28456417
6. Stančáková A, Laakso M. 2016. Genetics of type 2 diabetes. Endocr Dev 31 :203–220, PMID: , 10.1159/000439418.26824439
7. Kahn SE, Hull RL, Utzschneider KM. 2006. Mechanisms linking obesity to insulin resistance and type 2 diabetes. Nature 444 (7121 ):840–846, PMID: , 10.1038/nature05482.17167471
8. Zhang X, Zhang M, Zhao Z, Huang Z, Deng Q, Li Y, et al. 2020. Geographic variation in prevalence of adult obesity in China: results from the 2013–2014 National Chronic Disease and Risk Factor Surveillance. Ann Intern Med 172 (4 ):291–293, PMID: , 10.7326/M19-0477.31658469
9. Liu B, Du Y, Wu Y, Snetselaar LG, Wallace RB, Bao W. 2021. Trends in obesity and adiposity measures by race or ethnicity among adults in the United States 2011–2018: population based study. BMJ 5 (1 ):21–27, PMID: , 10.1136/bmj.n365.33727242
10. Wang L, Peng W, Zhao Z, Zhang M, Shi Z, Song Z, et al. 2021. Prevalence and treatment of diabetes in China, 2013–2018. JAMA 326 (24 ):2498–2506, PMID: , 10.1001/jama.2021.22208.34962526
11. Wang L, Li X, Wang Z, Bancks MP, Carnethon MR, Greenland P, et al. 2021. Trends in prevalence of diabetes and control of risk factors in diabetes among US adults, 1999–2018. JAMA 326 (8 ):1–13, PMID: , 10.1001/jama.2021.9883.34170288
12. Hales CN, Barker DJ. 1992. Type 2 (non-insulin-dependent) diabetes mellitus: the thrifty phenotype hypothesis. Diabetologia 35 (7 ):595–601, PMID: , 10.1007/BF00400248.1644236
13. Li Y, He Y, Qi L, Jaddoe VW, Feskens EJM, Yang X, et al. 2010. Exposure to the Chinese famine in early life and the risk of hyperglycemia and type 2 diabetes in adulthood. Diabetes 59 (10 ):2400–2406, PMID: , 10.2337/db10-0385.20622161
14. Lincoff AM, Nicholls SJ, Riesmeyer JS, Barter PJ, Brewer HB, Fox KAA, et al. 2017. Evacetrapib and cardiovascular outcomes in high-risk vascular disease. N Engl J Med 376 (20 ):1933–1942, PMID: , 10.1056/NEJMoa1609581.28514624
15. Li Y, Wang DD, Ley SH, Vasanti M, Howard AG, He Y, et al. 2017. Time trends of dietary and lifestyle factors and their potential impact on diabetes burden in China. Diabetes Care 40 (12 ):1685–1694, PMID: , 10.2337/dc17-0571.29046327
16. Landrigan PJ, Fuller R, Acosta NJR, Adeyi O, Arnold R, Basu NN, et al. 2018. The Lancet Commission on pollution and health. Lancet 391 (10119 ):462–512, PMID: , 10.1016/S0140-6736(17)32345-0.29056410
17. Ruiz D, Becerra M, Jagai JS, Ard K, Sargis RM. 2018. Disparities in environmental exposures to endocrine-disrupting chemicals and diabetes risk in vulnerable populations. Diabetes Care 41 (1 ):193–205, PMID: , 10.2337/dc16-2765.29142003
18. Yehuda AB, Zinger A, Durso S. 2014. The older patient with diabetes: a practical approach. Diabetes Metab Res Rev 30 (2 ):88–95, PMID: , 10.1002/dmrr.2485.24123811
19. Lind L, Zethelius B, Salihovic S, van Bavel B, Lind PM. 2014. Circulating levels of perfluoroalkyl substances and prevalent diabetes in the elderly. Diabetologia 57 (3 ):473–479, PMID: , 10.1007/s00125-013-3126-3.24337155
20. Lind PM, Zethelius B, Lind L. 2012. Circulating levels of phthalate metabolites are associated with prevalent diabetes in the elderly. Diabetes Care 35 (7 ):1519–1524, PMID: , 10.2337/dc11-2396.22498808
21. Choi YH, Kim JH, Hong YC. 2015. Sex-dependent and body weight-dependent associations between environmental PAHs exposure and insulin resistance: Korean urban elderly panel. J Epidemiol Community Health 69 (7 ):625–631, PMID: , 10.1136/jech-2014-204801.25669219
22. Blum A, Behl M, Birnbaum L, Diamond ML, Phillips A, Singla V, et al. 2019. Organophosphate ester flame retardants: are they a regrettable substitution for polybrominated diphenyl ethers? Environ Sci Technol Lett 6 (11 ):638–649, PMID: , 10.1021/acs.estlett.9b00582.32494578
23. Hu Z, Yin L, Wen X, Jiang C, Long Y, Zhang J, et al. 2021. Organophosphate esters in China: fate, occurrence, and human exposure. Toxics 9 (11 ):310, PMID: , 10.3390/toxics9110310.34822701
24. Hu L, Yu M, Li Y, Liu L, Li X, Song L, et al. 2022. Association of exposure to organophosphate esters with increased blood pressure in children and adolescents. Environ Pollut 295 :118685, PMID: , 10.1016/j.envpol.2021.118685.34923060
25. Hu W, Gao P, Wang L, Hu J. 2023. Endocrine disrupting toxicity of aryl organophosphate esters and mode of action. Crit Rev Environ Sci Technol 53 (1 ):1–18, 10.1080/10643389.2022.2050147.
26. Yang Y, Chen P, Ma S, Lu S, Yu Y, An T. 2022. A critical review of human internal exposure and the health risks of organophosphate ester flame retardants and their metabolites. Crit Rev Environ Sci Technol 52 (9 ):1528–1560, 10.1080/10643389.2020.1859307.
27. Wang C, Le Y, Lu D, Zhao M, Dou X, Zhang Q, et al. 2020. Triphenyl phosphate causes a sexually dimorphic metabolism dysfunction associated with disordered adiponectin receptors in pubertal mice. J Hazard Mater 388 :121732, PMID: , 10.1016/j.jhazmat.2019.121732.31796355
28. Du Z, Zhang Y, Wang G, Peng J, Wang Z, Gao S. 2016. TPhP exposure disturbs carbohydrate metabolism, lipid metabolism, and the DNA damage repair system in zebrafish liver. Sci Rep 6 :21827, PMID: , 10.1038/srep21827.26898711
29. Green AJ, Graham JL, Gonzalez EA, La Frano MR, Petropoulou SSE, Park JS, et al. 2017. Perinatal triphenyl phosphate exposure accelerates type 2 diabetes onset and increases adipose accumulation in UCD-type 2 diabetes mellitus rats. Reprod Toxicol 68 :119–129, PMID: , 10.1016/j.reprotox.2016.07.009.27421578
30. Bo Y, Zhu Y. 2022. Organophosphate esters exposure in relation to glucose homeostasis and type 2 diabetes in adults: a national cross-sectional study from the National Health and Nutrition Survey. Chemosphere 301 :134669, PMID: , 10.1016/j.chemosphere.2022.134669.35460677
31. Luo K, Aimuzi R, Wang Y, Nian M, Zhang J. 2020. Urinary organophosphate esters metabolites, glucose homeostasis and prediabetes in adolescents. Environ Pollut 267 :115607, PMID: , 10.1016/j.envpol.2020.115607.33254666
32. Wang S, Yong H, He XD. 2021. Multi-omics: opportunities for research on mechanism of type 2 diabetes mellitus. World J Diabetes 12 (7 ):1070–1080, PMID: , 10.4239/wjd.v12.i7.1070.34326955
33. Beulens JWJ, Pinho MGM, Abreu TC, den Braver NR, Lam TM, Huss A, et al. 2022. Environmental risk factors of type 2 diabetes—an exposome approach. Diabetologia 65 (2 ):263–274, PMID: , 10.1007/s00125-021-05618-w.34792619
34. Price EJ, Vitale CM, Miller GW, David A, Barouki R, Audouze K, et al. 2022. Merging the exposome into an integrated framework for “omics” sciences. iScience 25 (3 ):103976, PMID: , 10.1016/j.isci.2022.103976.35310334
35. Tang S, Li T, Fang J, Chen R, Cha Y, Wang Y, et al. 2021. The exposome in practice: an exploratory panel study of biomarkers of air pollutant exposure in Chinese people aged 60–69 years (China BAPE Study). Environ Int 157 :106866, PMID: , 10.1016/j.envint.2021.106866.34525388
36. Hou M, Fang J, Shi Y, Tang S, Dong H, Liu Y, et al. 2021. Exposure to organophosphate esters in elderly people: relationships of OPE body burdens with indoor air and dust concentrations and food consumption. Environ Int 157 :106803, PMID: , 10.1016/j.envint.2021.106803.34365320
37. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC. 1985. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28 (7 ):412–419, PMID: , 10.1007/BF00280883.3899825
38. Kim D, Langmead B, Salzberg SL. 2015. HISAT: a fast spliced aligner with low memory requirements. Nat Methods 12 (4 ):357–360, PMID: , 10.1038/nmeth.3317.25751142
39. Liao Y, Smyth GK, Shi W. 2014. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30 (7 ):923–930, PMID: , 10.1093/bioinformatics/btt656.24227677
40. Jia X, Jin Q, Fang J, Shi Y, Hou M, Dong H, et al. 2022. Emerging and legacy per- and polyfluoroalkyl substances in an elderly population in Jinan, China: the exposure level, short-term variation, and intake assessment. Environ Sci Technol 56 (12 ):7905–7916, PMID: , 10.1021/acs.est.2c00381.35584234
41. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ, et al. 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 128 (4 ):47004, PMID: , 10.1289/EHP5838.32255670
42. Millstein J, Zhang B, Zhu J, Schadt EE. 2009. Disentangling molecular relationships with a causal inference test. BMC Genet 10 :23, PMID: , 10.1186/1471-2156-10-23.19473544
43. Krämer A, Green J, Pollard J Jr, Tugendreich S. 2014. Causal analysis approaches in Ingenuity Pathway Analysis. Bioinformatics 30 (4 ):523–530, PMID: , 10.1093/bioinformatics/btt703.24336805
44. White IR, Royston P, Wood AM. 2011. Multiple imputation using chained equations: issues and guidance for practice. Stat Med 30 (4 ):377–399, PMID: , 10.1002/sim.4067.21225900
45. Boyle M, Buckley JP, Quirós-Alcalá L. 2019. Associations between urinary organophosphate ester metabolites and measures of adiposity among U.S. children and adults: NHANES 2013–2014. Environ Int 127 :754–763, PMID: , 10.1016/j.envint.2019.03.055.31003058
46. Ospina M, Jayatilaka NK, Wong LY, Restrepo P, Calafat AM. 2018. Exposure to organophosphate flame retardant chemicals in the U.S. general population: data from the 2013–2014 National Health and Nutrition Examination Survey. Environ Int 110 :32–41, PMID: , 10.1016/j.envint.2017.10.001.29102155
47. Storey JD. 2002. A direct approach to false discovery rates. J R Stat Soc Series B Stat Methodol 64 (3 ):479–498, 10.1111/1467-9868.00346.
48. Liu X, Zhao X, Wang Y, Hong J, Shi M, Pfaff D, et al. 2020. Triphenyl phosphate permeates the blood brain barrier and induces neurotoxicity in mouse brain. Chemosphere 252 :126470, PMID: , 10.1016/j.chemosphere.2020.126470.32443258
49. Wang D, Yan S, Yan J, Teng M, Meng Z, Li R, et al. 2019. Effects of triphenyl phosphate exposure during fetal development on obesity and metabolic dysfunctions in adult mice: impaired lipid metabolism and intestinal dysbiosis. Environ Pollut 246 :630–638, PMID: , 10.1016/j.envpol.2018.12.053.30605818
50. Wang L, Huang X, Laserna AKC, Li SFY. 2018. Untargeted metabolomics reveals transformation pathways and metabolic response of the earthworm Perionyx excavatus after exposure to triphenyl phosphate. Sci Rep 8 (1 ):16440, PMID: , 10.1038/s41598-018-34814-9.30401822
51. Philbrook NA, Restivo VE, Belanger CL, Winn LM. 2018. Gestational triphenyl phosphate exposure in C57Bl/6 mice perturbs expression of insulin-like growth factor signaling genes in maternal and fetal liver. Birth Defects Res 110 (6 ):483–494, PMID: , 10.1002/bdr2.1185.29316351
52. Wang D, Zhu W, Chen L, Yan J, Teng M, Zhou Z, et al. 2018. Neonatal triphenyl phosphate and its metabolite diphenyl phosphate exposure induce sex- and dose-dependent metabolic disruptions in adult mice. Environ Pollut 237 :10–17, PMID: , 10.1016/j.envpol.2018.01.047.29466770
53. Adams S, Wiersielis K, Yasrebi A, Conde K, Armstrong L, Guo GL, et al. 2020. Sex- and age-dependent effects of maternal organophosphate flame-retardant exposure on neonatal hypothalamic and hepatic gene expression. Reprod Toxicol 94 :65–74, PMID: , 10.1016/j.reprotox.2020.04.001.32360330
54. Walley SN, Krumm EA, Yasrebi A, Kwiecinski J, Wright V, Baker C, et al. 2021. Maternal organophosphate flame-retardant exposure alters offspring energy and glucose homeostasis in a sexually dimorphic manner in mice. J Appl Toxicol 41 (4 ):572–586, PMID: , 10.1002/jat.4066.32969501
55. Vail GM, Walley SN, Yasrebi A, Maeng A, Conde KM, Roepke TA. 2020. The interactions of diet-induced obesity and organophosphate flame retardant exposure on energy homeostasis in adult male and female mice. J Toxicol Environ Health A 83 (11–12 ):438–455, PMID: , 10.1080/15287394.2020.1777235.32546061
56. Krumm EA, Patel VJ, Tillery TS, Yasrebi A, Shen J, Guo GL, et al. 2018. Organophosphate flame-retardants alter adult mouse homeostasis and gene expression in a sex-dependent manner potentially through interactions with ERα. Toxicol Sci 162 (1 ):212–224, PMID: , 10.1093/toxsci/kfx238.29112739
57. Vail GM, Walley SN, Yasrebi A, Maeng A, Degroat TJ, Conde KM, et al. 2022. Implications of peroxisome proliferator-activated receptor gamma (PPARY) with the intersection of organophosphate flame retardants and diet-induced obesity in adult mice. J Toxicol Environ Health A 85 (9 ):381–396, PMID: , 10.1080/15287394.2021.2023716.35000574
58. Vail GM, Walley SN, Yasrebi A, Maeng A, Degroat TJ, Conde KM, et al. 2022. Implications of estrogen receptor alpha (ERa) with the intersection of organophosphate flame retardants and diet-induced obesity in adult mice. J Toxicol Environ Health A 85 (10 ):397–413, PMID: , 10.1080/15287394.2022.2026849.35045790
59. Liu Y, Le Y, Xu M, Wang W, Chen H, Zhang Q, et al. 2022. Remodeling on adipocytic physiology of organophosphorus esters in mature adipocytes. Environ Pollut 305 :119287, PMID: , 10.1016/j.envpol.2022.119287.35421551
60. Gu J, Su F, Hong P, Zhang Q, Zhao M. 2019. 1H NMR-based metabolomic analysis of nine organophosphate flame retardants metabolic disturbance in Hep G2 cell line. Sci Total Environ 665 :162–170, PMID: , 10.1016/j.scitotenv.2019.02.055.30772545
61. Cano-Sancho G, Smith A, La Merrill MA. 2017. Triphenyl phosphate enhances adipogenic differentiation, glucose uptake and lipolysis via endocrine and noradrenergic mechanisms. Toxicol In Vitro 40 :280–288, PMID: , 10.1016/j.tiv.2017.01.021.28163246
62. Hu W, Kang Q, Zhang C, Ma H, Xu C, Wan Y, et al. 2020. Triphenyl phosphate modulated saturation of phospholipids: induction of endoplasmic reticulum stress and inflammation. Environ Pollut 263 (pt A ):114474, PMID: , 10.1016/j.envpol.2020.114474.32259740
63. Liu Y, Li Y, Dong S, Han L, Guo R, Fu Y, et al. 2021. The risk and impact of organophosphate esters on the development of female-specific cancers: comparative analysis of patients with benign and malignant tumors. J Hazard Mater 404 (pt B ):124020, PMID: , 10.1016/j.jhazmat.2020.124020.33049558
64. Ya M, Yu N, Zhang Y, Su H, Tang S, Su G. 2019. Biomonitoring of organophosphate triesters and diesters in human blood in Jiangsu Province, eastern China: occurrences, associations, and suspect screening of novel metabolites. Environ Int 131 :105056, PMID: , 10.1016/j.envint.2019.105056.31369981
65. Wang X, Wang L, Li F, Teng Y, Ji C, Wu H. 2022. Toxicity pathways of lipid metabolic disorders induced by typical replacement flame retardants via data-driven analysis, in silico and in vitro approaches. Chemosphere 287 (pt 4 ):132419, PMID: , 10.1016/j.chemosphere.2021.132419.34600017
66. Batista TM, Jayavelu AK, Wewer Albrechtsen NJ, Iovino S, Lebastchi J, Pan H, et al. 2020. A cell-autonomous signature of dysregulated protein phosphorylation underlies muscle insulin resistance in type 2 diabetes. Cell Metab 32 (5 ):844–859.e5, PMID: , 10.1016/j.cmet.2020.08.007.32888406
67. Sacco F, Seelig A, Humphrey SJ, Krahmer N, Volta F, Reggio A, et al. 2019. Phosphoproteomics reveals the GSK3-PDX1 axis as a key pathogenic signaling node in diabetic islets. Cell Metab 29 (6 ):1422–1432.e3, PMID: , 10.1016/j.cmet.2019.02.012.30879985
68. Yu X, Yin H, Peng H, Lu G, Liu Z, Dang Z. 2019. OPFRs and BFRs induced A549 cell apoptosis by caspase-dependent mitochondrial pathway. Chemosphere 221 :693–702, PMID: , 10.1016/j.chemosphere.2019.01.074.30669111
69. Meng Y, Xu X, Niu D, Xu Y, Qiu Y, Zhu Z, et al. 2022. Organophosphate flame retardants induce oxidative stress and Chop/Caspase 3-related apoptosis via Sod1/p53/Map3k6/Fkbp5 in NCI-1975 cells. Sci Total Environ 819 :153160, PMID: , 10.1016/j.scitotenv.2022.153160.35051466
70. Deng Q, Jiang L, Mao L, Song XH, He CQ, Li XL, et al. 2020. The role of protein kinase C alpha in tri-ortho-cresyl phosphate-induced autophagy in human neuroblastoma SK-N-SH cells. J Appl Toxicol 40 (11 ):1480–1490, PMID: , 10.1002/jat.3999.33020912
71. Chen R, Hou R, Hong X, Yan S, Zha J. 2019. Organophosphate flame retardants (OPFRs) induce genotoxicity in vivo: a survey on apoptosis, DNA methylation, DNA oxidative damage, liver metabolites, and transcriptomics. Environ Int 130 :104914, PMID: , 10.1016/j.envint.2019.104914.31226563
72. de Guia RM, Rose AJ, Herzig S. 2014. Glucocorticoid hormones and energy homeostasis. Horm Mol Biol Clin Investig 19 (2 ):117–128, PMID: , 10.1515/hmbci-2014-0021.25390020
73. Kuo T, McQueen A, Chen TC, Wang JC. 2015. Regulation of glucose homeostasis by glucocorticoids. Adv Exp Med Biol 872 :99–126, PMID: , 10.1007/978-1-4939-2895-8_5.26215992
74. Kang S, Tsai LT, Zhou Y, Evertts A, Xu S, Griffin MJ, et al. 2015. Identification of nuclear hormone receptor pathways causing insulin resistance by transcriptional and epigenomic analysis. Nat Cell Biol 17 (1 ):44–56, PMID: , 10.1038/ncb3080.25503565
75. Umamaheswari S, Karthika P, Suvenitha K, Kadirvelu K, Ramesh M. 2021. Dose-dependent molecular responses of Labeo rohita to triphenyl phosphate. Chem Res Toxicol 34 (12 ):2500–2511, PMID: , 10.1021/acs.chemrestox.1c00281.34847329
76. Ramesh M, Angitha S, Haritha S, Poopal RK, Ren Z, Umamaheswari S. 2020. Organophosphorus flame retardant induced hepatotoxicity and brain AChE inhibition on zebrafish (Danio rerio). Neurotoxicol Teratol 82 :106919, PMID: , 10.1016/j.ntt.2020.106919.32853706
77. Canbaz D, Logiantara A, van Ree R, van Rijt LS. 2017. Immunotoxicity of organophosphate flame retardants TPHP and TDCIPP on murine dendritic cells in vitro. Chemosphere 177 :56–64, PMID: , 10.1016/j.chemosphere.2017.02.149.28282624
78. Van den Eede N, Cuykx M, Rodrigues RM, Laukens K, Neels H, Covaci A, et al. 2015. Metabolomics analysis of the toxicity pathways of triphenyl phosphate in HepaRG cells and comparison to oxidative stress mechanisms caused by acetaminophen. Toxicol In Vitro 29 (8 ):2045–2054, PMID: , 10.1016/j.tiv.2015.08.012.26318275
79. Mennillo E, Cappelli F, Arukwe A. 2019. Biotransformation and oxidative stress responses in rat hepatic cell-line (H4IIE) exposed to organophosphate esters (OPEs). Toxicol Appl Pharmacol 371 :84–94, PMID: , 10.1016/j.taap.2019.04.004.30974155
80. Chen G, Zhang S, Jin Y, Wu Y, Liu L, Qian H, et al. 2015. TPP and TCEP induce oxidative stress and alter steroidogenesis in TM3 Leydig cells. Reprod Toxicol 57 :100–110, PMID: , 10.1016/j.reprotox.2015.05.011.26049154
81. Wang X, Li F, Liu J, Ji C, Wu H. 2020. Transcriptomic, proteomic and metabolomic profiling unravel the mechanisms of hepatotoxicity pathway induced by triphenyl phosphate (TPP). Ecotoxicol Environ Saf 205 :111126, PMID: , 10.1016/j.ecoenv.2020.111126.32823070
82. Schang G, Robaire B, Hales BF. 2016. Organophosphate flame retardants act as endocrine-disrupting chemicals in MA-10 mouse tumor Leydig cells. Toxicol Sci 150 (2 ):499–509, PMID: , 10.1093/toxsci/kfw012.26794138
83. An J, Hu J, Shang Y, Zhong Y, Zhang X, Yu Z. 2016. The cytotoxicity of organophosphate flame retardants on HepG2, A549 and Caco-2 cells. J Environ Sci Health A Tox Hazard Subst Environ Eng 51 (11 ):980–988, PMID: , 10.1080/10934529.2016.1191819.27336727
84. Chen G, Jin Y, Wu Y, Liu L, Fu Z. 2015. Exposure of male mice to two kinds of organophosphate flame retardants (OPFRs) induced oxidative stress and endocrine disruption. Environ Toxicol Pharmacol 40 (1 ):310–318, PMID: , 10.1016/j.etap.2015.06.021.26183808
85. Yuan S, Zhu K, Ma M, Zhu X, Rao K, Wang Z. 2020. In vitro oxidative stress, mitochondrial impairment and G1 phase cell cycle arrest induced by alkyl-phosphorus-containing flame retardants. Chemosphere 248 :126026, PMID: , 10.1016/j.chemosphere.2020.126026.32006839
86. Yao Y, Li M, Pan L, Duan Y, Duan X, Li Y, et al. 2021. Exposure to organophosphate ester flame retardants and plasticizers during pregnancy: thyroid endocrine disruption and mediation role of oxidative stress. Environ Int 146 :106215, PMID: , 10.1016/j.envint.2020.106215.33113466
87. Wang Y, Hong J, Shi M, Guo L, Liu L, Tang H, et al. 2021. Triphenyl phosphate disturbs the lipidome and induces endoplasmic reticulum stress and apoptosis in JEG-3 cells. Chemosphere 275 :129978, PMID: , 10.1016/j.chemosphere.2021.129978.33662732
88. Hong J, Jiang M, Guo L, Lin J, Wang Y, Tang H, et al. 2022. Prenatal exposure to triphenyl phosphate activated PPARγ in placental trophoblasts and impaired pregnancy outcomes. Environ Pollut 301 :119039, PMID: , 10.1016/j.envpol.2022.119039.35192884
89. Wang X, Li F, Liu J, Li Q, Ji C, Wu H. 2021. New insights into the mechanism of hepatocyte apoptosis induced by typical organophosphate ester: an integrated in vitro and in silico approach. Ecotoxicol Environ Saf 219 :112342, PMID: , 10.1016/j.ecoenv.2021.112342.34023725
90. Bowen C, Childers G, Perry C, Martin N, McPherson CA, Lauten T, et al. 2020. Mitochondrial-related effects of pentabromophenol, tetrabromobisphenol A, and triphenyl phosphate on murine BV-2 microglia cells. Chemosphere 255 :126919, PMID: , 10.1016/j.chemosphere.2020.126919.32402876
91. Zhang Q, Wang J, Zhu J, Liu J, Zhao M. 2017. Potential glucocorticoid and mineralocorticoid effects of nine organophosphate flame retardants. Environ Sci Technol 51 (10 ):5803–5810, PMID: , 10.1021/acs.est.7b01237.28430429
92. Kojima H, Takeuchi S, Van den Eede N, Covaci A. 2016. Effects of primary metabolites of organophosphate flame retardants on transcriptional activity via human nuclear receptors. Toxicol Lett 245 :31–39, PMID: , 10.1016/j.toxlet.2016.01.004.26778350
93. Kojima H, Takeuchi S, Itoh T, Iida M, Kobayashi S, Yoshida T. 2013. In vitro endocrine disruption potential of organophosphate flame retardants via human nuclear receptors. Toxicology 314 (1 ):76–83, PMID: , 10.1016/j.tox.2013.09.004.24051214
94. Wang B, Chandrasekera PC, Pippin JJ. 2014. Leptin- and leptin receptor-deficient rodent models: relevance for human type 2 diabetes. Curr Diabetes Rev 10 (2 ):131–145, PMID: , 10.2174/1573399810666140508121012.24809394
95. Bani D, Pini A, Yue SKS. 2012. Relaxin, insulin and diabetes: an intriguing connection. Curr Diabetes Rev 8 (5 ):329–335, PMID: , 10.2174/157339912802083487.22698078
96. Yang YM, Kuen DS, Chung Y, Kurose H, Kim SG. 2020. Gα12/13 signaling in metabolic diseases. Exp Mol Med 52 (6 ):896–910, PMID: , 10.1038/s12276-020-0454-5.32576930
97. Cartee GD. 2015. Mechanisms for greater insulin-stimulated glucose uptake in normal and insulin-resistant skeletal muscle after acute exercise. Am J Physiol Endocrinol Metab 309 (12 ):E949–E959, PMID: , 10.1152/ajpendo.00416.2015.26487009
98. Shieh JM, Wu HT, Cheng KC, Cheng JT. 2009. Melatonin ameliorates high fat diet-induced diabetes and stimulates glycogen synthesis via a PKCζ-Akt-GSK3β pathway in hepatic cells. J Pineal Res 47 (4 ):339–344, PMID: , 10.1111/j.1600-079X.2009.00720.x.19817973
99. Saltiel AR, Kahn CR. 2001. Insulin signalling and the regulation of glucose and lipid metabolism. Nature 414 (6865 ):799–806, PMID: , 10.1038/414799a.11742412
100. Huang X, Liu G, Guo J, Su Z. 2018. The PI3K/AKT pathway in obesity and type 2 diabetes. Int J Biol Sci 14 (11 ):1483–1496, PMID: , 10.7150/ijbs.27173.30263000
101. Lehallier B, Gate D, Schaum N, Nanasi T, Lee SE, Yousef H, et al. 2019. Undulating changes in human plasma proteome profiles across the lifespan. Nat Med 25 (12 ):1843–1850, PMID: , 10.1038/s41591-019-0673-2.31806903
|
PMC010xxxxxx/PMC10104165.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37058435
EHP10959
10.1289/EHP10959
Research
Associations between Aircraft Noise Exposure and Self-Reported Sleep Duration and Quality in the United States-Based Prospective Nurses’ Health Study Cohort
https://orcid.org/0000-0003-0429-0247
Bozigar Matthew 1 *
https://orcid.org/0000-0001-8420-9167
Huang Tianyi 2
Redline Susan 2 3 4
https://orcid.org/0000-0002-0826-1163
Hart Jaime E. 2 5
https://orcid.org/0000-0002-1544-358X
Grady Stephanie T. 1
https://orcid.org/0000-0001-6748-4677
Nguyen Daniel D. 1
https://orcid.org/0000-0002-2858-1973
James Peter 5 6
https://orcid.org/0000-0002-6929-4305
Nicholas Bradley 7
https://orcid.org/0000-0002-1116-4006
Levy Jonathan I. 1
https://orcid.org/0000-0002-2813-2174
Laden Francine 2 3 5
https://orcid.org/0000-0003-4542-4563
Peters Junenette L. 1
1 Department of Environmental Health, Boston University School of Public Health, Boston, Massachusetts, USA
2 Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, Massachusetts, USA
3 Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
4 Beth Israel Deaconess Medical Center, Boston, Massachusetts, USA
5 Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
6 Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, Massachusetts, USA
7 Volpe National Transportation Systems Center, U.S. Department of Transportation, Cambridge, Massachusetts, USA
Address correspondence to Matthew Bozigar, 160 SW 26th St., Corvallis, OR 97331 USA. Email: [email protected]
14 4 2023
4 2023
131 4 04701017 1 2022
21 2 2023
03 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Sleep disruption is linked with chronic disease, and aircraft noise can disrupt sleep. However, there are few investigations of aircraft noise and sleep in large cohorts.
Objectives:
We examined associations between aircraft noise and self-reported sleep duration and quality in the Nurses’ Health Study, a large prospective cohort.
Methods:
Aircraft nighttime equivalent sound levels (Lnight) and day–night average sound levels (DNL) were modeled around 90 U.S. airports from 1995 to 2015 in 5-y intervals using the Aviation Environmental Design Tool and linked to geocoded participant residential addresses. Lnight exposure was dichotomized at the lowest modeled level of 45 A-weighted decibels [dB(A)] and at multiple cut points for DNL. Multiple categories of both metrics were compared with <45 dB(A). Self-reported short sleep duration (<7 h/24-h day) was ascertained in 2000, 2002, 2008, 2012, and 2014, and poor sleep quality (frequent trouble falling/staying asleep) was ascertained in 2000. We analyzed repeated sleep duration measures using generalized estimating equations and sleep quality by conditional logistic regression. We adjusted for participant-level demographics, behaviors, comorbidities, and environmental exposures (greenness and light at night) and examined effect modification.
Results:
In 35,226 female nurses averaging 66.1 years of age at baseline, prevalence of short sleep duration and poor sleep quality were 29.6% and 13.1%, respectively. In multivariable models, exposure to Lnight ≥45 dB(A) was associated with 23% [95% confidence interval (CI): 7%, 40%] greater odds of short sleep duration but was not associated with poor sleep quality (9% lower odds; 95% CI: −30%, 19%). Increasing categories of Lnight and DNL ≥45 dB(A) suggested an exposure–response relationship for short sleep duration. We observed higher magnitude associations among participants living in the West, near major cargo airports, and near water-adjacent airports and among those reporting no hearing loss.
Discussion:
Aircraft noise was associated with short sleep duration in female nurses, modified by individual and airport characteristics. https://doi.org/10.1289/EHP10959
Supplemental Material is available online (https://doi.org/10.1289/EHP10959).
* Current address: Matthew Bozigar, College of Public Health and Human Sciences, Oregon State University, Corvallis, Oregon, USA.
S.R. reports consulting fees and grant support from Jazz Pharma, unrelated to this article. The other authors declare they have no actual or potential competing financial interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Sleep is an essential, natural process needed for healthy brain and general body functioning.1 Disruption of sleep can cause drowsiness and poor concentration and adversely affect the metabolic, endocrine, and immune systems.2–5 Poor sleep quality and short sleep duration (alternatively referred to as insufficient sleep or reduced total sleep time) have been associated with many adverse health outcomes,6 including depression,7 metabolic disorders (e.g., obesity, type 2 diabetes),8–10 cardiovascular disease,11 incident ulcerative colitis,12 coronary events,13 cancer,14 kidney function decline,15 mental and physical functional decline,16–18 and mortality.19–21
Noise is unwanted or harmful sound,22 where sound is defined as repetitive variations in air pressure (i.e., vibrations) sensed by the human ear. Humans recognize, evaluate, and react to environmental sounds even when asleep.23 Noise can disrupt sleep architecture via arousals, sleep-stage changes, and awakenings.24,25 These, in turn, increase cortical excitations, indicative of an elevated stress response to noise, which may activate the sympathetic nervous system.24
Aircraft noise is unique for its multispectral acoustical properties that impact the human auditory system26 and has been shown to disrupt sleep.24,25,27 However, most studies of aircraft noise and sleep have taken place in a small collection of homes,28 human sleep laboratories,25,29,30 or around one or a few airports.28,31–37 Despite calls by researchers for more large-scale field studies,24 only two previous studies to our knowledge have used large-scale methods across many airports.38,39 In addition, most studies have been cross-sectional in design and therefore unable to assess associations over time. Sleep patterns may be disrupted by lifestyle factors such as shift work,40 as well as environmental exposures that include air pollution, greenness, inopportune light at night (LAN), and noise.41–45 But to our knowledge, there are no large-scale studies of aircraft noise that integrate and adjust for multiple environmental exposures such as greenness and LAN together, particularly in the United States,38 where the health effects of aircraft noise have been understudied. Although a legal framework for noise abatement was established for high aircraft noise levels [i.e., ≥65 dB(A)] in the United States,46,47 few studies have assessed potential health effects at lower thresholds in the country. Finally, there has been limited assessment of effect measure modification in previous studies to identify potential vulnerable and susceptible subpopulations.
We seek to add to existing knowledge by investigating associations between aircraft noise and sleep repeatedly self-reported in a U.S.-based prospective cohort living near 90 U.S. airports, controlling for individual and area-level factors. We also examine whether any relationships found are modified by individual, airport, or area characteristics.
Methods
Study Population and Airports
We used the Nurses’ Health Study (NHS), an ongoing nationwide U.S. prospective cohort started in 1976, in this research. The cohort was comprised of 121,701 female registered nurses who were 30–55 years of age at initial enrollment. Although recruitment originally focused on the 11 states with the most registered nurses, participants have moved during follow-up and now live throughout the United States. Questionnaires were self-administered biennially and contained questions about incident disease, medical history, and lifestyle factors. Response rates have been ≥90%.48,49 Home addresses indicated by the participants were geocoded every 2 y (corresponding with the biennial sequence of the NHS).
We used responses to questionnaires in survey years within the interval with aircraft noise assessment (1995–2015) that included questions about sleep—2000, 2002, 2008, 2012, and 2014—to define the study years. The year 2000, the first year for which both aircraft noise and sleep data were available, was the study baseline. To be included in this study, participants needed to have reported on sleep and have a valid residential address in the United States that could be used to successfully locate them geographically (i.e., geocoded). Residential moves among the participants during the study years were captured at the biennial assessment and aircraft noise exposure and environmental exposure estimates were updated, but participant-years were skipped when new addresses could not be geocoded. Participants were further excluded from the study if they did not live within a 22.2-mi (35.7-km) radius buffer around 1 of the 90 study airports at baseline, which represented the maximal empirical extent of aircraft noise above a day–night average sound level (DNL) of 45 decibels [A-weighted, dB(A)] surrounding any of the airports in this study. This exclusion criterion was used to limit the people unexposed to aircraft noise [here referring to those below our lowest modeled level of 45 dB(A)] to those who lived in places similar to where the exposed group lived and, therefore, reduce potential bias from unmeasured confounding factors associated with living close to an airport. Use of the 22.2-mi (35.7-km) buffers additionally helped limit exposure misclassification of participants living near airports that were not included in this study. Table S1 summarizes the number of participants excluded at each survey year for each successive criterion.
The 90 study airports were selected based on availability of model input (aircraft operations) data. They were located in 40 of 50 states plus the District of Columbia and captured 87% of all enplanements in the United States in 2010.50
The study protocol was approved by the institutional review board of Brigham and Women’s Hospital, Boston, Massachusetts. Consent was implied through the return of the questionnaires.
Outcome Assessment
Participants self-reported the average number of hours they slept in a 24-h day in 5 survey years [2000 (baseline), 2002, 2008, 2012, and 2014]. Response options included ≤5, 6, 7, 8, 9, and ≥10 h/24-h day. Self-reported sleep duration on the 2002 questionnaire was previously shown to correlate strongly (rSpearman=0.79) with sleep assessed by a 1-wk sleep diary in a validation study among a subset of NHS participants.21 The American Academy of Sleep Medicine and the Sleep Research Society reports that physiological and neurobehavioral deficits can occur and worsen over time with <7 h of sleep every 24 h.14,51 Therefore, we defined short sleep duration as <7 h/24-h day. For the sleep quality outcome, participants were asked a question at baseline (2000) about their frequency of having difficulty falling or staying asleep during the previous 4 wk. We defined poor sleep quality as participant answers to this question on a six-point Likert scale of: “all of the time,” “most of the time,” or “a good bit of the time.”
Exposure Assessment
Details about the generation of aircraft noise estimates are described elsewhere.52,53 Briefly, aircraft noise estimates around all 90 study airports were modeled comprehensively using analogous input data and modeling assumptions by the U.S. Department of Transportation’s John A. Volpe National Transportation Systems Center (Volpe) using the U.S. Federal Aviation Administration’s (FAA’s) Aviation Environmental Design Tool (AEDT; https://aedt.faa.gov/). Source data for the AEDT were aircraft operations from the Enhanced Traffic Management System, excluding helicopter operations. All annualized operations (e.g., commercial, cargo, military) were grouped by Aircraft Noise and Performance aircraft type, day or nighttime, operation airport, and stage length.
Aircraft noise level contours were modeled every 5 y from 1995 to 2015 (1995, 2000, 2005, 2010, and 2015) surrounding the 90 U.S. airports. Two noise metrics were estimated: aircraft nighttime equivalent sound level (Lnight, A-weighted) and DNL (A-weighted). Noise levels ranged from 45 to 71 dB(A). Lnight was the primary aircraft noise metric in this analysis because it captures aircraft noise occurring when people typically sleep. However, DNL incorporates nighttime estimates as well, and it is the primary metric the FAA uses to inform decision-making about aircraft noise.54 Where Lnight assesses aircraft noise from 2200 to 0700 hours, DNL is a 24-h annualized average, capturing the average day of a year’s operations with a 10 dB(A) penalty for nighttime aircraft noise from 2200 to 0700 hours, when levels of background noise tend to be lower compared with daytime. We therefore included DNL as a secondary aircraft noise exposure metric.
Spatially, each geocoded participant home address point was linked to the aircraft noise contour polygon in which it resided and assigned the respective aircraft noise level. If an address point was not within any contour, it was assigned an arbitrary value of 44 dB(A), just below the lowest modeled value of 45 dB(A), which indicated exposures estimated to be <45 dB(A), for both Lnight and DNL. DNL <45 dB(A) aligns with the recommendation and guideline for protecting health by limiting aircraft noise from the World Health Organization (WHO) Regional Office for Europe using the day–evening–night average sound level (Lden), an aircraft noise metric comparable to DNL that includes an extra penalty for evening-time aircraft noise. However, the 45 dB(A) threshold we used for Lnight is higher than the corresponding WHO guideline of limiting Lnight to <40 dB(A).55
Temporally, the aircraft exposure estimates for a given year were matched to current home addresses of participants of the same survey year when they coincided. When the 5-y aircraft noise estimates were not temporally coincident to a survey year, the most recent previous aircraft noise estimates were temporally matched.
Covariates and Potential Confounders
A directed acyclic graph (DAG) proposed by Billings et al. was used to guide our theoretical DAG (Figure S1) and an operational DAG (Figure S2) on the adjustment for covariates and potential confounding factors.42,56 Based on variables identified in the DAG, we additionally determined inclusion of environmental variables in final models after evaluating correlation magnitudes among these variables and whether they changed the effect estimates by at least 10%.
Demographic factors included age (continuous), age2 (continuous), U.S. Census region of residence (Northeast, Midwest, South, West), race (White, Black, American Indian, Asian, Hawaiian), and individual socioeconomic status (SES). Race was derived from an NHS algorithm using a 1992 question about major ancestry in combination with a 2004 question about best fitting race category and was non–time-varying. Race is a social construct, and we include it as proxy metric to aid in adjusting for socio-historical and multigenerational effects of discrimination and racism, which have been shown to affect sleep.57 The categories we analytically used for race mirror the survey response options. Age and census region of residence were reported every 2 y. Including age2 in statistical models adjusted for quadratic effects of age on our sleep disturbance-related outcomes. Individual SES was captured by two variables, currently living alone (yes, no) and spouse’s education (<high school, high school, >high school, missing/not married). In the NHS cohort, typical measures of individual SES are deemed less applicable given that participants are or were all nurses of moderate-to-high SES; instead, living alone and lower spousal education levels, indicating less or no joint financial resources, are used as markers of lower relative individual SES among this population.58 Spouse’s education was reported in 1992 and carried forward, whereas living alone was reported in 2000, 2008, and 2012 and was carried forward as necessary. A question asked about shift work in the 1988 survey year indicated that although 49.1% of participants reporting previously working night shifts at some point in their careers, in 1996 only 2.9% of NHS participants reported having rotating night shift work in the previous 6 months. Given the low frequency of shift work 4 y prior to this study’s baseline year (2000) and no further data points on shift work after 1996, shift work was not included as a potential confounder. Biannually varying postmenopausal status until 2002 (yes, no, missing) and hormone replacement therapy (never, current, former, missing) were included in a sensitivity analysis, but these individual metrics did not vary significantly, likely because 98.8% of the participants were already post menopausal at baseline (Table 1). Behaviors included smoking status (never, former, current, missing) and alcohol consumption (none, >0–4g/d, 5–9g/d, 10–14g/d, 15–29g/d, ≥30g/d, missing) and were reported every 4 y and carried forward 2 y as necessary. Comorbidities included diabetes and hypertension, determined by a self-report of a current clinician diagnosis, and were reported every 2 y. Potential environmental confounders included air pollution [particulate matter with an aerodynamic diameter of ≤2.5μm (PM2.5) per cubic meter], greenness, LAN, population density, and neighborhood SES (nSES).
Table 1 Age-standardized characteristics of women in the Nurses’ Health Study (NHS) at baseline (2000) overall, by nighttime aircraft sound (Lnight) exposure group, and by aircraft day–night sound level (DNL) exposure group.
Characteristic Overall Lnight <45 dB(A) Lnight ≥45 dB(A) DNL <45 dB(A) DNL 45–54 dB(A) DNL 55–64 dB(A) DNL ≥65 dB(A)
N=35,381 N=34,838 N=543 N=28,544 N=5,882 N=916 N=39
Demographics
Age [y (mean±SD)] 66.1±7.2y 66.1±7.2y 67.0±7.1 66.1±7.2 66.1±7.1 66.6±7.1 68.3±6.5
Region of residence (%)
Northeast 48.7 48.5 58.7 46.7 56.8 57.4 59.1
Midwest 14.8 14.8 10.9 14.8 15.5 10.3 8.7
South 19.3 19.4 16.2 19.9 16.9 18.2 9.1
West 17.2 17.3 14.2 18.6 10.9 14.1 23.0
Race (%)
White 96.1 96.2 89.1 96.6 94.5 90.7 86.6
Black 2.4 2.3 9.8 1.9 3.8 7.4 13.4
American Indian 0.2 0.2 0.0 0.2 0.3 0.2 0.0
Asian 1.3 1.3 1.1 1.2 1.3 1.6 0.0
Hawaiian 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Currently live alone 21.6 21.6 25.6 21.4 22.4 24.0 33.0
Spouse’s education (%)
<High school 4.0 4.0 2.9 3.8 4.6 4.5 0.0
High school 26.7 26.6 32.6 26.4 27.0 33.7 28.6
>High school 44.9 45.0 36.0 45.7 42.3 36.5 47.9
Missing 24.4 24.4 28.4 24.0 26.1 25.3 23.6
Postmenopausal (%)
Yes 98.8 98.7 99.1 98.7 98.9 99.0 100.0
No 1.2 1.2 0.9 1.2 1.1 1.0 0.0
Missing 0.1 0.1 0.0 0.1 0.0 0.0 0.0
Hormone replacement therapy (%)
Never 23.0 22.9 28.8 22.3 26.0 27.8 14.3
Former 27.2 27.2 26.9 27.2 27.5 27.4 39.5
Current 42.9 43.0 37.0 43.9 39.0 36.1 39.6
Missing 6.9 6.9 7.4 6.7 7.6 8.7 6.6
Behaviors
Smoking status (%)
Never 42.9 42.9 44.7 43.2 41.6 42.9 54.0
Former 47.8 47.8 47.5 47.7 48.0 47.1 43.1
Current 9.1 9.1 7.3 8.9 10.2 9.8 2.9
Missing 0.2 0.2 0.5 0.2 0.2 0.2 0.0
Alcohol consumption
Nonmissing [%; g/d (mean±SD)] 91.1; 5.3±9.1 91.2; 5.3±9.2 89.4; 4.0±7.5 91.3; 5.3±9.2 90.4; 5.1±9.0 90.0; 4.5±8.1 85.8; 3.1±5.0
Missing (%) 8.9 8.8 10.6 8.7 9.6 10.0 14.2
Comorbidities (%)
Diabetes 6.1 6.1 7.8 6.1 6.1 8.1 5.6
Hypertension 35.1 35.0 40.6 35.0 35.1 39.2 47.7
Environmental (mean±SD)
NDVI greenness index 0.4±0.1 0.4±0.1 0.3±0.1 0.4±0.1 0.3±0.1 0.3±0.1 0.3±0.1
LAN (nW/cm2/sr) 38.7±25.2 38.4±24.9 61.5±30.1 35.5±22.9 51.2±28.8 59.7±30.3 63.1±18.0
PM2.5 (μg/m3) 13.0±2.7 13.0±2.7 13.7±2.5 12.9±2.7 13.4±2.4 13.6±2.5 14.2±2.2
Population density (ppl/km2) 1,929.3±3,724.5 1,888.8±3,642.3 4,478.5±6,728.3 1,611.1±2,914.1 3,098.7±5,848.2 4,173.4±5,848.2 7,619.2±9,457.4
nSES (z-score) −0.4±3.3 −0.4±3.3 −1.0±2.5 −0.4±3.3 −0.4±3.0 −0.9±2.7 −0.8±1.9
Sleep (%)
Insufficient sleep 29.6 29.5 37.6 29.2 30.8 35.4 36.8
Poor sleep quality 13.1 13.1 12.2 13.2 12.5 13.5 21.6
Note: Environmental exposures considered included a 270-m buffer of greenness (NDVI), LAN measured in units of radiance (nanowatts per square centimeter per steradian, nW/cm2/sr), annual PM2.5 (μg/m3) at participant residential address, census tract population density (number of people per square kilometer, ppl/km2), and census tract socioeconomic status z-score (nSES). Missing values are excluded from calculation of descriptive statistics. The study sample was age-standardized to reflect the age distribution of the NHS cohort. LAN, light at night; NDVI, Normalized Difference Vegetation Index; nSES, neighborhood socioeconomic status; PM2.5, particulate matter with an aerodynamic diameter of ≤2.5μm; SD, standard deviation.
Air pollution, particularly PM2.5 and nitrogen oxides, has been associated with sleep apnea,59,60 likely through inflammatory effects on upper airway function.42 The annual average of monthly ambient PM2.5 (i.e., each 1-month average in the calendar year of survey year, averaged) was estimated at participant residential addresses using generalized additive mixed models developed for spatiotemporal prediction.61 The prediction models integrate monitored data from the U.S. Environmental Protection Agency’s Air Quality Monitoring System and other publicly available networks with geospatial predictors (e.g., road network data, land use, elevation). Predictive accuracy for PM2.5 was high (R2=0.77). PM2.5 estimates were made through 2007; estimates from 2007 were carried forward to the remaining study years for each participant.
Greenness from natural vegetation has been linked with longer sleep duration, improved cardiovascular biomarkers, and decreased sympathetic activation and may be associated with improved mental health and healthy behaviors promoting quality sleep, such as walking.42,62,63 However, greener areas have also been linked with more aircraft noise annoyance,45 potentially resulting from greener areas being quieter and/or pleasing and thus modifying the experience of noise from aircraft. Greenness was estimated via the Normalized Difference Vegetation Index (NDVI), or the ratio of the difference between the near-infrared region and red reflectance (isolating visible light absorbed by chlorophyll in plants) to the sum of the two measures64 from 30-m resolution Landsat data. We calculated annual average NDVI at a spatial resolution of 270m annually using R (version 4.0.3; R Development Core Team) and Google Earth Engine, matching the corresponding grid cell to the home address of each study participant. We chose to use a 270-m buffer, which is roughly the size of an average city block in Manhattan, New York, because this size is more representative of the immediate greenness environment that could affect sleep.
Light is a major input into the timing of the circadian system. LAN has been documented as having adverse effects on sleep.65–67 Assessment of outdoor LAN was described by James et al.68 Briefly, annual average LAN was estimated using nighttime radiance units (in nanowatts per centimeter squared per steradian) at a spatial resolution of 30 arc-seconds, or ∼1 km2. LAN estimates were developed from satellite imagery data from the U.S. Defense Meteorological Satellite Program’s Operational Linescan System, managed by the National Oceanic and Atmospheric Administration’s (NOAA’s) Earth Observation Group.69 LAN values were estimated for each survey year until 2010, after which estimates were carried forward.
Dense urban built environments have been linked to positive influences on health by promoting walking and other physical activity,70,71 but urban density may come at the cost of reduced and poorer quality sleep.72 Population density and neighborhood metrics of room crowding have previously been associated with worse sleep outcomes.72–74 Biannually varying census tract population density (in number of people per kilometer squared) metrics were derived from the 2000 U.S. Census for the 2000 and 2002 survey years and from the 2010 U.S. Census for the 2008, 2012, and 2014 survey years.
nSES is another potential confounder of aircraft noise affecting sleep, but previous findings have been mixed. Neighborhood disadvantage has been associated with adverse sleep outcomes in some studies75–77 but not in others.78–80 The social environment may be more highly associated with sleep disruption than nSES, directly.79 However, previous research has suggested that nSES can act as an upstream factor affecting downstream mediating environmental exposures influencing sleep, including aircraft noise.42,81 The nSES metric comprised a summed z-score of SES-related census variables (e.g., race, education, income, home value, nativity, unemployment) at the level of the census tract.82 Annual environmental/area exposures considered in statistical models (PM2.5, NDVI, LAN, population density, and nSES) were assigned to participant home addresses every 2 y to match the current home addresses of participants given that some of them moved residences throughout the study period.
Potential Effect Modifiers
Potential effect modifiers were determined a priori from hypothesized influences of individual, airport, and geographic characteristics on the aircraft noise and sleep relationship. We hypothesized that the association between aircraft noise and sleep would vary by census region (Northeast, Midwest, South, West) in the United States.83 based on potential underlying acoustical differences resulting from seasonal and weather (e.g., temperature, humidity) factors,84,85 geographical development patterns,86,87 housing types,88,89 and intensity and daily temporal distribution of air services and peak hours of operation,90 which were covariates for which we did not have direct measurements. We also hypothesized that living near a major cargo airport could result in a stronger effect because cargo aircraft are typically larger and older planes with less aircraft noise reduction technology and frequently operate during the night.91 Heavier aircraft climb more slowly and subsequently generate higher noise exposure on the ground. Most of these components of cargo aircraft operations are captured in AEDT modeling. However, AEDT uses the stage length method that uses takeoff-to-landing distance as a proxy for aircraft weight,92 which may more accurately estimate the weight of passenger aircraft than the weight of cargo aircraft. We identified the 24 largest cargo airports of the 90 study airports by total landed weight of all-cargo operations.50 The 24 cargo airports were consistently among the top 25 in landed weight over the study period.50 We further hypothesized that the association would be stronger for participants living near airports adjacent to a large water body (i.e., water-adjacent airports), where water and local weather could acoustically alter or enhance the experience of noise from aircraft in such areas.93,94 Our aircraft noise metrics likely did not capture the within-year variation in weather conditions that can occur near large water bodies. There were 21 water-adjacent airports, which were determined by assessing whether any existing runway configuration at an airport allowed for an overwater approach or departure. In 2008 and 2012 most participants answered questions about hearing loss. Regardless of the cause of hearing loss, we posited that greater hearing loss would decrease the magnitude of the association between aircraft noise and short sleep duration. We were unable to investigate potential effect modification of hearing loss relative to sleep quality because data on hearing loss were not available for the relevant survey year. Owing to small numbers in some strata, hearing loss was categorized as none, mild, moderate/severe, and missing.
Statistical Analysis
We created a map of the locations of the 90 airports included in this study as points using ArcGIS (version 10.8.1; ESRI). Counts of NHS participants around each airport included at study baseline were categorized into quartiles. Regions were indicated as the four U.S. Census regions in the context of state outlines using shapefiles from the U.S. Census (https://www.census.gov/geographies/mapping-files/time-series/geo/tiger-line-file.html).
We examined associations between aircraft noise and short sleep duration longitudinally using generalized estimating equations for repeated measures, and poor sleep quality cross-sectionally at baseline using conditional logistic regression. For repeated measures, an unstructured covariance matrix was used to maximize model flexibility given that we were not limited by degrees of freedom. We used SAS (version 9.4; SAS Institute Inc.) for all statistical analyses.
A participant was defined as exposed if their respective Lnight or DNL was estimated at ≥45 dB(A). We also examined cut points of Lnight 50 dB(A) and DNL 55 and 65 dB(A). We were unable to examine Lnight cut points ≥50 dB(A) and DNL ≥65 dB(A) owing to low counts of exposed participants. To evaluate potential exposure–response with each sleep outcome, we also used mutually exclusive increasing aircraft noise exposure categories compared with <45 dB(A). We used a three-category version of Lnight [<45 (reference), 45–49, and ≥50 dB(A)] and a four-category version of DNL [<45 (reference), 45–54, 55–64, and ≥65 dB(A)]. Significance of trends (pTrend) were estimated by modeling categories of aircraft noise exposures (e.g., four-category DNL) as continuous variables. We also examined continuous aircraft noise exposure levels >44 dB(A), but only as an ancillary analysis because 98.5% and 80.7% of participants were below the lowest available estimate of 45 dB(A) for Lnight and DNL, respectively.
We built five models for each outcome, starting with a crude model (model 0) adjusting only for age, age2, and calendar period. Model 1 further adjusted for demographic factors, such as region of residence, race, living alone, and spouse’s education. Models 2–4 then successively adjusted for behaviors (smoking status and alcohol consumption), comorbidities (diabetes and hypertension), and environmental factors (greenness and LAN), respectively, in which model 4 was the final, fully adjusted model. Given the high correlation between LAN and population density (Table S2), we chose to select only one of these variables for inclusion in the models. LAN was selected in favor of population density as a confounder in the final analyses owing to its possible direct effect on sleep in addition to its estimation on a relatively fine-scale, 1-km grid. Analyses included indicator variables for missing covariate data. We assessed effect measure modification hypotheses by including multiplicative interaction terms of potential effect modifiers and aircraft noise (pInteraction) in models and stratifying them by the categories of each effect modifier.
Results
Figure 1 shows the location of the 90 airports included in this study. Study population characteristics at baseline for the 35,381 participants are listed overall, by Lnight exposure, and by DNL exposure in Table 1. At baseline, the age [mean±standard deviation (SD)] of participants was 66.1±7.2y, and only 13.0% were still working full time as nurses. Most participants lived in the Northeast (48.7%). Approximately 21.6% of the participants were living alone at baseline, and over half of the nurses’ spouses had more than a high school education.
Figure 1. Map of 90 study airports across the United States symbolized by U.S. Census region and quartiles of Nurses’ Health Study (NHS) participants.
Figure 1 is a map of the United States of America, depicting the study of 90 airports symbolized by U.S. Census regions and quartiles of Nurses’ Health Study participants. A scale depicts miles ranging from 0 to 180 in increments of 90 and 180 to 360 in increments of 180. At the bottom-left, a map of Alaska is displayed with a scale depicting miles ranging from 0 to 460 in increments of 230 and 460 to 920 in increments of 460. At the bottom-center, a map of Hawaii is displayed with a scale depicting miles ranging from 0 to 80 in increments of 40 and 80 to 160 in increments of 80. At the bottom-right, the following information is given: The Census region includes the Midwest, Northeast, South, and West. Airport and N H S participants includes four ranges: 0 to 26, 27 to 108, 109 to 616, and 617 to 4,780.
Characteristics were similar for aircraft noise exposure groups, with a few notable exceptions. Participants exposed to Lnight ≥45 dB(A) tended to live more in the Northeast (58.7%) and less in the remaining regions compared with unexposed participants [Lnight <45 dB(A)]. Compared with unexposed participants, those exposed to Lnight ≥45 dB(A) were more likely to be Black (9.8% vs. 2.3%), more often lived alone (25.6% vs. 21.6%), less frequently had a spouse with more than a high school education (50.3% vs. 59.5%), and had higher rates of diabetes (7.8% vs. 6.1%) and hypertension (40.6% vs. 35.0%). Participants with high levels of exposure to either Lnight or DNL also experienced, on average, higher LAN and lived in more densely populated census tracts compared with participants exposed to low levels of aircraft noise.
Over time, participants provided an average of 3.48 survey-years of follow-up. Participant characteristics did not differ substantially for those who only provided data at baseline (N=2,901) vs. those who provided 2–5 survey years of data (N=32,480; see Table S3). The participants tended to live near the same airport throughout the study period (95.0%) and to have stable aircraft DNL exposure within 10 dB(A) (90.5%).
Table S4 describes levels of exposure to aircraft Lnight and DNL overall and by potentially effect-modifying airport characteristics for those exposed to ≥45 dB(A) of aircraft noise; 98.5% and 80.7% of participants were exposed to Lnight and DNL <45 dB(A) of aircraft noise, respectively. Exposures did not vary much by region, living near a major cargo airport, nor living near a water-adjacent airport.
Table 2 shows results for the estimated association between aircraft noise and sleep, adjusted for individual factors, behaviors, comorbidities, and other environmental exposures. In longitudinal analysis, we found 34% greater odds [95% confidence interval (CI): 17%, 53%] of short sleep duration among those exposed to Lnight ≥45 dB(A) compared with <45 dB(A) in the crude, age-adjusted model. Adjusting for demographics, behaviors, comorbidities, and environmental factors, the magnitude of association attenuated to 23% greater odds (95% CI: 7%, 40%) of short sleep duration among the exposed compared with the unexposed (Table 2). Using DNL as the exposure, in the fully adjusted model, we found 3% greater odds of short sleep duration for DNL ≥45 dB(A) relative to <45 dB(A) (95% CI: −1%, 8%). At the DNL 55-dB(A) cut point, there were stronger associations for all models, including the fully adjusted model (13% greater odds; 95% CI: 2%, 26%) (Table 2). The magnitude of the aircraft noise and short sleep duration association using a cut point of DNL 65 dB(A) was elevated (44% greater odds in the fully adjusted model) for those exposed compared with not exposed (Table 2), although the 95% CI was wide (−7%, 123%) given the small number of occurrences that any individual was an exposed case (n=39; case counts by exposure category are included in Figures 2–5).
Table 2 Estimated odds ratios (ORs) and 95% confidence intervals (CIs) for the relationship between aircraft nighttime sound level (Lnight), day–night sound level exposure (DNL), and short sleep duration (<7h/24-h d) and poor sleep quality (trouble falling/staying asleep ≥ “a good bit of the time”) cases at specified aircraft noise metric cut points and levels of adjustment in the Nurses’ Health Study (NHS) (2000–2014).
Model Short sleep duration
Nobs=123,023, Nparticipants=35,381, Ncases=35,497 Poor sleep quality
Nobs=35,226, Nparticipants=35,226, Ncases=4,617
Lnight ≥45 vs. <45 dB(A)
Crude: age-adjusted 1.34 (1.17, 1.53) 0.93 (0.72, 1.20)
1: Crude+other demographics 1.27 (1.11, 1.45) 0.94 (0.71, 1.21)
2: 1 + behaviors 1.26 (1.10, 1.44) 0.94 (0.72, 1.22)
3: 2 + comorbidities 1.26 (1.10, 1.44) 0.92 (0.71, 1.20)
4: 3 + environmental 1.23 (1.07, 1.40) 0.91 (0.70, 1.19)
DNL ≥45 vs. <45 dB(A)
Crude: age-adjusted 1.10 (1.05, 1.15) 0.96 (0.89, 1.04)
1: Crude + other demographics 1.06 (1.01, 1.10) 0.96 (0.89, 1.04)
2: 1 + behaviors 1.06 (1.01, 1.10) 0.96 (0.88, 1.04)
3: 2 + comorbidities 1.05 (1.01, 1.10) 0.96 (0.88, 1.03)
4: 3 + environmental 1.03 (0.99, 1.08) 0.94 (0.86, 1.02)
DNL ≥55 vs. <55 dB(A)
Crude: age-adjusted 1.23 (1.11, 1.37) 1.08 (0.89, 1.30)
1: Crude + other demographics 1.17 (1.05, 1.30) 1.08 (0.89, 1.30)
2: 1 + behaviors 1.16 (1.05, 1.29) 1.08 (0.90, 1.30)
3: 2 + comorbidities 1.16 (1.04, 1.29) 1.07 (0.89, 1.29)
4: 3 + environmental 1.13 (1.02, 1.26) 1.06 (0.87, 1.28)
DNL ≥65 vs. <65 dB(A)
Crude: age-adjusted 1.53 (0.98, 2.39) 1.94 (0.92, 4.08)
1: Crude + other demographics 1.50 (0.97, 2.31) 2.00 (0.95, 4.22)
2: 1 + behaviors 1.49 (0.96, 2.31) 2.04 (0.97, 4.30)
3: 2 + comorbidities 1.48 (0.96, 2.29) 2.04 (0.97, 4.30)
4: 3 + environmental 1.44 (0.93, 2.23) 2.01 (0.95, 4.25)
Note: Age-adjusted (age, age2) models were sequentially further adjusted as indicated with other demographics, behaviors, comorbidities, and environmental factors. Other demographics: U.S. region of residence, race, living alone, spouse’s education. Behaviors: smoking status, alcohol consumption. Comorbidities: diabetes, hypertension. Environmental: greenness (NDVI), LAN. Models of short sleep duration used generalized estimating equations to estimate odds from repeated measures in survey years 2000 (study baseline), 2002, 2008, 2012, and 2014. Conditional logistic regression models of sleep quality were used to estimate odds only for the baseline study year. dB(A), A-weighted decibel; LAN, light at night; Ncases, number of cases; NDVI, Normalized Difference Vegetation Index; Nobs, number of observations; Nparticipants, number of participants.
Figure 2. Odds ratio (OR) point estimates and 95% confidence intervals (CIs) investigating exposure–response relationship (pTrend<0.01) between categorical aircraft nighttime sound level (Lnight) exposure [<45 dB(A) (reference), 45–49 dB(A), and ≥50 dB(A)] and short sleep duration (<7 h/24-h day), using GEEs from repeated measures in survey years 2000 (study baseline), 2002, 2008, 2012, and 2014 in the Nurses’ Health Study (NHS). OR and CI estimates can be found in Table S6. Models adjusted for age (age, age2) were sequentially further adjusted for other demographics, behaviors, comorbidities, and environmental factors. Other demographics: U.S. region of residence, race, living alone, spouse’s education. Behaviors: smoking status, alcohol consumption. Comorbidities: diabetes, hypertension. Environmental: greenness (NDVI), LAN. Models of short sleep duration used GEEs to estimate odds from repeated measures in survey years 2000 (study baseline), 2002, 2008, 2012, and 2014. Note: dB(A), A-weighted decibel; GEE, generalized estimating equation; LAN, light at night; Ncases, number of cases; NDVI, Normalized Difference Vegetation Index; Nobs, number of observations.
Figure 2 is an error bar graph, plotting estimated odds ratio, ranging from 1.00 to 1.75 in increments of 0.25 (y-axis) across nighttime noise level exposure category A-weighted decibel, ranging, less than 45 (reference) with 34,987 cases and 121,595 observations, 45 to 50 with 407 cases and 1,148 observations, and 50 plus including 0: crude (age), 1: 0 plus other demographics, 2: 1 plus behaviors, 3: 2 plus comorbidities, 4: 3 with 103 cases and 280 observations (x-axis) for model, plus environmental.
Figure 3. Odds ratio (OR) point estimates and 95% confidence intervals (CIs) investigating exposure–response relationship (pTrend=0.03) between categorical aircraft day–night average sound level (DNL) exposure [<45 (reference), 45–54, 55–64, and ≥65 dB(A)] and short sleep duration (<7 h/24-h day), using GEEs from repeated measures in survey years 2000 (study baseline), 2002, 2008, 2012, and 2014 in the Nurses’ Health Study (NHS). OR and CI estimates can be found in Table S6. Models adjusted for age (age, age2) were sequentially further adjusted for other demographics, behaviors, comorbidities, and environmental factors. Other demographics: U.S. region of residence, race, living alone, spouse’s education. Behaviors: smoking status, alcohol consumption. Comorbidities: diabetes, hypertension. Environmental: greenness (NDVI), LAN. Models of short sleep duration used GEEs to estimate odds from repeated measures in survey years 2000 (study baseline), 2002, 2008, 2012, and 2014. Note: dB(A), A-weighted decibel; GEE, generalized estimating equation; LAN, light at night; Ncases, number of cases; NDVI, Normalized Difference Vegetation Index; Nobs, number of observations.
Figure 3 is an error bar graph, plotting estimated odds ratio, ranging from 1.0 to 2.5 in increments of 0.5 (y-axis) across day–night average sound level exposure category A-weighted decibel, ranging, less than 45 (reference) with 29,403 cases and 103,218 observations, 45 to 54 with 5,209 cases and 17,220 observations, 55 to 64 with 846 cases and 2,485 observations, and 65 plus with 39 cases and 100 observations (x-axis) for model, including 0: crude (age), 1: 0 plus other demographics, 2: 1 plus behaviors, 3: 2 plus comorbidities, 4: 3 plus environmental.
Figure 4. Odds ratio (OR) point estimates and 95% confidence intervals (CIs) investigating exposure–response relationship (pTrend=0.37) between categorical aircraft nighttime equivalent sound level (Lnight) exposure [<45 (reference), 45–49 dB(A), and ≥50 dB(A)] and poor sleep quality (trouble falling/staying asleep ≥ “a good bit of the time”) using conditional logistic regression at study baseline (2000) in the Nurses’ Health Study (NHS). OR and CI estimates can be found in Table S6. Models were adjusted for age (age, age2) other demographics, behaviors, comorbidities, and environmental factors. Other demographics: U.S. region of residence (removed from the region-specific models), race, living alone, spouse’s education. Behaviors: smoking status, alcohol consumption. Comorbidities: diabetes, hypertension. Environmental: greenness (NDVI), LAN. Conditional logistic regression models of sleep quality were used to estimate odds only for the baseline study year. Note: dB(A), A-weighted decibel; LAN, light at night; Ncases, number of cases; NDVI, Normalized Difference Vegetation Index; Nobs, number of observations.
Figure 4 is an error bar graph, plotting estimated odds ratio, ranging from 1.0 to 2.0 in increments of 0.5 (y-axis) across nighttime noise level exposure category A-weighted decibel, ranging, less than 45 (reference) with 4,551 cases and 34,689 observations, 45 to 50 with 50 cases and 433 observations, and 50 plus with 16 cases and 104 observations (x-axis) for model, including 0: crude (age), 1: 0 plus other demographics, 2: 1 plus behaviors, 3: 2 plus comorbidities, 4: 3 plus environmental.
Figure 5. Odds ratio (OR) point estimates and 95% confidence intervals (CIs) investigating exposure–response relationship (pTrend=0.37) between categorical aircraft day–night sound level (DNL) exposure [<45 (reference), 45–54, 55–64, and ≥65 dB(A)] and poor sleep quality (trouble falling/staying asleep ≥ “a good bit of the time”) using conditional logistic regression at study baseline (2000) in the Nurses’ Health Study (NHS). OR and CI estimates can be found in Table S6. Models were adjusted for age (age, age2) other demographics, behaviors, comorbidities, and environmental factors. Other demographics: U.S. region of residence (removed from the region-specific models), race, living alone, spouse’s education. Behaviors: smoking status, alcohol consumption. Comorbidities: diabetes, hypertension. Environmental: greenness (NDVI), LAN. Conditional logistic regression models of sleep quality were used to estimate odds only for the baseline study year. Note: dB(A), A-weighted decibel; LAN, light at night; Ncases, number of cases; NDVI, Normalized Difference Vegetation Index; Nobs, number of observations.
Figure 5 is an error bar graph, plotting estimated odds ratio, ranging from 1 to 4 in unit increments (y-axis) across day–night average sound level exposure category A-weighted decibel, ranging, less than 45 (reference) with 3,752 cases and 28,427 observations, 45 to 54 with 733 cases and 5,852 observations, 55 to 64 with 123 cases and 907 observations, and 65 plus with 9 cases and 40 observations (x-axis) for model, including 0: crude (age), 1: 0 plus other demographics, 2: 1 plus behaviors, 3: 2 plus comorbidities, 4: 3 plus environmental.
There was no evidence of a crude association between Lnight ≥45 dB(A) and poor sleep quality cross-sectionally (7% lower odds; 95% CI: −28%, 20%), and full adjustment for additional factors did not appreciably change the relationship (9% lower odds; 95% CI: −30%, 19%) (Table 2). Similarly, when using DNL as the exposure at 45 and 55 dB(A) cut points, we did not find evidence of associations between aircraft noise exposure and sleep quality. However, there were 101% greater odds of poor sleep quality from exposure to DNL ≥65 dB(A), although the 95% CI was wide and included the null (95% CI: −5%, 325%) (Table 2).
In sensitivity analyses, inclusion/exclusion of nSES, annual PM2.5, postmenopausal status, and hormone replacement therapy, and swapping LAN for census tract population density did not change effect estimates by >10% (Table S5); therefore, these variables were not included in final models.
Using continuous versions of aircraft noise in fully adjusted models, a 5-dB(A) increase in Lnight was associated with 23% greater odds (95% CI: 6%, 44%) of short sleep duration, whereas a 5-dB(A) increase in DNL was associated with 3% greater odds (95% CI: 0%, 6%). Figures 2 and 3 show the potential exposure–response relationships between aircraft noise exposure and short sleep duration from longitudinal modeling, whereas tabular results are included in Table S6. Similar to results for dichotomized and categorical aircraft noise exposures, controlling for covariates attenuated the estimated odds ratios within ordinal categories between either Lnight or DNL and short sleep duration (Figures 2 and 3). For the crude models, results indicate the presence of an exposure–response relationship between increasing aircraft noise exposures and short sleep duration. For the fully adjusted models, results similarly suggest the presence of monotonic exposure–response relationships (Lnight pTrend<0.01; DNL pTrend=0.03).
Using continuous versions of aircraft noise >45 dB(A) in fully adjusted models, 5-dB(A) increases in Lnight was associated (Lnight: 2% greater odds; 95% CI: −23%, 34%; DNL: 2% lower odds; 95% CI: −7%, 4%) with poor sleep quality. Figures 4 and 5 (and Table S6) show that there were some indications of effects for the highest respectively exposed groups but there was no evidence of exposure–response relationships for lower exposure categories for poor sleep quality cross-sectionally (Lnight pTrend=0.77; DNL pTrend=0.37).
Results from fully adjusted longitudinal models stratified by region, major cargo airport, water-adjacent airport, and hearing loss are shown in Table 3. They suggest that the association between nighttime aircraft noise and short sleep duration differed by region of residence (pInteraction=0.06). The association was strongest in the West, having 83% greater odds (95% CI: 32%, 152%). Participants living near a major cargo airport had 69% greater odds (95% CI: 21%, 136%) of short sleep duration associated with nighttime aircraft noise exposure, which was higher (pInteraction=0.09) than those not living near a major cargo airport (16% greater odds; 95% CI: 1%, 35%). For participants living near a water-adjacent airport, the association between Lnight and short sleep duration resulted in 36% greater odds (95% CI: 12%, 65%) compared with 11% greater odds (95% CI: −9%, 34%; pInteraction=0.14) for those living near a non–water-adjacent airport. For poor sleep quality in the cross-sectional analysis, there was little evidence of effect modification by region or by living near a water-adjacent airport (Table 3). However, for participants living near a major cargo airport, there were 45% greater odds (95% CI: −18%, 157%; pInteraction=0.09) of poor sleep quality among those exposed to nighttime aircraft noise, vs. 18% reduced odds (95% CI: −39%, 10%) for participants not living near a major cargo airport, but both intervals contained the null. Finally, for participants who reported no hearing loss, there were an estimated 50% higher odds (95% CI: 11%, 103%) of short sleep duration from exposure to nighttime aircraft noise. Estimated odds of short sleep duration were in progressively lower but also more imprecise with higher reported hearing loss [mild: 31% higher odds (95% CI: −19%, 114%); moderate/severe: 15% lower odds (95% CI: −61%, 86%)]. However, we did not find definitive evidence in the two survey years with hearing loss data available in our cohort (2008 and 2012) of an interaction between hearing loss and aircraft noise (pInteraction=0.33), although there was a suggested relationship.
Table 3 Estimated odds ratios (ORs) and 95% confidence intervals (CIs) for potential effect measure modifiers of the relationship between aircraft nighttime sound level [Lnight ≥45 dB(A)] and short sleep duration (<7 h/24-h day) and poor sleep quality (trouble falling/staying asleep ≥ “a good bit of the time”) cases from fully adjusted models (model 4) in the Nurses’ Health Study (NHS) (2000–2014).
Potential effect modifier Short sleep duration
Nobs=123,023, Nparticipants=35,381, Ncases=35,497 Poor sleep quality
Nobs=35,226, Nparticipants=35,226, Ncases=4,617
Nobs (cases) OR (95% CI) p-Value Nobs (cases) OR (95% CI) p-Value
Region 0.06 0.90
Northeast 59,358 (18,278) 1.15 (0.96, 1.37) 17,132 (2,251) 0.88 (0.62, 1.25)
Midwest 18,720 (4,831) 1.33 (0.90, 1.96) 5,211 (641) 1.08 (0.51, 2.31)
South 23,564 (6,533) 1.08 (0.77, 1.51) 6,803 (975) 1.05 (0.57, 1.90)
West 21,381 (5,855) 1.83 (1.32, 2.52) 6,080 (750) 0.77 (0.37, 1.60)
Water-adjacent airport 0.14 0.99
No 76,071 (20,781) 1.11 (0.91, 1.34) 21,401 (2,761) 0.91 (0.63, 1.31)
Yes 46,952 (14,716) 1.36 (1.12, 1.65) 13,749 (1,856) 0.95 (0.65, 1.38)
Major cargo airport 0.09 0.09
No 96,445 (27,760) 1.16 (1.01, 1.35) 27,431 (3,613) 0.82 (0.61, 1.10)
Yes 26,578 (7,737) 1.69 (1.21, 2.36) 7,795 (1,004) 1.45 (0.82, 2.57)
Hearing lossa 0.33 NA
None 30,422 (8,974) 1.50 (1.11, 2.03)
Mild 15,939 (4,899) 1.31 (0.81, 2.14)
Moderate/severe 9,379 (2,711) 0.85 (0.39, 1.86)
Note: Models were adjusted for age (age, age2), other demographics, behaviors, comorbidities, and environmental factors. Other demographics: U.S. region of residence (removed from the region-specific models), race, living alone, spouse’s education. Behaviors: smoking status, alcohol consumption. Comorbidities: diabetes, hypertension. Environmental: greenness (NDVI), LAN. Models of short sleep duration used generalized estimating equations to estimate odds from repeated measures in survey years 2000 (study baseline), 2002, 2008, 2012, and 2014. Conditional logistic regression models of sleep quality were used to estimate odds only for the baseline study year. dB(A), A-weighted decibel; LAN, light at night; Ncases, number of cases; NDVI, Normalized Difference Vegetation Index; Nobs, number of observations; Nparticipants, number of participants.
a Hearing loss was assessed and analyzed only for short sleep duration in 2008 and 2012. For potential effect modification by hearing loss: Nobs=55,740, Nparticipants=25,627, and Ncases=16,584.
Discussion
This study investigated the relationship between aircraft noise exposure and self-reported sleep in the NHS cohort participants living near 90 major U.S. airports. Adjusting for several individual and two environmental confounders (greenness and LAN), we found that exposure to aircraft noise was associated with short sleep duration in a repeated measures analysis, with an exposure–response relationship seemingly evident across all levels of aircraft noise included. Furthermore, we found evidence of potential effect modification by individual, area, and airport characteristics, with stronger associations with short sleep duration for participants living in the West, near major cargo airports, near water-adjacent airports, and among those reporting no hearing loss. Exposure to aircraft noise had a limited association with poor sleep quality cross-sectionally, with highly positive but uncertain associations seen only for the highest aircraft noise exposure category.
We found that short sleep duration was linked with two metrics of average annual aircraft noise exposure, Lnight and DNL. This is only partly consistent with existing literature on aircraft noise and sleep quantity. Most laboratory and residential studies have documented that aircraft noise events can shorten sleep duration,34,95 but some found associations with longer sleep duration in conjunction with poorer sleep quality.31,96 A national study of aircraft noise and self-reported sleep from the large-scale U.S. Behavioral Risk Factor Surveillance System surveys did not find an association between aircraft noise and short sleep duration,39 but the study differed in numerous ways from the present study. Although Holt et al. established feasibility of a large-scale study of self-reported sleep, the study differed in its cross-sectional design, aircraft noise exposure modeling (Integrated Noise Model), area-level (ZIP code) exposure assignment, absence of Lnight estimates, higher DNL cut points starting at 55 dB, dissimilar reference population that included participants far from study areas and possibly near non-study airports, and potential confounders.39 Another study around two airports in France found that a Lden 10-dB(A) increase in aircraft noise was associated with 63% greater odds (95% CI: 15%, 132%) of short total sleep time, after adjusting for individual level potential confounders.34 However, the study did not adjust for other environmental factors, potentially confounding the relationship between aircraft noise and sleep, nor did the researchers have data available to investigate potential effect measure modification in their sample of 1,244 adults. At the area level, elevated census block group environmental noise was not associated with total hours slept in a subsample of a national cross-sectional survey of urban individuals in the United States, although it was significantly associated with other adverse sleep outcomes.97 Although Rudolph et al. applied a novel nationwide environmental noise model, the study could not isolate the association with aircraft noise and had to rely on area-level (census block group) exposure assignment.97 The differences in study designs used in previous research may explain their mixed results in contrast to the longitudinal design used in this study that found a strong signal between aircraft noise and short sleep duration in the NHS.
For sleep assessments, most population-based studies used relatively high cut points of the DNL aircraft noise metric, such as 5034,62 or 55 dB,39 although some smaller studies have assessed the influences on sleep of aircraft noise exposures as low as 30 or 35 dB.28,31 Large-scale sleep studies investigating aircraft noise have mostly been unable to assess low potential thresholds or to incorporate nighttime specific metrics such as Lnight, particularly in the United States, with a few exceptions.38 We found limited evidence of thresholds across the range of exposures included. When considering exposed vs. unexposed people at a 45-dB(A) cut-point, we observed a stronger association between short sleep duration and the exposure metrics for the nighttime measure than the 24-h day–night metric. Given consistent associations between sleep and a wide range of health outcomes, this finding suggests value in including nighttime aircraft noise metrics in health assessments of airport noise. Furthermore, finding evidence of an exposure–response relationship using DNL, future studies should assess a range of exposures and consider the sensitivity of conclusions about exposure categorization (Table 1).
Our findings suggest a clear, monotonic exposure–response relationship between aircraft noise and short sleep duration that is consistent with the results of smaller studies analyzing a variety of sleep parameters.29,32,38,95 However, most of the studies documenting exposure–response relationships between transportation noise and sleep parameters often used laboratory- or home-based designs with limited adjustment for potential confounding factors that lacked generalizability to larger populations.
An additional strength of this study was the examination of potential effect modification of the nighttime aircraft noise–sleep relationship by individual, area, and airport characteristics. Previous studies assessed potential effect modification, but in different ways. For example, one study used total sleep duration as an effect modifier and various other sleep parameters as outcomes.38 Although greater risk for those near major cargo airports was expected, given that the aircraft noise metrics we used may not have sufficiently incorporated characteristics associated with nighttime aircraft noise dynamics, effect modification of the association with sleep had not been previously investigated. Our results suggest that participants living near major cargo airports had shorter and poorer quality sleep related to aircraft noise at night than those living near airports with less cargo activity. Evidence from Europe has shown that air cargo flights tend to use older, larger aircraft that commonly operate in the early morning hours when commercial flight activity is low.91 In the United States, the quantity of cargo shipped by air has increased from 56.4M ton-miles in 2003 to 76.6M ton-miles in 2020.98 If the trend continues, further indicated by the rapid growth in e-commerce-related cargo flight activity during the COVID-19 pandemic,99–101 these results suggest that impacts on sleep may also increase in magnitude and for additional populations. Although our analyses incorporated noise exposure estimates that should, in theory, capture differential contributions by cargo flights, either the noise metrics we used were not well suited for characterizing sleep disturbance (e.g., differences in flight patterns or sound frequency distributions for cargo flights) or the AEDT modeling has differential error for cargo flights relative to other flights. The AEDT does not directly incorporate cargo weight because aircraft takeoff weight is proprietary in the United States; instead, the AEDT uses stage length as a proxy.92 If true cargo weight is systematically greater than estimates from stage length formulas, such as the scenario when cargo flights have shorter stage lengths (short-haul operations) but are heavily laden, then we would expect to see differentially greater effects at airports with more cargo operations, as we indeed found.
Similarly, we found support for the hypothesis that the association between aircraft noise and short sleep duration was greater for participants living near a water-adjacent airport, but there was little evidence of a modifying effect on poor sleep quality. For water-adjacent airports, our sleep duration findings may indicate additional complexity in these acoustical environments whereby aircraft sound energy may propagate more easily over water,93,94 differentially influencing sleep. The AEDT assumes a soft ground surface (e.g., grass) near the sound receiver. Reflections off hard ground, such as pavement, or water generally cause higher sound levels. Under certain conditions, sound propagating over water can be channeled by the reflection off the surface and then refracted downward owing to cool air being immediately above the surface with warmer air above that.
It is not clear why aircraft noise seems to vary in its association with short sleep duration by region of the country, although it potentially relates to aspects of the surrounding environment and climate zones, as well as nationally and internationally influenced flight operations scheduling. Additional research is needed to disentangle potential place-based aircraft sound propagation mechanisms, such as varied surfaces (hard vs. grassy surfaces), local weather, seasonality, the influence of atmospheric conditions during overflights,102 and housing materials and types. Unique regional climates have been associated with patterns in heating/cooling system types and related behaviors (e.g., opening windows) to achieve thermal comfort in bedrooms while sleeping,103 which may further modify the effects of aircraft noise on sleep. Daily flight scheduling can be a function of time zones, levels of business activity, airline hub and spoke networks, and markets served (e.g., geographic proximity to national and international destinations).104
There are age-related changes in hearing as people age,105 but we found a trend suggestive of a lower magnitude relationship between aircraft noise and short sleep duration with greater hearing loss independent of age. However, this study was underpowered to robustly investigate potential effect modification by hearing loss. In the literature, hearing loss is usually mentioned as a physiological outcome of aircraft noise,55,106 used as an exclusion criterion,30,96 or, as in one study, considered as a confounder.35 However, we did not find sleep studies that assessed potential effect modification by hearing loss.
We did not find associations with sleep quality in most analyses. However, the NHS surveys did not capture longitudinal self-assessments of multiple dimensions of sleep quality. Thus, we were only able to use participant answers to one question in 2000, in which only 66 of the 541 participants who were exposed to Lnight ≥45 dB(A) reported poor sleep quality. Despite limited power, we cannot rule out a potential association with poor sleep quality at very high levels of DNL exposure. Other predominantly smaller-scale studies conducted in human sleep laboratories or participant homes have found deleterious effects of aircraft noise on sleep quality.31,35
There were several limitations of our study. Our study population did not include males, younger individuals, or many individuals from underrepresented groups owing to the construction of the original cohort. Noise sensitivity and annoyance, which may influence the effects of aircraft noise on sleep, were not included because they were not measured in the cohort. In addition, there may be residual confounding by weather conditions, such as temperature, which independently affects sleep.74,83 Furthermore, we could not adjust for sleep medication use, which might be a potential confounder. Sleep was subjectively self-reported, and it has been shown that subjective reporting of sleep duration may underestimate objective measures, although it can depend on how the question is asked. Only a single question about sleep quality was available on one biennial questionnaire. Although aircraft noise estimates used were average sound energies over time, single-event exposures of individual flights [e.g., maximum sound level (Lmax) or sound exposure level (SEL)] may be more relevant to sleep, yet such metrics are infrequently used.55 We could not capture within-5-y variations in noise exposure. Aircraft noise estimates were outside at home addresses, not in the actual bedrooms in which participants presumably slept, and they did not include noise from ground operations in and around airports. Although important housing and indoor home environments were not able to be captured, including sound insulation, window opening/closing or air conditioner use,107 address-level exposure assignment for a large-scale study was a novel contribution to aircraft noise and sleep research. Other sources of environmental noise (e.g., natural, community, road, rail) were not captured directly, although they were likely partly captured by environmental confounders included. In the present study, LAN likely acted as a proxy for population density and sources of noise (e.g., road) and light happening more frequently in denser (e.g., urban) areas. The cohort and study population were not randomly distributed, given that participants were originally recruited from 11 U.S. states in 1976. For example, the Boston area, and the Northeast more broadly, was initially overrepresented. However, few nationwide populations have been followed with repeated sleep and aircraft noise measurements over time. Similarly, airports were not randomly selected into this study, but were included where operations data were available. However, the respective geographic coverage of both participants and airports were still wide over the study period, and the airports included in the study captured the vast majority of enplanements annually in the United States.50 We did not directly account for noise reduction policies of individual airports. However, this would be indirectly reflected in the aircraft noise estimates. In addition, although the Aviation Safety and Noise Abatement Act of 1976 and the Airport Noise and Capacity Act of 1990 established a legal framework for abatement corresponding to a threshold of DNL 65 dB,46,47 “airport sponsors have limited proprietary authority to restrict access as a means of reducing aircraft noise impacts” to local communities,108 and, in practice, this authority is rarely exercised. There are currently no state or federal policies directly limiting aircraft noise in the United States.46
Conclusions
The increasing recognition of the importance of adequate sleep for maintaining health and optimal daytime functioning has spurred research aimed at identifying modifiable factors for improving sleep duration and quality. Environmental risk factors—including noise pollution—represent targets for improving sleep health that have been underinvestigated. Estimated at participant’s home addresses, multiple metrics of aircraft noise were associated with self-reported short sleep duration even after adjustment for environmental characteristics, including greenness and LAN. Short sleep duration was associated with both Lnight and DNL, and the Lnight association varied by individual, area, and airport characteristics, including region, living near a major cargo airport, living near a water-adjacent airport, and by self-reported level of hearing loss. We found evidence for adverse effects on sleep at exposures as low as DNL 45 dB(A), the lowest modeled noise level, and evidence further showed an exposure–response relationship between aircraft noise and short sleep duration. There was little evidence that aircraft noise was associated with sleep quality as assessed by questionnaire at study baseline across most levels of aircraft noise exposure.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We thank the participants and staff of the Nurses’ Health Study; this study would not be possible without their valuable contributions. E.J. Nelson helped prepare the numerous tables in this manuscript and its Supplemental Material.
This study was funded by the Federal Aviation Administration (FAA; 13-C-AJFE-BU-016, to J.L.P.), under the Aviation Sustainability Center. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the FAA. S.T.G. and D.D.N. were supported by the National Institute of Environmental Health Sciences T32 training grant T32ES014562. J.L.P., J.I.L., and D.D.N. were additionally supported on R01ES025791-01A1, and J.E.H. and F.L. were additionally supported on P30 ES000002. S.R. was partially supported by National Heart, Lung, and Blood Institute grant R35 HL135818. We acknowledge the contribution made by the John A. Volpe National Transportation Systems Center in developing the aircraft noise contours. The Nurses’ Health Studies are supported by UM1 CA186107, R01 HL150119, R01 HL034594, U01 CA176726, R01 HL35464, U01 HL145386.
==== Refs
References
1. Institute of Medicine Committee on Sleep Medicine and Research. 2006. Sleep Disorders and Sleep Deprivation: An Unmet Public Health Problem. Colten HR, Altevogt BM, eds. Washington, DC: National Academies Press.
2. Spiegel K, Tasali E, Leproult R, Van Cauter E. 2009. Effects of poor and short sleep on glucose metabolism and obesity. Nat Rev Endocrinol 5 (5 ):253–261, PMID: , 10.1038/nrendo.2009.23.19444258
3. Knutson KL, Spiegel K, Penev P, Van Cauter E. 2007. The metabolic consequences of sleep deprivation. Sleep Med Rev 11 (3 ):163–178, PMID: , 10.1016/j.smrv.2007.01.002.17442599
4. Besedovsky L, Lange T, Haack M. 2019. The sleep–immune crosstalk in health and disease. Physiol Rev 99 (3 ):1325–1380, PMID: , 10.1152/physrev.00010.2018.30920354
5. Spiegel K, Leproult R, Van Cauter E. 1999. Impact of sleep debt on metabolic and endocrine function. Lancet 354 (9188 ):1435–1439, PMID: , 10.1016/S0140-6736(99)01376-8.10543671
6. Sivertsen B, Lallukka T, Salo P, Pallesen S, Hysing M, Krokstad S, et al. 2014. Insomnia as a risk factor for ill health: results from the large population-based prospective HUNT study in Norway. J Sleep Res 23 (2 ):124–132, PMID: , 10.1111/jsr.12102.24635564
7. Salo P, Sivertsen B, Oksanen T, Sjösten N, Pentti J, Virtanen M, et al. 2012. Insomnia symptoms as a predictor of incident treatment for depression: prospective cohort study of 40,791 men and women. Sleep Med 13 (3 ):278–284, PMID: , 10.1016/j.sleep.2011.06.022.22177343
8. Cappuccio FP, D’Elia L, Strazzullo P, Miller MA. 2010. Quantity and quality of sleep and incidence of type 2 diabetes: a systematic review and meta-analysis. Diabetes Care 33 (2 ):414–420, PMID: , 10.2337/dc09-1124.19910503
9. Ayas NT, White DP, Al-Delaimy WK, Manson JE, Stampfer MJ, Speizer FE, et al. 2003. A prospective study of self-reported sleep duration and incident diabetes in women. Diabetes Care 26 (2 ):380–384, PMID: , 10.2337/diacare.26.2.380.12547866
10. Baden MY, Hu FB, Vetter C, Schernhammer E, Redline S, Huang T. 2020. Sleep duration patterns in early to middle adulthood and subsequent risk of type 2 diabetes in women. Diabetes Care 43 (6 ):1219–1226, PMID: , 10.2337/dc19-2371.32209646
11. Cappuccio FP, Cooper D, D’Elia L, Strazzullo P, Miller MA. 2011. Sleep duration predicts cardiovascular outcomes: a systematic review and meta-analysis of prospective studies. Eur Heart J 32 (12 ):1484–1492, PMID: , 10.1093/eurheartj/ehr007.21300732
12. Ananthakrishnan AN, Khalili H, Konijeti GG, Higuchi LM, de Silva P, Fuchs CS, et al. 2014. Sleep duration affects risk for ulcerative colitis: a prospective cohort study. Clin Gastroenterol Hepatol 12 (11 ):1879–1886, PMID: , 10.1016/j.cgh.2014.04.021.24780288
13. Ayas NT, White DP, Manson JE, Stampfer MJ, Speizer FE, Malhotra A, et al. 2003. A prospective study of sleep duration and coronary heart disease in women. Arch Intern Med 163 (2 ):205–209, PMID: , 10.1001/archinte.163.2.205.12546611
14. Luyster FS, Strollo PJ Jr, Zee PC, Walsh JK, Boards of Directors of the American Academy of Sleep Medicine and the Sleep Research Society. 2012. Sleep: a health imperative. Sleep 35 (6 ):727–734, PMID: , 10.5665/sleep.1846.22654183
15. McMullan CJ, Curhan GC, Forman JP. 2016. Association of short sleep duration and rapid decline in renal function. Kidney Int 89 (6 ):1324–1330, PMID: , 10.1016/j.kint.2015.12.048.27165820
16. Stenholm S, Kronholm E, Bandinelli S, Guralnik JM, Ferrucci L. 2011. Self-reported sleep duration and time in bed as predictors of physical function decline: results from the InCHIANTI study. Sleep 34 (11 ):1583–1593, PMID: , 10.5665/sleep.1402.22043129
17. Chen H, Schernhammer E, Schwarzschild MA, Ascherio A. 2006. A prospective study of night shift work, sleep duration, and risk of Parkinson’s disease. Am J Epidemiol 163 (8 ):726–730, PMID: , 10.1093/aje/kwj096.16495472
18. Tworoger SS, Lee S, Schernhammer ES, Grodstein F. 2006. The association of self-reported sleep duration, difficulty sleeping, and snoring with cognitive function in older women. Alzheimer Dis Assoc Disord 20 (1 ):41–48, PMID: , 10.1097/01.wad.0000201850.52707.80.16493235
19. Rod NH, Vahtera J, Westerlund H, Kivimaki M, Zins M, Goldberg M, et al. 2011. Sleep disturbances and cause-specific mortality: results from the GAZEL cohort study. Am J Epidemiol 173 (3 ):300–309, PMID: , 10.1093/aje/kwq371.21193534
20. Cappuccio FP, D’Elia L, Strazzullo P, Miller MA. 2010. Sleep duration and all-cause mortality: a systematic review and meta-analysis of prospective studies. Sleep 33 (5 ):585–592, PMID: , 10.1093/sleep/33.5.585.20469800
21. Patel SR, Ayas NT, Malhotra MR, White DP, Schernhammer ES, Speizer FE, et al. 2004. A prospective study of sleep duration and mortality risk in women. Sleep 27 (3 ):440–444, PMID: , 10.1093/sleep/27.3.440.15164896
22. Fink D. 2019. A new definition of noise: noise is unwanted and/or harmful sound. Noise is the new ‘secondhand smoke.’ Proc Meet Acoust 39 :050002. 10.1121/2.0001186.
23. Oswald I, Taylor AM, Treisman M. 1960. Discriminative responses to stimulation during human sleep. Brain 83 (3 ):440–453, PMID: , 10.1093/brain/83.3.440.13731563
24. Basner M, Griefahn B, van den Berg M. 2010. Aircraft noise effects on sleep: mechanisms, mitigation and research needs. Noise Health 12 (47 ):95–109, PMID: , 10.4103/1463-1741.63210.20472955
25. Raschke F. 2004. Arousals and aircraft noise–environmental disorders of sleep and health in terms of sleep medicine. Noise Health 6 (22 ):15–26, PMID: .15070525
26. Kryter KD. 1966. Psychological reactions to aircraft noise. Science 151 (3716 ):1346–1355, PMID: , 10.1126/science.151.3716.1346.5909784
27. Scheuch K. 2004. Special assessment of aircraft noise effects during night by the Council of Experts for Environmental Questions of FRG. Noise Health 6 (22 ):65–67, PMID: .15070531
28. Smith MG, Rocha S, Witte M, Basner M. 2020. On the feasibility of measuring physiologic and self-reported sleep disturbance by aircraft noise on a national scale: a pilot study around Atlanta airport. Sci Total Environ 718 :137368, PMID: , 10.1016/j.scitotenv.2020.137368.32092522
29. Elmenhorst EM, Griefahn B, Rolny V, Basner M. 2019. Comparing the effects of road, railway, and aircraft noise on sleep: exposure–response relationships from pooled data of three laboratory studies. Int J Environ Res Public Health 16 (6 ):1073, PMID: , 10.3390/ijerph16061073.30917492
30. McGuire S, Müller U, Elmenhorst EM, Basner M. 2016. Inter-individual differences in the effects of aircraft noise on sleep fragmentation. Sleep 39 (5 ):1107–1110, PMID: , 10.5665/sleep.5764.26856901
31. Nassur AM, Léger D, Lefèvre M, Elbaz M, Mietlicki F, Nguyen P, et al. 2019. The impact of aircraft noise exposure on objective parameters of sleep quality: results of the DEBATS study in France. Sleep Med 54 :70–77, PMID: , 10.1016/j.sleep.2018.10.013.30529780
32. Basner M, Witte M, McGuire S. 2019. Aircraft noise effects on sleep—results of a pilot study near Philadelphia International Airport. Int J Environ Res Public Health 16 (17 ):3178, PMID: , 10.3390/ijerph16173178.31480420
33. Nassur AM, Léger D, Lefèvre M, Elbaz M, Mietlicki F, Nguyen P, et al. 2019. Effects of aircraft noise exposure on heart rate during sleep in the population living near airports. Int J Environ Res Public Health 16 (2 ):269, PMID: , 10.3390/ijerph16020269.30669300
34. Nassur AM, Lefèvre M, Laumon B, Léger D, Evrard AS. 2019. Aircraft noise exposure and subjective sleep quality: the results of the DEBATS study in France. Behav Sleep Med 17 (4 ):502–513, PMID: , 10.1080/15402002.2017.1409224.29172716
35. Rocha S, Smith MG, Witte M, Basner M. 2019. Survey results of a pilot sleep study near Atlanta International Airport. Int J Environ Res Public Health 16 (22 ):4321, PMID: , 10.3390/ijerph16224321.31698800
36. Perron S, Plante C, Ragettli MS, Kaiser DJ, Goudreau S, Smargiassi A. 2016. Sleep disturbance from road traffic, railways, airplanes and from total environmental noise levels in Montreal. Int J Environ Res Public Health 13 (8 ):809, PMID: , 10.3390/ijerph13080809.27529260
37. Franssen EAM, van Wiechen CMAG, Nagelkerke NJD, Lebret E. 2004. Aircraft noise around a large international airport and its impact on general health and medication use. Occup Environ Med 61 (5 ):405–413, PMID: , 10.1136/oem.2002.005488.15090660
38. Brink M, Schäffer B, Vienneau D, Pieren R, Foraster M, Eze IC, et al. 2019. Self-reported sleep disturbance from road, rail and aircraft noise: exposure–response relationships and effect modifiers in the SiRENE study. Int J Environ Res Public Health 16 (21 ):4186, PMID: , 10.3390/ijerph16214186.31671890
39. Holt JB, Zhang X, Sizov N, Croft JB. 2015. Airport noise and self-reported sleep insufficiency, United States, 2008 and 2009. Prev Chronic Dis 12 (4 ):140551, PMID: , 10.5888/pcd12.140551.25880768
40. Knutsson A. 2003. Health disorders of shift workers. Occup Med (Lond) 53 (2 ):103–108, PMID: , 10.1093/occmed/kqg048.12637594
41. Muzet A. 2007. Environmental noise, sleep and health. Sleep Med Rev 11 (2 ):135–142, PMID: , 10.1016/j.smrv.2006.09.001.17317241
42. Billings ME, Hale L, Johnson DA. 2020. Physical and social environment relationship with sleep health and disorders. Chest 157 (5 ):1304–1312, PMID: , 10.1016/j.chest.2019.12.002.31870910
43. Grigsby-Toussaint DS, Turi KN, Krupa M, Williams NJ, Pandi-Perumal SR, Jean-Louis G. 2015. Sleep insufficiency and the natural environment: results from the US Behavioral Risk Factor Surveillance System survey. Prev Med 78 :78–84, PMID: , 10.1016/j.ypmed.2015.07.011.26193624
44. Astell-Burt T, Feng X, Kolt GS. 2013. Does access to neighbourhood green space promote a healthy duration of sleep? Novel findings from a cross-sectional study of 259 319 Australians. BMJ Open 3 (8 ):e003094, PMID: , 10.1136/bmjopen-2013-003094.23943772
45. Schäffer B, Brink M, Schlatter F, Vienneau D, Wunderli JM. 2020. Residential green is associated with reduced annoyance to road traffic and railway noise but increased annoyance to aircraft noise exposure. Environ Int 143 :105885, PMID: , 10.1016/j.envint.2020.105885.32619911
46. U.S. Congress. 1980. Public Law 96-193: Aviation Safety and Noise Abatement Act of 1979. https://www.govtrack.us/congress/bills/96/hr2440/text [accessed 25 July 2002].
47. U.S. Congress. 1990. Aircraft Noise and Capacity Act of 1990 (“ANCA”). https://www.lawa.org/-/media/lawa-web/environment/files/airport_noise_and_capacity_act_of_1990.ashx?la=en&hash=908B516999C1AE9F81A7852B0233081E707C1869 [accessed 25 July 2002].
48. Bao Y, Bertoia ML, Lenart EB, Stampfer MJ, Willett WC, Speizer FE, et al. 2016. Origin, methods, and evolution of the three Nurses’ Health Studies. Am J Public Health 106 (9 ):1573–1581, PMID: , 10.2105/AJPH.2016.303338.27459450
49. Morabia A. 2016. 120 000 Nurses who shook public health. Am J Public Health 106 (9 ):1528–1529, PMID: , 10.2105/AJPH.2016.303345.27509274
50. FAA (Federal Aviation Administration). 2021. Passenger boarding (enplanement) and all-cargo data for U.S. airports. https://www.faa.gov/airports/planning_capacity/passenger_allcargo_stats/passenger/ [accessed 17 August 2021].
51. Consensus Conference Panel, Watson NF, Badr MS, Belenky G, Bliwise DL, Buxton OM, et al. 2015. Recommended amount of sleep for a healthy adult: a joint consensus statement of the American Academy of Sleep Medicine and Sleep Research Society. J Clin Sleep Med 11 (6 ):591–592, PMID: , 10.5664/jcsm.4758.25979105
52. Kim CS, Grady ST, Hart JE, Laden F, VoPham T, Nguyen DD, et al. 2022. Long-term aircraft noise exposure and risk of hypertension in the Nurses’ Health Studies. Environ Res 207 :112195, PMID: , 10.1016/j.envres.2021.112195.34627796
53. Simon MC, Hart JE, Levy JI, VoPham T, Malwitz A, Nguyen D, et al. 2022. Sociodemographic patterns of exposure to civil aircraft noise in the United States. Environ Health Perspect 130 (2 ):027009, PMID: , 10.1289/EHP9307.35167327
54. FAA. 2020. Report to Congress: FAA Reauthorization Act of 2018 (Pub. L. 115-254): Section 188 and Sec 173. https://www.faa.gov/sites/faa.gov/files/2021-11/Day-Night_Average_Sound_Levels_COMPLETED_report_w_letters.pdf [accessed 6 October 2021].
55. World Health Organization (WHO) Regional Office for Europe. 2018. WHO Environmental Noise Guidelines for the European Region. https://www.euro.who.int/en/publications/abstracts/environmental-noise-guidelines-for-the-european-region-2018.
56. Textor J, van der Zander B, Gilthorpe MS, Liśkiewicz M, Ellison GTH. 2016. Robust causal inference using directed acyclic graphs: the R package ‘dagitty.’ Int J Epidemiol 45 (6 ):1887–1894, PMID: , 10.1093/ije/dyw341.28089956
57. Bethea TN, Zhou ES, Schernhammer ES, Castro-Webb N, Cozier YC, Rosenberg L. 2020. Perceived racial discrimination and risk of insomnia among middle-aged and elderly Black women. Sleep 43 (1 ):zsz208, PMID: , 10.1093/sleep/zsz208.31555803
58. Lidfeldt J, Li TY, Hu FB, Manson JE, Kawachi I. 2007. A prospective study of childhood and adult socioeconomic status and incidence of type 2 diabetes in women. Am J Epidemiol 165 (8 ):882–889, PMID: , 10.1093/aje/kwk078.17284723
59. Billings ME, Gold D, Szpiro A, Aaron CP, Jorgensen N, Gassett A, et al. 2019. The association of ambient air pollution with sleep apnea: the Multi-Ethnic Study of Atherosclerosis. Ann Am Thorac Soc 16 (3 ):363–370, PMID: , 10.1513/AnnalsATS.201804-248OC.30571166
60. Shen YL, Liu WT, Lee KY, Chuang HC, Chen HW, Chuang KJ. 2018. Association of PM2.5 with sleep-disordered breathing from a population-based study in northern Taiwan urban areas. Environ Pollut 233 :109–113, PMID: , 10.1016/j.envpol.2017.10.052.29059625
61. Yanosky JD, Paciorek CJ, Laden F, Hart JE, Puett RC, Liao D, et al. 2014. Spatio-temporal modeling of particulate air pollution in the conterminous United States using geographic and meteorological predictors. Environ Health 13 (1 ):1–15, PMID: , 10.1186/1476-069X-13-63.24405644
62. Johnson BS, Malecki KM, Peppard PE, Beyer KMM. 2018. Exposure to neighborhood green space and sleep: evidence from the Survey of the Health of Wisconsin. Sleep Health 4 (5 ):413–419, PMID: , 10.1016/j.sleh.2018.08.001.30241655
63. Yeager R, Riggs DW, DeJarnett N, Tollerud DJ, Wilson J, Conklin DJ, et al. 2018. Association between residential greenness and cardiovascular disease risk. J Am Heart Assoc 7 (24 ):e009117, PMID: , 10.1161/JAHA.118.009117.30561265
64. Kriegler FJ, Malila WA, Nalepka RF, Richardson W. 1969. Preprocessing transformations and their effects on multispectral recognition. In: Proceedings of the Sixth International Symposium on Remote Sensing of Environment, Volume II. 13–16 October 1969. Ann Arbor, MI: Infrared and Optics Laboratory, Willow Run Laboratories, Environmental Research Institute of Michigan, 97–131.
65. Ohayon MM, Milesi C. 2016. Artificial outdoor nighttime lights associate with altered sleep behavior in the American general population. Sleep 39 (6 ):1311–1320, PMID: , 10.5665/sleep.5860.27091523
66. Chepesiuk R. 2009. Missing the dark: health effects of light pollution. Environ Health Perspect 117 (1 ):A20–A27, PMID: , 10.1289/ehp.117-a20.19165374
67. Cho YM, Ryu SH, Lee BR, Kim KH, Lee E, Choi J. 2015. Effects of artificial light at night on human health: a literature review of observational and experimental studies applied to exposure assessment. Chronobiol Int 32 (9 ):1294–1310, PMID: , 10.3109/07420528.2015.1073158.26375320
68. James P, Bertrand KA, Hart JE, Schernhammer ES, Tamimi RM, Laden F. 2017. Outdoor light at night and breast cancer incidence in the Nurses’ Health Study II. Environ Health Perspect 125 (8 ):087010, PMID: , 10.1289/EHP935.28886600
69. NOAA (National Oceanic and Atmospheric Administration). 2014. Global radiance calibrated nighttime lights. https://ngdc.noaa.gov/eog/dmsp/download_radcal.html [accessed 8 August 2021].
70. Kredlow MA, Capozzoli MC, Hearon BA, Calkins AW, Otto MW. 2015. The effects of physical activity on sleep: a meta-analytic review. J Behav Med 38 (3 ):427–449, PMID: , 10.1007/s10865-015-9617-6.25596964
71. Hirsch JA, Diez Roux AV, Moore KA, Evenson KR, Rodriguez DA. 2014. Change in walking and body mass index following residential relocation: the Multi-Ethnic Study of Atherosclerosis. Am J Public Health 104 (3 ):e49–e56, PMID: , 10.2105/AJPH.2013.301773.24432935
72. Johnson DA, Hirsch JA, Moore KA, Redline S, Diez Roux AV. 2018. Associations between the built environment and objective measures of sleep. Am J Epidemiol 187 (5 ):941–950, PMID: , 10.1093/aje/kwx302.29547912
73. Johnson DA, Drake C, Joseph CLM, Krajenta R, Hudgel DW, Cassidy-Bushrow AE. 2015. Influence of neighbourhood-level crowding on sleep-disordered breathing severity: mediation by body size. J Sleep Res 24 (5 ):559–565, PMID: , 10.1111/jsr.12305.25950087
74. Hale L, Do DP. 2007. Racial differences in self-reports of sleep duration in a population-based study. Sleep 30 (9 ):1096–1103, PMID: , 10.1093/sleep/30.9.1096.17910381
75. Watson NF, Horn E, Duncan GE, Buchwald D, Vitiello MV, Turkheimer E. 2016. Sleep duration and area-level deprivation in twins. Sleep 39 (1 ):67–77, PMID: , 10.5665/sleep.5320.26285009
76. Fuller-Rowell TE, Curtis DS, El-Sheikh M, Chae DH, Boylan JM, Ryff CD. 2016. Racial disparities in sleep: the role of neighborhood disadvantage. Sleep Med 27–28 :1–8, PMID: , 10.1016/j.sleep.2016.10.008.27938909
77. Bassett E, Moore S. 2014. Neighbourhood disadvantage, network capital and restless sleep: is the association moderated by gender in urban-dwelling adults? Soc Sci Med 108 :185–193, PMID: , 10.1016/j.socscimed.2014.02.029.24650740
78. Johnson DA, Brown DL, Morgenstern LB, Meurer WJ, Lisabeth LD. 2015. The association of neighborhood characteristics with sleep duration and daytime sleepiness. Sleep Health 1 (3 ):148–155, PMID: , 10.1016/j.sleh.2015.06.002.29073435
79. Johnson DA, Lisabeth L, Hickson D, Johnson-Lawrence V, Samdarshi T, Taylor H, et al. 2016. The social patterning of sleep in African Americans: associations of socioeconomic position and neighborhood characteristics with sleep in the Jackson Heart Study. Sleep 39 (9 ):1749–1759, PMID: , 10.5665/sleep.6106.27253767
80. DeSantis AS, Diez Roux AV, Moore K, Baron KG, Mujahid MS, Javier Nieto FJ. 2013. Associations of neighborhood characteristics with sleep timing and quality: the Multi-Ethnic Study of Atherosclerosis. Sleep 36 (10 ):1543–1551, PMID: , 10.5665/sleep.3054.24082314
81. Schisterman EF, Cole SR, Platt RW. 2009. Overadjustment bias and unnecessary adjustment in epidemiologic studies. Epidemiology 20 (4 ):488–495, PMID: , 10.1097/EDE.0b013e3181a819a1.19525685
82. DeVille NV, Iyer HS, Holland I, Bhupathiraju SN, Chai B, James P, et al. 2022. Neighborhood socioeconomic status and mortality in the Nurses’ Health Study (NHS) and the Nurses’ Health Study II (NHSII). Environ Epidemiol 7 (1 ):e235, PMID: , 10.1097/EE9.0000000000000235.36777531
83. Liu Y, Wheaton AG, Chapman DP, Cunningham TJ, Lu H, Croft JB. 2016. Prevalence of healthy sleep duration among adults—United States, 2014. MMWR Morb Mortal Wkly Rep 65 (6 ):137–141, PMID: , 10.15585/mmwr.mm6506a1.26890214
84. Obradovich N, Migliorini R, Mednick SC, Fowler JH. 2017. Nighttime temperature and human sleep loss in a changing climate. Sci Adv 3 (5 ):e1601555, PMID: , 10.1126/sciadv.1601555.28560320
85. Liao C, Akimoto M, Bivolarova MP, Sekhar C, Laverge J, Fan X, et al. 2021. A survey of bedroom ventilation types and the subjective sleep quality associated with them in Danish housing. Sci Total Environ 798 :149209, PMID: , 10.1016/j.scitotenv.2021.149209.34332381
86. Rasker R, Gude PH, Gude JA, van den Noort J. 2009. The economic importance of air travel in high-amenity rural areas. J Rural Stud 25 (3 ):343–353, 10.1016/j.jrurstud.2009.03.004.
87. Grandner MA, Smith TE, Jackson N, Jackson T, Burgard S, Branas C. 2015. Geographic distribution of insufficient sleep across the United States: a county-level hotspot analysis. Sleep Health 1 (3 ):158–165, PMID: , 10.1016/j.sleh.2015.06.003.26989761
88. Johnson DA, Thorpe RJ, McGrath JA, Jackson WB, Jackson CL. 2018. Black–White differences in housing type and sleep duration as well as sleep difficulties in the United States. Int J Environ Res Public Health 15 (4 ):564, PMID: , 10.3390/ijerph15040564.29561769
89. U.S. Census Bureau. 1997. American housing brief: regional differences in housing. https://www.census.gov/content/dam/Census/programs-surveys/ahs/publications/ahb-9501.pdf [accessed 14 December 2022].
90. San Francisco International Airport. 2021. SFO flight patterns and operations. https://www.flysfo.com/about/community-noise/noise-office/flight-patterns-operations [accessed 9 June 2021].
91. Leleu C, Marsh D. 2009. Dependent on the dark: cargo and other night flights in European airspace. https://www.eurocontrol.int/sites/default/files/publication/files/tat5-night-freight-report.pdf [accessed 9 June 2021].
92. FAA (Federal Aviation Administration). 2020. Aviation Environmental Design Tool (AEDT). Version 3c. User Manual. https://aedt.faa.gov/Documents/AEDT3c_UserManual.pdf [accessed 18 December 2022].
93. Konishi K, Maekawa Z. 2001. Interpretation of long term data measured continuously on long range sound propagation over sea surfaces. Appl Acoust 62 (10 ):1183–1210, 10.1016/S0003-682X(00)00096-7.
94. Konishi K, Tanioku Y, Maekawa Z. 2000. Long time measurement of long range sound propagation over an ocean surface. Appl Acoust 61 (2 ):149–172, 10.1016/S0003-682X(00)00009-8.
95. Basner M, McGuire S. 2018. WHO Environmental Noise Guidelines for the European Region: a systematic review on environmental noise and effects on sleep. Int J Environ Res Public Health 15 (3 ):519, PMID: , 10.3390/ijerph15030519.29538344
96. Basner M, Samel A. 2005. Effects of nocturnal aircraft noise on sleep structure. Somnologie 9 (2 ):84–95, 10.1111/j.1439-054X.2005.00051.x.
97. Rudolph KE, Shev A, Paksarian D, Merikangas KR, Mennitt DJ, James P, et al. 2019. Environmental noise and sleep and mental health outcomes in a nationally representative sample of urban US adolescents. Environ Epidemiol 3 (4 ):e056, PMID: , 10.1097/EE9.0000000000000056.31538137
98. Bureau of Transportation Statistics. 2021. Air cargo summary data (all). https://www.transtats.bts.gov/freight.asp [accessed 26 October 2021].
99. Airports Council International–North America Cargo Committee. 2019. Air Cargo Guide. https://airportscouncil.org/wp-content/uploads/2020/03/Air-Cargo-Guide.pdf [accessed 30 November 2022].
100. Dožić S, Babić D. 2015. Trends in airline cargo fleet. In: 2nd Logistics International Conference. 21–23 May 2015. Belgrade, Serbia: LOGIC, 1–6.
101. IATA (International Air Transport Association). 2022. Annual Review 2022. https://www.iata.org/contentassets/c81222d96c9a4e0bb4ff6ced0126f0bb/annual-review-2022.pdf [accessed 15 December 2022].
102. Zaporozhets O. 2016. Aircraft noise models for assessment of noise around airports—improvements and limitations. https://www.icao.int/environmental-protection/Documents/EnvironmentalReports/2016/ENVReport2016_pg50-55.pdf [accessed 25 July 2022].
103. Liu J, Liu J, Lai D, Pei J, Wei S. 2021. A field investigation of the thermal environment and adaptive thermal behavior in bedrooms in different climate regions in China. Indoor Air 31 (3 ):887–898, PMID: , 10.1111/ina.12775.33350529
104. Wensveen JG. 2016. Managerial aspects of airlines. In: Air Transportation: A Management Perspective. 8th ed. New York, NY: Routledge, 423–452.
105. Lin FR, Niparko JK, Ferrucci L. 2011. Hearing loss prevalence in the United States. Arch Intern Med 171 (20 ):1851–1852, PMID: , 10.1001/archinternmed.2011.506.22083573
106. Basner M, Brink M, Bristow A, de Kluizenaar Y, Finegold L, Hong J, et al. 2015. ICBEN review of research on the biological effects of noise 2011–2014. Noise Health 17 (75 ):57–82, PMID: , 10.4103/1463-1741.153373.25774609
107. Hauptman M, Lockley SW, Landrigan CP. 2022. Invited commentary: there’s no place like home—integrating a place-based approach to understanding sleep. Am J Epidemiol 191 (9 ):1540–1543, PMID: , 10.1093/aje/kwac086.35524716
108. FAA. 2021. Airport noise and access restrictions. In: Airport Compliance Manual: Order 5190.6B Change 1. Washington, DC: FAA. https://www.federalregister.gov/documents/2021/12/03/2021-25936/airport-compliance-program [accessed 2 April 2023].
|
PMC010xxxxxx/PMC10104168.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37058434
EHP12829
10.1289/EHP12829
Science Selection
Inside Information: Black Carbon Exposure and the Early-Childhood Gut Microbiome
Schmidt Silke
14 4 2023
4 2023
131 4 04400101 2 2023
21 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
A young child crosses a crosswalk holding the hands of a man, in front, and a woman, behind, backlit by bright, low sun
==== Body
pmcEpidemiologic studies of fine particulate matter (PM2.5) often rely on external estimates of exposure that are derived from ambient air monitoring networks.1 Direct measurements of pollutants inside the body may offer greater insight into human health effects, but internal measurements can be more difficult to obtain.2 Black carbon is an especially toxic3,4 component of PM2.5 that can move from lung tissue into the bloodstream and reach multiple organs, including the gut.5 Black carbon can also cross the placenta during pregnancy6 and may accumulate in fetal organs during gestation.7
In a study recently published in Environmental Health Perspectives,8 investigators analyzed the relationship between black carbon exposure and gut biodiversity in Belgian children 4–6 years of age. “Among ambient air pollutants, black carbon particles—often considered a surrogate for traffic-related air pollutants—are one of the few materials that can be used as a biomarker of exposure,” says Joel D. Kaufman a professor of environmental health at the University of Washington and editor-in-chief of Environmental Health Perspectives, who was not involved in the study. “They can be visualized in tissues and ascribed to inhalation origin.”
Using internal measures of black carbon, the new study found associations between air pollution exposure and differences in the gut microbiome of young children. Image: © Halfpoint/stock.adobe.com.
A young child crosses a crosswalk holding the hands of a man, in front, and a woman, behind, backlit by bright, low sun
A less diverse microbiome in children has been associated with a greater risk of metabolic, cardiovascular, and cognitive disorders later in life.9,10 The researchers found that higher pre- and postnatal internal exposures to black carbon were associated with lower diversity in the children’s microbiome and differences in relative abundance of some bacterial families. Notably, the researchers found no association between ambient air concentrations of black carbon and gut biodiversity. (The ambient—or external—measures were estimated from modeled air levels at each mother–child pair’s residence during gestation and early childhood.) This finding suggests that relying on modeling-based external exposure measures may underestimate air pollution’s health risk.
Black carbon particles smaller than 1μm are formed during the combustion of wood, coal, and oil during transportation, power generation, forest fires, and agricultural burning activities.11 These microscopic particles adsorb metals and other chemicals, potentially increasing their toxicity.4
Factors such as age, sex, diet, and antibiotic use influence the composition of the gut microbiome, but more than 80% of variation among people remains unexplained.12 “The gut microbiome is like an organ of its own,” says Tim Nawrot, a professor of environmental epidemiology at Hasselt University, Belgium, and the study’s senior author. “We know that the trillions of microbes in our gut contribute to many health-related processes throughout life, but this is the first study of internally measured black carbon and the early-childhood gut microbiome.”
Nawrot and his colleagues analyzed a subset of 85 mother–child pairs enrolled in the ENVIRONAGE birth cohort.13 More than 2,000 mother–newborn pairs at the East-Limburg Hospital in Genk, Belgium, have been enrolled in the cohort since 2010. Recruitment is ongoing, and participants are followed longitudinally. The children in the new study were recruited in 2017 and 2018.8
The gut microbiome reaches its adult-like composition around 3 years of age and remains fairly stable thereafter14; the team collected fecal samples when the children were at least 4 years old. Following standard microbiome sequencing analysis, the researchers classified intestinal microbes into common taxonomic families and calculated bacterial richness and diversity indices: the Chao index, which reflects the number of detected taxa (richness), and the Shannon and Simpson indices, which account for richness as well as microbial distribution and diversity.15
Using recently developed technology,2 they quantified black carbon particles in placental tissue and cord blood collected at birth (reflecting prenatal exposures) and in the children’s urine collected around the same time as the fecal samples. Higher black carbon levels in all three sample types were associated with lower diversity based on the Shannon and Simpson (but not the Chao) indices; black carbon levels explained 6%–17% of their respective variance.
“This effect size is surprisingly large,” says Nawrot, noting that black carbon explained, on average, an estimated five times more variation in the Simpson index than antibiotic use during the previous month or soda intake during the previous 3 months.
Two bacterial families (Defluviitaleaceae and Marinifilaceae) became less common as black carbon levels in the placenta increased. Two other families (Christensenellaceae and Coriobacteriaceae) became less common as black carbon levels in urine increased. Three of these four bacterial families have previously been linked to a healthier, more diverse gut microbiome in both mice16 and humans.17
For Tanya Alderete, an assistant professor of integrative physiology at the University of Colorado Boulder, observing different associations in samples collected prenatally and in early childhood is plausible because they assess indirect and direct pollutant exposures, respectively. “The microbiome also changes rapidly between birth and age three, and different physiological mechanisms likely explain varying associations observed at different life stages,” says Alderete, who was not involved in the study.
Pilar Francino, a senior researcher in genomics and health at the Foundation for the Promotion of Health and Biomedical Research in the Valencian Community (FISABIO) in Valencia, Spain, applauds the authors’ use of internal black carbon levels. “An important contribution is that the associations were not detected with the standard external estimates, suggesting that previous studies may also have missed them,” says Francino, who was also not involved in the project.
Alderete agrees that the results are compelling and says the study adds to the growing body of evidence that air pollution may have a widespread negative impact on our microbes, perhaps starting early in life. Looking toward potential follow-up studies, Alderete says, “That makes it important to study whether breastfeeding or a healthy childhood diet can offset some of the potentially harmful effects of air pollution exposure on the gut microbiome.”
Silke Schmidt, PhD, writes about science, health, and the environment from Madison, Wisconsin.
==== Refs
References
1. Forehead H, Huynh N. 2018. Review of modelling air pollution from traffic at street-level—the state of the science. Environ Pollut 241 :775–786, PMID: , 10.1016/j.envpol.2018.06.019.29908501
2. Bové H, Steuwe C, Fron E, Slenders E, D’Haen J, Fujita Y, et al. 2016. Biocompatible label-free detection of carbon black particles by femtosecond pulsed laser microscopy. Nano Lett 16 (5 ):3173–3178, PMID: , 10.1021/acs.nanolett.6b00502.27104759
3. Janssen NAH, Hoek G, Simic-Lawson M, Fischer P, van Bree L, ten Brink H, et al. 2011. Black carbon as an additional indicator of the adverse health effects of airborne particles compared with PM10 and PM2.5. Environ Health Perspect 119 (12 ):1691–1699, PMID: , 10.1289/ehp.1003369.21810552
4. Niranjan R, Thakur AK. 2017. The toxicological mechanisms of environmental soot (black carbon) and carbon black: focus on oxidative stress and inflammatory pathways. Front Immunol 8 :763, PMID: , 10.3389/fimmu.2017.00763.28713383
5. Chow JC, Watson JG, Mauderly JL, Costa DL, Wyzga RE, Vedal S, et al. 2006. Health effects of fine particulate air pollution: lines that connect. J Air Waste Manag Assoc 56 (10 ):1368–1380, PMID: , 10.1080/10473289.2006.10464545.17063860
6. Bové H, Bongaerts E, Slenders E, Bijnens EM, Saenen ND, Gyselaers W, et al. 2019. Ambient black carbon particles reach the fetal side of human placenta. Nat Commun 10 (1 ):3866, PMID: , 10.1038/s41467-019-11654-3.31530803
7. Bongaerts E, Lecante LL, Bové H, Roeffaers MBJ, Ameloot M, Fowler PA, et al. 2022. Maternal exposure to ambient black carbon particles and their presence in maternal and fetal circulation and organs: an analysis of two independent population-based observational studies. Lancet Planet Health 6 (10 ):e804–e811, PMID: , 10.1016/S2542-5196(22)00200-5.36208643
8. Van Pee T, Hogervorst J, Dockx Y, Witters K, Thijs S, Wang C, et al. 2023. Accumulation of black carbon particles in placenta, cord blood, and childhood urine in association with the intestinal microbiome diversity and composition in four- to-six-year-old children in the ENVIRONAGE birth cohort. Environ Health Perspect 131 (1 ):17010, PMID: , 10.1289/EHP11257.36719212
9. Bajinka O, Tan Y, Abdelhalim KA, Özdemir G, Qiu X. 2020. Extrinsic factors influencing gut microbes, the immediate consequences and restoring eubiosis. AMB Express 10 (1 ):130, PMID: , 10.1186/s13568-020-01066-8.32710186
10. Vallès Y, Francino MP. 2018. Air pollution, early life microbiome, and development. Curr Environ Health Rep 5 (4 ):512–521, PMID: , 10.1007/s40572-018-0215-y.30269309
11. Fawole OG, Cai XM, MacKenzie AR. 2016. Gas flaring and resultant air pollution: a review focusing on black carbon. Environ Pollut 216 :182–197, PMID: , 10.1016/j.envpol.2016.05.075.27262132
12. Falony G, Joossens M, Vieira-Silva S, Wang J, Darzi Y, Faust K, et al. 2016. Population-level analysis of gut microbiome variation. Science 352 (6285 ):560–564, PMID: , 10.1126/science.aad3503.27126039
13. Janssen BG, Madhloum N, Gyselaers W, Bijnens E, Clemente DB, Cox B, et al. 2017. Cohort profile: the ENVIRonmental influence ON early AGEing (ENVIRONAGE): a birth cohort study. Int J Epidemiol 46 (5 ):1386–1387m, PMID: , 10.1093/ije/dyw269.28089960
14. Derrien M, Alvarez AS, de Vos WM. 2019. The gut microbiota in the first decade of life. Trends Microbiol 27 (12 ):997–1010, PMID: , 10.1016/j.tim.2019.08.001.31474424
15. Kim BR, Shin J, Guevarra R, Lee JH, Kim DW, Seol KH, et al. 2017. Deciphering diversity indices for a better understanding of microbial communities. J Microbiol Biotechnol 27 (12 ):2089–2093, PMID: , 10.4014/jmb.1709.09027.29032640
16. Ge X, Wang C, Chen H, Liu T, Chen L, Huang Y, et al. 2020. Luteolin cooperated with metformin hydrochloride alleviates lipid metabolism disorders and optimizes intestinal flora compositions of high-fat diet mice. Food Funct 11 (11 ):10033–10046, PMID: , 10.1039/d0fo01840f.33135040
17. Waters JL, Ley RE. 2019. The human gut bacteria Christensenellaceae are widespread, heritable, and associated with health. BMC Biol 17 (1 ):83, PMID: , 10.1186/s12915-019-0699-4.31660948
|
PMC010xxxxxx/PMC10104171.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37058433
EHP11559
10.1289/EHP11559
Research
High-Dimensional Mediation Analysis: A New Method Applied to Maternal Smoking, Placental DNA Methylation, and Birth Outcomes
Jumentier Basile 1
Barrot Claire-Cécile 1
Estavoyer Maxime 1
Tost Jorg 2
Heude Barbara 3
François Olivier 1 4 *
https://orcid.org/0000-0001-8907-197X
Lepeule Johanna 5 *
1 Université Grenoble-Alpes, Centre National de la Recherche Scientifique, Grenoble INP, TIMC CNRS UMR 5525, Grenoble, France
2 Laboratory for Epigenetics and Environment, Centre National de Recherche en Genomique Humaine, CEA – Institut de Biologie François Jacob, University Paris Saclay, Evry, France
3 Université Paris Cité et Université Sorbonne Paris Nord, Inserm, INRAE, Centre de Recherche en Épidémiologie et StatistiqueS (CRESS), F-75004 Paris, France
4 Inria Grenoble – Rhône-Alpes Inovallée, Montbonnot, France
5 Université Grenoble-Alpes, Inserm, CNRS, Team of Environmental Epidemiology Applied to Development and Respiratory Health, Institute for Advanced Biosciences, Grenoble, France
Address correspondence to Olivier François, [email protected]. And, Johanna Lepeule, Johanna.Lepeule@univ-grenoble-alpes
14 4 2023
4 2023
131 4 04701113 5 2022
22 2 2023
23 2 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
High-dimensional mediation analysis is an extension of unidimensional mediation analysis that includes multiple mediators, and increasingly it is being used to evaluate the indirect omics-layer effects of environmental exposures on health outcomes. Analyses involving high-dimensional mediators raise several statistical issues. Although many methods have recently been developed, no consensus has been reached about the optimal combination of approaches to high-dimensional mediation analyses.
Objectives:
We developed and validated a method for high-dimensional mediation analysis (HDMAX2) and applied it to evaluate the causal role of placental DNA methylation in the pathway between exposure to maternal smoking (MS) during pregnancy and gestational age (GA) and birth weight of the baby at birth.
Methods:
HDMAX2 combines latent factor regression models for epigenome-wide association studies with max2 tests for mediation and considers CpGs and aggregated mediator regions (AMRs). HDMAX2 was carefully evaluated using simulated data and compared to state-of-the-art multidimensional epigenetic mediation methods. Then, HDMAX2 was applied to data from 470 women of the Etude des Déterminants pré et postnatals du développement de la santé de l’Enfant (EDEN) cohort.
Results:
HDMAX2 demonstrated increased power in comparison with state-of-the-art multidimensional mediation methods and identified several AMRs not identified in previous mediation analyses of exposure to MS on birth weight and GA. The results provided evidence for a polygenic architecture of the mediation pathway with a posterior estimate of the overall indirect effect of CpGs and AMRs equal to 44.5g lower birth weight representing 32.1% of the total effect [standard deviation (SD)=60.7g]. HDMAX2 also identified AMRs having simultaneous effects both on GA and on birth weight. Among the top hits of both GA and birth weight analyses, regions located in COASY, BLCAP, and ESRP2 also mediated the relationship between GA and birth weight, suggesting reverse causality in the relationship between GA and the methylome.
Discussion:
HDMAX2 outperformed existing approaches and revealed an unsuspected complexity of the potential causal relationships between exposure to MS and birth weight at the epigenome-wide level. HDMAX2 is applicable to a wide range of tissues and omic layers. https://doi.org/10.1289/EHP11559
Supplemental Material is available online (https://doi.org/10.1289/EHP11559).
* These authors contributed equally to the study.
The authors declare that they have no competing interests.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Mediation analysis is a statistical tool used to gain insights into the causal mechanisms that relate an exposure to an outcome.1 It is increasingly used in environmental epidemiology, in particular in Developmental Origins of Health and Disease (DOHaD) research and in molecular epidemiology studies.2,3 With the development of high-throughput screening technologies, these methods have become key tools to investigate the pathways by which environmental exposures can affect health outcomes and more specifically those involving epigenetic mechanisms such as DNA methylation (DNAm) variations.4–7
High-dimensional mediation analysis is an extension of unidimensional mediation analysis including multiple mediators.2 A typical high-dimensional analysis for DNAm markers generally includes three main steps. The first step tests both the effects of exposure on DNAm levels and the effects of DNAm levels on the health outcome based on epigenome-wide association studies (EWAS). The second step combines significance values obtained from the two EWAS at the first step to perform mediation tests and assesses the mediator status of each marker. The third step quantifies the indirect effects of exposure on the health outcome through DNAm differences. Analyses involving a large set of mediators are difficult and raise numerous statistical issues.2 Current methods do not optimally control for unobserved exposure-mediator and mediator-outcome confounding. Those suboptimal procedures result in estimates that cannot be interpreted as direct and indirect effects. In this study, we proposed to use state-of-art methods to estimate unobserved confounders in exposure-mediator and mediator-outcome models and then to control the false discovery rate by using an empirical-null hypothesis testing approach. Classical approaches perform multidimensional analysis by performing unidimensional mediation analyses for each DNAm marker, for example using Sobel tests or by estimating Average Causal Mediation Effect (ACME).8,9 The Sobel test is overly conservative and makes wrong assumptions regarding the theoretical null distribution.2 ACME is a single mediator approach that fails to estimate overall indirect effects when spread over multiple mediators. Improvements of the Sobel test for indirect effects combine the significance values obtained from the two EWAS in various ways.10–15 However, there is no consensus on the most relevant combination of EWAS and mediation tests for a high-dimensional analysis. Furthermore, the overall indirect effect of multiple mediators remains poorly quantified from estimates of single mediator effects in a context of correlation among the mediators.
We addressed the above issues by developing HDMAX2, a method for high-dimensional mediation analysis, and systematically compared HDMAX2 to recently proposed approaches. HDMAX2 relies on latent factor regression models to evaluate associations of exposure and outcome with DNAm and on mediation tests that control the type I error when combining the significance values obtained in the exposure and outcome EWAS. We developed additional features to further consider methylation regions as mediators and to estimate an overall mediated effect of DNAm accounting for all identified mediators simultaneously. We then used HDMAX2 to evaluate the causal role of placental DNA methylation in the pathway between maternal smoking (MS) during pregnancy, gestational age (GA) at delivery and birth weight of the baby. Although several studies focused on cord blood DNAm,16–19 we focused our investigations on placental DNAm20–23 because it plays a key role in fetal programming. In this study, we evaluate CpG mediators and regions based on multidimensional approaches, and we propose to estimate an overall indirect effect of MS during pregnancy on newborn birth weight and on GA at delivery. Although our application involves placental DNAm data, the approach extends to various types of data and holds for other types of tissue and quantitative omics data.
Methods
Overview of the HDMAX2 Method
HDMAX2 is a new approach for high-dimensional mediation analysis structured in three main steps (Figure 1). The first step of HDMAX2 corresponds to an extension of regression models considered generally in unidimensional mediation analysis.1 The extension includes latent factors as covariates in the models to account for unobserved variables that confound multidimensional DNAm data analysis, such as batch effects or cell-type heterogeneity in samples. The second step identifies potential mediators by combining paired significance values that are obtained when testing the effect of exposure on DNAm and the effect of DNAm on outcome in step 1. Step 2 is not restricted to CpG markers, and it can also identify aggregated mediator regions (AMRs) based on the paired p-values. The third step quantifies indirect effects either separately (each identified mediator) or simultaneously with a cumulated indirect effect of all mediators called overall indirect effect.
Figure 1. Workflow of the high-dimensional mediation analysis procedure implemented in HDMAX2. Note: HDMAX2, high-dimensional mediation analysis.
Figure 1 is a flowchart with three steps. Step 1: epigenome-wide association studies using latent factor models, including methylation approximate to exposure and outcome approximate to methylation plus exposure. Step 2: cytosine–phosphate–guanine: Mediation test with maximum squared and detection of cytosine–phosphate–guanine by F D R control. Aggregated mediator regions: Detection of aggregated mediator regions using the comb-p algorithm on the uppercase p of the maximum squared test. Step 3: Estimate indirect effect: Estimation of the individual indirect effects of cytosine–phosphate–guanine and aggregated mediator regions via the mediation package. Estimation of the overall indirect effects (O I E) of cytosine–phosphate–guanine and aggregated mediator regions and both.
Step 1. Evaluating associations between exposure, mediators, and outcome.
The first step of HDMAX2 is to adjust latent factor mixed models (LFMMs) to estimate the effects of exposure, X, on a matrix M of CpG markers and the effect of each marker on outcome, Y.24,25 LFMMs belong to a class of estimation algorithms that adjust latent factor models and that encompass surrogate variable analysis (SVA),26 directed SVA,27 or confounder adjusted testing and estimation (CATE).28 Latent factor models differ from models based on a priori estimates of cell types29,30 and represent a more general approach to the issue of confounding in association studies.26 Within the latent factor regression framework, additional known covariates like maternal age or sex of the newborn can be included in the model to improve accuracy.
To estimate the effects of exposure (X) on a matrix of CpG markers (M), the following model was first adjusted to the centered data: (1) M=XaT+U1V1T+E1,
where a contains the vector of effect sizes of exposure on DNAm levels, U1 is a matrix formed of K latent factors estimated simultaneously with a, V1 contains the loadings associated with the latent factors, E1 is a matrix of residual errors, and T is the transposed of the given matrix. The K latent factors represent hidden confounders—e.g., unobserved cell types of tissue samples and batch effects. Using the latent factor regression defined in Equation 1, a significance value, Px, is computed for the test of a null effect size for exposure on DNAm at each CpG marker (H0: aj=0, for the jth marker).
A second EWAS was then performed to estimate effect sizes for the DNAm levels on the health outcome (Y) as follows: (2) Y=Xc+MbT+U2V2T+E2,
where c contains the direct effect of exposure on outcome, b contains the effect sizes of DNAm levels on outcome, U2 are latent factors from a latent factor regression model, V2 contains the corresponding loadings, and E2 is a matrix of errors. For each marker j, a significance value, Py, is computed for the test of a null effect size for DNAm on outcome (H0: bj=0, for the jth marker).
Step 2. Identifying potential CpG mediators and AMRs.
The second step of HDMAX2 combines the significance values Px and Py computed at each DNAm marker by using a new procedure called the max-squared (max2) test. The p-value for the max2 test was computed as p=max(Px,Py)². Like the Sobel test, the max2 test rejects the null hypothesis that either the effect of exposure on DNAm or the effect of DNAm on outcome is null. The square in the formula warrants that the distribution of p-values is uniform when Px and Py are independent and uniformly distributed. In HDMAX2, the max2 test was first used to identify potential CpG mediators. A combination of p-values along the methylome was then performed to identify potential AMRs using comb-p, a method relying on the Stouffer-Liptak-Kechris correction that combines adjacent CpG p-values in sliding windows.31 We considered methylated regions including at least two markers at a maximum distance of 1,000 bp and significant at the 10% false discovery rate (FDR) level. The mean value of DNAm levels for CpGs located in AMRs was retained to summarize information on methylated regions.
Step 3. Quantifying indirect effects with single and multiple mediators.
Mediation of exposure on the outcome was first assessed at the level of CpG markers and then at the level of aggregated regions. For CpG and for AMRs, estimates of indirect effect sizes and the proportion of mediated effect were computed in the R package mediation.8 For CpGs, the estimate of the indirect effect size for marker j was checked to be equivalent to the product of effect sizes, aj bj, computed in Equations 1 and 2. A novelty of HDMAX2 is to evaluate an overall (cumulated) indirect effect for all CpGs or AMRs identified in Step 3. The overall indirect effect (OIE) was estimated in a model including m mediator variables as follows: (3) yi=c xi+∑j=1m bjmij+∑k=1Kvkuik(2)+εi,
where (mij) represents methylation levels observed at m CpG mediators or AMRs (or both of them), and the terms (uik(2)) correspond to the latent factor coordinates estimated in Step 1. The overall indirect effect was then computed as (4) OIE=∑j=1majbj,
where (aj) represents the effect of exposure on methylation (Step 1). To account for correlation among mediators, the standard deviation of the OIE estimate was computed using a bootstrap approach (10,000 replicates). The bootstrap distribution represents an approximate noninformative posterior distribution of our parameter.32 Steps 1 and 2 meet the assumptions needed for the estimates of direct and indirect effects to be interpreted causally.6 First, control is made for exposure–outcome confounding by adjusting estimates on child and maternal covariates and on technical factors related to DNAm measurements (see below, Assumption A16). Second, control is made for mediator-outcome confounding and for exposure-mediator confounding (Assumptions A2 and A36). Those adjustments are realized through the estimation of latent factors and through additional corrections performed in an empirical-null hypothesis testing approach (see below). Finally, latent factors in U1 and U2, which are estimated separately, differ from each other, and they also differ from cell-type composition estimates that are commonly considered as confounding factors in methylation EWAS. Thus, estimates of direct and indirect effects are built in a way that minimizes the chance that mediator-outcome confounder is affected by the exposure (Assumption A46).
Simulation Studies
We performed simulations to compare the methods implemented in HDMAX2 with state-of-the-art approaches for EWAS in Step 1 and for mediation tests in Step 2.
Step-1 EWAS methods evaluated in simulations.
In HDMAX2 Step 1, several latent factor estimation algorithms could be implemented for performing the EWAS. A preliminary study was performed to decide which of LFMM2, SVA, and CATE was the best for our data set using precision (1-FDR) and F1-score (harmonic mean of precision and power), as evaluation measures (Figure S1; Excel Table S1). As another performance metric, we also measured the computational time of high-dimensional mediation methods as a function of the number of markers, varied from 102 to 106 in a typical analysis. Then we performed generative simulations to compare methods using latent factors with those based on estimates of cell-type composition. In this step, HDMAX2 was compared to two linear regression models including a priori estimates of cell-type composition obtained from RefFreeEWAS29 and ReFACTor.30 These two deconvolution methods provide proportions of putative cell types defined by a subset of the methylation matrix. Note that this is a major difference with LFMM, in which cell-type composition is replaced by latent factor estimates computed simultaneously with effect size estimates (vectors a and b in Equations 1 and 2). Generative simulations were built using a conditional simulation algorithm to simulate data used in EWAS and to evaluate the performance of statistical methods for latent factor regression models as defined in Equation 1. Consider a matrix of methylation profiles, M, obtained for n individuals and corresponding to M markers. We assumed that regression models mj=bjX + Ej describe the relationship between an unobserved exposure variable, X, and methylation levels observed for the jth marker (bj is the size of the effect on exposure on methylation level j, and Ej has variance σj2). Conditional on the matrix of methylation profiles, we can simulate the unobserved exposure X, of variance σX2, as follows: (5) X|M=(m1,…,mM)∼N(σx21+σx2‖b/σ‖22∑j=1mbjmjσj2,σx21+σx2‖b/σ‖22),
where ‖b/σ‖22=b12/σ12+b22/σ22+⋯+bM2/σM2.
Using data from chromosome 1 in the Etude des Déterminants pré et postnatals du développement de la santé de l’Enfant (EDEN) cohort data of placenta DNA methylation (see section “MS, Placental DNA Methylation, and Pregnancy Outcomes” below), we defined causal markers, for which bj is non-null, and the corresponding effects on methylation levels as follows: One hundred EWAS were generated from those data with 30 randomly chosen causal markers and σx=0.18. The effect sizes at causal markers were defined as bj=σj/0.1. We chose k=5 latent factors in SVA, CATE, and LFMM2.
Step-2 mediation methods evaluated in simulations.
We compared the max2 mediation test in Step 2 of HDMAX2 to methods based on direct application of Sobel tests and of univariate mediation analysis using the F1-score.20,21 Then we compared the max2 test to five recent methods for high-dimensional mediation: a multiple-testing procedure for high-dimensional mediation hypotheses, HDMT, similar to the max2 test,10 a two-step familywise error rate procedure called ScreenMin,11 an approach using familywise error rate and false discovery rate control when testing multiple mediators SBMH,14 a linear regression model combined with an ANOVA,33 and an approach using variable selection to reduce the number of mediators HIMA.15 The last two approaches combined the first steps of HDMAX2 in a single step.
Mediation model simulations.
The simulations were performed according to a generative model that reproduces the mediation pathways described in Equation 1 and Equation 2. Exposure and outcome (X and Y) and three confounding factors (U) were simulated according to a multivariate Gaussian model. The percentage of variance of exposure and outcome explained by the confounding factors, and the correlation between those variables, were set at 10%. The variances of confounding factors were equal to one. The number of DNAm markers was equal to m=38,000, approximately equal to the number of CpGs for a single chromosome in our empirical data, and the number of individuals was equal to n=500. The vectors of effect sizes (a for exposure and b for outcome) were generated by setting a proportion of effect sizes to zero. Non-null effect sizes were sampled according to a standard Gaussian distribution. The levels of parameters a and b, representing lower and higher values of effect sizes, were chosen so that the performance metrics result in enough variability across methods to allow useful interpretations. A residual error matrix E was simulated by using a multivariate Gaussian distribution with means equal to zero and standard deviations of one. In addition to the three confounding factors, six additional factors representing artificial cell proportions for six different cell types were simulated using a Dirichlet distribution. To consider values that are realistic with respect to our data analysis, the parameters of the Dirichlet distribution were equal to the cell-type proportions estimated on the EDEN placental DNAm data (described in Mediation analyses). A matrix of DNAm markers was built using Equation 1 and Equation 2 with three parameters: the mean of non-null effect size for exposure (X) on methylation M (a=0.2, 0.4), the mean of nonnull effect size for M on outcome (b=0.2, 0.4), and the number of putative causal markers (equal to 8, 16, or 32). For each set of parameters, 200 simulations were carried out. For each method tested, a subset of hits with a level of FDR=5% was selected as potential mediators.34 For each list of hits, we computed precision (1-FDR), sensitivity (power), and the harmonic mean of precision and sensitivity (F1-score). The highest value of an F1-score is 1, if precision and sensitivity are maximal, and the lowest value is zero, if either the precision or the sensitivity is null.
MS, Placental DNA Methylation, and Pregnancy Outcomes
Study population.
Our analysis included participants of the 2002 mother–child dyads from the EDEN cohort enrolled in the university hospitals of Nancy and Poitiers, France, between 2003 and 2006.35,36 Lifestyle, demographic, and clinical data were collected by questionnaires and interviews during pregnancy and after delivery. The EDEN cohort received approval from the ethics committee (CCPPRB) of Kremlin Bicêtre and from the French data privacy institution Commission Nationale de l’Informatique et des Libertés (CNIL). Written consent was obtained from the mothers for themselves and for the offspring.
DNAm measurements.
DNAm was measured from DNA extracted from 668 placental samples, collected by specifically trained midwives of the study using the following standardized procedure in both centers. Placenta was sampled (∼5 mm3) a few centimeters from the insertion of the cord under the chorio-amniotic membrane, washed in a saline solution, and immediately frozen at −80°C. Illumina’s Infinium HumanMethylation450 BeadChip (Illumina, Inc.) was used to assess the levels of methylation in samples following the manufacturer’s instructions. Protocols for placental DNA extraction and DNAm processing are detailed in.37 Briefly, DNAm was normalized using the beta-mixture quantile (BMIQ) method to ultimately obtain beta-methylation levels for 379,904 CpG probed CpG sites.38
MS, birth weight, and GA.
Among the 668 women, we excluded preterm deliveries (n=28, gestational duration <37 wk), women who reported quitting smoking in the 3 months before pregnancy (n=70), and women whose smoking status was unknown (n=100), leaving 470 women included in our analyses. Birth weight (in grams) was extracted from medical records. We computed the Pearson correlation coefficient between birth weight and GA. Prenatal maternal cigarette smoking was collected by questionnaires during prenatal and postpartum clinical examinations. Nonsmokers were defined as women who did not smoke during the 3 months before and during pregnancy (359 nonsmokers). Smokers were defined as women smoking more than one cigarette per day throughout the duration of the pregnancy (111 smokers). All smokers during pregnancy also smoked during the 3 months before pregnancy. GA was defined as GA at birth (in weeks).
Mediation analyses.
We hypothesized that maternal smoking during pregnancy could induce modifications of placental DNAm that result in differences in GA or in birth weight. To this aim, we investigated the relationships between MS, placental DNAm, and each pregnancy outcome. MS was encoded as a categorical variable (smokers/nonsmokers), and the outcomes were encoded as continuous variables. To identify mediators of the exposure–outcome relationship, we used the HDMAX2 approach to evaluate DNAm CpG mediators first and then to identify AMRs.
In HDMAX2 regression models, adjustment factors included child sex, parity (0, 1, ≥2 children; categorical covariate), maternal age at end of education (≤18, 19–20, 21–22, 23–24, ≥25 y; categorical covariate), maternal body mass index [BMI (kilograms per square meter); continuous] before pregnancy, and maternal age at delivery (years; continuous), collected during pregnancy and at delivery by maternal self-administered questionnaires or by the midwives during clinical examinations. Adjustment factors also included season of conception (categorical covariate); study center (Nancy/Poitiers); and batch, plate, and chip technical factors related to DNAm measurements (categorical covariates). We relied on the principal component analysis of the DNAm matrix to include six latent factors in the HDMAX2 regression models (Figure S2). This number was consistent with the six factors selected in a previous work to represent the cell types using the Reffree algorithm.22 We adopted an empirical null approach, which can correct for shift in the data to respect the shape of the theoretical null.39 FDR-corrected p-values were calculated for the 379,904 CpGs using the local FDR algorithm in fdrtool.40 Calibration of the max2 test p-values was evaluated through a direct examination of the histogram of p-values. The local FDR parameter (eta0) was computed to evaluate the proportion of null hypothesis among the 379,904 tests. This proportion was estimated at eta0=99.8%−99.9%, suggesting that an FDR level of 5% would be overly conservative (Figure S3). To agree with the value of eta0, candidate CpGs were selected at FDR levels <10%, corresponding to adjusted p<9.03×10−6 for birth weight and to adjusted p<3.27×10−6 for GA. Results obtained after considering FDR levels <20% and <5% are also reported.
Chained mediation of MS on birth weight.
To better understand the causal pathways involving (six) genic regions that mediate the effect of MS both on GA and on birth weight, we hypothesized that GA has reverse effect on DNAm levels. To assess reverse causality, we evaluated the indirect effects of targeted AMRs in a mediation analysis of GA on birth weight and of birth weight on GA. For AMRs having a significant mediation p-value, each indirect effect and an overall indirect effect were computed from the above-described procedures.
Bioinformatic analyses.
Promoter and enhancer regions were obtained from Illumina chip annotations. Gene annotations were obtained using the FDb.InfiniumMethylation.hg19 package.41 Placental gene expression of annotated genes was compared to their gene expression in other tissues according to the Expression Atlas database.42 For every gene, Chauvenet’s criterion was used to decide whether the gene was an outlier for placental expression in comparison with other tissues. Functional annotation was made from the KEGG and the Gene Ontology databases.43,44 The method presented in this study is available in the R package HDMAX2 at https://github.com/bcm-uga/hdmax2 GNU and reusable under General Public License (version 3.0). Scripts reproducing the simulations analyses are available at https://github.com/bcm-uga/HDMAX2_Simulation_Scripts. A tutorial is provided as a supplemental file. The R package lfmm is publicly available from CRAN.
Results
Simulations
HDMAX2 was compared to several recent combinations of methods for multidimensional mediation analysis using simulation experiments. First, we compared the performances of latent factor models to other regression methods in estimating the association between exposure, DNAm levels, and outcome (Step 1 of HDMAX2). Then we compared the max2 mediation test to recently proposed tests (Step 2 of HDMAX2).
Performances of regression methods in Step 1 of HDMAX2.
A preliminary simulation study evaluated which of SVA, LFMM, or CATE provided the best estimation algorithm of latent factors for our empirical data set. CATE and LFMM obtained better performance scores than SVA (Figure S1; Excel Table S1). LFMM run times were shorter than those of CATE, and LFMM performance scores were higher. Thus, we concluded that LFMM is the most appropriate for analysis of the EDEN cohort data, and we used it everywhere in subsequent assessments of HDMAX2. Using more general simulation experiments (see above, “Mediation model simulations”), we measured the relative performances of HDMAX2, that jointly estimates effect sizes and latent factors with LFMM, and linear regressions adjusted for a priori estimates of cell-type composition with RefFreeEWAS and ReFACTor (Figure 2; Excel Table S2). In all scenarios, the performances of the ReFACTor method were much lower than those of LFMM and RefFreeEWAS (Figure 2; Excel Table S2). For lower effect sizes of DNAm on outcome, LFMM and RefFreeEWAS reached close F1-scores, but LFMM obtained higher scores than RefFreeEWAS for higher effect sizes. All approaches obtained higher scores when more mediators were simulated or when both the effect of exposure on DNAm and the effect of DNAm on outcome were higher. The results indicated that latent factor regression models outperformed methods that directly attempt to estimate cell-type composition from the DNAm data.
Figure 2. Relative performances of Step 1 approaches estimating latent factors vs. inclusion of cell-type composition estimates. F1-score as a function of the number of mediators (16 or 32), effect size of exposure on DNA methylation (DNAm) (X→M; low=0.2, high=0.4), and effect size of DNAm on outcome (M→Y; low=0.2, high=0.4). Each simulation included 38,000 CpGs for 500 samples, with 6 cell types and 3 additional confounding factors (Excel Table S2). Boxes represent 25th to 75th percentiles, the middle horizontal line is the median, whiskers extend to the most extreme point within 1.5 interquartile ranges of the box and the dots outside boxes indicate outliers.
Figure 2 is a set of four box and whiskers plot. On the left, the two plots are titled Low effect size (exposure greater than or equal to methylation), plotting uppercase f 1-score, ranging from 0.00 to 1.00 in increments of 0.25 (left y-axis) and Low effect size (exposure greater than or equal to methylation) and high effect size (exposure greater than or equal to methylation) (right y-axis) across latent factor mixed model, lowercase l m plus reffree epigenome-wide association studies, lowercase l m plus refactor (x-axis) for numbers of mediators, including 16 and 32. On the right, the two plots are titled high effect size (exposure greater than or equal to methylation), plotting uppercase f 1-score, ranging from 0.00 to 1.00 in increments of 0.25 (left y-axis) and Low effect size (exposure greater than or equal to methylation) and high effect size (exposure greater than or equal to methylation) (right y-axis) across latent factor mixed model, lowercase l m plus reffree epigenome-wide association studies, lowercase l m plus refactor (x-axis) for numbers of mediators, including 16 and 32.
Performances of mediation tests in Step 2 of HDMAX2.
Next, we compared HDMAX2 to five recent tests for high-dimensional mediation: HDMT, ScreenMin, SBMH, linear models combined with analysis of variance (ANOVA) (lm+anova), and HIMA (Figure 3; Excel Table S3). In every scenario, HDMAX2 and HDMT reached similar scores, and those approaches were the best ones overall. In the specific case of high DNAm on outcome effect sizes and low exposure on DNAm effect sizes, lm+anova obtained the best scores, immediately followed by HDMAX2 and HDMT. The lowest performances were obtained with ScreenMin, SBMH, and HIMA. When both effect sizes were high, HIMA obtained the lowest performances. For low DNAm on outcome effect sizes, lm+anova and SBMH obtained the poorest performances. In addition, HDMAX2 outperformed mediation analyses combining EWAS with Sobel tests and with unidimensional mediation analyses repeated at each marker, especially when the number of mediators increased from 16 to 32 (Figure S4; Excel Table S4). Because the run time was much shorter for HDMAX2 than for HDMT and for other approaches (Figure S5), HDMAX2 was used in our analyses on empirical data.
Figure 3. Relative performances of Step 2 multidimensional mediation methods. F1-score as a function of the number of mediators (16 or 32), effect size of exposure on DNA methylation (DNAm) (X→M; low=0.2, high=0.4), and effect size of DNAm on outcome (M→Y; low=0.2, high=0.4). Each simulation included 38,000 CpGs for 500 samples, with 6 cell types and 3 additional confounding factors (Excel Table S3). Boxes represent 25th to 75th percentiles, the middle horizontal line is the median, whiskers extend to the most extreme point within 1.5 interquartile ranges of the box, and the dots outside boxes indicate outliers.
Figure 3 is a set of four box and whiskers plot. On the left, the two plots are titled Low effect size (exposure greater than or equal to methylation), plotting uppercase f 1-score, ranging from 0.00 to 1.00 in increments of 0.25 (left y-axis) and Low effect size (exposure greater than or equal to methylation) and high effect size (exposure greater than or equal to methylation) (right y-axis) across high-dimensional mediation analysis, high-dimensional mediation hypotheses, a two-step familywise error rate procedure, S B M H, lowercase l m plus analysis of variance, H I M A (x-axis) for numbers of mediators, including 16 and 32. On the right, the two plots are titled high effect size (exposure greater than or equal to methylation), plotting uppercase f 1-score, ranging from 0.00 to 1.00 in increments of 0.25 (left y-axis) and Low effect size (exposure greater than or equal to methylation) and high effect size (exposure greater than or equal to methylation) (right y-axis) across high-dimensional mediation analysis, high-dimensional mediation hypotheses, a two-step familywise error rate procedure, S B M H, lowercase l m plus analysis of variance, H I M A (x-axis) for numbers of mediators, including 16 and 32.
Mediation of Prenatal Exposure to Smoking on Pregnancy Outcomes
Among 470 mother–infant pairs, mean maternal age at enrollment was 29 y (SD=5y), BMI before pregnancy was 23kg/m2 (SD=4.4 kg/m2) and 23.6% of women smoked during pregnancy (Table 1). Term birth weight ranged between 2,010g and 4,960g, with a mean of 3,352g±435g. GA varied from 37 wk to 42 wk, with a mean of 40 wk ±1.20 wk (8.4 d). MS during pregnancy had a significant correlation with birth weight (r=−0.16, p=0.003) but not with GA (Figure S6). Birth weight and GA were significantly correlated in mother–infant pairs (r=0.31, p=1.6×10−12). After adjustment, the total effect of MS was 140g lower birth weight (SD=49.1g, p=0.004), and the total effect of MS was not significant for GA (effect size=0.12 wk, SD=0.14 wk, p=0.2434).
Table 1 Characteristics of the included study population (n=470) from the EDEN mother–child cohort, France (Nancy, Poitiers), 2003–2006. Included population corresponds to mother–child dyads who had placental DNAm levels, maternal smoking status during pregnancy, and baby’s birth weight available.
Characteristics Mean±SE n (%)
Center
Poitiers — 189 (40)
Nancy — 281 (60)
Sex of offspring
Male — 241 (51)
Female — 229 (49)
Parity
0 — 189 (40)
1 — 195 (41)
≥2 — 86 (18)
Age of the mother at the end of education (y)
≤18 — 89 (20)
19–20 — 70 (15)
21–22 — 115 (24)
23–24 — 109 (23)
≥25 — 87 (18)
Season of conception
January–March — 100 (21)
April–June — 103 (22)
July–September — 130 (28)
October–December — 137 (29)
Maternal smoking
Smoker — 111 (24)
Nonsmoker — 359 (76)
BMI (kg/m²) 23.0±4.4 — —
Maternal age (y) 29.4±5.0 — —
Birth weight (g) 3,352±435 — —
Gestational duration (wk) 40±1.2 — —
Note: —, no data; BMI, prepregnancy body mass index; SE, standard error.
Mediation of MS on birth weight.
A high-dimensional mediation analysis of MS on birth weight was performed using placental DNAm data from the EDEN mother–child cohort. At an FDR level of 10% (5%), 32 (20) CpGs were identified as mediators of MS on birth weight (Figure 4A, adjusted max2 p<9.11×10−6; Excel Table S5). Twenty CpGs were associated with a lower birth weight for the newborn [average ACME: −32.0g, SD=5.6g; average proportion mediated (PM): 22.8%, SD=4.0], and 12 CpGs were associated with a higher birth weight (average ACME: 32.6g, SD=10.3g; average PM: 23.3%, SD=7.4) (Figure S7; Excel Table S5). The 32 CpGs were associated with an overall indirect effect corresponding to 40.3g lower birth weight (SD=51.3g).
Figure 4. HDMAX2 of maternal smoking on birth weight. (A) Manhattan plot for CpG’s −log (p-values) obtained from HDMAX2. Gene names correspond to hits identified at the 10% FDR level (32 hits) (Excel Table S5). Colored bars without gene names correspond to hits identified at the 20% FDR level (164 hits) (Excel Table S6). Gray bars without dots correspond to CpGs above the 20% FDR level. (B) Manhattan plot of −log10 (p-values) for potential AMR at the 10% FDR level (28 hits). (C) Estimates of indirect effect (ACME) and proportions of mediated effect for confirmed AMRs (19 hits) (Excel Table S7). The effect estimate is represented by a dot and its 95% CI by the bar. Symbols on top of colored bars correspond to classification as enhancer, promoter, or unknown. Overall indirect effect of AMRs: 52g lower birth weight. Results are adjusted for child sex; parity; maternal age at end of education; maternal body mass index before pregnancy; maternal age at delivery; season of conception; study center; batch, plate, and chip technical factors; and six latent factors. Note: ACME, average causal mediation effect; AMR, aggregated mediator region; CI, confidence interval; FDR, false discovery rate; HDMAX2, high-dimensional mediation analysis.
Figures 4A and 4B are graphs, plotting negative log of uppercase p, ranging from 0 to 8, in increments of 2 and 0 to 15 in increments of 5 (y-axis) across chromosome position (x-axis) for information, including enhancer, promoter, and unknown, and Average Causal Mediation Effect, including negative and positive. Figure 4C is a graph, plotting B L C A P, C C D C 137, S P 6, M L X, C O A S Y, E S R P 2, M G R N 1, L O C 100134368, W T 1, G A T A 3-A S 1, W D R 11, C E P 41, M Y L K 4, F B N 2, Z F P 42, P R I C K L E 2, L I N C 00885, I T P K B, and S K I (y-axis) across Average Causal Mediation Effect (grams), ranging from negative 60 to 60 in increments of 30 (x-axis). Figure 4D is a graph, plotting 20: 36148579 to 36149235; 17: 79634966 to 79635196; 17: 45949799 to 45950001; 17: 40718932 to 40719777; 17: 40713862 to 40715404; 16: 68269417 to 68270510; 16: 433561 to 433838; 11: 32355446 to 32355832; 10: 8094553 to 8094802; 10: 122708861 to 122709152; 07: 130080968 to 130081362; 06: 2697880 to 2698160; 05: 127872049 to 127872452; 04: 188916790 to 188917066; 03: 64211113 to 64211587; 03: 195849490 to 195849797; 01: 226926913 to 226927091; 01: 2171078 to 2171378 (y-axis) across proportion mediated (percentage), ranging from negative 2 to 1 in unit increments (x-axis).
Examples of CpG mediators with the largest negative indirect effects include cg10624729 (adjusted p=5.15×10−8), in MIGA1 (Mitoguardin 1), a regulator of mitochondrial fusion, associated with 41g lower birth weight; cg19406975 (adjusted p=9.27×10−8), in SH3BP5L (SH3 Binding Domain Protein 5 Like), which functions as a guanine exchange factor, associated with 41g lower birth weight; cg01686933 (adjusted p=6.98×10−7), in NECTIN1 (Nectin Cell Adhesion Molecule 1) that encodes an adhesion protein that plays a role in the organization of epithelial and endothelial cells, associated with 41g lower birth weight; and cg14502606 (adjusted p=1.04×10−6), in MLX (MAX Dimerization Protein MLX), a transcription factor that plays a role in proliferation, determination, and differentiation, associated with 38g lower birth weight (Excel Table S5).
At an FDR level <20%, 164 mediators were discovered, including 55 CpGs within enhancer regions and 26 CpGs within promoter regions (Figure 4A; Excel Table S6). In comparison with the methylome, the list of mediators was enriched in hits corresponding to enhancer regions (33% of all hits, p=0.0003, Fisher test; Figure S8A), and it was depleted in hits corresponding to promoter regions (15% of all hits, p=0.04, Fisher test; Figure S8B). Several mediators were found in the body of a gene (109 hits), and some genes were hit more than once (AJAP1, ESRP2, SH3BP2, SKI, SRSF5, VAV2, and MLX). We additionally performed mediation analyses for CpG cg27402634 (between LINC00086 and LEKR1) and cg25585967 (TRIO) identified in Morales et al.21 and for one CpG (cg11280108) in the HumanMethylation450 BeadChip, which was among the seven CpGs identified in Cardenas et al.20 from the EPIC chip. Although associations of DNAm with exposure to MS were significant for those CpGs (adjusted p=9.07×10−14), none of those markers were mediators of MS on birth weight in our analysis (significant at FDR >0.93).
Regarding methylated regions, HDMAX2 detected 28 potential AMRs, including 4 within enhancer regions, 7 within promoter regions, and 20 within the body of a gene (FDR level <10%; Figure 4B). Nineteen AMRs were associated with statistically significant indirect effects ranging between 26.7g lower birth weight and 33.0g higher birth weight (Excel Table S7). Twelve AMRs were associated with a lower birth weight (average ACME: −19.7g, SD=4.6; average PM: 14.0%, SD=3.3%), and seven were associated with a higher birth weight (average ACME: 17.5g, SD=7.9; average PM: 12.5%, SD=5.6%; Figure 4C). The 19 AMRs were associated with an overall indirect effect corresponding to 52g lower birth weight (SD=45g). The overall indirect effect of both CpG mediators and AMRs was 44.5g lower birth weight (SD=60.7g). The strongest evidence corresponded to AMR chr17:40,713,862-40,715,404 (adjusted p=3.20×10−13) in COASY (Coenzyme A Synthase), which plays an important role in numerous synthetic and degradative metabolic pathways in all organisms, associated with 26g lower birth weight. This AMR was only 3 kb close to another AMR, chr17:40,718,932-40,719,777 (adjusted p=9.37×10−19, in MLX, which was associated with 27g lower birth weight (Figure 4; see Excel Table S7 for a full list of AMRs).
Mediation of MS on GA.
An independent mediation analysis was performed on the DNAm data to evaluate the indirect effects of MS on GA. At an FDR level <10%, 15 CpGs (2 CpGs at FDR level <5%) were identified as mediators of MS on GA (Figure 5A; adjusted max2 p<3.28×10−6; Excel Table S8). The 15 CpGs were associated with a weak overall indirect effect corresponding to 0.28 wk (2 d) lower GA (SD=0.12) (Figure S9; Excel Table S8).
Figure 5. HDMAX2 of maternal smoking on gestational age. (A) Manhattan plot for CpG’s −log (p-values) obtained from HDMAX2. Gene names correspond to hits identified at the 10% FDR level (15 hits) (Excel Table S8). Colored bars without gene names correspond to hits identified at the 20% FDR level (63 hits) (Excel Table S9). Gray bars without dots correspond to CpGs above the 20% FDR level. (B) Manhattan plot of −log10 (p-values) for potential AMR at 10% FDR level (31 hits). (C) Estimates of indirect effects (ACME) and proportions of mediated effect for confirmed AMRs (23 hits) (Excel Table S10). The effect estimate is represented by a dot, and its 95% CI by the bar. Symbols on top of colored bars correspond to classification as enhancer, promoter, or unknown. Overall indirect effect of AMRs: 0.12 wk lower gestational age. Results are adjusted for child sex; parity; maternal age at end of education; maternal body mass index before pregnancy; maternal age at delivery; season of conception; study center; batch, plate, and chip technical factors; and six latent factors. Note: ACME, average causal mediation effect; AMR, aggregated mediator region; CI, confidence interval; FDR, false discovery rate; HDMAX2, high-dimensional mediation analysis.
Figures 5A and 5B are graphs, plotting negative log of uppercase p, ranging from 0 to 6, in increments of 2 and 0 to 20 in increments of 5 (y-axis) across chromosome position (x-axis) for information, including enhancer, promoter, and unknown, and Average Causal Mediation Effect, including negative and positive. Figure 5C is a graph, plotting F A M 207 A, B L C A P, S C A M P 4, K L F 16, I G F 2 B P 1, M A P 3 K 14-A S 1, C O A S Y, R A P G E F L 1, E S R P 2, P R S S 41, Z F P 36 L 1, T M E M 134, P O N 1, B 3 G A L T 4, P R R T 1, P P P 1 R 10, F O X C 1, C D X 1, P C D H G A 1, F O X L 2, E M I L I N 1, S N H G 12 and S K I (y-axis) across Average Causal Mediation Effect (grams), ranging from negative 0.1 to 0.1 in unit increments (x-axis). Figure 5D is a graph, plotting 21: 46378365 to 46378748, 20: 36148579 to 36149354, 19: 1908254 to 1908511, 19: 1852004 to 1852288, 17: 47091461 to 47092395, 17: 43339450 to 43339954, 17: 40714100 to 40714374, 17: 38334055 to 38334562, 16: 68270251 to 68270510, 16: 2848919 to 2849267, 14: 69256799 to 69257100, 11: 67232391 to 67232634, 07: 95025733 to 95026625, 06: 33245303 to 33245927, 06: 32116086 to 32117211, 06: 30582296 to 30582539, 06: 1608518 to 1608790, 05: 149546195 to 149547069, 05: 140723577 to 140723807, 03: 138658365 to 138658677, 02: 27308999 to 27309276, 01: 28906332 to 28906661, 01: 2171078 to 2171376 (y-axis)across proportion mediated (percentage), ranging from negative 4 to 1 in increments of 4 (x-axis).
Examples of CpG mediators with the most negative effects included cg10298741 (adjusted p=4.82×10−7), in ZFHX3 (Zinc Finger Homeobox 3), a transcription factor that regulates myogenic and neuronal differentiation, associated with 0.08 wk lower GA; cg04908961 (adjusted p=9.19×10−7), in MIR17HG (MiR-17-92a-1 Cluster Host Gene), a host gene for the MiR17-92 cluster, a group microRNAs (miRNAs) that may be involved in cell survival, proliferation, and differentiation, associated with 0.09 wk lower GA; and cg08402058 (adjusted p=1.04×10−6), in BLCAP (Bladder cancer–associated protein) which reduces cell growth by stimulating apoptosis, associated with 0.09 wk lower GA (see Table S8 for a full list of CpG mediators). Ten CpGs were associated with a shorter GA (average indirect effect 0.09 wk lower GA, SD=0.02; PM: 74%, SD=14%), and 5 CpGs were associated with higher GA (average ACME: 0.09 wk, SD=0.01; PM: 71%, SD=10%; Figure S9; Excel Table S8). At an FDR level <20%, 63 mediators were identified, including 26 hits within an enhancer region (Figure 5A; Excel Table S9). This subset of CpG mediators was enriched in hits corresponding to enhancer regions (33% of all hits, p<2.2×10−16, Fisher test; Figure S8A).
The per-region analysis resulted in the detection of 31 potential AMRs, including 11 regions within enhancers, 5 within promoters, and 26 within the body of a gene (Figure 5B). Twenty-three AMRs were associated with small but statistically significant indirect effects ranging between −0.09 wk and 0.10 wk (none were associated with a significant mediated proportion) (Excel Table S10). Five regions were associated with a lower GA (average ACME: −0.06 week; SD=0.01; average PM: 54.1%, SD=13.1%), and 18 regions were associated with a higher GA (average ACME: 0.06 wk; SD=0.01; average PM: 49.4%, SD=11.1%).
The 23 AMRs were associated with a weak overall indirect effect corresponding to 0.12 wk (23 h) shorter GA (SD=0.11). The cumulative overall indirect effect of CpG mediators and AMRs was 0.09 wk (15 h) shorter GA (SD=0.14). The largest negative indirect effects corresponded to AMR chr1:28,906,332-28,906,661 (adjusted p=7.89×10−9) in SNHG12 (Small Nucleolar RNA Host Gene 12), an RNA gene that may promote tumorigenesis, associated with 0.09 wk lower GA; chr20:36,148,579-36,149,354 (adjusted p=1.13×10−10) in BLCAP, which encodes a protein that reduces cell growth by stimulating apoptosis, associated with 0.06 wk lower GA; and chr17:40,714,100-40,714,374 (adjusted p=2.84×10−6) in COASY, associated with 0.06 wk lower (Figure 5; see Excel Table S10 for a full list of AMRs).
Chained mediation of MS on birth weight through DNAm and GA.
Six genes, COASY, BLCAP, SKI, DECR1, ESRP2, and PRRT1, included AMRs that act as mediators both for MS on birth weight and for MS on GA. To better understand the causal pathways involving those genic regions, we tested the hypothesis that GA influences methylation levels in those regions and estimated the indirect effects in a mediation analysis of GA on birth weight (Figure 6; Figure S10; Table S1). In this analysis, GA had significant indirect effects on birth weight for two of the six AMRs, in COASY (ACME=6.9g, mediation p<10−3), BLCAP (ACME=5.1g, mediation p=0.01). The two AMRs were associated with an overall indirect effect corresponding to 10g higher birth weight (SD=3.91). We found no evidence that one of these six AMRs was present in the pathway from birth weight to GA.
Figure 6. Summary of mediation analysis of MS on gestational age GA and birth weight for AMRs. Nineteen AMRs mediate the relationship between MS and birth weight, with a total indirect effect corresponding to 52g lower birth weight. Twenty-three AMRs mediate the relationship between MS and GA, with a total indirect effect corresponding to 0.12 wk shorter GA. Two AMRs mediate the relationship between GA and birth weight with a total indirect effect corresponding to 10g higher birth weight. The color of each gene’s box indicates the sign of the indirect effect. The color of segments indicates the direction of association (blue full line means positive, red dotted lines means negative). Results are adjusted for child sex; parity; maternal age at end of education; maternal body mass index before pregnancy; maternal age at delivery; season of conception; study center; batch, plate, and chip technical factors; and six latent factors. Note: AMR, aggregated mediator region; BW, birth weight; GA, gestational age; MS, maternal smoking.
Figure 6 is an illustration depicting the synopsis of the mediation analysis of maternal smoking on gestational age and birth weight targeting aggregated mediator regions. The dotted line represent negative and the straight line represent positive. Between S P 6 and G AT A3-AS 1, the total indirect effect corresponds to a negative 52 gramme lower birth weight. Between the S C A M P 4 and the P R S S 41, the total indirect effect corresponding to 0.12 weeks is shorter on gestational age. The two aggregated mediator regions mediate the relationship between gestational age and birth weight, with a total indirect effect corresponding to a 10 grams higher birth weight.
In the genomic region surrounding the COASY gene (Figure S10), the AMRs were located in regions with low DNAm levels (Figure S11B), and MS decreased DNAm levels within AMRs (Figure S11C). The CpGs contained in AMRs mediated lower birth weight and were among the most negative observed indirect effects (Figure S11D–E). In the genomic region surrounding the BLCAP gene (Figure S12), AMRs were located in highly methylated gene body areas (Figure S12B), and MS decreased DNAm levels within AMRs (Figure S12C). The CpGs contained in AMRs mediated lower birth weight, and again, they were among the most negative observed indirect effects (Figure S12D–E). Figure 6 provides a summary of the chained mediation analysis (Figure S13 for a summary of CpG mediation analysis).
Discussion
Main Contributions
High-dimensional mediation analysis holds promising results for deciphering molecular mechanisms underlying the association between exposure and outcomes. We presented HDMAX2, a method combining estimates of latent factors in EWAS with max2 tests for mediation, which also evaluates an overall mediated effect for CpG or AMR. Using simulations, we performed an in-depth evaluation of the statistical performances of HDMAX2 and showed that HDMAX2 outperforms state-of-art methods and recent approaches proposed to identify mediators in a high-dimensional setting. HDMAX2 was applied to assess the indirect effects of exposure to MS on GA and birth weight in a study of 470 women from the EDEN mother–child cohort and confirmed the important role played by placental DNAm in the pathway between MS during pregnancy and fetal growth outcomes.3 In addition to single CpG mediators, our analysis examined AMRs and computed an overall indirect effect of all mediators considered simultaneously. The posterior means of the overall indirect effect of CpG and AMR were 44.5g lower birth weight (SD=60.7g, 32.1% of the total effect size) and 0.09 wk lower GA (SD=0.14 wk, 75% of the total effect size). With respect to the results based on single mediators on birth weight, the standard deviation estimate from the posterior distribution can be interpreted as mediation of smoking on lower birth weight in about 77% of cases and as mediation of smoking on higher birth weight in about 23% of cases (a similar interpretation holds for GA as well). These results support the hypothesis that the role of placental DNAm in the mediation of effect of exposure to MS on birth weight and on GA may be more polygenic than previously reported. In addition, a chained mediation analysis of MS on birth weight suggested the existence of reverse causal relationships for AMR located in the genes COASY and BLCAP, which mediate a proportion of the effect of MS on birth weight through an effect of GA on DNAm.
Simulation Studies
The main improvements of HDMAX2 over existing mediation methods is the use of latent factor models for estimating hidden confounders in Step 1, and the max2 test of mediation in Step 2. The combination of latent factors and max2 tests proposed by the HDMAX2 approach was carefully evaluated with intensive simulations and resulted in increased performances in comparison with five state-of-the-art methods evaluating multiple mediators.10,11,14,15,33 Latent factors increased statistical power in comparison with using a priori estimates of cell-type proportions from reference-free methods.29,30 The max2 tests showed considerably better performances in comparison to the univariate mediation or Sobel test approaches, which were used in previous studies analyzing the role of placental DNAm data in the pathway between MS and birth weight.20,21 Using HDMAX2, none of the mediating CpGs identified using univariate mediation or Sobel test approaches20,21 were mediators of MS on birth weight in our analysis (significant at FDR >0.93).
Mediation Analysis of Maternal Smoking on Birth Weight
Previous studies have shown a possibly overestimated mediated effect of MS on birth weight, sometimes greater than the total effect size.45 This overestimation is a limitation of univariate indirect effects estimated independently in a context of correlation between multiple mediators. In contrast, our approach estimated an overall indirect effect of the placental methylome representing 32% of the total effect size of MS on birth weight. In comparison with previous placental DNAm mediation analyses of MS on birth weight,20,21 the magnitude of each mediator indirect effect size estimated in our cohort represented smaller part (<24% for AMRs) of the total effect size, and it was spread over more mediators, suggesting that indirect effects are more polygenic than in previous estimates. With a Bayesian interpretation of the bootstrap distribution,32 the OIE estimates for CpGs, AMRs, and CpGs + AMRs, representing 40.3g, 52g, 44.5g lower birth weight, respectively, correspond to the mean of the posterior distribution. Estimates of standard error from the bootstrap distribution, ranging from 45g to 60.7g, indicate probabilities that DNAm could mediate higher birth weight ranging between 12%–23% for any mother–child pair in the EDEN cohort.
CpG Mediators
HDMAX2 identified 32 CpG mediators of MS on birth weight, for which a majority (20/32) of effects represented a lower birth weight. The results provided evidence for an enrichment in enhancer regions and for a depletion in promoter regions among mediators, which agrees with conclusions from an association study between MS and placental DNAm in the EDEN cohort.22 According to the Gene Ontology database,44 six mediators were located in genes linked to development or to the growth of tissues: cg24571086 in FGFR2, cg11362604 in MEIS2, cg00108098 in SEMA5B, cg10778780 in CCK, cg20482145 in MYH10, and cg07156115 in AHR. The genes FGFR2 and SEMA5B are linked to the development of multicellular organisms and to the growth of developmental organs, MEIS2 is linked to the development of the brain, eyes, and pancreas, CCK is linked to neuron migration, and AHR is linked to the development of blood vessels.
AMRs
Evidence for increased polygenicity of placental DNAm mediation was confirmed by examination of AMRs, which are seen as more robust and more biologically meaningful than isolated differentially methylated CpGs.46 HDMAX2 identified 19 AMRs of MS on birth weight, for which a majority of effects represented a lower birth weight (Figure 4C; Excel Table S7). The most negative effects corresponded to AMRs in COASY, which plays an important role in numerous synthetic and degradative metabolic pathways, and in MLX, a transcription factor physically close to COASY, which is coexpressed in the placenta.47 Four regions were located in genes linked to tissue development or growth, in FBN2 related to camera-type eye development; ZFP42 to gonad development; ESRP2 to fibroblast growth factor receptor signaling pathway; and SKI to roof of mouth development, olfactory bulb development, camera-type eye development, and skeletal muscle fibber development.47 The genes FBN2 and ZFP42 were overexpressed in the placenta in comparison with other tissues. Smoking-induced AMRs in FBN2 and ESRP2 were associated with higher birth weight, whereas AMRs in ZFP42 and SKI were associated with lower birth weight. Looking more closely at the biology of mediators, we found a large number of them located in genes related to preeclampsia, a pregnancy complication of placental origin characterized by high blood pressure and protein in the urine, causing about a third of very premature births. Preeclampsia-related genes included NECTIN1,48 AHR,49 FGFR2,50 COASY,51 BLCAP,52 SKI,51 AJAP1,53 and SH3BP5.54 The overrepresentation of preeclampsia-related genes supports a pleiotropic effect of mediators and highlights the difficulty of disentangling relationships between correlated outcomes.
Mediation Analysis of MA on GA and Potential for Reverse Causality
Our results provided evidence that DNAm (CpG + AMR) mediates a very small total indirect effect of MS on GA, representing 0.09 wk lower GA (15 h). The largest negative effects corresponded to AMRs located in SNHG12 and in BLCAP (Excel Table S10). The effect sizes observed for GA have a low clinical relevance. An interesting finding was that six genes contained DMRs mediating both the effect of MS on birth weight and the effect of MS on GA. Two of those AMRs, located in BLCAP and COASY, had among the largest negative effects on both GA and birth weight. We reported strong evidence that BLCAP and COASY were present in the pathway from GA to birth weight but no evidence that they were present in the pathway from birth weight to GA. This result indicates that the corresponding AMRs in BLCAP and COASY may be involved in complex causal relationships in which DNAm plays a role in the negative effect of MS on birth weight (and on GA), which is amplified by a lower GA (Figure 6). Knowing whether placental DNAm influences GA or GA influences placental DNAm remains an open question. A limitation to interpretation is the fact that GA and placental DNAm are co-occurring events. However, our results suggest a bidirectional association between placental DNAm and GA, with a feedback loop from GA to birth weight through placental DNAm.
Universally Applicable Framework for High-Dimensional Mediating Events
A large body of epigenetic research in perinatal health is dedicated to cord blood DNA methylation, although the placenta has attracted recent attention.20,21,55 The placenta exhibits a unique epigenetic profile because it is one of the tissues with lower DNA methylation levels that undergoes intense remodeling in early gestation and dynamic changes with increased DNA methylation as gestation advances.56,57 The placenta supports both the health of the mother and the development of the fetus: it produces hormones, ensures immune tolerance, provides nutrients to the fetus, and regulates the exchange of gases and wastes. The placenta contains key information on the intrauterine environment and is a highly relevant tissue to investigate within the DOHaD framework. Besides being associated with several prenatal exposures, placental DNA methylation is suggested to be a relevant proxy for neurodevelopmental outcomes58–60 and respiratory health61 of the child. Understanding the indirect effects of placental DNAm modifications on such outcomes will be an important objective, for which the HDMAX2 framework will be very helpful. Beyond the role of the placenta and DNA methylation, other tissues and omics markers are relevant to investigate in perinatal and more generally epidemiological studies. The HDMAX2 framework can be applied with other layers of mediators, basically any type of high-throughput data (i.e., gene expression data) or with data on any other tissue types.
Summary
We developed a novel algorithm for high-dimensional mediation, HDMAX2. Beyond our current application to placental DNAm data, HDMAX2 is applicable to a wide range of tissues and omic layers, including genomics, transcriptomics, and other types of omics. HDMAX2 showed better performances on simulations and increased power in comparison with existing approaches. We showed the strength of HDMAX2 by applying it to characterize associations between exposure to MS during pregnancy and birth weight and GA at birth of the baby. The mediation analysis suggested a causal relationship between MA during pregnancy and those outcomes underpinning many more epigenetic regions than previously found, suggesting a polygenic architecture for the pathways. Not limited to single CpG markers, HDMAX2 is extended to identifying AMRs. AMRs provided more robust evidence than single CpGs and allowed the characterization of regions mediating effects of MS during pregnancy both on GA and birth weight, suggesting that placental DNAm is an important biological mechanism. We further showed the overall indirect effect accounting simultaneously for all mediators identified as a plausible estimate of the mediated effect. AMRs located in COASY and BLCAP suggested reverse causality in the relationship between gestational and the methylome contributing to lower birth weight. Our study added several statistical improvements to high-dimensional mediation analyses and revealed an unsuspected complexity of the causal relationships between MS during pregnancy and birth weight at the epigenome-wide level. Limitations of the current work and thus future research avenues include a better characterization of interactions and of the polygenic architecture of phenotypes, especially when there is a high number of markers with small effect sizes, which will require much larger sample sizes.62
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Click here for additional data file.
Acknowledgments
The authors thank D. Vaiman (Inserm U1016) for his help with lab experiments. Tha authors also thank all the participants and members of the EDEN mother–child cohort study group.
This work was supported by a grant from the French National Cancer Institute (INCa), the French Institute for Public Health Research (IreSP) (INCa_13641), and the French Agency for National Research (ETAPE, ANR-18-CE36-0005). B.J. was partly supported by the Grenoble Alpes Data Institute, supported by the French National Research Agency under the Investissements d’Avenir program (ANR-15-IDEX-02), and by LabEx PERSYVAL Lab, ANR-11-LABX-0025-01. DNA methylation measurements were obtained thanks to grants from the Fondation de France (No. 2012-00031593 and 2012-00031617) and the French Agency for National Research (ANR-13-CESA-0011).
The EDEN mother–child study was supported by Foundation for Medical Research (FRM), National Agency for Research (ANR), National Institute for Research in Public health (IRESP), French Ministry of Health (DGS), French Ministry of Research, Inserm Bone and Joint Diseases National Research (PRO-A), and Human Nutrition National Research Programs, Nestlé, French National Institute for Population Health Surveillance (InVS), French National Institute for Health Education (INPES), the European Union FP7 programs (FP7/2007-2013, HELIX, ESCAPE, ENRIECO, Medall projects), Diabetes National Research Program, French Agency for Environmental Health Safety (ANSES), Mutuelle Générale de l’Education Nationale (MGEN), French National Agency for Food Security, and the French-speaking Association for the Study of Diabetes and Metabolism (ALFEDIAM).
B.J., C.C.B., and M.E. performed the statistical analyses. J.L. and O.F. designed the study, wrote the manuscript, and obtained funding. O.F. developed the statistical analysis plan. B.H. supervised the data collection and data management of the EDEN cohort and provided guidance on the project. J.T. supervised the methylation and lab arrays. All authors read, revised, and approved the final manuscript.
The EDEN data sets analyzed in the presented study are not publicly available because they contain information that could compromise the research participants’ privacy/consent. However, they are available from the corresponding author on reasonable request and with permission from the EDEN Steering Committee.
==== Refs
References
1. Baron RM, Kenny DA. 1986. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Pers Soc Psychol 51 (6 ):1173–1182, PMID: , 10.1037//0022-3514.51.6.1173.3806354
2. Blum MGB, Valeri L, François O, Cadiou S, Siroux V, Lepeule J, et al. 2020. Challenges raised by mediation analysis in a high-dimension setting. Environ Health Perspect 128 (5 ):55001, PMID: , 10.1289/EHP6240.32379489
3. Nakamura A, François O, Lepeule J. 2021. Epigenetic alterations of maternal tobacco smoking during pregnancy: a narrative review. Int J Environ Res Public Health 18 (10 ):5083, PMID: , 10.3390/ijerph18105083.34064931
4. MacKinnon DP, Fairchild AJ, Fritz MS. 2007. Mediation analysis. Annu Rev Psychol 58 (1 ):593–614, PMID: , 10.1146/annurev.psych.58.110405.085542.16968208
5. VanderWeele T. 2015. Explanation in Causal Inference: Methods for Mediation and Interaction. Oxford, UK: Oxford University Press.
6. VanderWeele TJ. 2016. Mediation analysis: a practitioner’s guide. Annu Rev Public Health 37 (1 ):17–32, PMID: , 10.1186/s40985-016-0032-5, 10.1146/annurev-publhealth-032315-021402.26653405
7. Zeng P, Shao Z, Zhou X. 2021. Statistical methods for mediation analysis in the era of high-throughput genomics: current successes and future challenges. Comput Struct Biotechnol J 19 :3209–3224, PMID: , 10.1016/j.csbj.2021.05.042.34141140
8. Imai K, Keele L, Tingley D. 2010. A general approach to causal mediation analysis. Psychol Methods 15 (4 ):309–334, PMID: , 10.1037/a0020761.20954780
9. Sobel ME. 1982. Asymptotic confidence intervals for indirect effects in structural equation models. Sociol Methodol 13 :290, 10.2307/270723.
10. Dai JY, Stanford JL, LeBlanc M. 2022. A multiple-testing procedure for high-dimensional mediation hypotheses. J Am Stat Assoc 117 (537 ):198–213, PMID: , 10.1080/01621459.2020.1765785.35400115
11. Djordjilović V, Hemerik J, Thoresen M. 2020. On optimal two-stage testing of multiple mediators. Biom J 64 (6 ):1090–1108, PMID: , 10.1002/bimj.202100190.35426161
12. Djordjilović V, Page CM, Gran JM, Nøst TH, Sandanger TM, Veierød MB, et al. 2019. Global test for high‐dimensional mediation: testing groups of potential mediators. Stat Med 38 (18 ):3346–3360, PMID: , 10.1002/sim.8199.31074092
13. Gao Y, Yang H, Fang R, Zhang Y, Goode EL, Cui Y. 2019. Testing mediation effects in high-dimensional epigenetic studies. Front Genet 10 :119, PMID: , 10.3389/fgene.2019.01195.30858864
14. Sampson JN, Boca SM, Moore SC, Heller R. 2018. FWER and FDR control when testing multiple mediators. Bioinformatics 34 (14 ):2418–2424, PMID: , 10.1093/bioinformatics/bty064.29420693
15. Zhang H, Zheng Y, Zhang Z, Gao T, Joyce B, Yoon G, et al. 2016. Estimating and testing high-dimensional mediation effects in epigenetic studies. Bioinformatics 32 (20 ):3150–3154, PMID: , 10.1093/bioinformatics/btw351.27357171
16. Agha G, Hajj H, Rifas-Shiman SL, Just AC, Hivert MF, Burris HH, et al. 2016. Birth weight-for-gestational age is associated with DNA methylation at birth and in childhood. Clin Epigenetics 8 :118, PMID: , 10.1186/s13148-016-0285-3.27891191
17. Joubert BR, Håberg SE, Nilsen RM, Wang X, Vollset SE, Murphy SK, et al. 2012. 450K Epigenome-Wide scan identifies differential DNA methylation in newborns related to maternal smoking during pregnancy. Environ Health Perspect 120 (10 ):1425–1431, PMID: , 10.1289/ehp.1205412.22851337
18. Küpers LK, Xu X, Jankipersadsing SA, Vaez A, la Bastide-van Gemert S, Scholtens S, et al. 2015. DNA methylation mediates the effect of maternal smoking during pregnancy on birthweight of the offspring. Int J Epidemiol 44 (4 ):1224–1237, PMID: , 10.1093/ije/dyv048.25862628
19. Xu R, Hong X, Zhang B, Huang W, Hou W, Wang G, et al. 2021. DNA methylation mediates the effect of maternal smoking on offspring birthweight: a birth cohort study of multi-ethnic US mother–newborn pairs. Clin Epigenetics 13 (1 ):47, PMID: , 10.1186/s13148-021-01032-6.33663600
20. Cardenas A, Lutz SM, Everson TM, Perron P, Bouchard L, Hivert MF. 2019. Placental DNA methylation mediates the association of prenatal maternal smoking on birth weight. Am J Epidemiol 188 (11 ):1878–1886, PMID: , 10.1093/aje/kwz184.31497855
21. Morales E, Vilahur N, Salas LA, Motta V, Fernandez MF, Murcia M, et al. 2016. Genome-wide DNA methylation study in human placenta identifies novel loci associated with maternal smoking during pregnancy. Int J Epidemiol 45 (5 ):1644–1655, PMID: , 10.1093/ije/dyw196.27591263
22. Rousseaux S, Seyve E, Chuffart F, Bourova-Flin E, Benmerad M, Charles MA, et al. 2019. Immediate and durable effects of maternal tobacco consumption alter placental DNA methylation in enhancer and imprinted gene-containing regions. BMC Med 18 (1 ):306, PMID: , 10.1186/s12916-020-01736-1.33023569
23. Thornburg KL, Marshall N. 2015. The placenta is the center of the chronic disease universe. Am J Obstet Gynecol 213 (suppl 4 ):S14–20, PMID: , 10.1016/j.ajog.2015.08.030.26428494
24. Caye K, Jumentier B, Lepeule J, François O. 2019. LFMM 2: fast and accurate inference of gene-environment associations in genome-wide studies. Mol Biol Evol 36 (4 ):852–860, PMID: , 10.1093/molbev/msz008.30657943
25. Jumentier B, Caye K, Heude B, Lepeule J, François O. 2022. Sparse latent factor regression models for genome-wide and epigenome-wide association studies. Stat Appl Genet Mol Biol 21 (1 ), PMID: , 10.1515/sagmb-2021-0035.35245419
26. Leek JT, Storey JD. 2007. Capturing heterogeneity in gene expression studies by surrogate variable analysis. PLoS Genet 3 (9 ):1724–1735, PMID: , 10.1371/journal.pgen.0030161.17907809
27. Lee S, Sun W, Wright FA, Zou F. 2017. An improved and explicit surrogate variable analysis procedure by coefficient adjustment. Biometrika 104 (2 ):303–316, PMID: , 10.1093/biomet/asx018.29430031
28. Wang J, Zhao Q, Hastie T, Owen AB. 2017. Confounder adjustment in multiple hypothesis testing. Ann Stat 45 (5 ):1863–1894, PMID: .31439967
29. Houseman EA, Kile ML, Christiani DC, Ince TA, Kelsey KT, Marsit CJ. 2016. Reference-free deconvolution of DNA methylation data and mediation by cell composition effects. BMC Bioinformatics 17 :259, PMID: , 10.1186/s12859-016-1140-4.27358049
30. Rahmani E, Zaitlen N, Baran Y, Eng C, Hu D, Galanter J, et al. 2016. Sparse PCA corrects for cell type heterogeneity in epigenome-wide association studies. Nat Methods 13 (5 ):443–445, PMID: , 10.1038/nmeth.3809.27018579
31. Xu Z, Niu L, Li L, Taylor JA. 2016. ENmix: a novel background correction method for Illumina HumanMethylation450 BeadChip. Nucleic Acids Res 44 (3 ):e20, PMID: , 10.1093/nar/gkv907.26384415
32. Hastie T, Tibshirani R, Friedman J. 2009. Model Inference and Averaging. In: The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Hastie T, Tibshirani R, Friedman J, eds. New York, NY: Springer, 261–294.
33. Tobi EW, Slieker RC, Luijk R, Dekkers KF, Stein AD, Xu KM, et al. 2018. DNA methylation as a mediator of the association between prenatal adversity and risk factors for metabolic disease in adulthood. Sci Adv 4 (1 ):eaao4364, PMID: , 10.1126/sciadv.aao4364.29399631
34. Benjamini Y, Hochberg Y. 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B Methodol 57 (1 ):289–300, 10.1111/j.2517-6161.1995.tb02031.x.
35. Abraham E, Rousseaux S, Agier L, Giorgis-Allemand L, Tost J, Galineau J, et al. 2018. Pregnancy exposure to atmospheric pollution and meteorological conditions and placental DNA methylation. Environ Int 118 :334–347, PMID: , 10.1016/j.envint.2018.05.007.29935799
36. Heude B, Forhan A, Slama R, Douhaud L, Bedel S, Saurel-Cubizolles MJ, et al. 2016. Cohort profile: the EDEN mother-child cohort on the prenatal and early postnatal determinants of child health and development. Int J Epidemiol 45 (2 ):353–363, PMID: , 10.1093/ije/dyv151.26283636
37. Jedynak P, Tost J, Calafat AM, Bourova-Flin E, Busato F, Forhan A, et al. 2021. Pregnancy exposure to synthetic phenols and placental DNA methylation – an epigenome-wide association study in male infants from the EDEN cohort. Environ Pollut 290 :118024, PMID: , 10.1016/j.envpol.2021.118024.34523531
38. Teschendorff AE, Marabita F, Lechner M, Bartlett T, Tegner J, Gomez-Cabrero D, et al. 2013. A beta-mixture quantile normalization method for correcting probe design bias in Illumina Infinium 450 k DNA methylation data. Bioinformatics 29 (2 ):189–196, PMID: , 10.1093/bioinformatics/bts680.23175756
39. Efron B. 2004. Large-scale simultaneous hypothesis testing: the choice of a null hypothesis. J Am Stat Assoc 99 (465 ):96–104, 10.1198/016214504000000089.
40. Strimmer K. 2008. A unified approach to false discovery rate estimation. BMC Bioinformatics 9 (1 ):303, PMID: , 10.1186/1471-2105-9-303.18613966
41. Triche TJ. 2014. FDb.InfiniumMethylation.hg19: annotation package for Illumina Infinium DNA methylation probes. R Package Version 2.2.0.
42. Papatheodorou I, Moreno P, Manning J, Fuentes AMP, George N, Fexova S, et al. 2020. Expression Atlas update: from tissues to single cells. Nucleic Acids Res 48 (D1 ):D77–D83, PMID: , 10.1093/nar/gkz947.31665515
43. Kanehisa M, Furumichi M, Sato Y, Ishiguro-Watanabe M, Tanabe M. 2021. KEGG: integrating viruses and cellular organisms. Nucleic Acids Res 49 (D1 ):D545–D551, PMID: , 10.1093/nar/gkaa970.33125081
44. Gene Ontology Consortium. 2021. The gene ontology resource: enriching a GOld mine. Nucleic Acids Res 49 (D1 ):D325–D334, PMID: , 10.1093/nar/gkaa1113.33290552
45. Valeri L, Reese SL, Zhao S, Page CM, Nystad W, Coull BA, et al. 2017. Misclassified exposure in epigenetic mediation analyses. Does DNA methylation mediate effects of smoking on birthweight? Epigenomics 9 (3 ):253–265, PMID: .28234025
46. Svendsen AJ, Gervin K, Lyle R, Christiansen L, Kyvik K, Junker P, et al. 2016. Differentially methylated DNA regions in monozygotic twin pairs discordant for rheumatoid arthritis: an epigenome-wide study. Front Immunol 7 :510, PMID: , 10.3389/fimmu.2016.00510.27909437
47. National Library of Medicine, National Center for Biotechnology Information. Gene. https://www.ncbi.nlm.nih.gov/gene/6945.
48. Ito M, Nishizawa H, Tsutsumi M, Kato A, Sakabe Y, Noda Y, et al. 2018. Potential role for nectin-4 in the pathogenesis of pre-eclampsia: a molecular genetic study. BMC Med Genet 19 (1 ):166, PMID: , 10.1186/s12881-018-0681-y.30217189
49. Wang K, Zhou Q, He Q, Tong G, Zhao Z, Duan T. 2011. The possible role of AhR in the protective effects of cigarette smoke on preeclampsia. Med Hypotheses 77 (5 ):872–874, PMID: , 10.1016/j.mehy.2011.07.061.21864991
50. Marwa BAG, Raguema N, Zitouni H, Feten HBA, Olfa K, Elfeleh R, et al. 2016. FGF1 and FGF2 mutations in preeclampsia and related features. Placenta 43 :81–85, PMID: , 10.1016/j.placenta.2016.05.007.27324104
51. Martin E, Ray PD, Smeester L, Grace MR, Boggess K, Fry RC. 2015. Epigenetics and preeclampsia: defining functional epimutations in the preeclamptic placenta related to the TGF-β pathway. PLoS One 10 (10 ):e0141294, PMID: , 10.1371/journal.pone.0141294.26510177
52. Li Y, Cui S, Shi W, Yang B, Yuan Y, Yan S, et al. 2020. Differential placental methylation in preeclampsia, preterm and term pregnancies. Placenta 93 :56–63, PMID: , 10.1016/j.placenta.2020.02.009.32250740
53. Yeung KR, Chiu CL, Pidsley R, Makris A, Hennessy A, Lind JM. 2016. DNA methylation profiles in preeclampsia and healthy control placentas. Am J Physiol Heart Circ Physiol 310 (10 ):H1295–H1303, PMID: , 10.1152/ajpheart.00958.2015.26968548
54. Kaartokallio T, Cervera A, Kyllönen A, Laivuori K, Kere J, Laivuori H, et al. 2015. Gene expression profiling of pre-eclamptic placentae by RNA sequencing. Sci Rep 5 :14107, PMID: , 10.1038/srep14107.26388242
55. Everson TM, Vives-Usano M, Seyve E, Cardenas A, Lacasaña M, Craig JM, et al. 2021. Placental DNA methylation signatures of maternal smoking during pregnancy and potential impacts on fetal growth. Nat Commun 12 (1 ):5095, PMID: , 10.1038/s41467-021-24558-y.34429407
56. Fuke C, Shimabukuro M, Petronis A, Sugimoto J, Oda T, Miura K, et al. 2004. Age related changes in 5-methylcytosine content in human peripheral leukocytes and placentas: an HPLC-based study. Ann Hum Genet 68 (pt 3 ):196–204, PMID: , 10.1046/j.1529-8817.2004.00081.x.15180700
57. Novakovic B, Yuen RK, Gordon L, Penaherrera MS, Sharkey A, Moffett A, et al. 2011. Evidence for widespread changes in promoter methylation profile in human placenta in response to increasing gestational age and environmental/stochastic factors. BMC Genomics 12 (1 ):529, PMID: , 10.1186/1471-2164-12-529.22032438
58. Jensen Peña C, Monk C, Champagne FA. 2012. Epigenetic effects of prenatal stress on 11β-hydroxysteroid dehydrogenase-2 in the placenta and fetal brain. PLoS One 7 (6 ):e39791, PMID: , 10.1371/journal.pone.0039791.22761903
59. Kundakovic M, Jaric I. 2017. The epigenetic link between prenatal adverse environments and neurodevelopmental disorders. Genes 8 (3 ):104, 10.3390/genes8030104.28335457
60. Lester BM, Marsit CJ. 2018. Epigenetic mechanisms in the placenta related to infant neurodevelopment. Epigenomics 10 (3 ):321–333, PMID: , 10.2217/epi-2016-0171.29381081
61. Chhabra D, Sharma S, Kho AT, Gaedigk R, Vyhlidal CA, Leeder JS, et al. 2014. Fetal lung and placental methylation is associated with in utero nicotine exposure. Epigenetics 9 (11 ):1473–1484, PMID: , 10.4161/15592294.2014.971593.25482056
62. Timpson NJ, Greenwood CMT, Soranzo N, Lawson DJ, Richards JB. 2018. Genetic architecture: the shape of the genetic contribution to human traits and disease. Nat Rev Genet 19 (2 ):110–124, PMID: , 10.1038/nrg.2017.101.29225335
|
PMC010xxxxxx/PMC10116877.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37074185
EHP11134
10.1289/EHP11134
Research
Prenatal Exposure to Air Pollution and Pre-Labor Rupture of Membranes in a Prospective Cohort Study: The Role of Maternal Hemoglobin and Iron Supplementation
Wu Lin 1 2 3 4 *
Yin Wan-jun 1 2 3 4 *
Yu Li-jun 1 2 3 4
Wang Yu-hong 1 2 3 4
Jiang Xiao-min 5
Zhang Ying 6
Tao Fang-biao 1 2 3 4
Tao Rui-xue 7
Zhu Peng 1 2 3 4
1 Department of Maternal, Child and Adolescent Health, School of Public Health, Anhui Medical University, Hefei, China
2 MOE Key Laboratory of Population Health Across Life Cycle, Hefei, China
3 NHC Key Laboratory of Study on Abnormal Gametes and Reproductive Tract, Anhui Medical University, Hefei, China
4 Anhui Provincial Key Laboratory of Population Health and Aristogenic, Anhui Medical University, Hefei, China
5 Department of Obstetrics and Gynecology, Anhui Women and Child Health Care Hospital, Hefei, China
6 Department of Obstetrics and Gynecology, The First Affiliated Hospital of Anhui Medical University, Hefei, China
7 Department of Gynecology and Obstetrics, Hefei First People’s Hospital, Hefei, China
Address correspondence to Peng Zhu, No. 81 Meishan Rd., Hefei, Anhui, China. Email: [email protected]. And, Rui-xue Tao, No. 390 Huaihe Rd., Hefei, Anhui, China. Email: [email protected].
19 4 2023
4 2023
131 4 04701320 2 2022
19 3 2023
20 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Exposure to air pollution in prenatal period is associated with prelabor rupture of membranes (PROM). However, the sensitive exposure time windows and the possible biological mechanisms underlying this association remain unclear.
Objective:
We aimed to identify the sensitive time windows of exposure to air pollution for PROM risk. Further, we examined whether maternal hemoglobin levels mediate the association between exposure to air pollution and PROM, as well as investigated the potential effect of iron supplementation on this association.
Method:
From 2015 to 2021, 6,824 mother–newborn pairs were enrolled in the study from three hospitals in Hefei, China. We obtained air pollutant data [particulate matter (PM) with aerodynamic diameter ≤2.5μm (PM2.5), PM with aerodynamic diameter ≤10μm (PM10), sulfur dioxide (SO2), and carbon monoxide (CO)] from the Hefei City Ecology and Environment Bureau. Information on maternal hemoglobin levels, gestational anemia, iron supplementation, and PROM was obtained from medical records. Logistic regression models with distributed lags were used to identify the sensitive time window for the effect of prenatal exposure to air pollutant on PROM. Mediation analysis estimated the mediated effect of maternal hemoglobin in the third trimester, linking prenatal air pollution with PROM. Stratified analysis was used to investigate the potential effect of iron supplementation on PROM risk.
Results:
We found significant association between prenatal exposure to air pollution and increased PROM risk after adjusting for confounders, and the critical exposure windows of PM2.5, PM10, SO2 and CO were the 21th to 24th weeks of pregnancy. Every 10-μg/m3 increase in PM2.5 and PM10, 5-μg/m3 increase in SO2, and 0.1-mg/m3 increase in CO was associated with low maternal hemoglobin levels [−0.94g/L (95% confidence interval (CI): −1.15, −0.73), −1.31g/L (95% CI: −1.55, −1.07), −2.96g/L (95% CI: −3.32, −2.61), and −1.11g/L (95% CI: −1.31, −0.92), respectively] in the third trimester. The proportion of the association between air pollution and PROM risk mediated by hemoglobin levels was 20.61% [average mediation effect (95% CI): 0.02 (0.01, 0.05); average direct effect (95%): 0.08 (0.02, 0.14)]. The PROM risk associated with exposure to low-medium air pollution could be attenuated by maternal iron supplementation in women with gestational anemia.
Conclusions:
Prenatal exposure to air pollution, especially in the 21st to 24th weeks of pregnancy, is associated with PROM risk, which is partly mediated by maternal hemoglobin levels. Iron supplementation in anemia pregnancies may have protective effects against PROM risk associated with exposure to low–medium air pollution. https://doi.org/10.1289/EHP11134
Supplemental Material is available online (https://doi.org/10.1289/EHP11134).
* Contributed equally to this work.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Air pollution is associated with health problems, including pregnancy-associated complications and increased adverse pregnancy outcomes.1–3 Prelabor rupture of membranes (PROM) affects 3% to 21% of all pregnant women globally4,5; it is a serious problem for maternal and infant health and may result in maternal mortality and premature birth.6–8 Thus, it is essential to identify factors that lead to PROM and their potential pathways.
Few studies have suggested a link between exposure to air pollution and PROM. A study conducted in Australia found that exposure to PM2.5 during the second trimester of pregnancy increased the risk of PROM by 3%.9 The prevalence of PROM has shown a small but significant decline in Australia; however, it continues to increase in China.9,10 A cohort study conducted in China between 2015 and 2017 indicated that daily PM2.5 levels (median concentrations: 61.8 μg/m3) were more than 2-fold the limit recommended by the World Health Organization (WHO) (10.00 μg/m3 annually)11,12; this could partly explain the rise in PROM in China.11 Because no consistent evidence is available, the association between chronic (during the entire period of pregnancy) exposure to air pollution and an increased risk of PROM needs to be investigated.
Recent studies have focused on the association of PROM with maternal factors, such as low maternal hemoglobin levels and anemia.13,14 Low maternal hemoglobin concentration may induce infections and predispose women to PROM.15 Several studies have indicated that maternal anemia may be responsible for the increased risk of PROM.13,15,16 In addition, few studies have suggested that fine particles might increase the risk of anemia and cause a decrease in hemoglobin levels in children and elder.17,18 However, evidence of the relationship between exposure to air pollution and hemoglobin levels and anemia in pregnant women is limited.
Therefore, this prospective cohort study aimed to investigate the associations of weekly exposure to air pollution during pregnancy with PROM risk, identify the windows of susceptibility, and calculate the cumulative effect of the window. Furthermore, we examined whether maternal hemoglobin levels have a mediator effect on these associations, as well as the potential effect of iron supplementation.
Research Design and Method
Study Participants
Our study was based on a maternal and infant health cohort study in Hefei (MIH-Hefei), China. Details of data collection and recruitment have been described previously.19,20 From March 2015 to September 2021, a total of 9,320 pregnant women were recruited for the study, provided they were between 13 and 23 gestational weeks, had a single pregnancy, lived in Hefei for at least 2 y, and gave birth at obstetrical examination hospitals.
Subjects who met the following criteria were excluded: participants pregnant using assisted reproductive techniques, participants experiencing serious pregnancy complications (including hyperemesis gravidarum and abnormal heart or liver function), pregnant mothers with uncertain address of residence, and pregnant women whose questionnaires showed missing data on birth outcome. Finally, 6,824 pregnant women were enrolled for the study at three hospitals in Hefei (Hefei First People’s Hospital, Anhui Maternal and Child Care Hospital, and the First Affiliated Hospital of Anhui Medical University). The annual outpatient volume of obstetrics in the three hospitals included in our study ranks in the top three in Hefei and accounts for ∼70% of all hospitals in Hefei.21 Protocols used in our study were approved by the Ethics Committee of Anhui Medical University (number: 2015002).
Exposure to Air Pollution
The ambient air pollution data for PM2.5, PM10, SO2, and CO levels in urban Hefei, estimated at 11 air monitoring points in five air monitoring stations (Shushanqu, Baohequ, Luyangqu, Yaohaiqu, and Xinzhanqu), are available on the Hefei Environment Projection Administration website (http://sthjj.hefei.gov.cn/index.html). We then converted the current address of the participants to map coordinates using Baidu Maps. The distance from the subjects’ home addresses to the matching monitor stations was calculated based on the coordinates, and 93.1% (mean: 3.0km, range: 0.9 to 8.2km) of participants lived within 5km of the nearest station. We estimated full gestational and trimester exposure to air pollution by calculating the daily averages and averaging the daily exposures throughout the total gestational and trimester-specific periods. Trimester periods were estimated based on the first day of the mother’s previous menstrual cycle and were verified by ultrasound examination of gestational age. The first, second, and third trimesters refer to gestational weeks (GW) 1–13, 14–26, and 27–40 or at birth, whichever was earlier.22
Assessment of PROM and Anemia during Pregnancy
Women with PROM (including preterm and term PROM) were identified from clinical and extract this information from the medical records from the three hospitals mentioned above; in our study, PROM was named code O42 and could be found in the International Classification of Diseases and Related Health Problems, 10th Revision (ICD-10). Preterm and term PROM is defined as rupture of fetal membranes prior to and at 37 wk or after 37 wk of gestation, respectively.
Maternal hemoglobin levels were measured during GW 24–28 and 32–36 in hospitals. Hemoglobin levels measured during 24–28 wk were used for the diagnosis of anemia according to the WHO guidelines.23 Women with hemoglobin levels lower than 110g/L were diagnosed as having anemia during pregnancy according to WHO criteria. Daily iron supplements (120mg/d of elemental iron) were recommended to pregnant women with anemia.24 Hemoglobin levels measured between 32–36 wk were used for subsequent analysis.
Confounding Variables
Demographic characteristic covariates were obtained through a face-to-face questionnaire used in a standardized interview of pregnant women at enrollment, including maternal age (<25, 25–34, ≥35 y), education (junior high school, high school, or bachelor’s degree and above), average family income [≤3,999, 4,000–7,999, ≥8,000 renminbi (RMB)/month], and parity (primipara, multipara). Self-reported information on lifestyle factors included the frequency of fruit and dessert intake (cake, ice cream, store-bought sweet rolls, etc.), vegetable intake during second trimester, maternal folic acid supplementation frequency (<3, ≥3 d/wk) during first trimester, maternal passive smoking status (ever or never) during second trimester, and maternal iron supplementation frequency (<3, ≥3 d/wk) during third trimester. The International Physical Activity Questionnaire25 was used to evaluate moderate physical activity (including table tennis, badminton, and vigorous walking) for at least 30 min per day. Pregnant women were divided into groups of <3 or ≥3 d/wk, based on iron supplementation frequency. Health-related characteristics were extracted from medical records during the second and third trimester, including prepregnancy body mass index (BMI) (<18.5 kg/m2, 18.5–23.9 kg/m2, or ≥24 kg/m2), gestational diabetes mellitus (GDM) (yes or no); hypertension during pregnancy, including chronic hypertension (hypertension that preceded pregnancy), and preeclampsia (hypertension along with thrombocytopenia, systemics impairment such as liver function damage, progressing renal insufficiency, etc.), gestational hypertension (hypertension happening after 20 GW without the systemic findings aforementioned or proteinuria),26 and vaginitis (yes or no). After delivery, further details such as the season of delivery (spring/summer/autumn/winter), gestational weeks of delivery, and preterm status (<37 GW) were collected from the medical records. Information on the daily temperature (°C) in the month before delivery was obtained from the China Meteorological Administration (https://data.cma.cn/).
Statistical Analysis
The demographic characteristics of the participants with and without PROM were summarized using descriptive statistics. Chi-square (χ2) tests were used to compare the characteristics of women with and without PROM and characteristics of women included and excluded in the analysis. Spearman’s correlation analysis was used to examine the associations between PM2.5, PM10, SO2, and CO levels. Confounders related to prenatal air pollution exposure, maternal hemoglobin levels, and PROM were determined using a directed acyclic graph (DAG) to visualize these relationships (Supplemental Figure 1).
Distributed lag nonlinear models were used to examine the sensitive windows of weekly PM2.5, PM10, SO2, and CO exposure throughout GW 1–37 related to PROM in unadjusted and adjusted models.27,28 Maternal age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature were incorporated in the models. We built cross-basis matrices to model the exposure–lag–response association. The exposure–response function was assumed to be linear, and the lag structure was modeled using a natural cubic spline with degrees of freedom (df) based on the Akaike information criterion. We used the distributed lag model (DLM) to calculate the cumulative estimates for the first trimester (GW 1–13), second trimester (GW 14–26), third trimester (after GW 27) by incorporating weekly exposure during pregnancy. The models were adjusted for several covariates, including maternal age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. We further constructed a logistic regression model to estimate the trimesters-specific exposure with PROM risk. In addition, a logistic regression model was used to estimate the associations of average exposure to air pollutant during the three trimesters with the risk of preterm PROM, term PROM, and PROM in unadjusted and adjusted models. The models were adjusted for several covariates, including maternal age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. The logistic models that additionally included the trimester adjustment were done. Furthermore, we preliminarily evaluated the collinearity of four air pollution concentrations using correlation analysis among four air pollutants. Considering the impact of collinearity on the effect estimate, the copollutant models were used in the sensitivity analysis.
Multiple linear regression was used to estimate the association of each air pollutant (per 10 μg/m3 in PM2.5 and PM10, per 5 μg/m3 in SO2, and per 0.1 mg/m3 in CO; per quartile increase) throughout the second and third trimesters with maternal hemoglobin levels in the third trimester. Adjustments for covariates included age, education, income, activity, passive smoking, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, and temperature. The odds ratio (OR) of PROM was calculated per unit increase in air pollution exposure (per 10 μg/m3 in PM2.5 and PM10, per 5 μg/m3 in SO2, and per 0.1 mg/m3 in CO) stratified by anemia status (110g/L) according to WHO standards23 based on logistic regression models. We examined the potential exposure–response relationships between maternal hemoglobin levels in the third trimester and PROM risk using logistic regression models after adjusting for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. Considering the WHO cutoff points of maternal hemoglobin levels (110g/L) for the diagnosis of anemia and the sample size, women were classified into nine groups according to 5g/L change. Women with hemoglobin levels of ≥130g/L (reference) and <100g/L were defined as the highest and lowest level groups, respectively.
Principal component analysis (PCA) accounts for interactions between all pollutants and attributes these interactions to a principal component.29–31 PCA was used as a tool capable of providing an overview of the interdependencies and variability of prenatal air pollutants. After calculating the principal component scores of four pollutants, a mediation analysis was performed using the “mediation” package in R to estimate the role of hemoglobin levels during the third trimester in association with exposure to four pollutants in the second trimester, the third trimester, and throughout the second trimester and third trimester as well as its contribution to PROM risk. The total effect, including the average mediation effect (AME) and average direct effect (ADE), was calculated.32 The AME referred to the indirect effect of prenatal exposure to four air pollutants on PROM mediated by hemoglobin in the third trimester, and the ADE referred to the effect of prenatal exposure to four air pollutants on PROM, excluding the effect of hemoglobin during the third trimester. The counterfactual framework for mediation analysis was used to examine the causal mediation assumptions. Mediator models (mediator=air pollution+covariates), outcome models (outcome=air pollution+hemoglobin+covariates) were specified, and the “treat” (air pollution) as well as “mediator” variables were specified (Supplemental Figure 1).33–34 Adjustments for covariates included age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature.
Logistic regression models were used to estimate the association between maternal iron supplementation and PROM risk stratified by air pollution in women with anemia. Low-medium air pollution was defined as air pollution concentration <50th percentile. Women with gestational anemia were classified into following four groups: a) low–medium air pollution (<50th percentile, P50) and iron supplementation, b) low–medium air pollution (<P50) and no iron supplementation, c) high air pollution (≥P50) and iron supplementation, and d) high air pollution (≥P50) and no iron supplementation. High air pollution (≥P50) and no iron supplementation group was considered as the reference group, and OR for PROM for the other three groups were calculated. The average exposure levels of air pollutants selected at P50 were similar to those seen in previous studies in China.11 The p-trend for maternal hemoglobin levels and the prevalence of PROM across four groups were calculated by general linear regression model and Mantel-Haenszel chi-square test, respectively. Furthermore, we conducted a sensitivity analysis based on the exposure levels of air pollution at P75 (75th percentile).
Statistical significance was considered as a two-sided p <0.05. We performed all analyses in R (version 3.5.0; R Core Development Team) using the R statistical packages “ggplot2,” “dlnm,” and “mediation” and the SPSS statistical software (Statistical Package for the Social Sciences version 23.0; IBM Corp.).
Results
The characteristics of the study population are summarized in Table 1. Of the 6,824 women who agreed to participate in the study, 1,439 (21.1%) had PROM. Women with PROM were more likely to be younger and to have a lower frequency of folic acid supplementation during pregnancy than those without PROM (Table 1). In comparison with women without PROM, higher proportions of primipara, GDM, gestational anemia, and delivery in winter and spring were observed in women with PROM (Table 1). The excluded women had incomplete covariates on birth outcomes and had a higher proportion of giving birth in winter and spring (Supplement Table 1).
Table 1 The general characteristics of the study population during 2015 to 2021 in Hefei [n (%)].
Characteristics All (n=6,824) PROM (n=1,439) Non-PROM (n=5,385) p-Valuea
Sociodemographic characteristics [n (%)]
Age (y) — — — <0.001
<25 1,666 (24.4) 406 (28.2) 1,260 (23.4) —
25–34 4,476 (65.6) 912 (63.4) 3,564 (66.2) —
≥35 682 (10.0) 121 (8.4) 561 (10.4) —
Education — — — 0.433
Junior high school 941 (13.8) 184 (12.8) 757 (14.1) —
High school 1,655 (24.3) 348 (24.2) 1,307 (24.3) —
Bachelor’s degree and above 4,228 (62.0) 907 (63.0) 3,321 (61.7) —
Family income (RMB/month) — — — 0.231
≤3,999 2,149 (31.5) 460 (32.0) 1,689 (31.4) —
4000–7999 4,132 (60.6) 878 (61.7) 3,254 (60.4) —
≥8,000 543 (8.0) 101 (7.0) 442 (8.2) —
Parity — — — <0.001
Primipara 2,611 (38.3) 642 (44.6) 1,969 (36.6) —
Multipara 4,213 (61.7) 797 (55.4) 3,416 (63.4) —
Season of delivery — — — 0.003
Spring 1,756 (25.7) 386 (26.8) 1,370 (25.4) —
Summer 1,800 (26.4) 331 (23.0) 1,469 (27.3) —
Autumn 1,730 (25.4) 403 (28.0) 1,327 (24.6) —
Winter 1,538 (22.5) 319 (22.2) 1,219 (22.6) —
Enrollment years — — — 0.080
2015–2016 2,385 (35.0) 538 (37.4) 1,846 (34.3) —
2017–2018 2,569 (37.6) 517 (35.9) 2,052 (38.1) —
2019–2021 1,870 (27.4) 384 (26.7) 1,486 (27.6) —
Perinatal health lifestyle factors [n (%)]b
Vegetable intake (times/week) — — 0.982
<3 231 (3.4) 49 (3.4) 182 (3.4) —
≥3 6,593 (96.6) 1,390 (96.6) 5,203 (96.6) —
Fruit intake (times/week) — — 0.917
<3 470 (6.9) 100 (6.9) 370 (6.9) —
≥3 6,354 (93.1) 1,339 (93.1) 5,015 (93.1) —
Dessert intake (times/week) — — 0.991
<3 5,633 (82.5) 1,188 (82.6) 4,445 (82.5) —
≥3 1,191 (17.5) 251 (17.4) 940 (17.5) —
Physical activity (days/week) — — — 0.281
<3 5,394 (79.0) 1,154 (80.2) 4,240 (78.7) —
≥3 1,430 (21.0) 285 (19.8) 1,145 (21.3) —
Folic acid supplementation (days/week) — — 0.047
<3 4,387 (64.2) 893 (62.1) 3,494 (64.9) —
≥3 2,437 (35.7) 546 (37.9) 1,891 (35.1) —
Iron supplementation (days/week) — — — 0.075
<3 5,847 (85.7) 1,254 (87.1) 4,593 (85.3) —
≥3 977 (14.3) 185 (12.9) 792 (14.7) —
Passive smoking — — — 0.113
Never 5,598 (82.0) 1,201 (83.5) 4,397 (81.7) —
Ever 1,226 (18.0) 238 (16.5) 988 (18.3) —
Perinatal health status [n (%)]
Prepregnancy BMI (kg/m2) — — — 0.709
<18.5 985 (14.4) 216 (15.0) 769 (14.3) —
18.5–23.9 4,816 (70.6) 1,014 (70.5) 3,802 (70.6) —
≥24.0 1,023 (15.0) 209 (14.5) 814 (15.1) —
Hypertension during pregnancy 139 (2.0) 37 (2.6) 102 (1.9) 1.106
Vaginitis 783 (11.5) 151 (10.5) 632 (11.7) 0.189
Gestational diabetes mellitus 1,420 (20.8) 255 (17.7) 1,165 (22.1) 0.001
Premature birth 229 (3.4) 111 (7.7) 118 (2.2) <0.001
Maternal anemia 2,290 (33.6) 591 (41.1) 1,699 (31.6) <0.001
Note: There were no missing values for covariates. —, no data; PROM, prelabor rupture of membranes; RMB, renminbi.
a Based on the chi-square test.
b The frequency of vegetable intake, fruit intake, dessert intake, physical activity was during second trimester. The frequency of folic acid supplementation intake was during the first trimester. The frequency of iron supplementation was during the third trimester.
A strong correlation between four air pollutants in three trimesters was observed (Spearman correlation ranged from 0.01 to 0.94; Supplement Table 2). The mean (SD) gestational exposure to PM2.5, PM10, SO2, and CO in the second and third trimester was 54.2 μg/m3 (15.1), 83.2 μg/m3 (13.5), 11.8 μg/m3 (4.0), and 0.9 mg/m3 (0.2), whereas that in the first trimester was 57.7 μg/m3 (18.6), 87.0 μg/m3 (17.0), 17.8 μg/m3 (4.9), and 0.9 mg/m3 (0.2) (Table 2).
Table 2 Cumulative and average effects between air pollutants exposure and PROM risk in distributed lag models and average exposure model.
Pollution Mean SD OR (95% CI) of PROMd
Distributed lag modele Average exposure modele
First trimester
PM2.5 (μg/m3)a 57.7 18.6 1.01 (0.97, 1.04) 1.01 (0.98, 1.04)
PM10 (μg/m3)a 87.0 17.0 1.02 (0.98, 1.06) 1.02 (0.98, 1.06)
SO2 (μg/m3)b 17.8 4.9 1.06 (0.98, 1.16) 1.08 (1.01, 1.14)
CO (mg/m3)c 0.9 0.2 1.03 (0.99, 1.07) 1.04 (1.01, 1.08)
Second trimester
PM2.5 (μg/m3)a 54.6 18.1 1.11 (1.04, 1.19) 1.14 (1.09, 1.19)
PM10 (μg/m3)a 84.1 16.3 1.14 (1.07, 1.22) 1.17 (1.12, 1.22)
SO2 (μg/m3)b 12.1 4.4 1.13 (1.03, 1.24) 1.22 (1.14, 1.31)
CO (mg/m3)c 0.9 0.2 1.10 (1.04, 1.16) 1.13 (1.09, 1.18)
Third trimester
PM2.5 (μg/m3)a 54.0 19.0 1.04 (1.01, 1.08) 1.06 (1.03, 1.09)
PM10 (μg/m3)a 82.2 16.9 1.05 (1.01, 1.09) 1.07 (1.04, 1.11)
SO2 (μg/m3)b 11.5 4.2 1.14 (1.05, 1.25) 1.16 (1.08, 1.24)
CO (mg/m3)c 0.9 0.2 1.06 (1.02, 1.10) 1.08 (1.04, 1.12)
Second and third trimesters
PM2.5 (μg/m3)a 54.2 15.1 1.13 (1.06, 1.22) 1.11 (1.06, 1.15)
PM10 (μg/m3)a 83.2 13.5 1.18 (1.10, 1.27) 1.14 (1.09, 1.20)
SO2 (μg/m3)b 11.8 4.0 1.16 (1.06, 1.28) 1.22 (1.13, 1.31)
CO (mg/m3)c 0.9 0.2 1.10 (1.04, 1.15) 1.11 (1.07, 1.16)
Note: BMI, body mass index; CI, confidence interval; CO, carbon monoxide; OR, odds ratio; PROM, prelabor rupture of membranes; SD, standard deviation.
a Per increase in 10 μg/m3.
b Per increase in 5 μg/m3.
c Per increase in 0.1 mg/m3.
d Estimated by distributed lag models using weekly mean exposures and by mean air pollution during specific exposure windows (average exposure model).
e The models were adjusted for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature.
Distributed lag nonlinear models were used to examine the sensitive windows of weekly PM2.5, PM10, SO2 and CO exposure related to PROM in the unadjusted and adjusted models (Figure 1; Supplement Table 3). The weekly air pollutant exposures were significantly associated with increased PROM risk: The window of susceptibility for PM2.5 (per 10-μg/m3 increase) was between GW 15 and 37, and the maximum effect was in the 21th–23th GW; for PM10 (per 10-μg/m3 increase) the susceptibility window was between GW 18 and 32, and the maximum effect was in the 24th GW; for SO2 (per 5-μg/m3 increase) the susceptibility window was between GW 10 and 37, and the maximum effect was in the 21th GW; and for CO (per 0.1-mg/m3 increase) the window of susceptibility was between GW 13 and 31, and the maximum effect was in the 21th–24th GW. Table 2 depicts the estimated trimester cumulative PROM risk in the DLMs and average models in logistic regression with covariates adjusted, as with the unadjusted model (Supplement Table 4). The estimated cumulative risk of PROM was significantly associated with prenatal exposure to air pollutants throughout the second and third trimesters (Table 2). For example, the cumulative OR for the second and third trimester was 1.13 per 10-μg/m3 increase in PM2.5 (95% CI: 1.06, 1.22). Similarly, significant associations were observed with per 10-μg/m3 increase in PM10 [OR=1.18 (95% CI: 1.10, 1.27)], per 5-μg/m3increase in SO2 [OR=1.16 (95% CI: 1.06, 1.28)], and per 0.1-mg/m3 increase in CO [OR=1.10 (95% CI: 1.04, 1.15)]. DLM estimates were consistent with the results from the average exposure models. We find similar effect estimates in the unadjusted model presented in Supplement Table 4. In addition, adjustment for trimesters air exposure in Supplement Table 5 did not change the pattern of estimates from those in Table 2. The associations between prenatal exposure to air pollution and term PROM and preterm PROM are presented in Supplement Table 6. However, prenatal exposure to air pollution was not associated with preterm PROM. For sensitivity analysis, we further evaluated the correlation in the copollution models during same exposure period, and the results are presented in Supplement Table 7. Results of copollution models for estimating the effect for PM2.5, PM10, CO, and SO2 were generally consistent with those of the single-pollutant model exposure to same period.
Figure 1. The PROM risk in association with week-specific prenatal air pollution exposure during pregnancy. Week-specific estimates are provided as the OR of PROM (with 95% CI) for a 10-μg/m3 increment of PM2.5 exposure (A). Week-specific estimates are provided as the OR of PROM (with 95% CI) for a 10-μg/m3 increment of PM10 exposure (B). Week-specific estimates are provided as the OR of PROM (with 95% CI) for a 5-μg/m3 increment of SO2 exposure (C). Week-specific estimates are provided as the OR of PROM (with 95% CI) for a 0.1-mg/m3 increment of CO exposure (D). Models were based on a distributed lag (nonlinear) model and adjusted for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. The numerical results are presented in Supplement Table 3. Note: BMI, body mass index; CI, confidence interval; CO, carbon monoxide; OR, odds ratio; PROM, prelabor rupture of membranes.
Figures 1A to 1D are ribbon plus line graphs titled particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, sulfur dioxide, and Carbon monoxide, plotting odds ratio of Prelabor rupture of membranes (95 percent confidence intervals), ranging from 0.96 to 1.08 in increments of 0.04; 0.96 to 1.08 in increments of 0.04; 0.95 to 1.25 in increments of 0.05; and 0.96 to 1.08 in increments of 0.04 (y-axis) across gestational week, ranging from 0 to 35 in increments of 5 (x-axis), respectively.
As presented in Figure 2A and Supplement Table 8, effect of each air pollutant (per 10 μg/m3 in PM2.5 and PM10, per 5 μg/m3 in SO2, and per 0.1 mg/m3 in CO) throughout the second trimester and third trimester was negatively associated with low maternal hemoglobin levels in third trimester [β with 95% CI for PM2.5: −0.94g/L (95% CI: −1.15, −0.73); β for PM10: −1.31g/L (95% CI: −1.55, −1.07); β for SO2: −2.96g/L (95% CI: −3.32, −2.61); and β for CO: −1.11g/L (95% CI: −1.31, −0.92)] upon adjustment for age, education, income, activity, passive smoking, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, and temperature. Significant associations with per quartile increase for each air pollutant were observed in Figure 2A. The results from the copollution models showed similar negative associations with the effect estimation for PM2.5, PM10, SO2, and CO in the single-pollutant model in Supplement Table 9. We further examined the potential exposure–response relationships between maternal hemoglobin levels during the third trimester and PROM risk (Figure 2B; Supplement Table 10). After adjusting for age, education, income, parity, activity, passive smoking folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature, higher risk of PROM was observed in women with hemoglobin below 100g/L (OR=2.44; 95% CI: 1.73, 3.46), women with a hemoglobin range of 100–104g/L (OR=2.42; 95% CI: 1.72, 3.41), women with a hemoglobin range of 105–109g/L (OR=2.40; 95% CI: 1.73, 3.34), women with a hemoglobin range of 110–114g/L (OR=1.97; 95% CI: 1.42, 2.73), women with a hemoglobin range of 115–119g/L (OR=1.77; 95% CI: 1.27, 2.45), women with a hemoglobin range of 120–124g/L (OR=1.72; 95% CI: 1.23, 2.42), and women with a hemoglobin range of 125–129g/L (OR=1.52; 95% CI: 1.05, 2.21) when compared with that in women in the conference group (hemoglobin higher than 130g/L).
Figure 2. The association among air pollution exposure, hemoglobin levels, and PROM risk. The estimated change in hemoglobin levels was calculated for each quartile and each unit increment in PM2.5, PM10, SO2, and CO during the second and third trimesters in linear regression model (A). The model was based on the line regression model and adjusted for age, education, income, activity, passive smoking, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, and temperature. The hemoglobin level per increase in PM2.5 and PM10 was 10 μg/m3, the hemoglobin level per increase in SO2 was 5 μg/m3, and the hemoglobin level per increase in CO was 0.1 mg/m3. The numerical results are presented in Supplement Table 8. The relationship between hemoglobin levels and PROM (B). The model was based on the logistic regression model and adjusted for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. Air pollution was in the second and third trimesters. The numerical results are presented in Supplement Table 10. Note: BMI, body mass index; CO, carbon monoxide; PROM, prelabor rupture of membranes.
Figure 2A is an error bar graph, plotting change in hemoglobin levels (95 percent confidence intervals), ranging from negative 8 to 0 in increments of negative 2 (y-axis) across particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, sulfur dioxide, and Carbon monoxide (x-axis) for by quartile increase and by per unit increase. Figure 2B is an error bar graph, plotting odds ratio of Prelabor rupture of membranes (95 percent confidence intervals), ranging from 0.5 to 4.0 in increments of 0.5 (y-axis) across hemoglobin levels (gram per liter), ranging as less than 100, 100 to 104, 105 to 109, 110 to 114, 115 to 119, 120 to 124, 125 to 129, and greater than or equal to 130 (x-axis).
The OR of PROM was calculated with per unit increase in exposure to air pollution (per 10 μg/m3 in PM2.5 and PM10, per 5 μg/m3 in SO2, and per 0.1 mg/m3 in CO) stratified by anemia status (Figure 3). Our results showed a decreased risk of PROM associated with per unit increase in exposure to PM2.5, PM10, SO2, and CO in women without anemia, in comparison with that in women with anemia. For example, for PM2.5, PROM risk in women without anemia [OR=1.05 (95% CI: 0.99, 1.11)] was lower than that in women with anemia [OR=1.12 (95% CI: 1.04, 1.20)].
Figure 3. The relationship between air pollution and PROM risk in different hemoglobin levels. Air pollution was in the second and third trimesters. The hemoglobin level per increase in PM2.5 and PM10 was 10 μg/m3, the hemoglobin level per increase in SO2 was 5 μg/m3, and the hemoglobin level per increase in CO was 0.1 mg/m3. Models adjusted for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. Note: BMI, body mass index; CO, carbon monoxide; Hb, hemoglobin; PROM, prelabor rupture of membranes.
Figure 3 is an error bar graph, plotting particulate matter begin subscript 2.5 end subscript, particulate matter begin subscript 10 end subscript, sulfur dioxide, and Carbon monoxide (y-axis) across odds ratio of Prelabor rupture of membranes (95 percent confidence intervals), ranging from 0.9 to 1.4 in increments of 0.1 (x-axis) for Anemia (hemoglobin less than 110 grams per liter) and non-anemia (hemoglobin greater than or equal to 110 grams per liter).
In PCA, the calculated eigenvalue of the first principal component (PC1) was 3.615 (>1), providing 90.38% composite information. PC1 is mainly driven by CO, PM10, PM2.5, and SO2 (Supplement Table 11). Maternal hemoglobin levels in the third trimester mediated 20.61% (AME=0.02; 95% CI: 0.01, 0.05) of the contribution to the association of principal component of exposure to air pollutants throughout the second and third trimesters with PROM after adjusting for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature (Table 3).
Table 3 Mediation effect by hemoglobin levels on third trimester air pollution associated with PROM risk.
Exposure window AME β (95% CI) p-Value ADE β (95% CI) p-Value Proportion (%)a
Second trimester 0.02 (0.01, 0.03) <0.001 0.08 (0.02, 0.10) <0.001 17.31
Third trimester 0.02 (0.01, 0.04) <0.001 0.06 (0.01, 0.08) 0.044 24.26
Second and third trimester 0.02 (0.01, 0.05) <0.001 0.08 (0.02, 0.14) 0.008 20.61
Note: Meditation effect by hemoglobin levels was calculated per 1g/L. Air pollution was based on the total score of principal components of four air pollutant exposures (PM10, PM2.5, SO2, and CO). ADE, average direct effect; AME, average mediation effect; BMI, body mass index; CI, confidence interval; CO, carbon monoxide; PROM, prelabor rupture of membranes.
a Proportion (%): The extent to which the association between four air pollution exposures and PROM was mediated through hemoglobin in the third trimester. The models were adjusted for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature.
The 50th and 75th percentiles for the exposures during second and third trimester was 52.8 μg/m3 and 66.5 μg/m3 for PM2.5, 82.4 μg/m3 and 92.3 μg/m3 for PM10, 11.9 μg/m3 and 14.9 μg/m3 for SO2, 0.90 mg/m3 and 1.03 mg/m3 for CO, respectively. (Table 4; Supplement Table 12). Data presented in Table 4 show that the risk of PROM was significantly attenuated by iron supplementation during pregnancy in women with gestational anemia under low–medium exposure to air pollution in adjusted and unadjusted models. Iron supplementation in women with low–medium exposure to air pollution resulted in a significantly lower incidence of PROM than that seen in women without iron supplementation. For PM2.5, our analysis indicated a pattern toward lower incidence of PROM in women who were taking iron supplements in comparison with that in women not taking iron supplements under low–medium air pollution (16.0% vs. 18.5%) and high air pollution (21.4% vs. 23.4%). The decreasing trend toward PROM incidence in the above-mentioned groups persisted with exposure to PM10 concentration of 10 μg/m3, SO2 concentration of 5 μg/m3, and CO concentration of 0.1 mg/m3. Results from the sensitivity analysis using the cutoff point of exposure to air pollution at P75 remained robust, as shown in Supplement Table 12. [For example, incidence of PROM under low–medium PM2.5 pollution was 17.9% and 19.5% for women with iron supplementation, without iron supplementation; high pollution (21.7% vs. 23.6%)] (Supplement Table 12). We did not observe discrepancy in the results of unadjusted model using cutoff point of P50 in Table 4. [For example, OR=0.65 (95% CI: 0.51, 0.83) for women with iron supplementation under low–medium PM2.5 pollution; OR=0.76 (95% CI: 0.56, 1.03) for women without iron supplementation under low–medium PM2.5 pollution; OR=0.90 (95% CI: 0.56, 1.43) for women with iron supplementation under high PM2.5 pollution] or P75 in Supplement Table 12 [OR with 95% CI: 0.70 (0.51, 0.98) for women with iron supplementation under low- to medium PM2.5 pollution; OR=0.78 (95% CI: 0.58, 1.06) for women without iron supplementation under low–medium PM2.5 pollution; OR=0.90 (95% CI: 0.55, 1.47) for women with iron supplementation under high PM2.5 pollution].
Table 4 The association between iron supplementation and PROM risk stratified by air pollution levels in women diagnosed with anemia.
Pollutionsa Exposure level Iron supplementation (days/week) n Hemoglobind PROMd
Mean±SD, g/L n (%) Unadjusted OR (95% CI) Adjusted OR (95% CI)e
PM2.5 <P50 f ≥3 531 115.3±10.3 b 85 (16.0)c 0.65 (0.51, 0.83) 0.62 (0.49, 0.79)
<3 817 113.2±11.1 151 (18.5) 0.76 (0.56, 1.03) 0.74 (0.53, 1.01)
≥P50 ≥3 309 112.1±10.3 66 (21.4) 0.90 (0.56, 1.43) 0.87 (0.54, 1.40)
<3 633 110.2±10.1 148 (23.4) 1.00f 1.00f
PM10 <P50 ≥3 555 115.3±10.3 b 90 (16.2)c 0.69 (0.52, 0.92) 0.68 (0.50, 0.93)
<3 821 112.7±10.9 156 (19.0) 0.79 (0.59, 1.06) 0.75 (0.56, 1.01)
≥P50 ≥3 285 111.6±10.1 60 (21.1) 0.92 (0.65, 1.30) 0.90 (0.62, 1.27)
<3 629 110.5±10.5 144 (22.9) 1.00f 1.00f
SO2 <P50 ≥3 587 115.9±9.7 b 95 (16.2)c 0.68 (0.51, 0.91) 0.66 (0.50, 0.87)
<3 790 113.8±10.7 149 (18.9) 0.75 (0.53, 1.05) 0.72 (0.51, 1.02)
≥P50 ≥3 253 109.6±10.7 50 (19.8) 0.80 (0.56, 1.14) 0.77 (0.54, 1.12)
<3 660 109.1±10.3 156 (23.6) 1.00f 1.00f
CO <P50 ≥3 596 115.1±10.3 b 103 (17.3)c 0.72 (0.54, 0.95) 0.74 (0.55, 0.99)
<3 843 112.7±10.7 158 (18.7) 0.79 (0.61, 1.02) 0.81 (0.62, 1.05)
≥P50 ≥3 244 111.5±10.3 52 (21.4) 0.93 (0.65, 1.34) 0.93 (0.65, 1.34)
<3 607 110.4±10.8 137 (22.6) 1.00f 1.00f
Note: CI, confidence interval; OR, odds ratio; PROM, prelabor rupture of membranes; SD, standard deviation.
a Air pollution was in the second and third trimesters.
b The p for trend of hemoglobin levels across four groups was <0.001, <0.001, <0.001, <0.001, respectively.
c The p for trend of PROM prevalence across four groups was <0.001, 0.006, 0.012, 0.004. respectively.
d The test for p-trend was performed using general linear regression model and Mantel-Haenszel chi-square test in hemoglobin levels and PROM prevalence across the above four groups.
e Models adjusted for age, education, income, parity, activity, passive smoking, folic acid supplementation, iron supplementation, prepregnancy BMI, hypertension during pregnancy, gestational diabetes mellitus, vaginitis, and temperature. The 50th percentile for the exposure during second and third trimester was 52.8 μg/m3 for PM2.5, 82.4 μg/m3 for PM10, 11.9 μg/m3 for SO2, and 0.90 mg/m3 for CO, respectively.
f Reference group.
Discussion
A positive association between exposure to air pollution and a statistically significance showed in this multicenter, prospective cohort study. Our mediation analysis suggested that maternal hemoglobin levels in the third trimester could partly mediate the association between prenatal exposure to air pollution and PROM. In addition, the potential association of iron supplementation on PROM risk was observed in women with gestational anemia under low–medium levels of air pollution. Therefore, iron supplementation during pregnancy could possibly affect PROM associated with prenatal exposure to air pollution in women with anemia.
The relationship between elevated air pollution and PROM has been examined but with inconsistent results.35–37 Contrasting results presented in these studies may be attributed to apparent differences in levels of air pollution, inconsistent domestic conditions, and differences in population characteristics. A prospective cohort study conducted in Wuhan, China, indicated that PM2.5 levels during pregnancy increased the risk of PROM, and the estimated PM2.5 exposure concentration was 61.8 μg/m3 during the entire pregnancy, which is similar to the finding in our current study.11 In recent years, some researchers have proposed that considering longer exposure periods does not take into account the potential windows of exposure that may span different periods of pregnancy, resulting in deviations in inferred susceptible windows of exposure. Therefore, it is necessary to further refine the exposure period to identify the window of the effect of exposure to air pollution more accurately.38 Results of the present study showed that the time windows for the maximum effect of PM2.5, PM10, CO, and SO2 were in the 21th to 24th week of pregnancy. Our study confirmed the association between a higher prenatal exposure to air pollution and PROM during pregnancy and more accurate time windows of maximum effects of exposure to pollutants. Accelerated fetal membrane aging and higher levels of immune responses from early- to mid-pregnancy could account for the sensitivity time windows of exposure to prenatal air pollution associated with risk to PROM.22,39,40
Anemia during pregnancy (hemoglobin <110g/L) is associated with poor maternal outcomes (such as postpartum infections) and infant outcomes (such as neonatal perinatal and mortality).41,42 Lower levels of hemoglobin have been considered as a marker in the condition of subclinical infections and inflammation.15 In our study, we found that exposure to air pollution during mid- to late pregnancy was negatively associated with maternal hemoglobin levels. This evidence indicates that changes in hemoglobin levels with increased prenatal exposure to air pollution could be an indicator of increased risk of adverse pregnancy outcomes. These findings are consistent with prior results reported for women with an average age of 69.6 y.17 Honda et al. reported that mean concentrations of PM2.5 exposure and hemoglobin levels were 10.39 μg/m3 and 13.5g/dL, respectively. They found that a 3.9-μg/m3 increase in annual-average PM2.5 was associated with a 0.81-g/dL decrease in hemoglobin levels. In accordance with our findings, these results indicate a significant negative association between air pollution and hemoglobin levels. Studies in Ethiopia have suggested that indoor air pollution is associated with anemia during pregnancy.43 Consistently, our study emphasized that exposure to air pollution during pregnancy is related to lower levels of maternal hemoglobin, thus contributing to prenatal anemia. Furthermore, our study found a significant correlation between anemia and an increased risk of PROM in pregnancy. Mediation analysis suggested that maternal hemoglobin levels could represent a potential factor contributing to increased risk of PROM posed by prenatal air pollution exposure. The assumption of unmeasured exposure–mediator confounding (e.g., wearing N95 masks for mitigating inhaled particulate air pollution, etc.),44 unmeasured mediator–outcome confounding (maternal diet, etc.), and unmeasured exposure–outcome confounding (genetic risk, etc.) could bias our results, although major covariates were adjusted for in our analyses.33,34,41
Iron deficiency may result from dramatically increased demands for iron in pregnancy to meet rapid fetal growth, as well as a decrease in serum iron triggered by inflammation, increased numbers of red blood cells, and increased plasma volume.45 Such factors may predispose pregnant women to anemia. Iron deficiency also accounts for adverse pregnancy and birth outcomes. In our study, we found a possible association of iron supplementation on the correction of anemia and PROM risk. Furthermore, in a stratified analysis using air pollution levels, the possible association of iron supplementation on PROM risk was detected only at low- to medium air pollution levels, showing that the potential positive effect of iron supplementation on PROM risk because of prenatal exposure to air pollution in women with anemia was limited. Our study suggests that PROM risk posed by exposure to low- to medium air pollution could be attenuated by maternal iron supplementation to some extent, especially in women with gestational anemia.
The mechanism through which hemoglobin may mediate the association of prenatal air pollution on PROM is unknown. We hypothesize the following potential mechanisms: Air pollutants cause a systemic inflammatory response and directly affect bone marrow function, and inflammation decreases renal erythropoietin secretion and increases bone marrow endogenous erythropoietin resistance, leading to the reduction in erythrocyte and hemoglobin production.46,47 In addition, air particles may cause hemolysis and destruction of red blood cells. Low hemoglobin levels is a known marker of inflammation,48 and therefore is associated with impaired immune function as well as dysregulation of functions of natural killer cells, T-cells, and neutrophils. Furthermore, impaired immune function has been identified in anemia, contributing to a higher susceptibility to bacterial infection, and it has been hypothesized to be associated with air pollution–related PROM.14
In the present study, the rates of preterm birth among included vs. excluded participants are 3.4% vs. 4.1%, respectively. The overall preterm rate in this study (3.4%) is lower than the rates in the overall China population49 (6.1%) but is similar with the rates in Anhui province50 (4.1%). Although several studies in China showed similar incidences of PROM from Nanjing5 (20.8%) and Shanghai51 (22.0%), the incidence of PROM in our study (21.0%) appears to be higher in comparison with that of the United States52 (12.0%). It could be explained by the discrepancies in air pollution levels, race, and socioeconomic factors.10,15,53 Moreover, considering the lower rate of preterm birth in this study, the proportion of preterm PROM (7.7%) of all PROM appears to be relatively low.
Our study is significant for several reasons. Our study could add to the growing body of literature pertaining to demonstrate the relationship between chronic ambient exposure to air pollution and PROM mediated by maternal anemia as well as decreasing hemoglobin levels based on results from our large-scale prospective cohort of pregnant women. Moreover, our finding that iron supplementation possibly alleviates the association of air pollution on PROM may drive further studies in this field. Second, we used data from our multicenter, prospective cohort study with adjustment for several confounding variables. Third, we simultaneously observed the influence of maternal anemia and possible association of iron supplementation. Finally, a more accurate time window for the effect of exposure of air pollution on PROM was identified in our study.
Nevertheless, this study has several limitations. First, we did not consider indoor air pollution, such as indoor dust, kitchen cooking oil fumes, indoor cigarette smoke, and sleep air quality. We failed to assess traffic-related air and noise pollution during pregnancy, which was demonstrated to be related to PROM.54 Second, we performed an observational study rather than a randomized controlled trial to elucidate the effect of iron supplementation. Moreover, the residual confounding and uncontrolled risk factors, such as prior history of PROM, could bias our results. In addition, given the cumulative effect of secular trends in air pollution before conception on hemoglobin levels in women, our study possibly overestimated the effect of air pollution exposure during pregnancy. Information on the change of addresses during pregnancy was not available, which may cause some bias in the effect estimates. Finally, the exposure data were assigned to the nearest air quality monitor rather than the estimated personal exposure to air pollution, which could provide incorrect results.
Conclusion
Prenatal exposure to ambient air pollution (PM2.5, PM10, SO2, and CO), especially in 21th to 24th week of pregnancy, is positively associated with PROM risk, partly mediated by maternal hemoglobin levels. Iron supplementation in anemia pregnancy potentially has a positive association on low- to medium air pollution-related PROM. Screening and treatment of gestational anemia could provide novel insights into the prevention of low-to medium air pollution–related PROM.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
This research received financial support from the National Key R&D Program of China (2022YFC2702901), the National Natural Science Foundation of China (81872631, 82173531) and Foundation for Scientific Research Improvement of Anhui Medical University (2021xkjT009).
L.W.: Analyzing-data, Writing—original draft. W.-J.Y.: Writing—original draft. L-J.Y., Y.-H.W., X.-M.J., Y.Z.: Investigation. F.-B.T.: Writing—review and editing. R.-X.T.: Supervision, project administration. P.Z.: Writing—review and editing; designing study.
==== Refs
References
1. Wang YY, Li Q, Guo Y, Zhou H, Wang X, Wang Q, et al. 2018. Association of long-term exposure to airborne particulate matter of 1 μm or less with preterm birth in China. JAMA Pediatr 172 (3 ):e174872, PMID: , 10.1001/jamapediatrics.2017.4872.29297052
2. Kwag Y, Kim MH, Oh J, Shah S, Ye S, Ha EH. 2021. Effect of heat waves and fine particulate matter on preterm births in Korea from 2010 to 2016. Environ Int 147 :106239, PMID: , 10.1016/j.envint.2020.106239.33341584
3. Smith RB, Beevers SD, Gulliver J, Dajnak D, Fecht D, Blangiardo M, et al. 2020. Impacts of air pollution and noise on risk of preterm birth and stillbirth in London. Environ Int 134 :105290, PMID: , 10.1016/j.envint.2019.105290.31783238
4. Committee on Practice Bulletins-Obstetrics. 2018. ACOG Practice Bulletin No. 188: Prelabor Rupture of Membranes. Obstet Gynecol 131 (1 ):e1–e14, PMID: , 10.1097/AOG.0000000000002455.29266075
5. Han Y, Wang W, Wang X, Dong T, van Donkelaar A, Martin RV, et al. 2020. Prenatal exposure to fine particles, premature rupture of membranes and gestational age: a prospective cohort study. Environ Int 145 :106146, PMID: , 10.1016/j.envint.2020.106146.32987218
6. Conde-Agudelo A, Belizán JM. 2000. Maternal morbidity and mortality associated with interpregnancy interval: cross sectional study. BMJ 321 (7271 ):1255–1259, PMID: , 10.1136/bmj.321.7271.1255.11082085
7. Marcellin L, Goffinet F. 2012. Are biological markers relevant for the diagnosis and the prognosis of preterm premature rupture of membranes (PPROM)? Clin Chem Lab Med 50 (6 ):1015–1019, PMID: , 10.1515/cclm-2011-1850.22706240
8. Morris JM, Roberts CL, Bowen JR, Patterson JA, Bond DM, Algert CS, et al. 2016. Immediate delivery compared with expectant management after preterm pre-labour rupture of the membranes close to term (PPROMT trial): a randomised controlled trial. Lancet 387 (10017 ):444–452, PMID: , 10.1016/S0140-6736(15)00724-2.26564381
9. Pereira G, Bell ML, Belanger K, de Klerk N. 2014. Fine particulate matter and risk of preterm birth and pre-labor rupture of membranes in Perth, Western Australia 1997–2007: a longitudinal study. Environ Int 73 :143–149, PMID: , 10.1016/j.envint.2014.07.014.25118087
10. Wallace ME, Grantz KL, Liu D, Zhu Y, Kim SS, Mendola P. 2016. Exposure to ambient air pollution and premature rupture of membranes. Am J Epidemiol 183 (12 ):1114–1121, PMID: , 10.1093/aje/kwv284.27188941
11. Wang K, Tian Y, Zheng H, Shan S, Zhao X, Liu C. 2019. Maternal exposure to ambient fine particulate matter and risk of premature rupture of membranes in Wuhan, Central China: a cohort study. Environ Health 18 (1 ):96, PMID: , 10.1186/s12940-019-0534-y.31727105
12. WHO Guidelines Approved by the Guidelines Review Committee. 2021. WHO Global Air Quality Guidelines: Particulate Matter (PM25 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide. Geneva, Switzerland: World Health Organization.
13. Chu FC, Shaw SW, Lo LM, Hsieh TT, Hung TH. 2020. Association between maternal anemia at admission for delivery and adverse perinatal outcomes. J Chin Med Assoc 83 (4 ):402–407, PMID: , 10.1097/JCMA.0000000000000215.32238782
14. Archabald K, Lopes V, Anderson B. 2015. Impact of antepartum anemia on the development of chorioamnionitis at term. Am J Perinatol 32 (3 ):283–288, PMID: , 10.1055/s-0034-1384638.25077469
15. Ferguson SE, Smith GN, Salenieks ME, Windrim R, Walker MC. 2002. Preterm premature rupture of membranes. Nutritional and socioeconomic factors. Obstet Gynecol 100 (6 ):1250–1256, PMID: , 10.1016/s0029-7844(02)02380-3.12468170
16. Gonzales GF, Tapia V, Gasco M, Carrillo CE, Fort AL. 2012. Association of hemoglobin values at booking with adverse maternal outcomes among Peruvian populations living at different altitudes. Int J Gynaecol Obstet 117 (2 ):134–139, PMID: , 10.1016/j.ijgo.2011.11.024.22356761
17. Honda T, Pun VC, Manjourides J, Suh H. 2017. Anemia prevalence and hemoglobin levels are associated with long-term exposure to air pollution in an older population. Environ Int 101 :125–132, PMID: , 10.1016/j.envint.2017.01.017.28153527
18. Harris AM, Sempértegui F, Estrella B, Narváez X, Egas J, Woodin M, et al. 2011. Air pollution and anemia as risk factors for pneumonia in Ecuadorian children: a retrospective cohort analysis. Environ Health 10 :93, PMID: , 10.1186/1476-069X-10-93.22050924
19. Yao M, Liu Y, Jin D, Yin W, Ma S, Tao R, et al. 2020. Relationship between temporal distribution of air pollution exposure and glucose homeostasis during pregnancy. Environ Res 185 :109456, PMID: , 10.1016/j.envres.2020.109456.32278159
20. Liu Y, Li L, Xie J, Jiao X, Hu H, Zhang Y, et al. 2021. Foetal 25-hydroxyvitamin D moderates the association of prenatal air pollution exposure with foetal glucolipid metabolism disorder and systemic inflammatory responses. Environ Int 151 :106460, PMID: , 10.1016/j.envint.2021.106460.33662886
21. Hefei Municipal Health Commission. 2019. The table of obstetric bed usage in the secondary midwifery medical institutions in Hefei. http://wjw.hefei.gov.cn/xwzx/tzgg/17313697.html [accessed 8 November 2022].
22. Martens DS, Cox B, Janssen BG, Clemente DBP, Gasparrini A, Vanpoucke C, et al. 2017. Prenatal air pollution and newborns’ predisposition to accelerated biological aging. JAMA Pediatr 171 (12 ):1160–1167, PMID: , 10.1001/jamapediatrics.2017.3024.29049509
23. Vogel JP, Souza JP, Mori R, Morisaki N, Lumbiganon P, Laopaiboon M, et al. 2014. Maternal complications and perinatal mortality: findings of the World Health Organization Multicountry Survey on Maternal and Newborn Health. BJOG 121 Suppl 1 :76–88, PMID: , 10.1111/1471-0528.12633.24641538
24. WHO (World Health Organization). 2012. Guideline: Daily Iron and Folic Acid Supplementation in Pregnant Women. Geneva, Switzerland: World Health Organization.
25. Craig CL, Marshall AL, Sjöström M, Bauman AE, Booth ML, Ainsworth BE, et al. 2003. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc 35 (8 ):1381–1395, PMID: , 10.1249/01.MSS.0000078924.61453.FB.12900694
26. Hypertension in pregnancy. 2013. Report of the American College of Obstetricians and Gynecologists’ Task Force on Hypertension in Pregnancy. Obstet Gynecol 122 (5 ):1122–1131, PMID: , 10.1097/01.aog.0000437382.03963.88.24150027
27. Gasparrini A, Armstrong B, Kenward MG. 2010. Distributed lag non-linear models. Stat Med 29 (21 ):2224–2234, PMID: , 10.1002/sim.3940.20812303
28. Neven KY, Wang C, Janssen BG, Roels HA, Vanpoucke C, Ruttens A, et al. 2021. Ambient air pollution exposure during the late gestational period is linked with lower placental iodine load in a Belgian birth cohort. Environ Int 147 :106334, PMID: , 10.1016/j.envint.2020.106334.33360673
29. Zeng Q, Fan L, Ni Y, Li G, Gu Q. 2020. Construction of AQHI based on the exposure relationship between air pollution and YLL in Northern China. Sci Total Environ 710 :136264, PMID: , 10.1016/j.scitotenv.2019.136264.31923661
30. Voukantsis D, Karatzas K, Kukkonen J, Räsänen T, Karppinen A, Kolehmainen M. 2011. Intercomparison of air quality data using principal component analysis, and forecasting of PM10 and PM2.5 concentrations using artificial neural networks, in Thessaloniki and Helsinki. Sci Total Environ 409 (7 ):1266–1276, PMID: , 10.1016/j.scitotenv.2010.12.039.21276603
31. Tran H, Kim J, Kim D, Choi M, Choi M. 2018. Impact of air pollution on cause-specific mortality in Korea: results from Bayesian model averaging and principle component regression approaches. Sci Total Environ 636 :1020–1031, PMID: , 10.1016/j.scitotenv.2018.04.273.29729505
32. Imai K, Keele L, Tingley D. 2010. A general approach to causal mediation analysis. Psychol Methods 15 (4 ):309–334, PMID: , 10.1037/a0020761.20954780
33. Richiardi L, Bellocco R, Zugna D. 2013. Mediation analysis in epidemiology: methods, interpretation and bias. Int J Epidemiol 42 (5 ):1511–1519, PMID: , 10.1093/ije/dyt127.24019424
34. VanderWeele TJ. 2016. Mediation analysis: a practitioner’s guide. Annu Rev Public Health 37 :17–32, PMID: , 10.1146/annurev-publhealth-032315-021402.26653405
35. Dadvand P, Basagaña X, Figueras F, Martinez D, Beelen R, Cirach M, et al. 2014. Air pollution and preterm premature rupture of membranes: a spatiotemporal analysis. Am J Epidemiol 179 (2 ):200–207, PMID: , 10.1093/aje/kwt240.24125920
36. Song J, Lu M, An Z, Liu Y, Zheng L, Li Y, et al. 2019. Estimating the acute effects of ambient ozone pollution on the premature rupture of membranes in Xinxiang, China. Chemosphere 227 :191–197, PMID: , 10.1016/j.chemosphere.2019.04.062.30986601
37. Zhang C, Li S, Guo GL, Hao JW, Cheng P, Xiong LL, et al. 2021. Acute associations between air pollution on premature rupture of membranes in Hefei, China. Environ Geochem Health 43 (9 ):3393–3406, PMID: , 10.1007/s10653-021-00833-1.33555491
38. Chen J, Li PH, Fan H, Li C, Zhang Y, Ju D, et al. 2022. Weekly-specific ambient fine particular matter exposures before and during pregnancy were associated with risks of small for gestational age and large for gestational age: results from project ELEFANT. Int J Epidemiol 51 (1 ):202–212, PMID: , 10.1093/ije/dyab166.34432047
39. Lannon SM, Vanderhoeven JP, Eschenbach DA, Gravett MG, Adams Waldorf KM. 2014. Synergy and interactions among biological pathways leading to preterm premature rupture of membranes. Reprod Sci 21 (10 ):1215–1227, PMID: , 10.1177/1933719114534535.24840939
40. Menon R, Richardson LS, Lappas M. 2019. Fetal membrane architecture, aging and inflammation in pregnancy and parturition. Placenta 79 :40–45, PMID: , 10.1016/j.placenta.2018.11.003.30454905
41. Young MF, Oaks BM, Tandon S, Martorell R, Dewey KG, Wendt AS. 2019. Maternal hemoglobin concentrations across pregnancy and maternal and child health: a systematic review and meta-analysis. Ann NY Acad Sci 1450 (1 ):47–68, PMID: , 10.1111/nyas.14093.30994929
42. Kemppinen L, Mattila M, Ekholm E, Pallasmaa N, Törmä A, Varakas L, et al. 2021. Gestational iron deficiency anemia is associated with preterm birth, fetal growth restriction, and postpartum infections. J Perinat Med 49 (4 ):431–438, PMID: , 10.1515/jpm-2020-0379.33554586
43. Andarge SD, Areba AS, Kabthymer RH, Legesse MT, Kanno GG. 2021. Is indoor air pollution from different fuel types associated with the anemia status of pregnant women in Ethiopia? J Prim Care Community Health 12 :21501327211034374, PMID: , 10.1177/21501327211034374.34328038
44. Kodros JK, O’Dell K, Samet JM, L’Orange C, Pierce JR, Volckens J. 2021. Quantifying the health benefits of face masks and respirators to mitigate exposure to severe air pollution. Geohealth 5 (9 ):e2021GH000482, PMID: , 10.1029/2021GH000482.34541439
45. Georgieff MK. 2020. Iron deficiency in pregnancy. Am J Obstet Gynecol 223 (4 ):516–524, PMID: , 10.1016/j.ajog.2020.03.006.32184147
46. Bárány P. 2001. Inflammation, serum C-reactive protein, and erythropoietin resistance. Nephrol Dial Transplant 16 (2 ):224–227, PMID: , 10.1093/ndt/16.2.224.11158392
47. Hsu CY, Bates DW, Kuperman GJ, Curhan GC. 2001. Relationship between hematocrit and renal function in men and women. Kidney Int 59 (2 ):725–731, PMID: , 10.1046/j.1523-1755.2001.059002725.x.11168955
48. Sears DA. 1992. Anemia of chronic disease. Med Clin North Am 76 (3 ):567–579, PMID: , 10.1016/s0025-7125(16)30340-6.1578957
49. Jing S, Chen C, Gan Y, Vogel J, Zhang J. 2020. Incidence and trend of preterm birth in China, 1990–2016: a systematic review and meta-analysis. BMJ Open 10 (12 ):e039303, PMID: , 10.1136/bmjopen-2020-039303.33310797
50. Li ZJ, Liang CM, Xia X, Huang K, Yan SQ, Tao RW, et al. 2019. Association between maternal and umbilical cord serum cobalt concentration during pregnancy and the risk of preterm birth: the Ma’anshan birth cohort (MABC) study. Chemosphere 218 :487–492, PMID: , 10.1016/j.chemosphere.2018.11.122.30497031
51. Wang C, Yu G, Menon R, Zhong N, Qiao C, Cai J, et al. 2022. Acute and chronic maternal exposure to fine particulate matter and prelabor rupture of the fetal membranes: a nation-wide survey in China. Environ Int 170 :107561, PMID: , 10.1016/j.envint.2022.107561.36209598
52. Pereira G, Evans KA, Rich DQ, Bracken MB, Bell ML. 2016. Fine particulates, preterm birth, and membrane rupture in Rochester, NY. Epidemiology 27 (1 ):66–73, PMID: , 10.1097/EDE.0000000000000366.26247489
53. Shree R, Caughey AB, Chandrasekaran S. 2018. Short interpregnancy interval increases the risk of preterm premature rupture of membranes and early delivery. J Matern Fetal Neonatal Med 31 (22 ):3014–3020, PMID: , 10.1080/14767058.2017.1362384.28764570
54. Yorifuji T, Naruse H, Kashima S, Murakoshi T, Doi H. 2015. Residential proximity to major roads and obstetrical complications. Sci Total Environ 508 :188–192, PMID: , 10.1016/j.scitotenv.2014.11.077.25478655
|
PMC010xxxxxx/PMC10117635.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37079391
EHP11963
10.1289/EHP11963
Invited Perspective
Invited Perspective: Toxic Metals and Hypertensive Disorders of Pregnancy
Eaves Lauren A. 1 2
Fry Rebecca C. 1 2 3 4
1 Department of Environmental Sciences and Engineering, Gillings School of Global Public Health, University of North Carolina at Chapel Hill (UNC-Chapel Hill), Chapel Hill, North Carolina, USA
2 Institute for Environmental Health Solutions, Gillings School of Global Public Health, UNC-Chapel Hill, Chapel Hill, North Carolina, USA
3 Curriculum in Toxicology and Environmental Medicine, UNC-Chapel Hill, Chapel Hill, North Carolina, USA
4 Department of Pediatrics, UNC-Chapel Hill, Chapel Hill, North Carolina, USA
Address correspondence to Rebecca C. Fry. Email: [email protected]
20 4 2023
4 2023
131 4 04130303 8 2022
24 10 2022
16 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
Refers to https://doi.org.10.1289/EHP10825
==== Body
pmcThe worldwide prevalence of gestational hypertension and preeclampsia is estimated at 10% and 2%–8%, respectively.1 In the United States alone, the incidence of preeclampsia—a leading cause of maternal mortality—increased by 25% between 1987 and 2004.2 Despite the prevalence and severity of these conditions, definitive causes remain elusive, hindering risk reduction interventions.3 There has been increasing attention to the role of environmental chemicals, including toxic metals, in the development of gestational hypertension and preeclampsia (together termed hypertensive disorders of pregnancy).4,5 The recent study by Borghese et al. significantly contributes to the growing literature establishing associations between toxic metal exposure and these disorders.6
The authors examined mixture effects, highlighted modification of toxic metal effects by essential metals, evaluated exposure windows of susceptibility, measured various species of toxic metals (e.g., arsenic), and assessed confounding by seafood consumption and air pollution, all within one of the largest study populations to address this topic. They found an increased risk of preeclampsia with elevated third-trimester blood lead levels. They also observed an increased risk of preeclampsia and gestational hypertension with elevated first-trimester blood arsenic concentrations. These data underscore arsenic and lead as perinatal toxicants that remain an urgent public health concern. Lead has been previously found to increase the risk of preeclampsia7; however, there is more mixed evidence with regard to arsenic’s contribution to hypertensive disorders of pregnancy.4 The findings by Borghese et al. expand upon prior work that has also documented other metals of concern, including cadmium, as potential etiologic factors underlying hypertensive disorders of pregnancy.4,8
Although toxic metals have been studied for hundreds of years, these chemicals have received relatively less research attention than newer, engineered chemicals in relation to hypertensive disorders of pregnancy—which is unfortunate given their omnipresence. Exposure to toxic metals, such as arsenic and lead, predominately occurs via contaminated drinking water,9,10 geogenic and industrial sources,11–13 and contaminated food sources.14,15 Despite the established toxicity of lead and governmental efforts to reduce exposure,16 measured biomarker levels remain concerningly high among reproductive-age women around the world.17–21 In fact, >500,000 pregnant women in the United States were predicted to have blood lead levels >5μg/dL between 2011 and 2017.19 This is particularly salient when considering that the median lead levels in the study by Borghese et al. were orders of magnitude lower (0.52–0.64μg/dL).6
Arsenic also continues to be a contaminant of concern, particularly in federally unregulated private well water, but also in public community water systems. In the United States, concentrations hundreds of times over the maximum contaminant level (MCL) set by the U.S. Environmental Protection Agency (EPA; 10μg/L) have been reported in private well water.22 Although public community water systems are regulated by the Safe Drinking Water Act,23 evidence shows that arsenic remains a problem in these systems as well, with exceedances especially likely in the Southwestern United States, in communities that are smaller or predominantly Hispanic and systems that rely on groundwater.24 However, placing the findings of blood arsenic from this study in the broader public health context is slightly more challenging than with lead given that there were inconsistent findings across the different arsenic biomarkers evaluated. In addition, the half-life of blood arsenic is several hours (thereby reflecting recent exposure that may or may not be chronic) and there are no specific public health guidelines on arsenic exposure for pregnant women, as exist for lead.25 Thus, more research is needed to validate the findings on arsenic from the study by Borghese et al.6 and to more fully grasp the clinical and public health implications.
Currently, it is not standard prenatal clinical care to test for maternal body concentrations of toxic metals or assess for potential exposure sources, although movement in this direction is endorsed by the American College of Obstetricians and Gynecologists and the International Federation of Gynecology and Obstetrics.26–28 With studies such as Borghese et al. bolstering the evidence that these toxicants contribute to hypertensive disorders of pregnancy,6 the foundation is strong for motivating change. Cultural shifts, such as clinicians asking patients about their drinking water sources and providing information on effective, low-cost water testing and filters,29,30 could improve outcomes for women at high risk of exposure. Clinics could incorporate biomonitoring of arsenic, lead, cadmium, and mercury, among other metals, and offer interventions as needed. Moving forward, health insurance companies should consider environmental health as preventative care, including covering the costs of biomonitoring and water/air filters. Achieving these changes may require evidence from clinical trials that evaluate the impact of such interventions on perinatal outcomes. Of course, action at the patient–provider level must be coupled with continued pressure to improve and tighten environmental regulations to reduce water, food, and air contamination in the first place.31,32 In fact, tighter federal regulations have been documented to reduce body burdens and disease incidence for both lead and arsenic.10,33 For example, the U.S. EPA’s more stringent MCL for arsenic implemented in 2006 reduced urinary arsenic levels by an average of 17% among public community water system users, which was predicted to reduce bladder and lung cancer incidence by 200–900 cases per year.10
One particularly striking finding in the study by Borghese et al. is that the toxicity of blood arsenic was reduced at higher blood manganese concentrations.6 Previous studies have also demonstrated the capacity of essential metals and other dietary factors to reduce the effects or body burden of toxic metals.8,34–37 These studies suggest that nutritional factors that act on common toxicity pathways could be used in clinical practice. For example, appropriate supplementation with manganese may improve outcomes in patients at high risk for metals exposure, a hypothesis worthy of further investigation. Future epidemiologic research on this topic should be encouraged to stratify by essential metals to test whether this specific finding is repeated in different populations. In addition, animal model experimentation would likely be required to evaluate safety of manganese supplementation, given its toxicity at higher doses, before proceeding to trial therapeutic use in pregnant populations. However, the use of essential metal supplementation is feasible, as evidenced by the fact that calcium supplementation has been shown to lower lead body burden in mouse models and among pregnant and lactating women in human studies.38,39 In turn, calcium supplementation is known to reduce the risk of preeclampsia; to our knowledge, the potential mediating role of the reduction in blood lead levels in this relationship has not been investigated.37,40
In addition to offering an avenue of potential clinical intervention for risk reduction, the antagonistic effect of manganese on arsenic toxicity furthers the toxicologic evidence that oxidative stress, particularly within the placenta, plays a critical role in the pathogenesis of hypertensive disorders of pregnancy.41 Manganese is a component of superoxidase dismutase, an antioxidant enzyme.42 Pathways related to oxidative stress and inflammation may play a role in poor placentation, one of the hallmarks of preeclampsia.43–45 The activation of these pathways by toxicants such as arsenic may be attenuated by the antioxidant capacities of chemicals such as manganese. Interestingly, data support environmentally responsive epigenetic control of these key pathways as part of the complex biological underpinnings of preeclampsia, offering another avenue for therapeutic strategies to be investigated.43,44,46,47 Continued epidemiologic studies along with in vivo and in vitro research into the mechanisms of environmentally induced hypertensive disorders of pregnancy may lead to further insights for risk-reducing interventions.
Last, the findings in the study by Borghese et al.6 take on added importance when considering the appalling racial disparities in maternal and infant mortality in the United States, where Black women are more likely to develop preeclampsia than their White counterparts.48,49 Although disparities are less extreme in Canada, where this study was conducted, they still persist.50 These disparities are likely in part driven by several forms of environmental injustice that result in women of color having greater exposure to harmful chemicals, including toxic metals.51,52 For example, municipal underbounding leaves periurban communities of color more likely to rely on unregulated private well water.53,54 Cultural and commercial pressure to attain White beauty standards often pushes women of color to use toxic skin and hair care products.55,56 Further, Superfund sites and other contaminating sources are disproportionately likely to adjoin communities of color.13,57 Thus, mounting evidence of toxic metals’ impact on adverse perinatal health outcomes behooves us to confront environmental racism to tackle maternal–child health disparities.
Taken together, the evidence raises three critical points to consider for improving perinatal environmental health. First, we must remain vigilant in focusing on toxic metals as chemicals of concern for perinatal health. Second, it is imperative that clinical care of pregnant patients include an assessment of environmental health history, perhaps moving toward biomonitoring of toxic metals and, ultimately, the implementation of exposure-reducing interventions. Finally, to translate these findings into improved perinatal health, we must advocate for solution-oriented changes, such as subsidizing and distributing water filters to families at high risk of exposure, ensuring community water systems comply with federal regulations, expanding clinical trials of nutritional interventions, and tackling environmental racism to promote clean drinking water for all.
Acknowledgments
The authors are grateful to T. Manuck for her review and edits on this manuscript.
==== Refs
References
1. Duley L. 2009. The global impact of pre-eclampsia and eclampsia. Semin Perinatol 33 (3 ):130–137, PMID: , 10.1053/j.semperi.2009.02.010.19464502
2. Wallis AB, Saftlas AF, Hsia J, Atrash HK. 2008. Secular trends in the rates of preeclampsia, eclampsia, and gestational hypertension, United States, 1987–2004. Am J Hypertens 21 (5 ):521–526, PMID: , 10.1038/ajh.2008.20.18437143
3. Rana S, Lemoine E, Granger JP, Karumanchi SA. 2019. Preeclampsia: pathophysiology, challenges, and perspectives. Circ Res 124 (7 ):1094–1112, PMID: , 10.1161/CIRCRESAHA.118.313276.30920918
4. Rosen EM, Muñoz MI, McElrath T, Cantonwine DE, Ferguson KK. 2018. Environmental contaminants and preeclampsia: a systematic literature review. J Toxicol Environ Health B Crit Rev 21 (5 ):291–319, PMID: , 10.1080/10937404.2018.1554515.30582407
5. Stone J, Sutrave P, Gascoigne E, Givens MB, Fry RC, Manuck TA. 2021. Exposure to toxic metals and per- and polyfluoroalkyl substances and the risk of preeclampsia and preterm birth in the United States: a review. Am J Obstet Gynecol MFM 3 (3 ):100308, PMID: , 10.1016/j.ajogmf.2021.100308.33444805
6. Borghese MM, Fisher M, Ashley-Martin J, Fraser WD, Trottier H, Lanphear B, et al. 2023. Individual, independent, and joint associations of toxic metals and manganese on hypertensive disorders of pregnancy: results from the MIREC Canadian pregnancy cohort. Environ Health Perspect 131 (4 ):047014, 10.1289/EHP10825.37079392
7. Poropat AE, Laidlaw MAS, Lanphear B, Ball A, Mielke HW. 2018. Blood lead and preeclampsia: a meta-analysis and review of implications. Environ Res 160 :12–19, PMID: , 10.1016/j.envres.2017.09.014.28938191
8. Laine JE, Ray P, Bodnar W, Cable PH, Boggess K, Offenbacher S, et al. 2015. Placental cadmium levels are associated with increased preeclampsia risk. PLoS One 10 (9 ):e0139341, PMID: , 10.1371/journal.pone.0139341.26422011
9. Mitra A, Chatterjee S, Gupta DK. 2020. Environmental arsenic exposure and human health risk. In: Arsenic Water Resources Contamination: Challenges and Solutions. Fares A, Singh SK, eds. Cham: Springer International Publishing, 103–129.
10. Nigra AE, Sanchez TR, Nachman KE, Harvey D, Chillrud SN, Graziano JH, et al. 2017. The effect of the Environmental Protection Agency maximum contaminant level on arsenic exposure in the USA from 2003 to 2014: an analysis of the National Health and Nutrition Examination Survey (NHANES). Lancet Public Health 2 (11 ):e513–e521, PMID: , 10.1016/S2468-2667(17)30195-0.29250608
11. Harkness JS, Sulkin B, Vengosh A. 2016. Evidence for coal ash ponds leaking in the southeastern United States. Environ Sci Technol 50 (12 ):6583–6592, PMID: , 10.1021/acs.est.6b01727.27286270
12. Amini M, Abbaspour KC, Berg M, Winkel L, Hug SJ, Hoehn E, et al. 2008. Statistical modeling of global geogenic arsenic contamination in groundwater. Environ Sci Technol 42 (10 ):3669–3675, PMID: , 10.1021/es702859e.18546706
13. Zota AR, Schaider LA, Ettinger AS, Wright RO, Shine JP, Spengler JD. 2011. Metal sources and exposures in the homes of young children living near a mining-impacted Superfund site. J Expo Sci Environ Epidemiol 21 (5 ):495–505, PMID: , 10.1038/jes.2011.21.21587306
14. Chung JY, Yu SD, Hong YS. 2014. Environmental source of arsenic exposure. J Prev Med Public Health 47 (5 ):253–257, PMID: , 10.3961/jpmph.14.036.25284196
15. Lynch HN, Greenberg GI, Pollock MC, Lewis AS. 2014. A comprehensive evaluation of inorganic arsenic in food and considerations for dietary intake analyses. Sci Total Environ 496 :299–313, PMID: , 10.1016/j.scitotenv.2014.07.032.25089691
16. Dignam T, Kaufmann RB, LeStourgeon L, Brown MJ. 2019. Control of lead sources in the United States, 1970–2017: public health progress and current challenges to eliminating lead exposure. J Public Health Manag Pract 25 (suppl 1, Lead Poisoning Prevention ):S13–S22, PMID: , 10.1097/PHH.0000000000000889.30507765
17. Bulka CM, Bommarito PA, Fry RC. 2019. Predictors of toxic metal exposures among US women of reproductive age. J Expo Sci Environ Epidemiol 29 (5 ):597–612, PMID: , 10.1038/s41370-019-0152-3.31235790
18. Sanders AP, Flood K, Chiang S, Herring AH, Wolf L, Fry RC. 2012. Towards prenatal biomonitoring in North Carolina: assessing arsenic, cadmium, mercury, and lead levels in pregnant women. PLoS One 7 (3 ):e31354, PMID: , 10.1371/journal.pone.0031354.22427803
19. Ettinger AS, Egan KB, Homa DM, Brown MJ. 2020. Blood lead levels in U.S. women of childbearing age, 1976–2016. Environ Health Perspect 128 (1 ):17012, PMID: , 10.1289/EHP5925.31944143
20. Callan AC, Hinwood AL, Ramalingam M, Boyce M, Heyworth J, McCafferty P, et al. 2013. Maternal exposure to metals—concentrations and predictors of exposure. Environ Res 126 :111–117, PMID: , 10.1016/j.envres.2013.07.004.23896418
21. Lozano M, Murcia M, Soler-Blasco R, Casas M, Zubero B, Riutort-Mayol G, et al. 2022. Exposure to metals and metalloids among pregnant women from Spain: levels and associated factors. Chemosphere 286 (pt 2 ):131809, PMID: , 10.1016/j.chemosphere.2021.131809.34388877
22. Eaves LA, Keil AP, Rager JE, George A, Fry RC. 2022. Analysis of the novel NCWELL database highlights two decades of co-occurrence of toxic metals in North Carolina private well water: public health and environmental justice implications. Sci Total Environ 812 :151479, PMID: , 10.1016/j.scitotenv.2021.151479.34767890
23. U.S. Congress. 1974. Safe Drinking Water Act. Pub L No. 93-523 (93rd U.S. Congress, 16 December 1974). https://www.congress.gov/bill/93rd-congress/house-bill/13002 [accessed 3 August 2022].
24. Nigra AE, Chen Q, Chillrud SN, Wang L, Harvey D, Mailloux B, et al. 2020. Inequalities in public water arsenic concentrations in counties and community water systems across the United States, 2006–2011. Environ Health Perspect 128 (12 ):127001, PMID: , 10.1289/EHP7313.33295795
25. CDC (Centers for Disease Control and Prevention). 2010. CDC Guidelines for the Identification and Management of Lead Exposure in Pregnant and Lactating Women. https://stacks.cdc.gov/view/cdc/11854 [accessed 19 May 2020].
26. ACOG Committee Opinion No 575. 2013. Exposure to toxic environmental agents. Fertil Steril 100 (4 ):931–934, PMID: , 10.1016/j.fertnstert.2013.08.043.24070500
27. Di Renzo GC, Conry JA, Blake J, DeFrancesco MS, DeNicola N, Martin JN Jr, et al. 2015. International Federation of Gynecology and Obstetrics opinion on reproductive health impacts of exposure to toxic environmental chemicals. Int J Gynaecol Obstet 131 (3 ):219–225, PMID: , 10.1016/j.ijgo.2015.09.002.26433469
28. Stotland NE, Sutton P, Trowbridge J, Atchley DS, Conry J, Trasande L, et al. 2014. Counseling patients on preventing prenatal environmental exposures—a mixed-methods study of obstetricians. PLoS One 9 (6 ):e98771, PMID: , 10.1371/journal.pone.0098771.24964083
29. Tomlinson MS, Bommarito P, George A, Yelton S, Cable P, Coyte R, et al. 2019. Assessment of inorganic contamination of private wells and demonstration of effective filter-based reduction: a pilot-study in Stokes County, North Carolina. Environ Res 177 :108618, PMID: , 10.1016/j.envres.2019.108618.31419714
30. Mulhern R, Gibson JM. 2020. Under-sink activated carbon water filters effectively remove lead from private well water for over six months. Water (Basel) 12 (12 ):3584, 10.3390/w12123584.
31. Sutton PM, Giudice LC, Woodruff TJ. 2016. Moving from awareness to action on preventing patient exposure to toxic environmental chemicals. Am J Obstet Gynecol 214 (5 ):555–558, PMID: , 10.1016/j.ajog.2016.03.029.27126615
32. Lubick N. 2011. Advising parents in the face of scientific uncertainty: an environmental health dilemma. Environ Health Perspect 119 (10 ):A437–A441, PMID: , 10.1289/ehp.119-a436.22069779
33. McFarland MJ, Hauer ME, Reuben A. 2022. Half of US population exposed to adverse lead levels in early childhood. Proc Natl Acad Sci USA 119 (11 ):e2118631119, PMID: , 10.1073/pnas.2118631119.35254913
34. Clark J, Bommarito P, Stýblo M, Rubio-Andrade M, García-Vargas GG, Gamble MV, et al. 2022. Maternal serum concentrations of one-carbon metabolism factors modify the association between biomarkers of arsenic methylation efficiency and birth weight. Environ Health 21 (1 ):68, PMID: , 10.1186/s12940-022-00875-7.35836250
35. Hall MN, Gamble MV. 2012. Nutritional manipulation of one-carbon metabolism: effects on arsenic methylation and toxicity. J Toxicol 2012 :595307, PMID: , 10.1155/2012/595307.22523489
36. Gamble MV, Liu X, Slavkovich V, Pilsner JR, Ilievski V, Factor-Litvak P, et al. 2007. Folic acid supplementation lowers blood arsenic. Am J Clin Nutr 86 (4 ):1202–1209, PMID: , 10.1093/ajcn/86.4.1202.17921403
37. Ettinger AS, Hu H, Hernandez-Avila M. 2007. Dietary calcium supplementation to lower blood lead levels in pregnancy and lactation. J Nutr Biochem 18 (3 ):172–178, PMID: , 10.1016/j.jnutbio.2006.12.007.17296490
38. Christensen K. 2022. Nutritional multitasking? Exploring calcium supplementation to reduce toxic metal effects. Environ Health Perspect 130 (12 ):124002, 10.1289/EHP12341.36541789
39. Li H-B, Xue R-Y, Chen X-Q, Lin X-Y, Shi X-X, Du H-Y, et al. 2022. Ca minerals and oral bioavailability of Pb, Cd, and As from indoor dust in mice: mechanisms and health implications. Environ Health Perspect 130 (12 ):127004, 10.1289/EHP11730.36541774
40. Hofmeyr GJ, Lawrie TA, Atallah ÁN, Torloni MR. 2018. Calcium supplementation during pregnancy for preventing hypertensive disorders and related problems. Cochrane Database Syst Rev 10 (10 ):CD001059, PMID: , 10.1002/14651858.CD001059.pub5.30277579
41. Roberts JM, Escudero C. 2012. The placenta in preeclampsia. Pregnancy Hypertens 2 (2 ):72–83, PMID: , 10.1016/j.preghy.2012.01.001.22745921
42. Li L, Yang X. 2018. The essential element manganese, oxidative stress, and metabolic diseases: links and interactions. Oxid Med Cell Longev 2018 :7580707, PMID: , 10.1155/2018/7580707.29849912
43. Brooks SA, Martin E, Smeester L, Grace MR, Boggess K, Fry RC. 2016. miRNAs as common regulators of the transforming growth factor (TGF)-β pathway in the preeclamptic placenta and cadmium-treated trophoblasts: links between the environment, the epigenome and preeclampsia. Food Chem Toxicol 98 (pt A ):50–57, PMID: , 10.1016/j.fct.2016.06.023.27375191
44. Brooks SA, Fry RC. 2017. Cadmium inhibits placental trophoblast cell migration via miRNA regulation of the transforming growth factor beta (TGF-β) pathway. Food Chem Toxicol 109 (pt 1 ):721–726, PMID: , 10.1016/j.fct.2017.07.059.28774740
45. Alvarez MM, Chakraborty C. 2011. Cadmium inhibits motility factor-dependent migration of human trophoblast cells. Toxicol In Vitro 25 (8 ):1926–1933, PMID: , 10.1016/j.tiv.2011.06.016.21745561
46. Martin E, Ray PD, Smeester L, Grace MR, Boggess K, Fry RC. 2015. Epigenetics and preeclampsia: defining functional epimutations in the preeclamptic placenta related to the TGF-β pathway. PLoS One 10 (10 ):e0141294, PMID: , 10.1371/journal.pone.0141294.26510177
47. Kamrani A, Alipourfard I, Ahmadi-Khiavi H, Yousefi M, Rostamzadeh D, Izadi M, et al. 2019. The role of epigenetic changes in preeclampsia. Biofactors 45 (5 ):712–724, PMID: , 10.1002/biof.1542.31343798
48. Fasanya HO, Hsiao CJ, Armstrong-Sylvester KR, Beal SG. 2021. A critical review on the use of race in understanding racial disparities in preeclampsia. J Appl Lab Med 6 (1 ):247–256, PMID: , 10.1093/jalm/jfaa149.33227139
49. Tanaka M, Jaamaa G, Kaiser M, Hills E, Soim A, Zhu M, et al. 2007. Racial disparity in hypertensive disorders of pregnancy in New York State: a 10-year longitudinal population-based study. Am J Public Health 97 (1 ):163–170, PMID: , 10.2105/AJPH.2005.068577.17138931
50. McKinnon B, Yang S, Kramer MS, Bushnik T, Sheppard AJ, Kaufman JS. 2016. Comparison of black–white disparities in preterm birth between Canada and the United States. CMAJ 188 (1 ):E19–E26, PMID: , 10.1503/cmaj.150464.26553860
51. Boyles AL, Beverly BE, Fenton SE, Jackson CL, Jukic AMZ, Sutherland VL, et al. 2021. Environmental factors involved in maternal morbidity and mortality. J Womens Health (Larchmt) 30 (2 ):245–252, PMID: , 10.1089/jwh.2020.8855.33211615
52. Morello-Frosch R, Shenassa ED. 2006. The environmental “riskscape” and social inequality: implications for explaining maternal and child health disparities. Environ Health Perspect 114 (8 ):1150–1153, PMID: , 10.1289/ehp.8930.16882517
53. Nigra AE. 2020. Environmental racism and the need for private well protections. Proc Natl Acad Sci USA 117 (30 ):17476–17478, PMID: , 10.1073/pnas.2011547117.32641505
54. Leker HG, MacDonald Gibson J. 2018. Relationship between race and community water and sewer service in North Carolina, USA. PLoS One 13 (3 ):e0193225, PMID: , 10.1371/journal.pone.0193225.29561859
55. Zota AR, Shamasunder B. 2017. The environmental injustice of beauty: framing chemical exposures from beauty products as a health disparities concern. Am J Obstet Gynecol 217 (4 ):418.e1–418.e6, PMID: , 10.1016/j.ajog.2017.07.020.28822238
56. James-Todd TM, Chiu YH, Zota AR. 2016. Racial/ethnic disparities in environmental endocrine disrupting chemicals and women’s reproductive health outcomes: epidemiological examples across the life course. Curr Epidemiol Rep 3 (2 ):161–180, PMID: , 10.1007/s40471-016-0073-9.28497013
57. U.S. EPA (U.S. Environmental Protection Agency). 2021. Population surrounding 1,866 Superfund sites. https://www.epa.gov/superfund/population-surrounding-1866-superfund-sites [accessed 14 July 2022].
|
PMC010xxxxxx/PMC10117658.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37079392
EHP10825
10.1289/EHP10825
Research
Individual, Independent, and Joint Associations of Toxic Metals and Manganese on Hypertensive Disorders of Pregnancy: Results from the MIREC Canadian Pregnancy Cohort
https://orcid.org/0000-0003-0007-1565
Borghese Michael M. 1
Fisher Mandy 1
Ashley-Martin Jillian 1
Fraser William D. 2
Trottier Helen 3
Lanphear Bruce 4
Johnson Markey 5
Helewa Michael 6
Foster Warren 7
Walker Mark 8
Arbuckle Tye E. 1
1 Environmental Health Science and Research Bureau, Health Canada, Ottawa, Ontario, Canada
2 Department of Obstetrics and Gynecology, University of Sherbrooke, Sherbrooke, Quebec, Canada
3 Department of Social and Preventive Medicine, Université de Montreal, Montreal, Quebec, Canada
4 Faculty of Health Sciences, Simon Fraser University, Burnaby, British Columbia, Canada
5 Water and Air Quality Bureau, Health Canada, Ottawa, Ontario, Canada
6 Department of Obstetrics, Gynecology and Reproductive Sciences, University of Manitoba, Winnipeg, Manitoba, Canada
7 Department of Obstetrics and Gynecology, McMaster University, Hamilton, Ontario, Canada
8 Department of Obstetrics, Gynecology, University of Ottawa, Ottawa, Ontario, Canada
Address correspondence to Michael M. Borghese, Environmental Health Science and Research Bureau, Health Canada, 101 Tunney’s Pasture Driveway, Ottawa, ON, Canada, K1A 0K9. Telephone: (343) 542 3282. Email: [email protected]
20 4 2023
4 2023
131 4 04701420 12 2021
11 2 2023
16 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Background:
Toxic metals, such as lead (Pb), cadmium (Cd), arsenic (As), and mercury (Hg), may be associated with a higher risk of gestational hypertension and preeclampsia, whereas manganese (Mn) is an essential metal that may be protective.
Objectives:
We estimated the individual, independent, and joint associations of Pb, Cd, As, Hg, and Mn on the risk of developing gestational hypertension and preeclampsia in a cohort of Canadian women.
Methods:
Metal concentrations were analyzed in first and third trimester maternal blood (n=1,560). We measured blood pressure after 20 wk gestation to diagnose gestational hypertension, whereas proteinuria and other complications defined preeclampsia. We estimated individual and independent (adjusted for coexposure) relative risks (RRs) for each doubling of metal concentrations and examined interactions between toxic metals and Mn. We used quantile g-computation to estimate the joint effect of trimester-specific exposures.
Results:
Each doubling of third trimester Pb (RR=1.54; 95% CI: 1.06, 2.22) and first trimester blood As (RR=1.25; 95% CI: 1.01, 1.58) was independently associated with a higher risk of developing preeclampsia. First trimester blood As (RR=3.40; 95% CI: 1.40, 8.28) and Mn (RR=0.63; 95% CI: 0.42, 0.94) concentrations were associated with a higher and lower risk, respectively, of developing gestational hypertension. Mn modified the association with As such that the deleterious association with As was stronger at lower concentrations of Mn. First trimester urinary dimethylarsinic acid concentrations were not associated with gestational hypertension (RR=1.31; 95% CI: 0.60, 2.85) or preeclampsia (RR=0.92; 95% CI: 0.68, 1.24). We did not observe overall joint effects for blood metals.
Discussion:
Our results confirm that even low blood Pb concentrations are a risk factor for preeclampsia. Women with higher blood As concentrations combined with lower Mn in early pregnancy were more likely to develop gestational hypertension. These pregnancy complications impact maternal and neonatal health. Understanding the contribution of toxic metals and Mn is of public health importance. https://doi.org/10.1289/EHP10825
Supplemental Material is available online (https://doi.org/10.1289/EHP10825).
The authors declare they have nothing to disclose.
Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact [email protected]. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
==== Body
pmcIntroduction
Gestational hypertension and preeclampsia are major contributors to maternal1,2 and newborn3,4 morbidity and mortality. Women with these conditions have an increased risk of developing hypertension and other cardiovascular diseases later in life.5–7 Based on recent reviews, toxic metals such as lead (Pb), cadmium (Cd), arsenic (As), and mercury (Hg) are associated with a higher risk of developing these conditions.8–10 However, these reviews have also noted that this evidence is derived from small samples using nonlongitudinal study designs with a single exposure biomarker and at higher levels of exposure than most pregnant women in developed countries experience today.8,10–12 Further studies are needed to address these limitations.
Oxidative stress is one of several mechanisms through which Pb, Cd, As, and Hg may operate,13,14 either by generating reactive oxygen species15,16 or by interfering with antioxidant enzymes, such as superoxide dismutase (SOD).17–21 In contrast, manganese (Mn), an essential metal necessary for optimal fetal and placental development,22 may reduce the risk of preeclampsia23,24 through its antioxidant capacity as a constituent of the enzyme SOD (i.e., MnSOD).25,26 Reduced concentrations of SOD have been observed in both the plasma27 and placentae27,28 of women with preeclampsia vs. women with otherwise healthy pregnancies. Epidemiological29,30 and experimental31,32 evidence supports the potential for interaction between toxic metals and Mn. However, evidence for interactions with perinatal outcomes is mixed.33–36 Further knowledge of potential metal interactions may be beneficial for elucidating mechanisms of action and developing public health interventions.37
In addition, studies are needed to estimate the health effects of chemical mixtures,38 especially for chemicals with suspected common or opposing mechanisms of action. As noted in a 2022 review,39 two studies have examined associations between mixtures of toxic metals and preeclampsia. In a case–control study of pregnant women in China using weighted-quantile sum regression, Wang et al. reported that each tertile sum increase in blood metal concentrations was associated with a higher risk of preeclampsia40; the largest weights were for chromium, Hg, Pb, and As. Using data from a nested case–control study, Bommarito et al. used principal components analysis to examine associations between late-pregnancy urinary concentrations of metals and preeclampsia.41 None of the metals were associated with preeclampsia individually, but the authors found that the mixture of Cd, Mn, and Pb was positively associated with preeclampsia among individuals with low levels of essential metals (copper, selenium, and zinc). Although not specific to gestational hypertension, Yim et al.39 noted in their review that several papers have identified associations between mixtures of toxic metals and hypertension in nonpregnant adults.
Our primary objective was to estimate individual, independent, and joint associations for blood Pb, Cd, As, Hg, and Mn in both early and late pregnancy with the risk of developing gestational hypertension or preeclampsia in a large prospective pregnancy cohort. Examining and comparing the individual, independent, and joint associations will improve our ability to disentangle the potential health effects of these metals and reduce the potential for coexposure confounding.42 To expand on our primary analysis of blood As, we also examined associations with methylated As metabolites and organic As species measured in urine. Fetal sex differences have been identified in studies examining prenatal exposure to toxic metals and pregnancy/birth outcomes.33,43–51 Placental antioxidant defense mechanisms, implicated in the development of hypertensive disorders of pregnancy,27,28 may be reduced for male fetuses.52 Therefore, we explored the potential for effect modification by fetal sex. Finally, in a sensitivity analysis, we estimated associations between changes in these metals and measured blood pressure from the first to third trimesters.
Methods
Study Design and Participants
We analyzed data from the Maternal–Infant Research on Environmental Chemicals (MIREC) study, a prospective Canadian pregnancy cohort. Detailed information on study design, recruitment, and inclusion/exclusion criteria have been published.53 We recruited 2001 pregnant women from 10 sites across Canada during the first trimester between 2008 and 2011. Women were excluded if they were <18 years of age, at >14 wk gestation, reported illicit drug use, or could not communicate in either English or French. Women were also excluded if their fetus had known fetal abnormalities or fetal chromosomal malformations or if the woman currently had any of the following conditions: molar pregnancy, threatened spontaneous abortion (women with previous bleeding in the first trimester were included if the site documented a viable fetus at the time of recruitment), renal disease with altered renal function, epilepsy, any collagen disease (e.g., lupus erythematosus, scleroderma), active and chronic liver disease (hepatitis), heart disease, serious pulmonary disease, cancer, or hematological disorder (women with anemia or thombophilias were included).
Women who withdrew from the study (n=18), experienced a miscarriage or stillbirth (n=74), were carrying multiple fetuses (n=48), or with chronic hypertension (n=57) were excluded from this analysis, leaving an eligible sample size of 1,804 participants. In addition, participants who did not provide both first and third trimester blood samples (n=244, including 14 participants with gestational hypertension and 8 participants with preeclampsia) were excluded to facilitate comparisons across trimesters, resulting in a final analytical sample of 1,560 participants with complete outcome data. This study was approved by the research ethics boards of Health Canada/Public Health Agency of Canada and Sainte-Justine University Hospital, as well as those of all MIREC-affiliated study sites. Informed consent was obtained from all participants.
Measurement of Toxic Metals and Mn
We collected maternal whole blood samples during first (6–13 wk gestation) and third (27–40 wk gestation) trimester clinic visits, as well as a urine sample during the first trimester clinic visit. The Laboratoire de Toxicologie, Institut National de Santé Publique du Québec (Quebec City, Quebec, Canada), which has ISO/CEI 17025 accreditation of the Standards Council of Canada for human toxicology analysis, performed the metal analyses. The accuracy and precision of the analyses are evaluated on a regular basis through the laboratory’s participation in external quality assessment programs. Internal quality assurance was ensured by running validated reference materials after calibration after every 10th sample and at the end of each analytical sequence. Whole blood was analyzed using inductively coupled plasma mass spectrometry (ICP-MS; PerkinElmer ELAN ICP-MS DRC II, Norwalk CT, USA), as previously described.53–55 Limits of detection (LODs) and the percentages >LOD are provided in Table S1.
Urine samples were analyzed for the As species arsenate, arsenite, dimethylarsinic acid (DMA), monomethylarsonic acid (MMA), and arsenobetaine by high-performance liquid chromatography coupled with ICP-MS (Varian Inc., Palo Alto CA, USA), as previously described.54,56 The LOD for all urinary As species was 0.75μg As/L. DMA and arsenobetaine were detectable in 1,337 (86%) and 733 (50%) samples (Table S1), respectively, and were included in the present analysis. Arsenobetaine was dichotomized as above or below the LOD because 787 (50.4%) samples had undetectable concentrations. Arsenate, arsenite, and MMA were detected in few samples (26–249 samples, or 1.7%–16.0%; Table S1), so these data were not considered further. In descriptive analyses, arsenobetaine and DMA were standardized by specific gravity (SG) and measured in thawed urine samples by refractometry (Atago U.S.A. Inc., Bellevue WA) using the following formula57: Pc=Pi[(SGm−1)/(SGi−1)], where Pc is the SG-adjusted metabolite concentration, Pi is the observed metabolite concentration, SGi is the SG of the ith urine sample, and SGm is the median SG for the MIREC cohort. Arsenobetaine and DMA were not standardized for SG in regression models; instead SG included as a covariate in these models.
Definitions of Gestational Hypertension and Preeclampsia
Systolic and diastolic blood pressure (SBP and DBP, respectively) were assessed by clinical staff using a sphygmomanometer at three prenatal clinic visits (at 6–13, 14–26, and 27–40 wk gestation). SBP and DBP were measured twice, about 1 min apart, and were averaged. Using the Society of Obstetricians and Gynaecologists of Canada guidelines,58 women were considered as having gestational hypertension if SBP was ≥140 mmHg or DBP was ≥90 mmHg at a gestational age of 20 wk or later. Women with preexisting (i.e., chronic) hypertension, diagnosed if hypertension was present at a gestational age of <20wk, were excluded from this analysis even if they developed superimposed preeclampsia. Preeclampsia was defined as the presence of gestational hypertension with the addition of either a) proteinuria (defined as protein dipstick test ≥1+ or proteinuria in urine ≥300mg/24h or ≥0.3g/L) or b) related maternal “adverse conditions,” including abdominal right upper quadrant pain, headaches, visual disturbances, shortness of breath with SpO2 <97%, elevated creatinine, or “adverse complications,” such as abruptio placenta, disseminated intravascular coagulation, pulmonary edema, convulsions/eclampsia, stroke or coma, blood transfusion, elevated liver enzyme levels, and/or platelet count <50×109/L. This information was collected by clinical staff at the study visits or was abstracted from medical charts following delivery. Participants who did not meet any of these conditions were considered normotensive. Gestational age in weeks was based on last menstrual period and ultrasound dating. Last menstrual period was the preferred method. If these two methods differed by >7 d, gestational age was determined using ultrasound owing to concerns over recall and reliability of the last menstrual period estimate. Participants were categorized into three outcome groups for the present analysis: normotensive, gestational hypertension (without preeclampsia), and preeclampsia.
Measurement of Fine Particulate Matter
Fine particulate matter [PM ≤2.5μm in aerodynamic diameter (PM2.5)] was considered in our analyses as a confounder of the associations between metals and hypertensive disorders of pregnancy. PM2.5 is a complex mixture of airborne particles, including toxic metals,59 and has been associated with the development of gestational hypertension and preeclampsia.60,61 Coexposure to components of PM2.5 other than metals may confound associations between toxic metals and gestational hypertension or preeclampsia. Air pollution exposures were assigned to each participant based on the first three digits of their postal code (Forward Sortation Area) at delivery. As previously described,62 these estimates were produced by combining satellite-based retrievals of aerosol optical depth with chemical transport model output to develop estimates of near-surface PM2.5 concentrations. These estimates were further refined using geographically weighted regression to produce 1-km annual concentrations of PM2.5. Temporal resolution was added to the surface data by using daily values from National Air Pollution Surveillance monitoring stations located within a 30-km centroid of the participants’ Forward Sortation Area.63
Covariates
We identified covariates a priori based on previous assessments of risk factors for gestational hypertension and preeclampsia among Canadian women64; previously identified predictors of blood and urine concentrations of Pb, Cd, As, Hg, and Mn among pregnant women54,55; as well as previous reports of associations between these metals and gestational hypertension/preeclampsia.8,9,24 We used a directed acyclic graph to identify the minimally sufficient set of confounders65 (Figure S1). Participants completed in-person questionnaires during each trimester to gather sociodemographic and lifestyle information. We abstracted clinical data from medical charts.
Covariates included maternal age at delivery (in years, continuous), parity (nulliparous vs. multiparous), level of education [university (i.e., bachelor’s degree), college (i.e., 2-y diploma, less than college)], maternal smoking status at the first and third trimester (never, quit before pregnancy, quit during pregnancy, currently smoking), country of birth (Canada vs. elsewhere), prepregnancy body mass index [BMI; determined as weight in kilograms divided by height in meters (squared), continuous], self-reported race and ethnicity (White vs. other), and trimester-specific estimates of exposure to PM2.5. Participants self-reported their race and ethnicity from the following options: White, Chinese, South Asian, Black, Filipino, Southeast Asian, Latin American, Arab, West Asian, Japanese, Korean, and Aboriginal. For the purpose of this analysis, participants who exclusively identified their race or ethnicity as White were coded as being White, whereas participants selecting any other category or combination of categories were coded as other. Disaggregated data on race and ethnicity are presented descriptively in Table S2 by collapsing the categories above into those participants who exclusively reported their race or ethnicity as White, Black, Indigenous, Chinese, or Latin American, as well as other, including those response options with <20 respondents as well as anyone selecting multiple response options.
In analyses for blood Hg, we also adjusted for the potential confounding effect of fish consumption. Fish is a common source of Hg exposure as well as a primary source of omega-3 fatty acids, protein, vitamin D, and other essential minerals important for optimal fetal development and maternal health.66 During the first and third trimester, women completed a food frequency questionnaire indicating daily, weekly, or monthly consumption of 25 species of fish. We summarized these data according to the frequency (i.e., weekly vs. monthly consumption) of consuming all fish species vs. only those species with higher levels of Hg (i.e., tuna, marlin, orange roughy, shark, swordfish, mackerel, and escolar67,68). We adjusted for the reported consumption of fish species high in Hg on an ordinal scale ranging from none to ≥6 times per month.
Statistical Analysis
Descriptive statistics.
We performed statistical analyses using SAS Enterprise Guide (version 7.1; SAS institute, Cary, NC) and R (version 3.6.2; R Development Core Team). Participant characteristics and metal concentrations are presented using frequencies or median and interquartile range (IQR), as appropriate, by study outcome group. Relationships among metals within and between trimesters were assessed using Spearman and intra-class correlations (ICCs), respectively.
Individual associations.
We derived relative risks (RRs) and 95% confidence intervals (95% CIs) using Poisson regression with robust variance estimation69 for the association between metals and the development of gestational hypertension or preeclampsia (in separate models). Analyses were adjusted for the aforementioned covariates as well as for reported consumption of fish high in Hg in the analyses of Hg and As. For models of urinary As species, we adjusted for the aforementioned covariates (including fish consumption) as well as for urinary SG. Given the lognormal distribution of the observed data and the high frequency of detection, for blood metals and urinary DMA concentrations below the LOD, we assigned a value of the LOD divided by the square root of 2.70,71 Metal concentrations were log2 transformed to normalize the distribution and to allow for interpretation of parameter estimates as per a doubling (i.e., 2-fold increase) of concentration. Missing covariate data were imputed using multiple imputation with the fully conditional specification method using a logistic model for categorical variables and a linear model for continuous variables (5 imputations),72,73 and iteration-specific parameter estimates were combined to provide appropriate variance estimation.
We examined the potential for interactions between toxic metals and Mn on the multiplicative scale by using product terms. We visualized statistically significant continuous interactions using predicted probabilities from a multivariable binomial model. We also calculated the relative excess risk due to interaction (RERI)74,75 to assess additive interactions for each 2-fold increase in toxic metal concentrations and each 2-fold decrease in Mn concentration.76 A RERI >0 indicates the presence of an additive interaction between a toxic metal and Mn.
For As and Hg models, we explored adjusting these models for reported consumption of 25 species of fish rather than only species of fish high in Hg. In analyses for blood Cd, we explored adjusting for trimester-specific concentrations of plasma cotinine,77 the predominant metabolite of nicotine, instead of, and along with, self-reported smoking status. Higher levels of plasma cotinine reflect exposure from not only primary tobacco smoke but also from secondhand exposure. Finally, we explored restricting the model with first trimester blood As to those with arsenobetaine concentrations <1μg/L (n=812 normotensive, n=64 with gestational hypertension, and n=23 with preeclampsia), as well as separately adjusting for first trimester urinary concentrations of arsenobetaine to control for organic As species and metabolites that originate from fish rather than from inorganic As.78–80
To evaluate the dose–response relationship, these exposures were also analyzed according to tertiles. The adverse effects of Mn tend to be observed at extreme high and low levels of exposure.25 Therefore, in a separate sensitivity analysis, we categorized participants into three groups and estimated the effect of having low (<10th percentile) or high (>90th percentile) Mn concentrations relative to the middle group (10th–90th percentiles). A previous analysis of participants in the MIREC study using this approach reported that Mn concentrations <10th percentile were associated with low birth weight.81
Infant sex, determined by chart review, was explored as an effect modifier. Although parity is another important effect modifier, it was not considered in this paper because only 10 multiparous women developed preeclampsia.
Independent and joint associations.
To account for potential confounding by coexposure to multiple toxic metals, we examined the independent associations with each blood metal while adjusting for other trimester-specific blood metals. To estimate the joint associations for exposure to these metals, we used quantile g-computation, a generalized linear model based implementation of g-computation.82 This method estimates the parameters of a marginal structural model that characterizes the change in the expected potential outcome for a simultaneous, 1-quartile increase in all of the exposures in the specified mixture. This approach provides an estimate of the overall joint effect along with weights that can be interpreted as adjusted, independent effect sizes for quantized exposures. When the weights are in the same direction, they will sum to one and can be directly compared relatively to one another. If the weights are in different directions, they represent partial negative and positive effects, which cannot be directly compared and will not sum to one. Log2-transformed metal concentrations were rescaled with a mean of zero and standard deviation (SD) of one prior to being included in the model. We ran the quantile g-computation model on each individual imputed data set (m=5) that was generated for the regression analyses and averaged the results to obtain our final estimates. We also ran the quantile g-computation models for the toxic metals only.
Analyses of measured blood pressure.
We examined the associations between changes in individual log2-transformed blood metals and changes in continuous measures of SBP and DBP between first and third trimester using adjusted linear mixed models with a first-order autoregressive [AR(1)] covariance matrix. We explored the potential for nonlinearity using 3-knot restricted cubic splines. Given that the use of antihypertensive medications would directly impact participants’ blood pressure measures and could mask potential associations, participants who reported using these medications at a gestational age of 20 wk or later (n=32) were excluded from the blood pressure analyses only. These participants were not excluded from the categorical analyses for gestational hypertension and preeclampsia.
Results
Descriptive Results
Of the 1,560 participants, 1,403 (89.9%) women were normotensive, 114 (7.3%) developed gestational hypertension (without preeclampsia), and 43 (2.8%) developed preeclampsia (Table 1). There were no cases of eclampsia in the cohort. Gestational age was determined using last menstrual period for 1,390 women and using ultrasound dating for 170 women. Thirty-three of the 43 women who developed preeclampsia had proteinuria and, of the remaining 10 women, 8 had elevated liver enzymes and 2 received a blood transfusion. Participants included in this analysis were similar to those excluded from the analysis (Table S2), with the exception of a higher proportion of missing values for characteristics assessed in the third trimester because participation in both first and third trimester visits was a criterion for inclusion in this analysis. Overall, the percentage missing was <1% for most covariates, except for prepregnancy BMI, for which 114 (7%) participants had missing data. The relative efficiency for imputing prepregnancy BMI with five imputations was 0.995, suggesting that there was little to be gained from additional imputations.
Table 1 Participant descriptive characteristics [n (%) or median (IQR)] by study outcome group among 1,560 Canadian women in the MIREC study (2008–2011).
Variable Normotensive (n=1,403) Gestational hypertension (n=114)a Preeclampsia (n=43)
Maternal age (y) 32 (29, 36) 32 (28, 55) 31 (27, 36)
Prepregnancy BMIb
Under/normal weight 879 (62.7) 46 (40.4) 12 (27.9)
Overweight 270 (19.2) 26 (22.8) 8 (18.6)
Obese 157 (11.2) 29 (25.4) 19 (44.2)
Missing 97 (6.9) 13 (11.4) 4 (9.3)
Place of birth
Canada 1,126 (80.3) 101 (88.6) 38 (88.4)
Other 277 (19.7) 13 (11.4) 5 (11.6)
Race/ethnicity
White 1,171 (83.5) 103 (90.3) 38 (88.4)
Black 38 (2.7) 6 (5.3) 1 (2.3)
Indigenous 29 (2.1) 1 (0.9) 2 (4.6)
Chinese 32 (2.3) 0 0
Latin American 42 (3.0) 0 1 (2.3)
Other 91 (6.5) 4 (3.5) 1 (2.3)
Education
High school diploma or less 186 (13.3) 11 (9.7) 15 (34.9)
Some college/trade school 309 (22.0) 37 (32.5) 11 (25.6)
University degree 906 (64.6) 66 (57.9) 17 (39.5)
Missing 2 (0.1) 0 0
Smoking status—first trimester
Never 868 (61.9) 74 (64.9) 22 (51.2)
Quit before pregnancy 378 (26.9) 28 (24.6) 12 (27.9)
Quit during pregnancy 82 (5.8) 7 (6.1) 7 (16.3)
Current smoker 75 (5.4) 5 (4.4) 2 (4.7)
Smoking status—third trimester
Never 866 (61.7) 74 (64.9) 22 (51.2)
Quit before pregnancy 372 (26.5) 28 (24.6) 12 (27.9)
Quit during pregnancy 94 (6.7) 8 (7.0) 7 (16.3)
Current smoker 68 (4.8) 4 (3.5) 2 (4.7)
Missing 3 (0.2) 0 0
Parity
Nulliparous 607 (43.3) 52 (45.6) 33 (76.7)
Multiparous 796 (56.7) 62 (54.4) 10 (23.4)
Consumption of fish high in Hg—first trimesterc
None 574 (40.9) 38 (33.3) 13 (30.2)
At least once/month 385 (27.4) 36 (31.6) 17 (39.5)
Once/month 166 (11.8) 13 (11.4) 3 (7.0)
2–3 times/month 73 (5.2) 6 (5.3) 2 (4.7)
4–5 times/month 126 (9.0) 13 (11.4) 4 (9.3)
≥6 times/month 74 (5.3) 7 (6.1) 4 (9.3)
Missing 5 (0.4) 1 (0.9) 0 (0)
Consumption of fish high in Hg—third trimesterc
None 625 (44.5) 47 (41.2) 22 (51.2)
At least once/month 344 (24.5) 32 (28.1) 9 (20.9)
Once/month 163 (11.6) 11 (9.3) 4 (9.3)
2–3 times/month 74 (5.3) 4 (3.5) 3 (7.0)
4–5 times/month 136 (9.7) 13 (11.4) 5 (11.6)
≥6 times/month 58 (4.1) 6 (5.3) 0 (0)
Missing 3 (0.2) 1 (0.9) 0 (0)
SBP (mmHg) 125 (118, 135) 146 (139, 151) 152 (144, 169)
DBP (mmHg) 82 (77, 88) 93 (88, 100) 100 (93, 108)
Note: Categorical values are presented using frequency (column percentage) and continuous values as median (IQR). %, percentage; BMI, body mass index; DBP, highest measured diastolic blood pressure after 20 wk gestation; first trimester, 6–13 wk gestation; Hg, mercury; IQR, interquartile range; MIREC, Maternal–Infant Research on Environmental Chemicals; SBP, highest measured systolic blood pressure after 20 wk gestation; third trimester, 27–40 wk gestation.
a Includes participants with gestational hypertension (not preexisting hypertension) but without preeclampsia.
b Under/normal weight, overweight, and obese, with cutoffs at 25 and 30kg/m2, respectively.83
c Includes reported consumption of tuna, marlin, orange roughy, shark, swordfish, mackerel, and escolar.
Median (IQR) values for blood metal concentrations and urinary As species are presented in Table 2. Relative to the other groups, women with preeclampsia had slightly higher first trimester concentrations of blood As and urinary arsenobetaine and lower blood concentrations of Mn. Trimester-specific Spearman correlations among blood metals were generally weak; the strongest correlation was observed between Hg and As (r=0.38 in both trimesters; Table S3). ICCs between trimesters were high for most blood metals (ICC=0.61–0.76), except for blood As (ICC=0.35). First trimester blood As was moderately correlated with first trimester urinary DMA (Spearman r=0.22) and arsenobetaine (Spearman r=0.48); urinary DMA and arsenobetaine were also moderately correlated with each other (Spearman r=0.45).
Table 2 Median (IQR) maternal concentrations of blood metals [first trimester (6–13 wk gestation) and third trimester (27–40 wk gestation)] and urinary As species (first trimester) among 1,560 Canadian women in the MIREC study (2008–2011).
Normotensive (n=1,403) Gestational hypertension (n=114)a Preeclampsia (n=43)
First trimester urine (μg As/L)b
DMA 2.43 (1.60, 3.97) 2.27 (1.51, 3.33) 2.32 (1.62, 3.38)
Arsenobetaine <LOD (<LOD, 3.90) 0.51 (<LOD, 2.78) 1.56 (<LOD, 6.96)
First trimester blood (μg/dL)
Pb 0.62 (0.46, 0.85) 0.57 (0.41, 0.75) 0.54 (0.41, 0.81)
Cd 0.20 (0.13, 0.30) 0.19 (0.12, 0.27) 0.19 (0.15, 0.31)
As 0.82 (0.52, 1.20) 0.75 (0.51, 1.05) 0.90 (0.58, 1.42)
Hg 0.72 (0.34, 1.36) 0.64 (0.34, 1.20) 0.66 (0.18, 1.04)
Mn 8.79 (7.14, 10.99) 8.24 (7.14, 10.44) 8.24 (7.14, 9.34)
Third trimester blood (μg/dL)
Pb 0.56 (0.41, 0.79) 0.53 (0.44, 0.77) 0.58 (0.44, 0.91)
Cd 0.20 (0.13, 0.29) 0.18 (0.12, 0.27) 0.19 (0.13, 0.26)
As 0.70 (0.41, 1.12) 0.57 (0.39, 0.97) 0.67 (0.49, 1.05)
Hg 0.56 (0.28, 1.02) 0.43 (0.20, 0.82) 0.42 (0.12, 0.78)
Mn 12.64 (9.89, 15.38) 12.36 (9.89, 15.38) 11.54 (9.34, 15.38)
Note: As, arsenic; Cd, cadmium; DMA, dimethylarsinic acid; Hg, mercury; IQR, interquartile range (25th and 75th percentiles); LOD, limit of detection; MIREC, Maternal–Infant Research on Environmental Chemicals; Mn, manganese; Pb, lead.
a Includes participants with gestational hypertension but without preeclampsia.
b Specific-gravity standardized values.
Individual Models
In single-metal multivariable models (Table 3), women were at a higher risk of developing preeclampsia with each doubling of third trimester Pb concentrations (RR=1.44; 95% CI: 0.99, 1.98) or first trimester blood As (RR=1.24; 95% CI: 1.01, 1.51). Women were at a higher risk of developing gestational hypertension with each doubling of first trimester As concentrations (RR=3.26; 95% CI: 1.11, 9.58) and at a lower risk with each doubling of Mn concentrations (RR=0.67; 95% CI: 0.44, 0.99). For these two metals, we observed an interaction on the multiplicative scale (product term RR=0.68; 95% CI: 0.48, 0.96), as well as an excess RR due to interaction on the additive scale for developing gestational hypertension (RERI=0.46; 95% CI: 0.03, 0.89). Figure 1 displays the predicted probabilities of developing gestational hypertension according to As concentrations. At the mean concentration of Mn, the probability of developing gestational hypertension aligned with the prevalence of diagnoses in this sample (7.3%) and was stable across values of As. The deleterious association with As was observed at lower concentrations of Mn (−1 SD). There were no other interactions between blood metals and Mn (Table S4).
Table 3 Adjusted RR (95% CI) for individual associations between first trimester (6–13 wk gestation) blood metals and urinary As species concentrations, as well as third trimester (27–40 wk gestation) blood metal concentrations, and gestational hypertension without preeclampsia (n=114) or with preeclampsia (n=43) vs. having normal blood pressure among 1,560 Canadian women in the MIREC study (2008–2011).
Gestational hypertension Preeclampsia
First trimester urine (μg As/L)
DMA 1.31 (0.60, 2.85) 0.92 (0.68, 1.24)
Arsenobetaine
<LOD 1.00 1.00
≥LOD 1.12 (0.77, 1.63) 1.20 (0.66, 2.18)
First trimester blood (μg/dL)
Pb 0.91 (0.68, 1.20) 1.05 (0.69, 1.59)
Cd 0.90 (0.75, 1.09) 0.91 (0.65, 1.28)
As 3.26 (1.11, 9.58) 1.24 (1.01, 1.51)
Hg 0.97 (0.85, 1.12) 0.93 (0.74, 1.18)
Mn 0.67 (0.44, 0.99) 0.87 (0.46, 1.66)
Third trimester blood (μg/dL)
Pb 1.05 (0.81, 1.37) 1.44 (0.99, 2.08)
Cd 0.96 (0.81, 1.13) 1.02 (0.80, 1.29)
As 0.91 (0.79, 1.04) 0.97 (0.78, 1.21)
Hg 0.88 (0.77, 1.01) 0.85 (0.67, 1.08)
Mn 0.99 (0.65, 1.51) 0.94 (0.49, 1.81)
Note: RRs represent a doubling (per log2 increase) in whole blood or urinary concentration derived from Poisson regression models with robust variance estimation with multiple imputation (m=5) for missing covariate information. For arsenobetaine, RRs are modeled as ≥LOD (0.75μg As/L) vs. <LOD (reference). All models are adjusted for maternal age (continuous), education (university, college, less than college), first or third trimester-specific smoking status (never, quit before pregnancy, quit during pregnancy, currently smoking), prepregnancy BMI (continuous), parity (nulliparous vs. multiparous), race/ethnicity (White vs. other), PM2.5 (continuous), and country of birth (Canada vs. elsewhere). Models for As and Hg are additionally adjusted for reported consumption of fish high in Hg during the 30 d prior to the first or third trimester visit (none, at least once/month, once/month, 2–3 times/month, 4–5 times/month, ≥6 times/month). Models for urinary As species are additionally adjusted for specific gravity. For gestational hypertension, models for first trimester blood As and Mn are additionally adjusted for their multiplicative (product) interaction term (product term RR=0.68; 95% CI: 0.48, 0.96). For gestational hypertension, models for first trimester urinary DMA are additionally adjusted for first trimester Mn [main effect RR=0.91; 95% CI: (0.54, 1.54)], as well as the multiplicative (product) interaction term between urinary DMA and Mn [RR=0.91; 95% CI: (0.71, 1.17)]. As, arsenic; BMI, body mass index; Cd, cadmium; CI, confidence interval; DMA, dimethylarsinic acid; Hg, mercury; LOD, limit of detection; MIREC, Maternal–Infant Research on Environmental Chemicals; Mn, manganese; Pb, lead; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RR, relative risk.
Figure 1. Interaction effect and 95% confidence intervals of first trimester (6–13 wk gestation) blood As concentrations on the predicted probability of developing gestational hypertension by levels of first trimester blood Mn while holding all covariates centered at their mean among 1,560 Canadian women in the MIREC study (2008–2011). Predicted probabilities are derived from a multivariable binomial model. Lines intersect at the mean log2 concentration of As. All models are adjusted for maternal age (continuous), education (university, college, less than college), first trimester-specific smoking status (never, quit before pregnancy, quit during pregnancy, currently smoking), prepregnancy BMI (continuous), parity (nulliparous vs. multiparous), race/ethnicity (White vs. other), PM2.5 (continuous), country of birth (Canada vs. elsewhere), and reported consumption of fish high in Hg during the 30 d prior to the first trimester visit (none, at least once/month, once/month, 2–3 times/month, 4–5 times/month, ≥6 times/month). For gestational hypertension, models for first trimester As and Mn are additionally adjusted for their multiplicative (product) interaction term. Note: As, arsenic; BMI, body mass index; Hg, mercury; MIREC, Maternal–Infant Research on Environmental Chemicals; Mn, manganese; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; SD, standard deviation.
Figure 1 is a ribbon plus line graph, plotting predicted probability of developing gestational hypertension, ranging from 0.0 to 0.3 in increments of 0.1 (y-axis) across log to the base 2 arsenic concentrations, ranging from negative 2 to 4 in increments of 2 (x-axis) for log to the base 2 manganese concentrations, including plus 1 standard deviation, mean, negative 1 standard deviation.
The associations for As and Hg were similar when we adjusted for reported consumption of 25 species of fish (Table S5) rather than only fish high in Hg. Associations for Cd were similar when we adjusted for trimester-specific plasma cotinine concentrations with, or without, smoking status (Table S6).
Additionally adjusting for urinary arsenobetaine attenuated the association between first trimester blood As and gestational hypertension (RR=2.58; 95% CI: 1.10, 6.05) and had little impact on the association for blood Mn (RR=0.68; 95% CI: 0.47, 0.98) or their interaction (RR=0.68; 95% CI: 0.48, 0.97). Similarly, this had little effect on the association between first trimester blood As and preeclampsia (RR=1.27; 95% CI: 1.02, 1.58). When we restricted the analysis to those women with <1μg/L arsenobetaine concentrations, the associations with gestational hypertension for first trimester blood As (RR=15.1; 95% CI: 3.69, 61.9), first trimester blood Mn (RR=0.43; 95% CI: 0.23, 0.77), and their interaction (product term RR=0.43; 95% CI: 0.28, 0.66) were considerably stronger, but more imprecise. We observed a similar association between first trimester As and preeclampsia in this restricted analysis (RR=1.41; 95% CI: 0.99, 2.01). Finally, neither urinary DMA concentrations nor arsenobetaine detection were associated with the risk of developing the outcomes (Table 3). We also observed null associations for urinary DMA concentrations with gestational hypertension (RR=1.34; 95% CI: 0.40, 4.50) and preeclampsia (RR=0.85; 95% CI: 0.48, 1.50) when restricting the analysis to those women with arsenobetaine concentrations <1μg/L. We also did not observe interactions between urinary As species and Mn (Table S4).
We observed a monotonic dose–response pattern for the aforementioned associations when exposures were analyzed according to tertiles (Table S7). In addition, participants in the second tertile of third trimester blood Pb concentrations had a higher risk of developing gestational hypertension, but this association did not follow a monotonic dose–response pattern. Mn exposure categorized according to the 10th and 90th percentiles was not associated with either condition (Table S8). We observed effect modification by fetal sex for first trimester Mn only (Table S9), whereby each doubling of concentration was associated with a lower risk of developing gestational hypertension among women carrying male (RR=0.47; 95% CI: 0.29, 0.75), but not female, fetuses [RR=1:45 (95% CI: 0.58, 3.62)]; [interaction RR=2:42 (95% CI: 1.07, 5.46)]. No other sex-specific differences were observed.
Independent Models
In multivariable models estimating independent associations (i.e., adjusting for other metals), the associations for third trimester Pb and first trimester As remained relatively unchanged (Table 4). Similarly, the associations for first trimester As and Mn with gestational hypertension were retained, as were the interactions on both the multiplicative (product term RR=0.68; 95% CI: 0.48, 0.96) and additive (RERI=0.43; 95% CI: 0.03, 0.85) scales. None of the other metals in either trimester were independently associated with the risk of developing the outcomes.
Table 4 Adjusted RR (95% CI) for the association between first trimester (6–13 wk gestation) and third trimester (27–40 wk gestation) blood metal concentrations and gestational hypertension without preeclampsia (n=114) or with preeclampsia (n=43) vs. having normal blood pressure while controlling for other trimester-specific blood metals in the model among 1,560 Canadian women in the MIREC study (2008–2011).
Gestational hypertension Preeclampsia
First trimester (μg/dL)
Pb 0.95 (0.72, 1.24) 1.07 (0.70, 1.64)
Cd 0.92 (0.76, 1.12) 0.93 (0.66, 1.30)
As 3.20 (1.09, 9.37) 1.26 (1.01, 1.60)
Hg 1.00 (0.87, 1.15) 0.89 (0.69, 1.14)
Mn 0.72 (0.46, 1.12) 0.84 (0.43, 1.65)
Third trimester (μg/dL)
Pb 1.10 (0.84, 1.45) 1.58 (1.08, 2.3)
Cd 0.96 (0.81, 1.13) 0.98 (0.77, 1.25)
As 0.93 (0.80, 1.08) 1.03 (0.81, 1.31)
Hg 0.89 (0.77, 1.03) 0.80 (0.62, 1.04)
Mn 1.00 (0.64, 1.55) 0.84 (0.43, 1.66)
Note: RRs represent a doubling (per log2 increase) in whole blood concentration derived from Poisson regression models with robust variance estimation with multiple imputation (m=5) for missing covariate information. All models are adjusted for maternal age (continuous), education (university, college, less than college), first or third trimester-specific smoking status (never, quit before pregnancy, quit during pregnancy, currently smoking), prepregnancy BMI (continuous), parity (nulliparous vs. multiparous), race/ethnicity (White vs. other), PM2.5 (continuous), country of birth (Canada vs. elsewhere), and reported consumption of fish high in Hg during the 30 d prior to the first or third trimester visit (none, at least once/month, once/month, 2–3 times/month, 4–5 times/month, ≥6 times/month) and other trimester-specific blood metal concentrations. For gestational hypertension, models for first trimester As and Mn are additionally adjusted for the multiplicative (product) interaction term between As and Mn (product term RR=0.68; 95% CI: 0.48, 0.96). As, arsenic; BMI, body mass index; Cd, cadmium; CI, confidence interval; Hg, mercury; MIREC, Maternal–Infant Research on Environmental Chemicals; Mn, manganese; Pb, lead; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RR, relative risk.
Mixture Models
Using quantile g-computation, we did not observe an association for a simultaneous 1-quartile increase in metal concentrations within each trimester on the risk of developing either condition (Table 5); the overall joint relative risks were all close to the null. Blood As was the only chemical with a positive weight (i.e., contributing to a higher RR) in the first trimester model for preeclampsia, whereas Pb had the highest positive weight in the third trimester models; however, interpreting these weights in the context of null overall effects is difficult and should be done with caution. We reran the quantile g-computation models for the toxic metals only (excluding Mn) and observed similar null overall effect estimates (Table S10).
Table 5 Associations between a simultaneous 1-quartile increase in the trimester-specific blood metal mixture and adjusted RR (95% CI) for gestational hypertension (without preeclampsia) or preeclampsia among 1,560 Canadian women in the MIREC study (2008–2011).
RR (95% CI) Mixture weightsa
Pb Cd As Hg Mn
First trimester (6–13 wk gestation)
Gestational hypertension 0.79 (0.56, 1.11) −0.45 −0.16 −0.21 1 −0.17
Preeclampsia 1.07 (0.61, 1.85) −0.40 0.02 0.98 −0.34 −0.25
Third trimester (27–40 wk gestation)
Gestational hypertension 0.85 (0.62, 1.16) 0.61 −0.20 −0.23 −0.57 0.39
Preeclampsia 0.99 (0.60, 1.66) 0.93 −0.44 0.07 −0.55 −0.01
Note: RRs represent a simultaneous 1-quartile change in all exposures derived from quantile g-computation models with multiple imputation (m=5) for missing covariate information. All models are adjusted for maternal age (continuous), education (university, college, less than college), first or third trimester-specific smoking status (never, quit before pregnancy, quit during pregnancy, currently smoking), prepregnancy BMI (continuous), parity (nulliparous vs. multiparous), race/ethnicity (White vs. other), PM2.5 (continuous), country of birth (Canada vs. elsewhere), and reported consumption of fish high in Hg during the 30 d prior to the first or third trimester visit (none, at least once/month, once/month, 2–3 times/month, 4–5 times/month, ≥6 times/month). As, arsenic; BMI, body mass index; Cd, cadmium; CI, confidence interval; Hg, mercury; MIREC, Maternal–Infant Research on Environmental Chemicals; Mn, manganese; Pb, lead; PM2.5, particulate matter ≤2.5μm in aerodynamic diameter; RR, relative risk.
a Positive mixture weights indicate contributions to higher risk of conditions and sum to positive one; negative weights contribute to lower risk of conditions and sum to negative one.
Models for Measured Blood Pressure
Changes in either blood Pb or Hg concentrations were not prospectively associated with changes in measured blood pressure between the first and third trimester (Table S11). Changes in blood Cd concentrations were positively associated with DBP, but not SBP. Changes in As concentrations were positively associated with SBP, whereas changes in Mn concentration were negatively associated with SBP; neither were associated with DBP. The magnitude of all of these associations were weak (i.e., <1 mmHg change from first to third trimester per doubling of exposure from first to third trimester). Using 3-knot restricted cubic splines, we did not observe any nonlinear associations with measured blood pressure (Figure S2). There were no interactions between toxic metals and Mn for either SBP or DBP (Table S12), and we did not observe effect modification by fetal sex for any of the blood metals (Table S13).
Discussion
We examined the individual, independent, and joint associations between exposure to four toxic metals and Mn and the risk of developing gestational hypertension or preeclampsia in a pan-Canadian sample of pregnant women. We showed that both third trimester Pb and first trimester As blood concentrations were independently associated with a higher risk of developing preeclampsia. We also showed that higher first trimester As and Mn concentrations were associated with higher and lower risks, respectively, of developing gestational hypertension. Similar directions of association were observed in the prospective analysis of changes in metals on changes in SBP (but not DBP), although the magnitude of these associations was small. In addition, we observed an interaction between first trimester concentrations of As and Mn on both the multiplicative and additive scales such that the deleterious association for a 2-fold change in As concentration on the risk of gestational hypertension was strongest among women with lower Mn concentrations. The tertile-specific associations support a monotonic, or linear, dose–response relationship for these associations. Neither first nor third trimester Cd and Hg concentrations were associated with either condition.
Interpretation of Mixture Results
We did not observe an overall joint effect for these metals in the quantile g-computation models, despite observing individual effects. This finding could be explained by the modest correlation among blood metal concentrations and limited likelihood of observing simultaneous, or joint, changes in these exposures. Although these toxic metals and Mn have multiple shared sources of exposure (e.g., drinking water and air), they also have independent, or at least more prominent, sources of exposure. These include Pb-based paints,84 as well as resorption from bone stores during pregnancy85 for Pb, cigarette smoke for Cd,86 shellfish87 and rice88 consumption for As, and predatory fish consumption66 and dental amalgams89 for Hg. Our results are similar to two recent cross-sectional analyses using quantile g-computation. Xu et al. did not find an association between a mixture of toxic metals and Mn on the prevalence of hypertension (prevalence ratio=0.96; 95% CI: 0.73, 1.27) among adults involved in cleanup activities for the Deepwater Horizon oil spill.90 Similarly, in another study, Xu et al. did not find an association between a mixture of toxic metals and Mn measured in ambient air and the prevalence of hypertension (prevalence ratio=1.02; 95% CI: 0.99, 1.04) among 47,595 women enrolled in the Sister Study cohort across the United States.91
Pb and Preeclampsia
After adjusting for coexposure to other metals, each doubling of third trimester blood Pb concentration was associated with a 58% higher risk of developing preeclampsia. Pb is an established risk factor for preeclampsia,9 and our results extend this literature by providing evidence for the deleterious effects of Pb at concentrations comparable to, or even lower than, those typically experienced in the general population. For example, the median Pb concentration among MIREC participants was slightly lower than that observed among similarly aged women sampled around the same time in the Canadian Health Measures Survey (median blood Pb concentration=0.86μg/dL).92
In contrast, we did not observe associations with first trimester blood Pb concentrations with either condition. In this sample of women, average blood Pb concentrations declined by 10% between the first and third trimesters among women with normal blood pressure, which could be the result of plasma volume expansion,93 but concentrations increased by 6% among women with preeclampsia. This temporal pattern could be influenced by the lower plasma volume expansion that women with preeclampsia experience in pregnancy,94 which may be due to a loss of intravascular water volume into interstitial areas.95 This temporal pattern is also similar to findings from an analysis of repeated blood Pb samples during pregnancy by Sowers et al.96 These authors showed that, relative to women without preeclampsia, early pregnancy blood Pb levels were similar among women who would eventually present with preeclampsia but that blood Pb levels were consistently higher among these women at every time point afterward, even after adjusting for calcium intake.96 Another possible explanation is the resorption of Pb from bone during pregnancy. Ninety percent of Pb is stored in bone97 and bone serves as a novel source of maternal Pb exposure during pregnancy,85 especially during the third trimester.98 It is possible that women who developed preeclampsia had higher levels of bone but not blood Pb in early pregnancy. Bone Pb reflects decades of prior exposure,99 whereas blood Pb reflects only exposure within the past month.84 This hypothesized toxicokinetic pattern of Pb in women with preeclampsia may help explain why we did not observe an association between first trimester Pb and preeclampsia.
As and Hypertensive Disorders of Pregnancy
Most of the previous evidence of the association between As and gestational hypertension or preeclampsia is limited to small nonlongitudinal studies with exposure assessed late in pregnancy or at delivery. Evidence from these studies is equivocal. One case–control study (n=88 cases, 88 matched controls) in the Democratic Republic of the Congo found that women with preeclampsia had higher concentrations of urinary As,100 whereas a Mexican case–control study (n=104 cases, 202 unmatched controls) observed no such difference.101 A South African case–control study (n=23 cases, 43 unmatched controls) had mixed findings with slightly higher hair As concentrations in women with preeclampsia but much higher serum concentrations of As in controls.102 In a larger case–control study of pregnant women in China (n=427 cases, 427 matched controls), Wang et al. showed that As concentrations were associated with a higher risk of preeclampsia both individually and as a mixture.40 Using prospective data from a large sample of pregnant women, we showed that each doubling of As concentrations in early pregnancy was associated with >3 times the risk of developing gestational hypertension, as well as a 23% higher risk of developing preeclampsia. These results are in line with evidence from nonpregnant populations highlighting As as a risk factor for developing hypertension103 and other cardiovascular diseases.104 Moreover, our finding is consistent with evidence previously generated from this same cohort demonstrating associations between As, in multiple matrices and forms, and gestational diabetes51,105 or small-for-gestational age birth.106
Previous studies analyzing repeated measures of As exposure during pregnancy have provided equivocal evidence for a relevant window of susceptibility, with studies reporting that both earlier107 or later46,108 measures may be more relevant for pregnancy complications. It is plausible that early pregnancy may be the more relevant critical window of exposure for total blood As given that As methylation efficiency increases throughout pregnancy, which may reduce the toxicity of total blood As in later pregnancy.109 However, at least some of the toxic effect of total As is likely derived from methylated urinary metabolites.110,111 In our analysis, adjusting for urinary arsenobetaine attenuated, but did not nullify, the association between first trimester blood As and gestational hypertension, and we observed considerably stronger, although more imprecise, associations when restricting this analysis to women with <1μg/L urinary arsenobetaine concentrations. This suggests that the associations with blood As were not driven by fish consumption. We did not observe an association with either condition for DMA, which is consistent with a recent analysis of first trimester urinary As species and hypertensive disorders of pregnancy from a Chinese birth cohort.112 In both this study and that by Wang et al.,112 estimates of exposure could be imprecise given that As species were analyzed in a single spot urine sample for which the within-person variance has been shown to be larger than the between-person variance46 and reflects exposure to As over only a short interval.113 MMA, arsenate, and arsenite had low rates of detection in our study. The LODs in our study are similar to or lower than national-level surveys conducted around the same time as our study in Canada92 and the United States.114,115 Studies with more sensitive LODs116 (i.e., 0.1 vs. 0.75μg As/L, as in our study) have reported higher detection rates. The use of more sensitive laboratory methods would be beneficial in future studies to better characterize exposure to these As species.
Mn and Gestational Hypertension
Mn is an essential metal but can be toxic at high levels of exposure25; associations with health are thought to be characterized by an inverted U-shaped dose–response pattern. Several prospective studies have shown that infant birth weight is higher with increasing Mn blood levels up to inflection points ranging from 21 to 42μg/L,117–119 after which birth weight is then lower with increasing levels. The levels of Mn in the present study’s sample were within the range where beneficial effects on birth weight have been observed,81 but nearly all participants in the present study had levels below these inflection points.
Within the context of this exposure range, our results suggest that higher concentrations of Mn were associated with a lower risk of developing gestational hypertension. This is in line with evidence in nonpregnant adults, which shows that Mn (measured in urine,120 blood,121 toenails,122 estimated via diet recall,123 or through occupational exposure124) is associated with lower blood pressure. The likely mechanism for the association is through the antioxidant enzyme MnSOD, which scavenges reactive oxygen species, such as superoxide anions, that are associated with hypertension.125 However, we did not observe associations with preeclampsia, which is contrary to the existing literature. Using data from two birth cohorts, Liu et al. showed that higher levels of Mn in red blood cells (RBCs) in the first trimester and at delivery were protective against developing preeclampsia.23,24 Mn measured in RBCs is somewhat comparable to whole blood (about 66% of Mn in whole blood is bound to RBCs126), and because of the expected life span of RBCs (∼120 d127), these measures likely correspond to exposures around the first and third trimesters. These authors examined preeclampsia only, rather than separating preeclampsia from gestational hypertension, limiting a direct comparison with this work.
The low number of cases of preeclampsia in our study (n=43) may have contributed to null association between Mn and preeclampsia observed in our study. The number of cases of preeclampsia was higher in the analysis from the Boston Birth Cohort study (n=115),24 which likely provided more statistical power for the observed negative association (per SD change in Mn, PR=0.68; 95% CI: 0.54, 0.86). However, the number of cases in the analysis from Project Viva was similar to our study (n=48),23 and the authors observed a negative association (third vs. first tertile RR=0.50; 95% CI: 0.25, 0.99).
We found that the association between Mn and gestational hypertension was present among women carrying male fetuses only. Evidence for fetal sex-specific effects of maternal levels of Mn is limited and contradictory. Studies examining higher levels of Mn exposure have identified inverted U-shaped associations with risk of small-for-gestational age among male infants,119 as well as head circumference33 and low birth weight128 among female infants. These sex-specific differences could be explained by sexually dimorphic placental antioxidant defense mechanisms. One study of human placentas showed that MnSOD concentrations were markedly reduced in the placentas for male vs. female fetuses.52 Similar to much of this literature, our sex-specific findings are based on a modest number of cases and will require replication in future work.
Interaction between As and Mn
To our knowledge, this is the first time that an interaction between As and Mn has been observed for hypertensive disorders of pregnancy. Ultimately, this finding will need to be replicated in separate prospective pregnancy cohorts, but it is plausible based on epidemiological and experimental literature. Authors of a Bangladeshi birth cohort have observed interactions between prenatal exposure to As and Mn in regard to child neurodevelopment.30 However, the evidence for interaction in studies examining birth outcomes is equivocal, with some studies observing interactions33,34 and others not observing interactions.35,36 One possible explanation for this interaction is the potential effect of As on MnSOD. In a study of nonpregnant Taiwanese adults,29 higher urinary As concentrations were associated with higher odds of developing hypertension, and this association was strongest among individuals with a single nucleotide polymorphism that reduces MnSOD activity. Our finding is supported by experimental work in rodents demonstrating that exposure to As can reduce MnSOD activity31,32 and that this effect can be attenuated with coexposure to Mn.32 In addition, Biswas et al. also found that serum concentrations of the liver enzymes aspartate aminotransferase, alanine aminotransferase, and alkaline phosphatase were elevated among mice exposed to As but not elevated among mice that were coexposed to both As and Mn.32 Elevations in these liver enzymes are one component of the definitions of both preeclampsia and Hemolysis, Elevated Liver enzymes and Low Platelets (HELLP) syndrome.
Cd and Hypertensive Disorders of Pregnancy
We did not observe associations between Cd and gestational hypertension or preeclampsia in our study. The epidemiological evidence for Cd is limited to a few nonlongitudinal studies that have demonstrated mixed results. Liu et al. identified an association between Cd measured in RBCs and preeclampsia, although the association was small and crossed the null (prevalence ratio=1.15; 95% CI: 0.98, 1.36).24 In a case–control study among South African women, those with or without preeclampsia had similar Cd concentrations measured in both serum (0.10 vs. 0.05mg/L, respectively) and hair (3.75 vs. 3.96μg/g, respectively).102 Two case–control studies of women with higher exposure to Cd found that those with preeclampsia had higher concentrations of Cd in urine (geometric mean=1.78 vs. 0.53μg/L),100 as well as in maternal blood (median=1.21 vs. 1.09μg/L)129 or placental tissue (median=4.28 vs. 3.61μg/kg).129 Neither of these studies adjusted for relevant confounders in their analysis, although one study matched participants for maternal age, gestational age, and parity.100 A larger case–control study observed a protective association between blood Cd and preeclampsia [high vs. low tertiles odds ratio (OR)=0.64; 95% CI: 0.45, 0.91] with a monotonic dose–response pattern across tertiles that persisted after adjusting for several other toxic metals.40 Finally, a nested case–control study observed a deleterious association between placental Cd and preeclampsia (OR=1.50; 95% CI: 1.10, 2.20).130 These authors found that the association for Cd was stronger in women with lower placental selenium (OR=2.0; 95% CI: 1.1, 3.5) and null in women with higher placental selenium (OR=0.98; 95% CI: 0.5, 1.9).
Our results are generally consistent with the studies with similarly low blood concentrations of Cd.24,102 Further research is needed to elucidate associations between Cd and hypertensive disorders of pregnancy. It may be helpful for this research to consider potential interactions with trace metals other than Mn, including selenium and zinc, which may be implicated in the development of hypertensive disorders of pregnancy.131
Hg and Hypertensive Disorders of Pregnancy
We also did not observe associations between total blood Hg and gestational hypertension or preeclampsia in our study. A previous analysis from this cohort found no association between the presence or replacement of dental amalgams, which is a source of inorganic Hg exposure, and gestational hypertension.132 In the Boston Birth Cohort, Liu et al. found no association between preeclampsia and Hg measured in RBCs,24 which is more reflective of exposure to methylmercury.133 Moreover, our findings are generally consistent with meta-analyses of the published literature demonstrating null associations between Hg and cardiovascular diseases, including hypertension, at the lower levels typically observed in the general population.134,135 However, a separate meta-analysis by Hu et al.135 supports an association between Hg and hypertension at higher levels of exposure (blood Hg >4.88μg/L). In their case–control study, Wang et al. found that women in the highest tertile of blood Hg concentrations (≥1.89μg/L, or 2–3 times higher than in our study) had higher odds of having preeclampsia.40 Similarly, a prospective occupational study of pregnant dental staff with high exposure to Hg through dental amalgams and dental administrators without such exposure found that the dental staff had higher urinary concentrations of Hg and higher odds of developing preeclampsia (RR=3.67; 95% CI: 1.25, 10.78).18 The dental staff also had lower blood concentrations of two antioxidant enzymes, including SOD. The null associations for Hg observed in our study are consistent with previous analyses of similarly low levels of exposure.
Interpretations of Time Windows of Exposure When Using Whole Blood as a Matrix
Whole blood is a widely used matrix for biomonitoring of metals,136 but this likely reflects different time windows of exposure for different metals. For Pb, Cd, Hg, and Mn, the concentrations reported in our paper provide reasonable estimates of trimester-specific exposures. For Pb, whole blood is the most commonly used matrix for biomonitoring84 and represents exposure in the previous 30 d.99 For Cd, whole blood is considered the most valid marker of recent exposure,86 with a half-life of 3–4 months.137 Following long-term, low-dose exposure, blood Cd can serve as a good reflection of Cd body burden.138,139 Whole blood is commonly used to measure total Hg, with a half-life ranging from 44140 to 80 d.141,142 Although there is no accepted matrix for measuring Mn, whole blood Mn is thought to be a good indicator of environmental exposure,143 with a half-life ranging from 2 to 5 wk.144–146 However, urine is the preferred matrix for measuring total As, although with a half-life of 4 d.147 Whole blood provides a measure of As exposure from several hours prior to collection,148 which may not reflect longer term exposure.113 With continuous exposure, such as through drinking water, blood As concentrations may reach a steady state.147,149 Use of total blood As is a limitation of our study and likely contributes to nondifferential exposure misclassification, reducing the precision of the observed associations. Analyses from several other pregnancy cohorts have identified deleterious associations for perinatal health or birth outcomes with total As in whole blood,34,35,40,150 although others have not.36 Given that As has a relatively low rate of bioaccumulation, even in urine151 when compared with Pb, Cd, or Hg, future studies should employ repeated measures to better characterize chronic exposures. Future studies should ideally include high-quality speciation with sensitive methods to disentangle inorganic and organic As to elucidate sources of exposure that will facilitate translating findings into public health action.
Studies of chemical mixtures, including metals, have typically used a single matrix to evaluate exposure,39 which can be imperfect if the chemicals have different pharmacokinetic properties. Using repeated, paired blood and urine samples from Project Viva, Ashrap et al. recently showed that measuring metals in either blood or urine are equally good approaches for assessing metals as a mixture in relation to the risk of preterm birth.152
Strengths and Limitations
This study has several strengths. We examined exposure to toxic metals and Mn at two time points during pregnancy using data from a large prospective pregnancy cohort. This allowed us to establish temporality, to compare time windows of exposure, and to control our analyses for several potential static and time-varying confounders. As part of this investigation, we used a novel statistical approach to estimate the joint effect of a mixture of metals. In contrast to a similar method, weighted-quantile sum regression, quantile g-computation does not require that the exposures have the same direction of association (directional homogeneity), does not require sample splitting into training/validation sets, and is computationally more efficient. In addition, we examined associations with urinary As species alongside total blood As concentrations. Finally, the richness of the exposure information in the MIREC cohort allowed us to leverage trimester-specific air pollution exposure data to account for potential confounding of PM2.5, which has not typically been done in studies examining blood metals and perinatal health.
This study has some limitations. First, concentrations of toxic metals were similar, or slightly lower, in this sample of women compared with nationally representative Canadian data from similarly aged women collected around the same time as this cohort (e.g., median Pb=0.86μg/dL, Cd=0.28μg/L, Hg=0.69μg/L; median urinary DMA=6.7μg/L).92 These results should be interpreted within the context of these low exposure levels. For instance, the health effects of As and Mn may be more apparent in populations with lower exposure to Pb,153 as is the case with our study. There is limited evidence examining associations in contemporary populations with low levels of exposure, and our work contributes to filling this gap in the literature. Second, in this study, we examined total blood Hg rather than the potentially unique contribution of inorganic and methylmercury; however, because most of the Hg in humans is in the methyl form,154 total blood Hg should reflect mostly methyl Hg exposure. Wells et al. showed that total blood Hg at delivery was not associated with maternal blood pressure at delivery155 but, rather, that methyl and inorganic Hg were associated with higher, and lower, blood pressures, respectively. Our results for total blood Hg are in line with those from Wells et al.,155 but a direct comparison with our results is difficult because it is not uncommon for blood pressure to increase transiently during early labor.156,157 Third, there is growing evidence to suggest that essential trace metals, such as calcium,158 selenium,131 copper,131,159 and zinc,159 may be implicated in blood pressure changes or development of hypertension/preeclampsia in pregnancy. We did not measure these essential trace metals in this study. Fourth, there may have been random error in the measurement of blood pressure, but this would be expected to be nondifferential because participants and study staff were not aware of the participants’ blood metal concentrations. Fifth, change in paternal partner is another potential risk factor for preeclampsia160,161 that could confound associations. We did not collect information on paternal partners from prior pregnancies. Sixth, although women with chronic kidney disease were excluded from enrollment, it is possible that alteration in kidney function could confound the association between first trimester blood As concentrations and gestational hypertension, if this were differential. Finally, we likely had greater statistical power to detect associations with gestational hypertension than we did for preeclampsia because of the larger numbers of cases, which is expected based on the prevalence of these conditions. Owing to the relatively small number of women with preeclampsia in our cohort (n=43), we were unable to estimate associations separately for women with early vs. late onset preeclampsia, which should be prioritized in future work.
Conclusion
Our results confirm that even low blood Pb concentrations are a risk factor for preeclampsia. This work contributes to a growing body of evidence supporting the deleterious role of As on the risk of developing gestational hypertension and preeclampsia, as well as the protective role of Mn for gestational hypertension. We observed an interaction between As and Mn in early pregnancy such that the deleterious association with higher As concentrations was stronger at lower concentrations of Mn. This paper highlights the importance of considering individual, independent, and mixture analyses to examine multiple chemical associations and reinforces the value of incorporating multiple chemicals and multiple measures of exposure throughout pregnancy.
Supplementary Material
Click here for additional data file.
Click here for additional data file.
Acknowledgments
We are grateful to the Maternal–Infant Research on Environmental Chemicals (MIREC) study families for their participation and to the dedicated site and coordinating center staff for recruiting the participants, as well as for collecting and managing the data and biospecimens. We thank R. Hocking (Health Canada) for conducting a literature search. We acknowledge that L.R. Camara conducted an analysis of metals and hypertensive disorders in pregnancy using the MIREC data for her PhD dissertation; however, the work presented here is an entirely new analysis and does not include any of the work conducted by L.R. Camara.
The MIREC study was funded by Health Canada’s Chemicals Management Plan (to T.E.A.), the Canadian Institute of Health Research (grant MOP-81285), and the Ontario Ministry of the Environment.
==== Refs
References
1. Lo JO, Mission JF, Caughey AB. 2013. Hypertensive disease of pregnancy and maternal mortality. Curr Opin Obstet Gynecol 25 (2 ):124–132, PMID: , 10.1097/GCO.0b013e32835e0ef5.23403779
2. Nakimuli A, Nakubulwa S, Kakaire O, Osinde MO, Mbalinda SN, Kakande N, et al. 2016. The burden of maternal morbidity and mortality attributable to hypertensive disorders in pregnancy: a prospective cohort study from Uganda. BMC Pregnancy Childbirth 16 (1 ):205, PMID: , 10.1186/s12884-016-1001-1.27492552
3. Jelin AC, Cheng YC, Shaffer BL, Kaimal AJ, Little SE, Caughey AB. 2009. Early-onset preeclampsia and neonatal outcomes. J Matern Fetal Neonatal Med 23 (5 ):389–392, 10.3109/14767050903168416.
4. van Esch JJA, van Heijst AF, de Haan AFJ, van der Heijden OWH. 2017. Early-onset preeclampsia is associated with perinatal mortality and severe neonatal morbidity. J Matern Fetal Neonatal Med 30 (23 ):2789–2794, 10.1080/14767058.2016.1263295.28282780
5. Brown MC, Best KE, Pearce MS, Waugh J, Robson SC, Bell R. 2013. Cardiovascular disease risk in women with pre-eclampsia: systematic review and meta-analysis. Eur J Epidemiol 28 (1 ):1–19, PMID: , 10.1007/s10654-013-9762-6.23397514
6. Benschop L, Duvekot JJ, van Lennep JER. 2019. Future risk of cardiovascular disease risk factors and events in women after a hypertensive disorder of pregnancy. Heart 105 (16 ):1273–1278, PMID: , 10.1136/heartjnl-2018-313453.31175138
7. Ying W, Catov JM, Ouyang P. 2018. Hypertensive disorders of pregnancy and future maternal cardiovascular risk. J Am Heart Assoc 7 (17 ):e009382, PMID: , 10.1161/JAHA.118.009382.30371154
8. Kahn LG, Trasande L. 2018. Environmental toxicant exposure and hypertensive disorders of pregnancy: recent findings. Curr Hypertens Rep 20 (10 ):87, PMID: , 10.1007/s11906-018-0888-5.30090982
9. Poropat AE, Laidlaw MAS, Lanphear B, Ball A, Mielke HW. 2018. Blood lead and preeclampsia: a meta-analysis and review of implications. Environ Res 160 :12–19, PMID: , 10.1016/j.envres.2017.09.014.28938191
10. Pollack AZ, Ranasinghe S, Sjaarda LA, Mumford SL. 2014. Cadmium and reproductive health in women: a systematic review of the epidemiologic evidence. Curr Environ Health Rep 1 (2 ):172–184, PMID: , 10.1007/s40572-014-0013-0.27453808
11. Stone J, Sutrave P, Gascoigne E, Givens MB, Fry RC, Manuck TA. 2021. Exposure to toxic metals and per- and polyfluoroalkyl substances and the risk of preeclampsia and preterm birth in the United States: a review. Am J Obstet Gynecol MFM 3 (3 ):100308, PMID: , 10.1016/j.ajogmf.2021.100308.33444805
12. Padula AM, Ma C, Huang H, Morello-Frosch R, Woodruff TJ, Carmichael SL. 2021. Drinking water contaminants in California and hypertensive disorders in pregnancy. Environ Epidemiol 5 (2 ):e149, PMID: , 10.1097/EE9.0000000000000149.33870020
13. da Cunha Martins A Jr, Carneiro MFH, Grotto D, Adeyemi JA, Barbosa F Jr. 2018. Arsenic, cadmium, and mercury-induced hypertension: mechanisms and epidemiological findings. J Toxicol Environ Health B Crit Rev 21 (2 ):61–82, PMID: , 10.1080/10937404.2018.1432025.29446707
14. Andrade VM, Mateus ML, Batoréu MC, Aschner M, Marreilha dos Santos AP. 2015. Lead, arsenic, and manganese metal mixture exposures: focus on biomarkers of effect. Biol Trace Elem Res 166 (1 ):13–23, PMID: , 10.1007/s12011-015-0267-x.25693681
15. Vaziri ND. 2008. Mechanisms of lead-induced hypertension and cardiovascular disease. Am J Physiol Heart Circ Physiol 295 (2 ):H454–H465, PMID: , 10.1152/ajpheart.00158.2008.18567711
16. Lopes ACBA, Peixe TS, Mesas AE, Paoliello MMB. 2016. Lead exposure and oxidative stress: a systematic review. Rev Environ Contam Toxicol 236 :193–238, PMID: , 10.1007/978-3-319-20013-2_3.26423075
17. Samir AM, Aref WM. 2011. Impact of occupational exposure to elemental mercury on some antioxidative enzymes among dental staff. Toxicol Ind Health 27 (9 ):779–786, PMID: , 10.1177/0748233710397420.21427135
18. El-Badry A, Rezk M, El-Sayed H. 2018. Mercury-induced oxidative stress may adversely affect pregnancy outcome among dental staff: a cohort study. Int J Occup Environ Med 9 (3 ):113–119, PMID: , 10.15171/ijoem.2018.1181.29995016
19. Houston MC. 2011. Role of mercury toxicity in hypertension, cardiovascular disease, and stroke. J Clin Hypertens (Greenwich) 13 (8 ):621–627, PMID: , 10.1111/j.1751-7176.2011.00489.x.21806773
20. Jomova K, Jenisova Z, Feszterova M, Baros S, Liska J, Hudecova D, et al. 2011. Arsenic: toxicity, oxidative stress and human disease. J Appl Toxicol 31 (2 ):95–107, PMID: , 10.1002/jat.1649.21321970
21. Moitra S, Brashier BB, Sahu S. 2014. Occupational cadmium exposure-associated oxidative stress and erythrocyte fragility among jewelry workers in India. Am J Ind Med 57 (9 ):1064–1072, PMID: , 10.1002/ajim.22336.24812026
22. Mistry HD, Williams PJ. 2011. The importance of antioxidant micronutrients in pregnancy. Oxid Med Cell Longev 2011 :841749, PMID: , 10.1155/2011/841749.21918714
23. Liu T, Hivert MF, Rifas-Shiman SL, Rahman ML, Oken E, Cardenas A, et al. 2020. Prospective association between manganese in early pregnancy and the risk of preeclampsia. Epidemiology 31 (5 ):677–680, PMID: , 10.1097/EDE.0000000000001227.32618710
24. Liu T, Zhang M, Guallar E, Wang G, Hong X, Wang X, et al. 2019. Trace minerals, heavy metals, and preeclampsia: findings from the Boston Birth Cohort. J Am Heart Assoc 8 (16 ):e012436, PMID: , 10.1161/JAHA.119.012436.31426704
25. Li L, Yang X. 2018. The essential element manganese, oxidative stress, and metabolic diseases: links and interactions. Oxid Med Cell Longev 2018 :7580707, PMID: , 10.1155/2018/7580707.29849912
26. Matsubara K, Higaki T, Matsubara Y, Nawa A. 2015. Nitric oxide and reactive oxygen species in the pathogenesis of preeclampsia. Int J Mol Sci 16 (3 ):4600–4614, PMID: , 10.3390/ijms16034600.25739077
27. Madazli R, Benian A, Aydin S, Uzun H, Tolun N. 2002. The plasma and placental levels of malondialdehyde, glutathione and superoxide dismutase in pre-eclampsia. J Obstet Gynaecol 22 (5 ):477–480, PMID: , 10.1080/0144361021000003573.12521411
28. Vanderlelie J, Venardos K, Clifton VL, Gude NM, Clarke FM, Perkins AV. 2005. Increased biological oxidation and reduced anti-oxidant enzyme activity in pre-eclamptic placentae. Placenta 26 (1 ):53–58, PMID: , 10.1016/j.placenta.2004.04.002.15664411
29. Chen SC, Chen CC, Kuo CY, Huang CH, Lin CH, Lu ZY, et al. 2012. Elevated risk of hypertension induced by arsenic exposure in Taiwanese rural residents: possible effects of manganese superoxide dismutase (MnSOD) and 8-oxoguanine DNA glycosylase (OGG1) genes. Arch Toxicol 86 (6 ):869–878, PMID: , 10.1007/s00204-011-0797-8.22193621
30. Valeri L, Mazumdar MM, Bobb JF, Claus Henn B, Rodrigues E, Sharif OIA, et al. 2017. The joint effect of prenatal exposure to metal mixtures on neurodevelopmental outcomes at 20–40 months of age: evidence from rural Bangladesh. Environ Health Perspect 125 (6 ):067015, PMID: , 10.1289/EHP614.28669934
31. Prakash C, Soni M, Kumar V. 2015. Biochemical and molecular alterations following arsenic-induced oxidative stress and mitochondrial dysfunction in rat brain. Biol Trace Elem Res 167 (1 ):121–129, PMID: , 10.1007/s12011-015-0284-9.25764338
32. Biswas S, Anjum A, Banna HUI, Rahman M, Siddique AE, Karim Y, et al. 2019. Manganese attenuates the effects of arsenic on neurobehavioral and biochemical changes in mice co-exposed to arsenic and manganese. Environ Sci Pollut Res Int 26 (28 ):29257–29266, PMID: , 10.1007/s11356-019-06112-y.31396869
33. Signes-Pastor AJ, Doherty BT, Romano ME, Gleason KM, Gui J, Baker E, et al. 2019. Prenatal exposure to metal mixture and sex-specific birth outcomes in the New Hampshire Birth Cohort Study. Environ Epidemiol 3 (5 ):e068, PMID: , 10.1097/EE9.0000000000000068.31844832
34. Rahman ML, Oken E, Hivert MF, Rifas-Shiman S, Lin PID, Colicino E, et al. 2021. Early pregnancy exposure to metal mixture and birth outcomes—a prospective study in Project Viva. Environ Int 156 :106714, PMID: , 10.1016/j.envint.2021.106714.34147999
35. Claus Henn B, Ettinger AS, Hopkins MR, Jim R, Amarasiriwardena C, Christiani DC, et al. 2016. Prenatal arsenic exposure and birth outcomes among a population residing near a mining-related Superfund site. Environ Health Perspect 124 (8 ):1308–1315, PMID: , 10.1289/ehp.1510070.26859631
36. Ashrap P, Watkins DJ, Mukherjee B, Boss J, Richards MJ, Rosario Z, et al. 2020. Maternal blood metal and metalloid concentrations in association with birth outcomes in northern Puerto Rico. Environ Int 138 :105606, PMID: , 10.1016/j.envint.2020.105606.32179314
37. Bours MJL. 2021. Tutorial: a nontechnical explanation of the counterfactual definition of effect modification and interaction. J Clin Epidemiol 134 :113–124, PMID: , 10.1016/j.jclinepi.2021.01.022.33548464
38. Braun JM, Gennings C, Hauser R, Webster TF. 2016. What can epidemiological studies tell us about the impact of chemical mixtures on human health? Environ Health Perspect 124 (1 ):A6–A9, PMID: , 10.1289/ehp.1510569.26720830
39. Yim G, Wang Y, Howe CG, Romano ME. 2022. Exposure to metal mixtures in association with cardiovascular risk factors and outcomes: a scoping review. Toxics 10 (3 ):116, PMID: , 10.3390/toxics10030116.35324741
40. Wang Y, Wang K, Han T, Zhang P, Chen X, Wu W, et al. 2020. Exposure to multiple metals and prevalence for preeclampsia in Taiyuan, China. Environ Int 145 :106098, PMID: , 10.1016/j.envint.2020.106098.32916414
41. Bommarito PA, Kim SS, Meeker JD, Fry RC, Cantonwine DE, McElrath TF, et al. 2019. Urinary trace metals, maternal circulating angiogenic biomarkers, and preeclampsia: a single-contaminant and mixture-based approach. Environ Health 18 (1 ):63, PMID: , 10.1186/s12940-019-0503-5.31300062
42. Weisskopf MG, Seals RM, Webster TF. 2018. Bias amplification in epidemiologic analysis of exposure to mixtures. Environ Health Perspect 126 (4 ):047003, PMID: , 10.1289/EHP2450.29624292
43. Kippler M, Wagatsuma Y, Rahman A, Nermell B, Persson LÅ, Raqib R, et al. 2012. Environmental exposure to arsenic and cadmium during pregnancy and fetal size: a longitudinal study in rural Bangladesh. Reprod Toxicol 34 (4 ):504–511, PMID: , 10.1016/j.reprotox.2012.08.002.22985739
44. Kippler M, Tofail F, Gardner R, Rahman A, Hamadani JD, Bottai M, et al. 2012. Maternal cadmium exposure during pregnancy and size at birth: a prospective cohort study. Environ Health Perspect 120 (2 ):284–289, PMID: , 10.1289/ehp.1103711.21862444
45. Taylor CM, Golding J, Emond AM. 2016. Moderate prenatal cadmium exposure and adverse birth outcomes: a role for sex-specific differences? Paediatr Perinat Epidemiol 30 (6 ):603–611, PMID: , 10.1111/ppe.12318.27778365
46. Liu H, Lu S, Zhang B, Xia W, Liu W, Peng Y, et al. 2018. Maternal arsenic exposure and birth outcomes: a birth cohort study in Wuhan, China. Environ Pollut 236 :817–823, PMID: , 10.1016/j.envpol.2018.02.012.29462776
47. Romano ME, Enquobahrie DA, Simpson C, Checkoway H, Williams MA. 2016. Maternal body burden of cadmium and offspring size at birth. Environ Res 147 :461–468, PMID: , 10.1016/j.envres.2016.02.029.26970900
48. Röllin HB, Kootbodien T, Channa K, Odland JØ. 2015. Prenatal exposure to cadmium, placental permeability and birth outcomes in coastal populations of South Africa. PLoS One 10 (11 ):e0142455, PMID: , 10.1371/journal.pone.0142455.26544567
49. Wang J, Gao ZY, Yan J, Ying XL, Tong SL, Yan CH. 2017. Sex differences in the effects of prenatal lead exposure on birth outcomes. Environ Pollut 225 :193–200, PMID: , 10.1016/j.envpol.2017.03.031.28371734
50. Shih YH, Chen HY, Christensen K, Handler A, Turyk ME, Argos M. 2021. Prenatal exposure to multiple metals and birth outcomes: an observational study within the National Children’s Study cohort. Environ Int 147 :106373, PMID: , 10.1016/j.envint.2020.106373.33422966
51. Ashley-Martin J, Dodds L, Arbuckle TE, Bouchard MF, Shapiro GD, Fisher M, et al. 2018. Association between maternal urinary speciated arsenic concentrations and gestational diabetes in a cohort of Canadian women. Environ Int 121 (pt 1 ):714–720, PMID: , 10.1016/j.envint.2018.10.008.30321846
52. Evans L, Myatt L. 2017. Sexual dimorphism in the effect of maternal obesity on antioxidant defense mechanisms in the human placenta. Placenta 51 :64–69, PMID: , 10.1016/j.placenta.2017.02.004.28292470
53. Arbuckle TE, Fraser WD, Fisher M, Davis K, Liang CL, Lupien N, et al. 2013. Cohort profile: the maternal-infant research on environmental chemicals research platform. Paediatr Perinat Epidemiol 27 (4 ):415–425, PMID: , 10.1111/ppe.12061.23772943
54. Ettinger AS, Arbuckle TE, Fisher M, Liang CL, Davis K, Cirtiu CM, et al. 2017. Arsenic levels among pregnant women and newborns in Canada: results from the Maternal-Infant Research on Environmental Chemicals (MIREC) cohort. Environ Res 153 :8–16, PMID: , 10.1016/j.envres.2016.11.008.27880879
55. Arbuckle TE, Liang CL, Morisset AS, Fisher M, Weiler H, Cirtiu CM, et al. 2016. Maternal and fetal exposure to cadmium, lead, manganese and mercury: the MIREC study. Chemosphere 163 :270–282, PMID: , 10.1016/j.chemosphere.2016.08.023.27540762
56. Belanger P, Dumas P. 2010. Fast determination of urine arsenic species by ion exchange chromatography and HPLC-ICP-MS using collision-reactive interface. At Spectrosc 31 (6 ):175–181.
57. Meeker JD, Hu H, Cantonwine DE, Lamadrid-Figueroa H, Calafat AM, Ettinger AS, et al. 2009. Urinary phthalate metabolites in relation to preterm birth in Mexico City. Environ Health Perspect 117 (10 ):1587–1592, PMID: , 10.1289/ehp.0800522.20019910
58. Magee LA, Pels A, Helewa M, Rey E, von Dadelszen P, Canadian Hypertensive Disorders of Pregnancy Working Group. 2014. Diagnosis, evaluation, and management of the hypertensive disorders of pregnancy. Pregnancy Hypertens 4 (2 ):105–145, PMID: , 10.1016/j.preghy.2014.01.003.26104418
59. Kampa M, Castanas E. 2008. Human health effects of air pollution. Environ Pollut 151 (2 ):362–367, PMID: , 10.1016/j.envpol.2007.06.012.17646040
60. Bai W, Li Y, Niu Y, Ding Y, Yu X, Zhu B, et al. 2020. Association between ambient air pollution and pregnancy complications: a systematic review and meta-analysis of cohort studies. Environ Res 185 :109471, PMID: , 10.1016/j.envres.2020.109471.32276169
61. Sun M, Yan W, Fang K, Chen D, Liu J, Chen Y, et al. 2020. The correlation between PM2.5 exposure and hypertensive disorders in pregnancy: a meta-analysis. Sci Total Environ 703 :134985, PMID: , 10.1016/j.scitotenv.2019.134985.31731170
62. van Donkelaar A, Martin RV, Li C, Burnett RT. 2019. Regional estimates of chemical composition of fine particulate matter using a combined geoscience-statistical method with information from satellites, models, and monitors. Environ Sci Technol 53 (5 ):2595–2611, PMID: , 10.1021/acs.est.8b06392.30698001
63. Johnson M, Shin HH, Roberts E, Sun L, Fisher M, Hystad P, et al. 2022. Critical time windows for air pollution exposure and birth weight in a multicity Canadian pregnancy cohort. Epidemiology 33 (1 ):7–16, PMID: , 10.1097/EDE.0000000000001428.34669628
64. Shen M, Smith GN, Rodger M, White RR, Walker MC, Wen SW. 2017. Comparison of risk factors and outcomes of gestational hypertension and pre-eclampsia. PLoS One 12 (4 ):e0175914, PMID: , 10.1371/journal.pone.0175914.28437461
65. Shrier I, Platt RW. 2008. Reducing bias through directed acyclic graphs. BMC Med Res Methodol 8 (1 ):70, PMID: , 10.1186/1471-2288-8-70.18973665
66. Health Canada. 2007. Human Health Risk Assessment of Mercury in Fish and Health Benefits of Fish Consumption. https://www.canada.ca/en/health-canada/services/food-nutrition/reports-publications/human-health-risk-assessment-mercury-fish-health-benefits-fish-consumption.html [accessed 1 March 2021].
67. Health Canada. 2019. Mercury in Fish. Last modified 27 November 2019. https://www.canada.ca/en/health-canada/services/food-nutrition/food-safety/chemical-contaminants/environmental-contaminants/mercury/mercury-fish.html [accessed 1 March 2021].
68. FDA (U.S. Food and Drug Administration). 2020. Technical Information on Development of FDA/EPA Advice about Eating Fish for Those Who Might Become or Are Pregnant or Breastfeeding Children 1-11 Years. https://www.fda.gov/food/metals-and-your-food/technical-information-development-fdaepa-advice-about-eating-fish-women-who-are-or-might-become [accessed 1 March 2021].
69. Zou G. 2004. A modified Poisson regression approach to prospective studies with binary data. Am J Epidemiol 159 (7 ):702–706, PMID: , 10.1093/aje/kwh090.15033648
70. Hornung RW, Reed LD. 1990. Estimation of average concentration in the presence of nondetectable values. Appl Occup Environ Hyg 5 (1 ):46–51, 10.1080/1047322X.1990.10389587.
71. Glass DC, Gray CN. 2001. Estimating mean exposures from censored data: exposure to benzene in the Australian petroleum industry. Ann Occup Hyg 45 (4 ):275–282, 10.1016/S0003-4878(01)00022-9.11378148
72. Rubin DB. 1987. Multiple Imputation for Nonresponse in Surveys. New York, NY: John Wiley & Sons, Inc.
73. White IR, Royston P, Wood AM. 2011. Multiple imputation using chained equations: issues and guidance for practice. Stat Med 30 (4 ):377–399, PMID: , 10.1002/sim.4067.21225900
74. Rothman KJ. 2012. Measuring interactions. In: Epidemiology, an Introduction. 2nd ed. New York, NY: Oxford University Press, 198–210.
75. VanderWeele TJ, Knol MJ. 2014. A tutorial on interaction. Epidemiol Methods 3 (1 ):33–72, 10.1515/em-2013-0005.
76. Knol MJ, VanderWeele TJ, Groenwold RHH, Klungel OH, Rovers MM, Grobbee DE. 2011. Estimating measures of interaction on an additive scale for preventive exposures. Eur J Epidemiol 26 (6 ):433–438, PMID: , 10.1007/s10654-011-9554-9.21344323
77. Arbuckle TE, Liang CL, Fisher M, Caron NJ, Fraser WD, MIREC Study Group. 2018. Exposure to tobacco smoke and validation of smoking status during pregnancy in the MIREC study. J Expo Sci Environ Epidemiol 28 (5 ):461–469, PMID: , 10.1038/s41370-017-0011-z.29296002
78. Howe CG, Farzan SF, Garcia E, Jursa T, Iyer R, Berhane K, et al. 2020. Arsenic and birth outcomes in a predominately lower income Hispanic pregnancy cohort in Los Angeles. Environ Res 184 :109294, PMID: , 10.1016/j.envres.2020.109294.32145549
79. Navas-Acien A, Francesconi KA, Silbergeld EK, Guallar E. 2011. Seafood intake and urine concentrations of total arsenic, dimethylarsinate and arsenobetaine in the US population. Environ Res 111 (1 ):110–118, PMID: , 10.1016/j.envres.2010.10.009.21093857
80. Navas-Acien A, Silbergeld EK, Pastor-Barriuso R, Guallar E. 2008. Arsenic exposure and prevalence of type 2 diabetes in US adults. JAMA 300 (7 ):814–822, PMID: , 10.1001/jama.300.7.814.18714061
81. Ashley-Martin J, Dodds L, Arbuckle TE, Ettinger AS, Shapiro GD, Fisher M, et al. 2018. Maternal and cord blood manganese (Mn) levels and birth weight: the MIREC birth cohort study. Int J Hyg Environ Health 221 (6 ):876–882, PMID: , 10.1016/j.ijheh.2018.05.015.29886104
82. Keil AP, Buckley JP, O’Brien KM, Ferguson KK, Zhao S, White AJ. 2020. A quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect 3 (4 ):047004, PMID: , 10.1289/EHP5838.32255670
83. WHO (World Health Organization). 2000. Obesity: Preventing and Managing the Global Epidemic. Report of a WHO Consultation. WHO Technical Report Series No. 894. https://apps.who.int/iris/bitstream/handle/10665/42330/WHO_TRS_894.pdf? sequence=1&isAllowed=y [accessed 10 November 2022].
84. Health Canada. 2013. Final Human Health State of the Science Report on Lead. https://www.canada.ca/content/dam/hc-sc/migration/hc-sc/ewh-semt/alt_formats/pdf/pubs/contaminants/dhhssrl-rpecscepsh/dhhssrl-rpecscepsh-eng.pdf [accessed 2 October 2020].
85. Gulson B, Mizon K, Korsch M, Taylor A. 2016. Revisiting mobilisation of skeletal lead during pregnancy based on monthly sampling and cord/maternal blood lead relationships confirm placental transfer of lead. Arch Toxicol 90 (4 ):805–816, PMID: , 10.1007/s00204-015-1515-8.25877328
86. Järup L, Åkesson A. 2009. Current status of cadmium as an environmental health problem. Toxicol Appl Pharmacol 238 (3 ):201–208, PMID: , 10.1016/j.taap.2009.04.020.19409405
87. Taylor V, Goodale B, Raab A, Schwerdtle T, Reimer K, Conklin S, et al. 2017. Human exposure to organic arsenic species from seafood. Sci Total Environ 580 :266–282, PMID: , 10.1016/j.scitotenv.2016.12.113.28024743
88. Davis MA, Signes-Pastor AJ, Argos M, Slaughter F, Pendergrast C, Punshon T, et al. 2017. Assessment of human dietary exposure to arsenic through rice. Sci Total Environ 586 :1237–1244, PMID: , 10.1016/j.scitotenv.2017.02.119.28233618
89. Golding J, Steer CD, Gregory S, Lowery T, Hibbeln JR, Taylor CM. 2016. Dental associations with blood mercury in pregnant women. Community Dent Oral Epidemiol 44 (3 ):216–222, PMID: , 10.1111/cdoe.12208.26688340
90. Xu J, Engel LS, Rhoden J, Jackson WB II, Kwok RK, Sandler DP. 2021. The association between blood metals and hypertension in the GuLF study. Environ Res 202 :111734, PMID: , 10.1016/j.envres.2021.111734.34303682
91. Xu J, White AJ, Niehoff NM, O’Brien KM, Sandler DP. 2020. Airborne metals exposure and risk of hypertension in the Sister Study. Environ Res 191 :110144, PMID: , 10.1016/j.envres.2020.110144.32898563
92. Health Canada. 2013. Second Report on Human Biomonitoring of Environmental Chemicals in Canada. Results of the Canadian Health Measures Survey Cycle 2 (2009–2011). https://www.canada.ca/content/dam/hc-sc/migration/hc-sc/ewh-semt/alt_formats/pdf/pubs/contaminants/chms-ecms-cycle2/chms-ecms-cycle2-eng.pdf [accessed 3 May 2019].
93. Aguree S, Gernand AD. 2019. Plasma volume expansion across healthy pregnancy: a systematic review and meta-analysis of longitudinal studies. BMC Pregnancy Childbirth 19 (1 ):508, PMID: , 10.1186/s12884-019-2619-6.31856759
94. de Haas S, Ghossein-Doha C, van Kuijk SMJ, van Drongelen J, Spaanderman MEA. 2017. Physiological adaptation of maternal plasma volume during pregnancy: a systematic review and meta-analysis. Ultrasound Obstet Gynecol 49 (2 ):177–187, PMID: , 10.1002/uog.17360.28169502
95. Heilmann L, Siekmann U, Schmid-Schönbein H, Ludwig H. 1981. Hemoconcentration and pre-eclampsia. Arch Gynecol 231 (1 ):7–21, PMID: , 10.1007/BF02110019.7332361
96. Sowers M, Jannausch M, Scholl T, Li W, Kemp FW, Bogden JD. 2002. Blood lead concentrations and pregnancy outcomes. Arch Environ Health 57 (5 ):489–495, PMID: , 10.1080/00039890209601442.12641194
97. Barry PS. 1975. A comparison of concentrations of lead in human tissues. Br J Ind Med 32 (2 ):119–139, PMID: , 10.1136/oem.32.2.119.1131339
98. Manton WI, Angle CR, Stanek KL, Kuntzelman D, Reese YR, Kuehnemann TJ. 2003. Release of lead from bone in pregnancy and lactation. Environ Res 92 (2 ):139–151, PMID: , 10.1016/s0013-9351(03)00020-3.12854694
99. Rabinowitz MB. 1991. Toxicokinetics of bone lead. Environ Health Perspect 91 (2 ):33–37, PMID: , 10.1289/ehp.919133.2040248
100. Elongi Moyene JP, Scheers H, Tandu-Umba B, Haufroid V, Buassa-Bu-Tsumbu B, Verdonck F, et al. 2016. Preeclampsia and toxic metals: a case–control study in Kinshasa, DR Congo. Environ Health 15 :48, PMID: , 10.1186/s12940-016-0132-1.27044488
101. Sandoval-Carrillo A, Méndez-Hernández EM, Antuna-Salcido EI, Salas-Pacheco SM, Vázquez-Alaniz F, Téllez-Valencia A, et al. 2016. Arsenic exposure and risk of preeclampsia in a Mexican mestizo population. BMC Pregnancy Childbirth 16 (1 ):153, PMID: , 10.1186/s12884-016-0946-4.27401918
102. Maduray K, Moodley J, Soobramoney C, Moodley R, Naicker T. 2017. Elemental analysis of serum and hair from pre-eclamptic South African women. J Trace Elem Med Biol 43 :180–186, PMID: , 10.1016/j.jtemb.2017.03.004.28325649
103. Abhyankar LN, Jones MR, Guallar E, Navas-Acien A. 2012. Arsenic exposure and hypertension: a systematic review. Environ Health Perspect 120 (4 ):494–500, PMID: , 10.1289/ehp.1103988.22138666
104. Moon K, Guallar E, Navas-Acien A. 2012. Arsenic exposure and cardiovascular disease: an updated systematic review. Curr Atheroscler Rep 14 (6 ):542–555, PMID: , 10.1007/s11883-012-0280-x.22968315
105. Shapiro GD, Dodds L, Arbuckle TE, Ashley-Martin J, Fraser W, Fisher M, et al. 2015. Exposure to phthalates, bisphenol A and metals in pregnancy and the association with impaired glucose tolerance and gestational diabetes mellitus: the MIREC study. Environ Int 83 :63–71, PMID: , 10.1016/j.envint.2015.05.016.26101084
106. Thomas S, Arbuckle TE, Fisher M, Fraser WD, Ettinger A, King W. 2015. Metals exposure and risk of small-for-gestational age birth in a Canadian birth cohort: the MIREC study. Environ Res 140 :430–439, PMID: , 10.1016/j.envres.2015.04.018.25967284
107. Xia X, Liang C, Sheng J, Yan S, Huang K, Li Z, et al. 2018. Association between serum arsenic levels and gestational diabetes mellitus: a population-based birth cohort study. Environ Pollut 235 :850–856, PMID: , 10.1016/j.envpol.2018.01.016.29348076
108. Mullin AM, Amarasiriwardena C, Cantoral-Preciado A, Claus Henn B, Leon Hsu HH, Sanders AP, et al. 2019. Maternal blood arsenic levels and associations with birth weight-for-gestational age. Environ Res 177 :108603, PMID: , 10.1016/j.envres.2019.108603.31357156
109. Gardner RM, Nermell B, Kippler M, Grandér M, Li L, Ekström EC, et al. 2011. Arsenic methylation efficiency increases during the first trimester of pregnancy independent of folate status. Reprod Toxicol 31 (2 ):210–218, PMID: , 10.1016/j.reprotox.2010.11.002.21078382
110. Watanabe T, Hirano S. 2013. Metabolism of arsenic and its toxicological relevance. Arch Toxicol 87 (6 ):969–979, PMID: , 10.1007/s00204-012-0904-5.22811022
111. Thomas DJ. 2021. Arsenic methylation—lessons from three decades of research. Toxicology 457 :152800, PMID: , 10.1016/j.tox.2021.152800.33901604
112. Wang X, Wu Y, Sun X, Guo Q, Xia W, Wu Y, et al. 2021. Arsenic exposure and metabolism in relation to blood pressure changes in pregnant women. Ecotoxicol Environ Saf 222 :112527, PMID: , 10.1016/j.ecoenv.2021.112527.34311426
113. ATSDR (Agency for Toxic Substances and Disease Registry). 2007. Toxicological Profile for Arsenic. https://wwwn.cdc.gov/TSP/ToxProfiles/ToxProfiles.aspx?id=22&tid=3 [accessed 1 June 2021].
114. CDC (Centers for Disease Control and Prevention). 2015. Fourth National Report on Human Exposure to Environmental Chemicals. https://www.cdc.gov/biomonitoring/pdf/FourthReport_UpdatedTables_Feb2015.pdf [accessed 22 April 2022].
115. Jones MR, Tellez-Plaza M, Vaidya D, Grau M, Francesconi KA, Goessler W, et al. 2016. Estimation of inorganic arsenic exposure in populations with frequent seafood intake: evidence from MESA and NHANES. Am J Epidemiol 184 (8 ):590–602, PMID: , 10.1093/aje/kww097.27702745
116. Scheer J, Findenig S, Goessler W, Francesconi KA, Howard B, Umans JG, et al. 2012. Arsenic species and selected metals in human urine: validation of HPLC/ICPMS and ICPMS procedures for a long-term population-based epidemiological study. Anal Methods 4 (2 ):406–413, PMID: , 10.1039/C2AY05638K.22685491
117. Zota AR, Ettinger AS, Bouchard M, Amarasiriwardena CJ, Schwartz J, Hu H, et al. 2009. Maternal blood manganese levels and infant birth weight. Epidemiology 20 (3 ):367–373, PMID: , 10.1097/EDE.0b013e31819b93c0.19289966
118. Chen L, Ding G, Gao Y, Wang P, Shi R, Huang H, et al. 2014. Manganese concentrations in maternal–infant blood and birth weight. Environ Sci Pollut Res Int 21 (9 ):6170–6175, PMID: , 10.1007/s11356-013-2465-4.24477335
119. Yamamoto M, Sakurai K, Eguchi A, Yamazaki S, Nakayama SF, Isobe T, et al. 2019. Association between blood manganese level during pregnancy and birth size: the Japan Environment and Children’s Study (JECS). Environ Res 172 :117–126, PMID: , 10.1016/j.envres.2019.02.007.30782531
120. Wu C, Woo JG, Zhang N. 2017. Association between urinary manganese and blood pressure: results from National Health and Nutrition Examination Survey (NHANES), 2011–2014. PLoS One 12 (11 ):e0188145, PMID: , 10.1371/journal.pone.0188145.29141052
121. Bulka CM, Scannell Bryan M, Persky VW, Daviglus ML, Durazo-Arvizu RA, Parvez F, et al. 2019. Changes in blood pressure associated with lead, manganese, and selenium in a Bangladeshi cohort. Environ Pollut 248 :28–35, PMID: , 10.1016/j.envpol.2019.01.129.30771745
122. Mordukhovich I, Wright RO, Hu H, Amarasiriwardena C, Baccarelli A, Litonjua A, et al. 2012. Associations of toenail arsenic, cadmium, mercury, manganese, and lead with blood pressure in the Normative Aging Study. Environ Health Perspect 120 (1 ):98–104, PMID: , 10.1289/ehp.1002805.21878420
123. Lee YK, Lyu ES, Oh SY, Park HR, Ro HK, Heo YR, et al. 2015. Daily copper and manganese intakes and their relation to blood pressure in normotensive adults. Clin Nutr Res 4 (4 ):259–266, PMID: , 10.7762/cnr.2015.4.4.259.26566521
124. Jiang Y, Zheng W. 2005. Cardiovascular toxicities upon manganese exposure. Cardiovasc Toxicol 5 (4 ):345–354, PMID: , 10.1385/ct:5:4:345.16382172
125. Schlichte SL, Pekas EJ, Bruett TJ, Kosmacek EA, Hackfort BT, Rasmussen JM, et al. 2021. Sympathoinhibition and vasodilation contribute to the acute hypotensive response of the superoxide dismutase mimic, MnTnBuOE-2-PyP5+, in hypertensive animals. Adv Redox Res 3 :100016, 10.1016/j.arres.2021.100016.
126. Milne DB, Sims RL, Ralston NV. 1990. Manganese content of the cellular components of blood. Clin Chem 36 (3 ):450–452, PMID: , 10.1093/clinchem/36.3.450.2311212
127. Arias CF, Arias CF. 2017. How do red blood cells know when to die? R Soc Open Sci 4 (4 ):160850, PMID: , 10.1098/rsos.160850.28484605
128. Xia W, Zhou Y, Zheng T, Zhang B, Bassig BA, Li Y, et al. 2016. Maternal urinary manganese and risk of low birth weight: a case–control study. BMC Public Health 16 (1 ):142, PMID: , 10.1186/s12889-016-2816-4.26869268
129. Wang F, Fan F, Wang L, Ye W, Zhang Q, Xie S. 2018. Maternal cadmium levels during pregnancy and the relationship with preeclampsia and fetal biometric parameters. Biol Trace Elem Res 186 (2 ):322–329, PMID: , 10.1007/s12011-018-1312-3.29651732
130. Laine JE, Ray P, Bodnar W, Cable PH, Boggess K, Offenbacher S, et al. 2015. Placental cadmium levels are associated with increased preeclampsia risk. PLoS One 10 (9 ):e0139341, PMID: , 10.1371/journal.pone.0139341.26422011
131. Liu T, Zhang M, Rahman ML, Wang X, Hinkle SN, Zhang C, et al. 2021. Exposure to heavy metals and trace minerals in first trimester and maternal blood pressure change over gestation. Environ Int 153 :106508, PMID: , 10.1016/j.envint.2021.106508.33901931
132. Louopou RC, Trottier H, Arbuckle TE, Fraser WD. 2020. Dental amalgams and risk of gestational hypertension in the MIREC study. Pregnancy Hypertens 21 :84–89, PMID: , 10.1016/j.preghy.2020.04.015.32447273
133. Berglund M, Lind B, Björnberg KA, Palm B, Einarsson Ö, Vahter M. 2005. Inter-individual variations of human mercury exposure biomarkers: a cross-sectional assessment. Environ Health 4 :20, PMID: , 10.1186/1476-069X-4-20.16202128
134. Chowdhury R, Ramond A, O’Keeffe LM, Shahzad S, Kunutsor SK, Muka T, et al. 2018. Environmental toxic metal contaminants and risk of cardiovascular disease: systematic review and meta-analysis. BMJ 362 :k3310, PMID: , 10.1136/bmj.k3310.30158148
135. Hu XF, Singh K, Chan HM. 2018. Mercury exposure, blood pressure, and hypertension: a systematic review and dose–response meta-analysis. Environ Health Perspect 126 (7 ):076002, PMID: , 10.1289/EHP2863.30073953
136. Santonen T, Aitio A, Fowler BA, Nordberg M. 2015. Biological monitoring and biomarkers. In: Handbook on the Toxicology of Metals, vol 1 . Nordberg GF, Fowler BA, Nordberg M, eds. 4th ed. San Diego, CA: Academic Press, 155–171.
137. Järup L, Rogenfelt A, Elinder CG, Nogawa K, Kjellström T. 1983. Biological half-time of cadmium in the blood of workers after cessation of exposure. Scand J Work Environ Health 9 (4 ):327–331, PMID: , 10.5271/sjweh.2404.6635611
138. Järup L, Berglund M, Elinder CG, Nordberg G, Vahter M. 1998. Health effects of cadmium exposure—a review of the literature and a risk estimate. Scand J Work Environ Health 24 (suppl. 1 ):1–51, PMID: .9569444
139. Nordberg GF, Nogawa K, Nordberg M. 2015. Cadmium. In: Handbook on the Toxicology of Metals, vol 2 . Nordberg GF, Fowler BA, Nordberg M, eds. 4th ed. San Diego, CA: Academic Press, 667–716.
140. Smith JC, Allen PV, Turner MD, Most B, Fisher HL, Hall LL. 1994. The kinetics of intravenously administered methyl mercury in man. Toxicol Appl Pharmacol 128 (2 ):251–256, PMID: , 10.1006/taap.1994.1204.7940540
141. Jo S, Woo HD, Kwon HJ, Oh SY, Park JD, Hong YS, et al. 2015. Estimation of the biological half-life of methylmercury using a population toxicokinetic model. Int J Environ Res Public Health 12 (8 ):9054–9067, PMID: , 10.3390/ijerph120809054.26264017
142. Miettinen J, Rahola T, Hattula T, Rissanen K, Tillander M. 1971. Elimination of 203Hg-methylmercury in man. Ann Clin Res 3 (2 ):116–122, PMID: .4997252
143. ATSDR. 2012. Toxicological Profile for Manganese. https://wwwn.cdc.gov/TSP/ToxProfiles/ToxProfiles.aspx?id=102&tid=23 [accessed 21 April 2022].
144. Davidsson L, Cederblad A, Lönnerdal B, Sandström B. 1989. Manganese retention in man: a method for estimating manganese absorption in man. Am J Clin Nutr 49 (1 ):170–179, PMID: , 10.1093/ajcn/49.1.170.2912001
145. Johnson PE, Lykken GI, Korynta ED. 1991. Absorption and biological half-life in humans of intrinsic and extrinsic 54Mn tracers from foods of plant origin. J Nutr 121 (5 ):711–717, PMID: , 10.1093/jn/121.5.711.2019880
146. Lucchini RG, Aschner M, Kim Y, Šarić M. 2015. Manganese. In: Handbook on the Toxicology of Metals, vol 2 . Nordberg GF, Fowler BA, Nordberg M, eds. 4th ed. San Diego, CA: Academic Press, 975–1011.
147. National Research Council. 1999. Arsenic in Drinking Water. Washington DC: National Academies Press.
148. Pomroy C, Charbonneau SM, McCullough RS, Tam GKH. 1980. Human retention studies with 74As. Toxicol Appl Pharmacol 53 (3 ):550–556, PMID: , 10.1016/0041-008x(80)90368-3.7385250
149. Hall M, Chen Y, Ahsan H, Slavkovich V, van Geen A, Parvez F, et al. 2006. Blood arsenic as a biomarker of arsenic exposure: results from a prospective study. Toxicology 225 (2–3 ):225–233, PMID: , 10.1016/j.tox.2006.06.010.16860454
150. Luo Y, McCullough LE, Tzeng JY, Darrah T, Vengosh A, Maguire RL, et al. 2017. Maternal blood cadmium, lead and arsenic levels, nutrient combinations, and offspring birthweight. BMC Public Health 17 (1 ):354, PMID: , 10.1186/s12889-017-4225-8.28438148
151. Hughes MF. 2006. Biomarkers of exposure: a case study with inorganic arsenic. Environ Health Perspect 114 (11 ):1790–1796, PMID: , 10.1289/ehp.9058.17107869
152. Ashrap P, Watkins DJ, Mukherjee B, Rosario-Pabón Z, Vélez-Vega CM, Alshawabkeh A, et al. 2021. Performance of urine, blood, and integrated metal biomarkers in relation to birth outcomes in a mixture setting. Environ Res 200 :111435, PMID: , 10.1016/j.envres.2021.111435.34097892
153. Rodrigues EG, Bellinger DC, Valeri L, Hasan MOSI, Quamruzzaman Q, Golam M, et al. 2016. Neurodevelopmental outcomes among 2- to 3-year-old children in Bangladesh with elevated blood lead and exposure to arsenic and manganese in drinking water. Environ Health 15 (1 ):44, PMID: , 10.1186/s12940-016-0127-y.26968381
154. Hong YS, Kim YM, Lee KE. 2012. Methylmercury exposure and health effects. J Prev Med Public Health 45 (6 ):353–363, PMID: , 10.3961/jpmph.2012.45.6.353.23230465
155. Wells EM, Herbstman JB, Lin YH, Hibbeln JR, Halden RU, Witter FR, et al. 2017. Methyl mercury, but not inorganic mercury, associated with higher blood pressure during pregnancy. Environ Res 154 :247–252, PMID: , 10.1016/j.envres.2017.01.013.28110211
156. Cohen J, Vaiman D, Sibai BM, Haddad B. 2015. Blood pressure changes during the first stage of labor and for the prediction of early postpartum preeclampsia: a prospective study. Eur J Obstet Gynecol Reprod Biol 184 :103–107, PMID: , 10.1016/j.ejogrb.2014.11.020.25483991
157. Ohno Y, Terauchi M, Tamakoshi K, Shiozaki A, Saito S. 2016. The risk factors for labor onset hypertension. Hypertens Res 39 (4 ):260–265, PMID: , 10.1038/hr.2015.112.26490090
158. Sun X, Li H, He X, Li M, Yan P, Xun Y, et al. 2019. The association between calcium supplement and preeclampsia and gestational hypertension: a systematic review and meta-analysis of randomized trials. Hypertens Pregnancy 38 (2 ):129–139, PMID: , 10.1080/10641955.2019.1593445.30935246
159. Lewandowska M, Sajdak S, Marciniak W, Lubiński J. 2019. First trimester serum copper or zinc levels, and risk of pregnancy-induced hypertension. Nutrients 11 (10 ):2479, PMID: , 10.3390/nu11102479.31623110
160. Wikström AK, Gunnarsdóttir J, Cnattingius S. 2012. The paternal role in pre-eclampsia and giving birth to a small for gestational age infant; a population-based cohort study. BMJ Open 2 (4 ):e001178, PMID: , 10.1136/bmjopen-2012-001178.22936817
161. Trupin LS, Simon LP, Eskenazi B. 1996. Change in paternity: a risk factor for preeclampsia in multiparas. Epidemiology 7 (3 ):240–244, PMID: .8728435
|
PMC010xxxxxx/PMC10124596.txt
|
==== Front
Environ Health Perspect
Environ Health Perspect
EHP
Environmental Health Perspectives
0091-6765
1552-9924
Environmental Health Perspectives
37093218
EHP12868
10.1289/EHP12868
Science Selection
Failure to Connect: Experimental Evidence that Benzo(a)pyrene Impedes Embryonic Implantation
Schmidt Charles W.
21 4 2023
4 2023
131 4 04400208 2 2023
08 3 2023
https://ehp.niehs.nih.gov/about-ehp/license EHP is an open-access journal published with support from the National Institute of Environmental Health Sciences, National Institutes of Health. All content is public domain unless otherwise noted.
Small, bumpy, oblong object nestled on a gently folded surface
==== Body
pmcGlobally, about 1 in 10 people will experience a miscarriage at some point during their childbearing years.1 These traumatic events are associated with risk factors such as infections, fetal chromosomal abnormalities, maternal age, and certain environmental exposures.2 But their causes are often unclear, and that is especially true for recurrent miscarriage, which is defined as two or more pregnancy losses experienced by the same person.3 About half of all recurrent miscarriages have unknown origins,4 and a better understanding of how they occur is an important priority for maternal and child health researchers.1 Now, experimental research in Environmental Health Perspectives sheds light on how one ubiquitous pollutant, benzo(a)pyrene [B(a)P], might elevate risk for recurrent miscarriage.5
A by-product of incomplete combustion, B(a)P is a polyaromatic hydrocarbon (PAH) found in tobacco and wood smoke, car exhaust, and grilled foods, among other sources.6 According to the new study,5 maternal exposure to B(a)P increased the expression of certain genes and proteins that may interfere with the attachment of a very early stage embryo—called a blastocyst—to the uterine wall. The research was led by Huidong Zhang, a professor at the Eighth Affiliated Hospital, Sun Yat-sen University, in Shenzhen, China.
The new study found that BPDE, a metabolite of benzo(a)pyrene, interfered with processes involved in embryonic implantation in mice. In people, such interference could lead to miscarriage or other pregnancy complications, according to the authors. This scanning electron micrograph shows an 8-day-old human embryo implanted in the uterus. Note: BPDE, benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide. Image: © Lennart Nilsson/Science Photo Library.
Small, bumpy, oblong object nestled on a gently folded surface
The current experimental work builds on evidence from epidemiological and animal studies associating BaP with higher miscarriage risk.7–9 During the new study, Zhang and colleagues focused specifically on developmental effects of the ultimate metabolite of B(a)P: benzo(a)pyrene-7,8-dihydrodiol-9,10-epoxide (BPDE).5 Previous research by this team and others showed that BPDE inhibits the normal biology of cells called trophoblasts, which form the outer layer of the blastocyst and give rise to the placenta.10,11 For the developing embryo to establish a blood supply, trophoblasts must first migrate to and then invade the endometrial lining of the uterus.12,13 But BPDE-exposed trophoblasts have exhibited a diminished capacity for migration and invasion,14 “and that increases the risk of miscarriage,” says Kelly Bakulski, an assistant professor of epidemiology at the University of Michigan’s School of Public Health in Ann Arbor, who was not involved in the study.
Zhang and colleagues focused specifically on how BPDE disrupts the activity of long noncoding RNAs (lncRNAs),5 which have key roles in trophoblast functioning. In an earlier cell study, the team identified 22 lncRNAs that are more active in human trophoblasts after BPDE exposure.11 But how lncRNAs might be involved in BPDE’s inhibition of trophoblast migration and invasion was not clear.
The team used a novel lncRNA, called lnc-HZ09, that they identified in this study.5 For the first set of experiments, they engineered two sets of human trophoblast cell lines, one in which lnc-HZ09 was overexpressed and another in which it was silenced. Results showed that the cells that overexpressed lnc-HZ09 were less able to move normally through a gel matrix, “suggesting that trophoblasts with high levels of lnc-HZ09 may have similar difficulties migrating and invading into uterine tissues,” Bakulski says. Furthermore, these cells contained abnormally low levels of three proteins suspected of playing key roles in trophoblast migration and invasion: PLD1, RAC1, and CDC42. The cells in which lnc-HZ09 was silenced showed “significantly greater migration and invasion,” the authors wrote.
Next, the researchers analyzed placental tissues collected from two groups of Chinese women: 15 who had experienced unexplained recurrent miscarriages, and a control group of 15 who had undergone abortions. The results supported the role of lnc-HZ09: Compared with tissues from the control group, samples collected from women who miscarried had higher lnc-HZ09 gene expression levels, as well as higher levels of PBDE-DNA adducts, the sites where the chemical attached to DNA.
For a final experiment, the team treated pregnant mice with B(a)P at doses of 0.05 or 2.0mg/kg body weight to induce miscarriage. The researchers found that in the animals receiving the higher dose, levels of PDL1, RAC1, and CDC42—the three proteins observed in the human trophoblast cell experiment—were lower in placental tissues.
According to the authors, the results suggest that lncHZ09 suppressed trophoblast cell migration and invasion through its effects on the PLD1/RAC1/CDC42 pathway. “We believe that the combination of these experiments suggests that, in humans, B(a)P/PBDE exposure may induce miscarriage,” Zhang says.
Ping-Kun Zhou, president of the Chinese Society of Toxicology and a professor at the Beijing Institute of Radiation Medicine, agrees this is an interesting finding. “The molecular mechanism might be further verified using primary trophoblast cells from villous [placental] tissues from individuals experiencing a miscarriage,” says Zhou, who was not involved in the study. In the meantime, Zhou adds, pregnant women should reduce their exposure to PAHs as much as possible by, for example, avoiding smoking, supporting adoption of clean energy, eating steamed rather than fried food, and ensuring their gas stove vents properly.
Charles W. Schmidt, MS, is an award-winning journalist in Portland, ME, whose work has appeared in Scientific American, Nature, Science, Discover Magazine, Undark, The Washington Post, and many other publications.
==== Refs
References
1. Ha S, Ghimire S, Martinez V. 2022. Outdoor air pollution and pregnancy loss: a review of recent literature. Curr Epidemiol Rep 9 (4 ):387–405, 10.1007/s40471-022-00304-w.
2. Quenby S, Gallos ID, Dhillon-Smith RK, Podesek M, Stephenson MD, Fisher J, et al. 2021. Miscarriage matters: the epidemiological, physical, psychological, and economic costs of early pregnancy loss. Lancet 397 (10285 ):1658–1667, PMID: , 10.1016/S0140-6736(21)00682-6.33915094
3. American Society for Reproductive Medicine. 2016. What is recurrent pregnancy loss (RPL)? https://www.reproductivefacts.org/news-and-publications/patient-fact-sheets-and-booklets/documents/fact-sheets-and-info-booklets/what-is-recurrent-pregnancy-loss-rpl/ [accessed 12 April 2023].
4. ESHRE Capri Workshop Group. 2008. Genetic aspects of female reproduction. Hum Reprod Update 14 (4 ):293–307, PMID: , 10.1093/humupd/dmn009.18385259
5. Dai M, Huang W, Huang X, Ma C, Wang R, Tian P, et al. 2023. BPDE, the migration and invasion of human trophoblast cells, and occurrence of miscarriage in humans: roles of a novel lncRNA-HZ09. Environ Health Perspect 131 (1 ):17009, PMID: , 10.1289/EHP10477.36719213
6. IARC (International Agency for Research on Cancer). 2012. Chemical agents and related occupations. IARC Monogr Eval Carcinog Risks Hum. 100 (pt F ):9–562, PMID: .23189753
7. Neal MS, Zhu J, Holloway AC, Foster WG. 2007. Follicle growth is inhibited by benzo-[a]-pyrene, at concentrations representative of human exposure, in an isolated rat follicle culture assay. Hum Reprod 22 (4 ):961–967, PMID: , 10.1093/humrep/del487.17218370
8. Ptashekas J, Ciuniene E, Barkiene M, Zurlyte I, Jonauskas G, Sliachtic N, et al. 1996. Environmental and health monitoring in Lithuanian cities: exposure to heavy metals and benz(a)pyrene in Vilnius and Siauliai residents. J Environ Pathol Toxicol Oncol 15 (2–4 ):135–141, PMID: .9216796
9. Zhao Y, Chen X, Liu X, Ding Y, Gao R, Qiu Y, et al. 2014. Exposure of mice to benzo(a)pyrene impairs endometrial receptivity and reduces the number of implantation sites during early pregnancy. Food Chem Toxicol 69 :244–251, PMID: , 10.1016/j.fct.2014.04.021.24769007
10. Xu Z, Tian P, Guo J, Mi C, Liang T, Xie J, et al. 2021. Lnc-HZ01 with m6A RNA methylation inhibits human trophoblast cell proliferation and induces miscarriage by up-regulating BPDE-activated lnc-HZ01/MXD1 positive feedback loop. Sci Total Environ 776 :145950, PMID: , 10.1016/j.scitotenv.2021.145950.33647641
11. Liang T, Xie J, Zhao J, Huang W, Xu Z, Tian P, et al. 2021. Novel lnc-HZ03 and miR-hz03 promote BPDE-induced human trophoblastic cell apoptosis and induce miscarriage by upregulating p53/SAT1 pathway. Cell Biol Toxicol 37 (6 ):951–970, PMID: , 10.1007/s10565-021-09583-3.33566220
12. Kliman HJ. 1998. From Trophoblast to Human Placenta. In: The Encyclopedia of Reproduction, Vol. 4. Knobil E, Neill JD, eds. 1st ed. United Kingdom: Elsevier Science, Vol. 4. https://medicine.yale.edu/obgyn/kliman/placenta/research/trophoblast%20to%20placenta%20eor_163163_284_18220_v1.pdf [accessed 12 April 2023].
13. Knöfler M, Haider S, Saleh L, Pollheimer J, Gamage TKJB, James J, et al. 2019. Human placenta and trophoblast development: key molecular mechanisms and model systems. Cell Mol Life Sci 76 (18 ):3479–3496, PMID: , 10.1007/s00018-019-03104-6.31049600
14. Ye Y, Jiang S, Zhang C, Cheng Y, Zhong H, Du T, et al. 2020. Environmental pollutant benzo[a]pyrene induces recurrent pregnancy loss through promoting apoptosis and suppressing migration of extravillous trophoblast. Biomed Res Int 2020 :8983494, PMID: , 10.1155/2020/8983494.33123590
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.